UNIVERSITY OF MALAGA

PhD THESIS

Photovoltaic Lifetime Forecast: Models for long-term

photovoltaic degradation prediction and forecast

Ismail Kaaya

Málaga, February, 2020

 PhD THESIS

Photovoltaic Lifetime Forecast: Models for long-term

photovoltaic degradation prediction and forecast

Author:
Ismail Kaaya

Supervisors:
Mariano Sidrach de Cardona Ortín
Karl-Anders Weiss

PhD program:
Mechanical Engineering and Energy Efficiency

PhD thesis presented in the SCHOOL OF INDUSTRIAL ENGINEERING

of MALAGA UNIVERSITY to obtain the Doctor Degree

Director:
Mariano Sidrach de Cardona Ortín

Málaga, February, 2020

Dr. Mariano Sidrach de Cardona Ortín, Professor in the Applied Physics department
at the University of Málaga and Dr. Karl-Anders Weiss head of service life analysis in
the Fraunhofer Institut of Solar Energy as Directors of the Doctoral Thesis: Photovoltaic
Lifetime Forecast: Models for long-term photovoltaic degradation prediction and forecast,
presented by Mr. Ismail Kaaya in the School of Industrial Engineering of the University
of Málaga to obtain the Doctor Degree.
They state that this thesis is supported by the following research articles:

Paper 1. I. Kaaya, M. Koehl, A.P. Mehilli, S. de Cardona Mariano, & K.A. Weiss, ”Mod-
eling Outdoor Service Lifetime Prediction of PV Modules: Effects of Combined
Climatic Stressors on PV Module Power Degradation”. IEEE Journal of Photo-
voltaics,vol. 9, no. 4, pp. 1105–1112, May. 2019.
This paper is reprinted from the Open Access publication available at https:
//ieeexplore.ieee.org/document/8720037. All rights remain with the au-
thors and editors.
Impact factor: 3.398

Paper 2. S. Lindig, I. Kaaya, K.A. Weiss, D. Moser, and M. Topic, “Review of statistical
and analytical degradation models for photovoltaic modules and systems as well
as related improvements,” IEEE J. Photovoltaic., vol. 8, no. 6, pp. 1773–1786,
Nov. 2018.
This paper is reprinted from the Open Access publication available at https://
ieeexplore.ieee.org/document/8472886. All rights remain with the authors
and editors.
Impact factor: 3.398

Paper 3. J. A-Vásqueza, I. Kaaya, K. Brecla, K-A. Weiss and M.Topic, ”Global climate
data processing and mapping of degradation mechanisms and degradation rates
of PV modules”. energies MDPI ,vol. 12, no. 4749 Dec 2019.
This paper is reprinted from the Open Access publication available at https://
www.mdpi.com/1996-1073/12/24/4749/htm. All rights remain with the authors
and editors. Impact factor: 2.707

Málaga, April, 2020

Signed:

Mariano Sidrach de Cardona Ortín Karl-Anders Weiss

Acknowledgements
The author has carried out a big part of the research reported in this thesis at Fraunhofer
Institute of Solar Energy (ISE), Freiburg, Germany, in the research group Service Life Anal-
ysis and Material Characterization. This research has received funding from the European
Unions Horizon 2020 research and innovation program under the Marie Sklodowska-Curie
grant agreement No 721452-Project Solar-Train.

Firstly, I would like to thank both my research supervisors: Dr. Mariano Sidrach de
Ortin of the University of Malaga and Dr. Karl-Anders Weiss of Fraunhofer ISE, for their
insight and mentorship, and for creating a collaborative and congenial working atmosphere
for me during the process of my research.

Secondly, I would like to acknowledge the different sources of data used in this research
as: Fraunhofer Instute of solar energy for sharing the module data from Negev, Gran-
Canaria and Zugspitze as well as indoor test data, the Scuola universitaria professionale
della Svizzera italiana (SUPSI) for sharing the TISO plant data, the Institute for Re-
newable Energy, EURAC Research for sharing the Bolzano system data and the Desert
Knowledge Australia Solar Centre (DKA Solar Centre) for the continuous effort to provide
free access photovoltaic systems data. These data sources are properly referenced in the
notes and bibliography.

Special thanks go to Micheal Koehl of Fraunhofer ISE, Lindig Sascha and David Moser
of EURAC Research, Julián Ascencio-Vásquez of University of Ljubljana and Alessandro
Virtuani of École Polytechnique Fédérale de Lausanne (EPFL) for their fruitful discus-
sions and collaboration during different stages of my research. I cannot forget to thank my
wife Sophie Marchand of Fraunhfer ISE for her continuous support and encouragements
throughout the entire research period. I could not have done it smoothly without you. I
also thank my daddy Mr.Nsubuga Med and my siblings for their support in my entire aca-
demic career. Lastly I extend my gratitude to the entire Solar-Train project team for the
wonderful discussions, training and informal meetings. It has been grateful to work with
and learn from everyone, indeed, it has been a time of tremendous learning and growth for
me both scientifically and personally.
ii
Abstract

As the share of photovoltaic keeps increasing in the global electricity mix, it becomes
critical to assess how the overall system and module performance (power) decreases over
time. This is not only important for financial reasons but also technically, because it is
crucial to understand the effects of the local climate on the performance degradation. Al-
though different models have been proposed to quantify the impact of climatic stresses
on performance degradation based on indoor ageing tests, less has been done to quantify
these effects in outdoor operations. The available methods for outdoor application are
mainly data-driven, meaning that the performance losses are evaluated from monitored
performance data without an understanding of the influencing environmental variables.
Moreover, these models are suitable for performance loss rates and not for degradation
rates evaluations.

Therefore, in the first part of this research, a physical model to determine the degra-
dation rates of photovoltaic modules based on outdoor climatic variables is proposed.
Through it, the impact of combined climatic loads on the module’s maximum power out-
put is quantified. In this approach, three degradation precursor mechanisms, namely,
hydrolysis, photodegradation, and thermo-mechanical, are assumed to be necessary for
service lifetime prediction. For each mechanism, an empirical model that describe well the
physical/chemical kinetics is selected or proposed. To validate the selected or proposed
models, experimental data from accelerated ageing tests are used. A generalized model
to quantify the effects of combined climatic loads for outdoor applications is then derived
from the three models. The generalized model is calibrated and validated using outdoor
experimental data of three identical mono-crystalline silicon modules installed in three
benchmarking climates: maritime (Gran Canaria, Spain), arid (Negev, Israel), and alpine
(Zugspitze, Germany). Finally, using the public climate database (ERA5), climatic data is
processed to extract the climatic stresses necessary for the calculation of the degradation
rate. These stresses are then applied to evaluate the degradation rates based on the three
precursor mechanisms and also to evaluate the total degradation rates. Therefore, global
degradation risk maps based on specific precursor mechanisms as well as total degradation
rate are provided. We believe that these risk maps are useful to understand the dominating
degradation mechanisms according to geographical locations and hence could be used to
develop photovoltaic materials depending on the operating geographical locations.

Other fundamental challenge of the available methods is their accuracy when long-term
forecasts are needed after a short operation time and with limited data points. The second
part of this research, addresses this challenge where a new data-driven method is proposed
so as to improve the accuracy of long-term prediction with small degradation history. The
model depends on the degradation patterns and a new concept of time dependent degra-
dation rate is introduced. The model has been calibrated and validated using different
iv Abstract

photovoltaic modules and systems data with 5 to 35 years of field exposure. The new
model is benchmarked against existing data-driven methods. The proposed model lowered
the long-term forecast uncertainties when forecasts are made after a small performance
degradation. Through this, the effects of long-term degradation to lifetime yield predic-
tion are assessed. It has been shown that using the proposed approach, the lifetime yield
predictions are more reliable due to more accurate long-term degradation forecast.

Finally, the two approaches are combined to form a hybrid model based on both the
physical and data-driven methods. Indeed, the hybrid modelis aimed to provide more
reliable long-term degradation forecast as well as having a physical understanding of the
dominating degradation mechanisms influencing the performance degradation. We believe
such a model is useful to calculate more reliable levelized cost of energy and thus the eco-
nomic viability of solar energy as well as to improve the development of new PV materials
according to the operating climatic conditions.
Contents

Abstract iii

1 Introduction 1

2 Literature review 5
2.1 Photoltaic degradation and degradation modes . . . . . . . . . . . . . . . . . 5
2.2 Degradation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 A review of the available physical models . . . . . . . . . . . . . . . . . . . . 7
2.3.1 Degradation models for corrosion . . . . . . . . . . . . . . . . . . . . . 8
2.3.2 Models for potential induced degradation . . . . . . . . . . . . . . . . 14
2.3.3 Models for ultraviolet irradiance degradation . . . . . . . . . . . . . . 16
2.3.4 Degradation models for delamination, fatigue solder failure and cell
cracks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.5 Physical models for combined degradation modes/stresses . . . . . . 17
2.4 Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 Methodology 21
3.1 Methodology for the proposed physical model . . . . . . . . . . . . . . . . . . 22
3.1.1 Combined climatic stresses degradation rate model . . . . . . . . . . 22
3.1.2 Degradation indicator model . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.3 Experimental part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.1.4 Global climate data processing . . . . . . . . . . . . . . . . . . . . . . 28
3.1.5 Statistical error analysis methods . . . . . . . . . . . . . . . . . . . . . 31
3.2 Methodology for data-driven model . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.2 Modelling assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.3 Modeling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.4 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.5 Statistical errors analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3 Methodology of a hybrid model . . . . . . . . . . . . . . . . . . . . . . . . . . 40

4 Results and Discussion 41

4.1 Results for the physical model . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.1.1 Properties of the proposed power degradation function . . . . . . . . 42
4.1.2 Calibration and validation of the specific degradation rate models . 43
4.1.3 Calibration and validation of the combined degradation rate model 44
4.1.4 Degradation rates and lifetime prediction . . . . . . . . . . . . . . . 45
4.1.5 Comparison of the proposed physical model with other physical model 46
vi CONTENTS

4.1.6 Global degradation rates mapping: A global PV degradation risk

analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.1.7 Uncertainties evaluation of the physical model . . . . . . . . . . . . . 50
4.2 Results for the data-driven model . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2.1 ”Time and degradation pattern” dependent models as well as 3%
degradation threshold . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2.2 Model calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.2.3 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2.4 Model application and comparison with statistical models . . . . . . 56
4.2.5 Model limits and uncertainties . . . . . . . . . . . . . . . . . . . . . . 59
4.2.6 Assessing the effects of PV modules long-term degradation on life-
time energy yield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2.7 Simplified User Interface (PVLife Toolbox) . . . . . . . . . . . . . . . 62
4.3 Results for the hybrid model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 Application of the hybrid model . . . . . . . . . . . . . . . . . . . . . 64

5 Conclusion 65

Mapping of global climate stress factors 75

Supporting articles 77
Chapter 1

Introduction

PV modules as well as systems are affected by continuous cycles of temperature, humidity,

irradiation, mechanical stress, and soiling. These environmental stresses cause different
degradation modes that take place within a PV module and reduce the performance of
the system. In order to accurately determine the degradation rates of PV modules and
to understand the effects of the operating conditions, the evolution of power in real life
operation must be monitored simultaneously with the local operating condition. However,
this requires waiting a considerable amount of time and huge financial investments. In this
case, two main approaches are currently used to mimic the outdoor degradation rates.
The first approach is using indoor accelerated ageing tests where modules are exposed
to extreme controlled operating conditions in climatic chambers in order to degrade the
modules within a short time. Although these tests are carried out based on the established
International Electrotechnical Commission (IEC) standards Wohlgemuth (2012), an open
question remains if they can really represent outdoor degradation.
The second approach is the utilization of mathematical models to determine the degra-
dation rates in shorter periods and to predict the lifetime of PV modules. Two modelling
techniques are commonly used; physical and data driven techniques. Physical models
are developed to describe the link between the modules performance degradation to en-
vironmental variables Bala Subramaniyan et al. (2018). Data-driven techniques utilize
monitored operational data related to system’s performance to evaluate the degradation
rate Meyers et al. (2019). The biggest challenge of mathematical models is the reliability
of the predicted results Jordan et al. (2020). Usually, to valid degradation models, one
needs to have enough degradation data available, which is not always the case.

On one hand, according to the state-of-the-art of degradation models, most physical

models are developed for accelerated indoor ageing application. Physical models to quan-
tify the effects of outdoor climatic stresses are very helpful to understand the correlation of
the different local climates with the degradation processes. This understanding is useful to
develop new materials based on the operation local climates. However, such models have
received relatively little attention in the industry. According to our knowledge, it is of
recent that, Bala Subramaniyan et al. (2018) proposed a model to link the module perfor-
mance degradation to environmental variables for outdoor application. They applied their
model to predict the degradation rates of a mono-crystalline silicon module in different
climatic locations. However, since PV modules are characterized by numerous degradation
modes Köntges et al. (2017) and that PV modules are of different technologies, further
developments for a model that take into account multiple climatic factors are still needed.
Moreover, degradation models are developed on basic assumptions as well as simplifica-
2 1. Introduction

tions based on the available or known degradation patterns, this makes generalization of
a given model a challenge. In this case, it is important to develop different models using
different assumptions describing different degradation patterns. Therefore, in the first part
of research, a new physical model to quantify the effects of combined climatic stresses is
proposed as a contribution to this challenge.
The new model is based on the assumptions that, three major degradation mecha-
nisms; hydrolysis, photo-degradation and thermo-mechanical degradation are necessary
for lifetime prediction. Empirical models are proposed to describe the kinetics of these
three mechanism and hence to calculate the degradation rates. Through accelerated age-
ing tests, these mechanisms are reproduced and hence the models are validated. A model
to evaluate the impacts of the combined climatic stresses is then derived from the three
mechanisms. The derived model is calibrated and validated using identical mono-crystalline
silicon modules installed in three different climates: maritime (Gran Canaria, Spain), arid
(Negev, Israel), and alpine (Zugspitze, Germany) using real monitored meteorological data.
To extend the analysis on a global scale, global climate reanalysis data is used to extract
the model’s climatic input variables. The extracted variables are then applied to simulate
maps of degradation rates based on specific degradation mechanisms as well as the total
degradation rates.

On the other hand, data-driven techniques Phinikarides et al. (2014), Meyers et al.
(2019) that utilize monitored operational data related to system’s are commonly used es-
pecially to evaluate the degradation rates as well as the performance loss rates of PV
systems in outdoor performance. Generally, in spite of the wide application and the rec-
ognized potential of the data-driven techniques, limitations still exist for their use in PV
degradation analysis. The lack of a systematic and flexible approach to select parame-
ters of these models and their black-box character limit their understanding. Moreover,
data-driven techniques for lifetime forecast are based on fitting the available systems degra-
dation data by regression models and then extrapolating the evolution up to the time of
failure. However, in practice, the system’s degradation history available may be short and
incomplete, and a simple extrapolation may lead to large uncertainties. Also, one other
serious challenge is the accuracy of the models when applied after a shorter time interval.
Data-driven models usually require enough degradation history for their training. When
applied to forecast long-term degradation using a short degradation history, they often
provide unrealistic degradation scenarios. For example, Taylor and Letham Taylor and
Letham (2018) performed a comparison of the forecasting accuracy of different statistical
models at different time horizons. In their study, most models displayed large uncertainties
when applied after shorter time interval. To address these challenges, a new data-driven
model is proposed specifically for long-term degradation forecast. The proposed model is
aimed at improving the long-term forecasting accuracy for a shorter operation time. To
achieve this, time dependent degradation factors are assumed instead of a constant one.
To further improve the accuracy, different degradation factor models are proposed based
on the degradation patterns. Moreover, the model has been calibrated and validated using
degradation data from different sources and of different PV technologies both on module
and system level.

One other drawback of the data-driven methods is the lack of a correlation of the evalu-
ated rates with the triggering degradation mechanisms. To solve this problem, we combine
both the physical and the data-driven approaches proposed in this research into a hybrid
model. The hybrid model will aid to have more reliable long-term degradation forecast as
 3

well as having a physical understanding of the dominating degradation mechanisms influ-

encing the performance degradation. We believe such a model is useful to calculate more
reliable levelized cost of energy and thus the economic viability of solar energy as well as
to improve the PV materials according to the operating conditions.

The structure of the thesis is as follows:

Chapter 1. Introduction of the research, the novelty of the research work is introduced
with regard to the start-of-the-art.

Chapter 2. The available physical models are reviewed and Were possible a comparison of
the performance of the different models is carried out using experimental data
from indoor tests. At the end of the chapter the gaps for improvements are
identified.

Chapter 3. In this chapter, the methodologies of the proposed models are presented. The
chapter contains three major sections corresponding to methodologies for a
physical, data-driven and hybrid models respectively. In the first two sections,
the experimental data used for model calibration and validation are also de-
scribed.

Chapter 4. Here, the results are presented and discussed in order of the sections in chapter
3.

Chapter 5. Summarises the main conclusions of the research. The references follows the
chapter and the annex with the copies of the published articles that supports
the research are attached after the references.
4 1. Introduction
Chapter 2

Literature review

Table 2.1: Chapter Nomenclatures

Symbols
Ea activation energy [eV]
I Intensity
I SC short circuit current [A]
kB Boltzmann constant (8.62 × 10−5 eV/K)
P max power at maximum power point [W]
RD degradation rate [%/hour]
T max maximum temperature [K]
T min minimum temperature [K]
Rs Series resistance [Ω]
R sh Shunt resistance [Ω]
U voltage [volt]
∆T temperature cycles [K]
Abbreviations
BoM bill of materials
DH damp heat
EVA Vinyl-acetate monomerspresent in Ethylene-vinyl acetate
LID light induced degradation
PID potential induced degradation
PV Photovoltaic
RH relative humidity [%]
RMSE root means square error
t time
T temperature [K]
TOW Time of wetness
UV Ultraviolet [W /m2 ]

2.1 Photoltaic degradation and degradation modes

Photovoltaic degradation is the reduction in efficiency with which a PV module/system
converts sunlight into electricity Jordan and Kurtz (2013). Degradation modes are effects
that irreversibly degrade the performance of a PV module or of a system and cloud cause
6 2. Literature review

safety problems Köntges et al. (2017). A great number of different degradation modes are
observed in PV modules, both under outdoor operation and and also using indoor accel-
erated ageing testing. The most commonly observed degradation modes include Köntges
et al. (2017): LID, solder fatigue failure, silver grid finger delamination, bypass diode fail-
ure, delamination, cell cracks, corrosion, polymeric discoloration, UV degradation of the
cell, polymeric mechanical failure, and PID. Each of these degradation modes has different
causes and is triggered by various stress factors.
Apart from the modes listed, different technical risks, which affect the PV performance
and the resulting costs, were found by Moser et al. (2017) as: glass breakage, snail track,
defective backsheet, hotspot, soiling, overheating, and junction box failure. An occurring
degradation mode can have an increasing impact on the PV performance over time. It can
develop in isolation as well as in combination with other degradation modes or technical
risks and might lead to the failure of a PV module. However, despite the definitions above,
a clear understanding of the definition of a degradation mode is also still a challenge and
stress should be put on common nomenclature to define a degradation mode with the same
terminology.

2.2 Degradation models

In general terms, degradation models are used to relate a test item’s estimated failure
time with the wear and tear during its usage period. It is important to note that for a
PV module, failure can be related to both performance degradation and safety issues. In
this regard, it is a prerequisite to describe in which context it is being used. To add on,
degradation models can either be developed to evaluate electrical performance or material
degradation, it is also necessary to clarify the context in this regard. Therefore, in this
section, the term failure (as used throughout the entire document) is defined only in terms
of performance degradation. The described as well as the proposed models quantify the
evolution of PV modules performance over time.
Two modeling approaches; physical and data-driven are commonly used to evaluate the
degradation rates and lifetime of PV modules.

Data-driven models are often employed to estimate degradation rates based on analysis
of given data sets. Data-driven techniques can be divided into two categories: statisti-
cal techniques (regression methods, ARMA models, etc.) and artificial intelligence (AI)
techniques (neural networks (NNs), fuzzy systems (FSs), etc.) Chen Xiongzi et al. (2011).
The goal of the statistical analysis is to calculate the trend of the PV performance time
series and to translate the slope of the trend to an annual loss rate, in units of %/year
Phinikarides et al. (2014). Although these models can provide consistent performance loss
rates (PLRs), which are useful for data extrapolation and service life predictions, they do
not directly provide evidence for the degradation modes taking place in the module. Other
effects such as diffuse soiling, snow, shading or module mismatch have also a direct impact
on the performance trend.

Physical models are based on the physical/chemical understanding and assumptions of

a specific degradation mode. These models represent the mechanism involved in complex
physical/chemical processes. For well-known PV module degradation modes, several ana-
lytical models to predict PV module degradation are available. All these models are based
on the principle of understanding the underlying process, but they are still only heuristic
2.3 A review of the available physical models 7

models, which do not include the influence of material parameters.

A significant part of the research is devoted to physical models, therefore, the literature
discussion is inclined towards this directions. The statistical methods are also reviewed in
the supporting article attached in the annex.

2.3 A review of the available physical models

On one hand, physical models are developed to explain experimental observations
of different degradation modes and can also be used in experimental designs. A model
can be applied to analyze the experimental observations by fitting experimental data to
extract parameters that cannot be directly measured (e.g kinetic energy), used to study
the sensitivity of performance degradation to climatic stress factors such as T, RH, ∆T ,
UV and other influence factors.
On the other hand, models are also used to predictive or forecast future behavior.
Models are used to predict in situations where a given module needs to be analyzed under
different test conditions. In this case, the measurements of the module under a given con-
dition can be applied to calibrate the degradation models and apply the calibrated model
to predict the effects of other different test conditions as illustrated in 2.1. Degradation
models are used as forecasting models to know the progress of degradation for an extended
time horizon. In this case, the models are calibrated on experimental data after a shorter
time and then applied to extrapolate the degradation for a longer time.

Figure 2.1: Schematic diagram showing the different applications of degradation models:
forecasting, predictive, sensitivity analysis and experimental evaluation. S1 are the applied
experimental stress conditions and S2 are the modeled stress conditions. At model cali-
bration stage, the extracted parameters can be used for extrapolation, results assessment
by correlating them to material properties or used as inputs to predict the degradation at
different test condition as well as to perform a sensitivity analysis

Electrical parameters such as: P max , I SC , R sh and R s resistances are commonly mod-
eled as degradation indicators. Hence, the environmental stresses and their interactions
with the PV module components are assessed based on the reduction of the initial electri-
cal parameter at time ( t = 0) before aging and at time ( t = t) after aging or in the field.
The degradation models can be divided into two categories: models for the degradation
indicators and models for the degradation rates. The degradation indicator models are
8 2. Literature review

mathematical functions proposed to evaluate the degradation trends of a given degrada-

tion indicator over time. Since each degradation mode or technology under investigation
can have a different degradation curve, different authors have proposed models to fit the
observed degradation curves. On contrast, degradation rate models evaluate the rate of
change of degradation indicators. They are function of the applied stress factors.
The structure of the review is as follows; models are discussed according to specific
degradation modes, how they are applied and where necessary their performances are
compared and discussed.

2.3.1 Degradation models for corrosion

Corrosion is one of the most occurring degradation modes in PV modules Ndiaye et al.
(2013). Corrosion is caused by the presence of high temperature and high humidity in
the module. Humidity can enter the module through the backsheet or the layers of the
encapsulant and spread into the module Park et al. (2013). One hypothesis is that humidity
leads to the formation of acetic acid through the hydrolysis of EVA Kempe et al. (2007);
Whitfield et al. (2012); Masuda et al. (2015) resulting to corrosion. Corrosion attacks the
metallic connections of PV cells and results in a loss of adhesive strength between the
cells and the metallic frame, as well as an increased leakage current and therefore a loss in
performance Kempe et al. (2007).
Degradation indicator models based on power at maximum power point and series
resistance have been proposed to fit the degradation patterns for corrosion as:
(1) Model of Pan et al. (2011)
P max
= exp(−R D tβ ) (2.1)
P max(0)

(2) P max and R s models according to Braisaz et al. (2014)

1 − exp(−B)
P max = (2.2)
1 + exp(R D t − B)

R s = R S(0) = exp(R D t − B) (2.3)

where P max [W] and R s [Ω] are the output power and series resistance at time ( t), P max(0)
& R S(0) are the power output and series resistance at time ( t = 0), β is the experimental
parameter, B is a coefficient to be calibrated and R D [%/hour] is the degradation rate.
These functions can be applied in two way; to extract the degradation rates and also for
extrapolation. The first application is used to compare the degradation rates of a similar
module but with different stress conditions or to analyze the impact of applied stresses to
different BoM. In this case, different modules with different BoM are tested under similar
conditions and the extracted degradation rates are compared. The second application is
used to forecast the evolution of degradation at an extended time horizon.

A. Comparison of Pan and Braisaz Model

The two models are applied to fit time series datasets from DH ageing tests at different
conditions: DH/750 C/85%RH, DH/900 C/50%RH and DH/900 C/70%RH. Since the loss
in power could take different shapes as reported by Köntges et al. (2014), it is unlikely that
a single degradation function can fit all the different shapes. Here we compare how well
2.3 A review of the available physical models 9

the two function can fit three time series dataset from DH measurements with different
degradation profiles as shown in figure 2.2. The fitting is done using the least squares
fitting function inbuilt in Python. In both cases the derived degradation rates, the model
parameters as well as the RMSE are presented in table 2.2 . The RMSE is calculated as:
sP
n
j =1 ( f i − m i )
2
RMSE = (2.4)
n
where, m is a measured value and f is the fit value, n are the number of observations.

Figure 2.2: Pan and Braisaz model fit for different DH measurements, the dotted lines
represent the experimental data and the thick lines are the corresponding models fit respec-
tively. The colors represents the different experimental condition as well as the different
failure profiles.

Table 2.2: Derived model parameters for different damp heat conditions
Experimental condition R D [%/hour] Model parameter (β) model parameter (B) RMSE [%]

Pan Model
DH 75 o C/85% RH 5.66e-8 1.50 - 0.35
DH 90 o C/50% RH 6.40e-4 0.43 - 0.19
DH 90 o C/70% RH 3.15e-9 2.13 - 4.99
Braisaz Model
DH 75 o C/85% RH 4.43e-4 - 6.34 0.39
DH 90 o C/50% RH 4.66e-4 - 5.42 0.39
DH 90 o C/70% RH 2.58e-3 - 10.75 1.33

Figure 2.2 shows the experimental data with the respective models fit. It is can be seen
that, depending on the degradation shape, the fitting accuracy varies for both models.
For example, the model of Pan was able to fit well the red and blue curves and could not
provide perfect fitting of the black curve. On contrary, the model according to Braizas
was able to perfectly fit the black curve but not the the blue curve. Overall the model
according to Braizas is preferable to that of Pan since the extracted degradation rates are
consistent with the applied stresses. However, a model that include a shape parameter
that can perfectly fit/optimize, the different degradation shapes is still needed and will be
10 2. Literature review

part of this research work.

Another application of degradation models, as mentioned before, is to extrapolate the

measured performance degradation to longer time horizons. Therefore, the two models are
calibrated on the same dataset (i.e. DH/750 C/85%RH) but only after 4000 hours. In each
case the model parameters are extracted as in Table 2.3. Then the calibrated models are
applied to forecast the power degradation from 4000 hours until 6500 hours. The results
for all the three models are plotted with the extended measurements 2.3

Table 2.3: Model parameters for Pan and Braisaz model after 4000 hours of
DH/75C/85%RH
Model R D [%/hour] Model parameter (β) model parameter (B) RMSE [%]

Pan Model
Pan 6.44e-10 2.07 - 0.13
Braisaz 9.00e-4 - 7.56 0.10

Figure 2.3: Extrapolation after 4000 hours of DH/75C/85%RH using Pan (blue) and
Braisaz (red ) models

Figure 2.3 shows the extrapolation of power degradation from 4000 hours to 6500 hours
using of Pan and Braisaz models. Indeed, it is hard to conclude that a particular model
is better than the other as this might highly depend on the degradation shape under
evaluation. However, two main conclusions can be drawn:

• A perfectly fitting model does not guarantee good forecasting. This is visible for
2.3 A review of the available physical models 11

the case of Braisaz’s model, with small residual deviation but instead the model
demonstrates the worst performance for this data set.

• Simple extrapolation after a given interval might not be reliable for longer time ex-
trapolation. This is visible in all the models. Indeed, this can be influenced by mea-
surement uncertainties, nature of the degradation pattern and many other factors.
Although many researchers use this kind of extrapolation, the reliability of extrapo-
lated performance is questionable and yet an unresolved challenge. Therefore, part
of this research focuses on developing methods to improve long-term extrapolation
accuracy.

The degradation rates of indicator models (equations 2.1, 2.2 and 2.3 can be calculated
as functions of T and RH using three common kinetics models namely; the Peck’s model,
the Eyring model and the exponential model Escobar and Meeker (2007); Jordan and
Kurtz (2010). These models are developed on the primary assumption that the rate of
degradation is proportional to the concentration of water in PV modules and that the rate
constant has Arrhenius temperature dependence. The models are written as:
(3) Peck’s model
³ E ´
a
R D.P eck = A.exp − RH n (2.5)
kB T
(4) Eyring Model
³ −E b ´
a
R D.E yring = A.exp − (2.6)
kB T RH
(5) Exponential model
³ −E ´
a
R D.Exp = A.exp .exp( m × RH ) (2.7)
kB T

where A and n, b and m are model parameters. In order to obtain A , E a , n, b and m

in the equations (2.5), (2.6) and (2.7), the equations can be fitted to experimental data or
represented on a logarithmic scale by a straight line, using the following equations:
³ E ´
a
ln(R D.P eck ) = ln( A ) − + n × ln(RH ) (2.8)
kB T

³ E ´ b
a
ln(R D.E yring ) = ln( A ) − − (2.9)
kB T RH
³ E ´
a
ln(R D.Exp ) = ln( A ) − + m × RH (2.10)
kB T
A plot of ln(R D ) versus 1/T (0 K ) gives an Arrhenius plot with a slope E a / k B and an intercept
ln( A ).

B. Comparison of Peck’s, Eyring and Exponential models

Here we performed a sensitivity analysis of the degradation rate models 2.5, 2.6 and 2.7 to
compare how the degradation rate varies with input climatic variables (temperature and
humidity). To perform unbiased comparison, all the three models where fitted using the
same dataset (in this case DH/75 o C/85% RH ). Also since the Braizas model was found to
have consistent results, it was selected for the calibration. For all the datasets, the model
12 2. Literature review

parameter B was keep the same in order to evaluate the same degradation rate. Moreover
since the activation energy should be the same in the physical point of view for a similar
reaction, it was kept constant for the three models. This is also good for a comparison
purpose, given that, this parameter is very sensitive. The extracted models parameters
used in the sensitivity analysis simulations are presented in table 2.4.
Figures 2.4, 2.5 and 2.6 show the plots of the sensitivity analysis of the Pecks, Eyring
and exponential models respectively. It is clearly visible that despite a uniform calibration
procedure, the sensitivity of the model varies to a big extent. This can be explained by
the assumptions taken during the models formulation. Although all the models have an
Arrhenius temperature dependence, the formulation of the relative humidity contribution
changes the variation of the degradation rate. For example, the exponential model, assumes
an exponential dependence of the relative humidity and this explain why the model is
too sensitive to relative humidity compared to the Pecks and Eyring model. This high
sensitivity of relative humidity together with the Arrhenius temperature dependence could
deteriorate the model if used for extrapolation. Overall, the Pecks model shows a good
dependence of both temperature and relative humidity.

Table 2.4: Derived model parameters used in the sensitivity analysis

Model A Parameters (n,b & m) (E a ) [eV] RMSE [%]

Peck 9.46 2.07 0.57 0.39

Eyring 9.97e4 6.01 0.57 0.39
Exponential 3.64e-4 0.23 0.57 0.39

Figure 2.4: Sensitivity analysis plot of the degradation rate with relative humidity and
temperature using the Pecks model (note the y axis in multiplied by 1e-4)
2.3 A review of the available physical models 13

Figure 2.5: Sensitivity analysis plot of the degradation rate with relative humidity and
temperature using the Eyring model (note the y axis in multiplied by 1e-4)

Figure 2.6: Sensitivity analysis plot of the degradation rate with relative humidity and
temperature using the exponential model (note the y axis in multiplied by 1e-3)
14 2. Literature review

2.3.2 Models for potential induced degradation

Potential induced degradation (PID) has been observed in all PV technologies and in al-
most all operating climates. It does not occur so frequently, but if it does, its effect can lead
to a dramatic performance loss within a short period Moser et al. (2017); Hacke et al. (2016,
2015); Köntges et al. (2014). In general terms, PID is caused by the difference in potential
between the cells and the support structure of the module. This difference drives a leakage
current that can lead to power degradation. Different types of PID occur depending on the
module technology. For crystalline silicon PV, two degradation modes can be identified,
PID-p (for polarisation or passivation) and PID-s (for shunting). PID-p is a temporary
and reversible degradation of the passivation layer, which reduces the performance due to
a surface recombination increase Naumann et al. (2014); Halm et al. (2015); Swanson et al.
(2005). PID-s is due to a leakage current involving an ionic flow of sodium ions (Na+)
from the glass, encapsulant or cell surface into the cell, diffusing into the silicon stacking
faults and shunting the cell Lausch et al. (2014). The sodium incorporation in the silicon
surface degrades primarily the fill factor (FF) , the open circuit voltage (Voc ), and lastly
the short circuit current ( I sc ). The relevant stress factors for PID-s include Köntges et al.
(2014): high temperature, relative humidity, system voltage, light, bias-junction potential
and injected carriers.

Different authors Braisaz et al. (2014); Hacke et al. (2015); Annigoni et al. (2017);
Hacke et al. (2016); Hattendorf et al. (2012); Taubitz et al. (2014) have proposed models
to evaluate PID degradation patterns using different indicators as;

(6) PIDHACKE model

P max ³ E ´
a
= 1 − A.exp − RH n × t2 (2.11)
P max(0) kB T

A parabolic model was proposed by Hacke et al. (2015) to fit the power degradation of c-Si
modules. The constants A and n are determined by fitting the equation to experimental
results. The parameters have to be determined for each module type. This parabolic model
is applicable to the beginning of the degradation phases of PID-s, as it can fit the beginning
of a sigmoid and does not describe the stabilization phase of the sigmoidal curve.

(7) PID model according to Annigoni et al. (2017)

P max ³ E ´
a
= 1 − A.exp − RH n × t2 × U (2.12)
P max(0) kB T

Annigoni et al. (2017) re-wrote the PIDHACKE model (2.11) including a voltage term (U)
(2.12) and used the indoor data to determine coefficients of the model for distinct aging
contributions (temperature, relative humidity and time) and then applied the model (2.12)
to outdoor PID degradation for different climates.

(8) Hattendorf model

P max (U, T, RH, t) = P max(0) (1 − P ( t)) (2.13)

³ ´
1 − exp τt1 ³ ³ U − U ´´−2
0
P ( t) = P ∞ ³ ´ ; P (U ) = 1 + exp (2.14)
t− t 0
1 − exp τ2 Φ
2.3 A review of the available physical models 15

t 0 = a × b × tˆ0 ; τ1 (T ) = b2 × τˆ1 ; τ2 = τˆ2 (2.15)

H0 ³T −T ´
0
a( H ) = ; b(T ) = exp (2.16)
H ϕ

The model of Hattendorf et al. (2012) is based on a matrix of indoor experiments where
modules are exposed to varying voltage, module temperature and ambient humidity. The
conditions are varied to determine the model parameters for the module power. The
model includes six adaptation parameters: tˆ0 , U0 , τˆ1 , τˆ2 , Φ and ϕ. H0 and T0 are scaling
parameters. The function P ( t) describes the power loss caused by degradation. P∞ (U )
is its limit for t → ∞, a(H ) and b(T ) are the acceleration functions of relative humidity
and temperature. For T = 0 and H = 0, they are equal to 1, therefore τˆ1 , τˆ2 are the time
constants under these conditions. τ2 remains constant for a given module. To determine
the model’s parameter, the power degradation is measured as a function of time with the
system voltage U as parameter and a fixed humidity H as well as temperature T . The
saturating power P∞ is extracted by fitting P ( t) to the measured data.
(9) Taubitz model Taubitz et al. (2014)

Shunting phase
³ −t ´
R sh ( t) = a S exp (2.17)
b S ( t)
Regeneration phase
³ t ´
R sh ( t) = C R + a R exp (2.18)
b R ( t)
Transition phase
R sh ( t) = a T (T )( t + b T (T ))2 + C T (2.19)

Taubitz et al. (2014) proposed a regeneration model for shunt resistance evolution over
time due to PID degradation. The shunt resistance was modeled in three phases; shunting
phase, regeneration phase and transition phase. Where a S , b S , C T , b T , a T (T ), b S (T ),
b T (T ), C R and a R are constants and have to be determined for a specific module type.
Some of them are dependent on the module temperature T . The constants are determined
by measuring the times t S , t T , and t R for reaching certain target values.

(10) PID model according to Braisaz et al. (2014)

R sh(0)
R sh ( t) = (2.20)
1 + aR D t

B ³ E ´
a
RD = A × U exp − (2.21)
1 + exp(−C (RH ) + D ) kB T

The model is based on shunt resistance R sh degradation as an indicator, as it is the most

important parameter for PID. The evolution of R sh as a function of voltage, temperature
and relative humidity was modeled as (2.20). Where R sh is the shunt resistance at time
( t), R sh(0) is the initial shunt resistance, A, B, C and D are model coefficients, U is the
applied voltage and R D is the degradation rate dependent on RH, U and T.
16 2. Literature review

2.3.3 Models for ultraviolet irradiance degradation

Ultraviolet (UV) light exposure has been reported to cause PV module degradation in
a number of ways. As an example, it could result in discoloration of the encapsulant
material Dunn et al. (2013), delamination at the glass encapsulant or cell encapsulant
interface Munoz et al. (2011). The parameter most impacted by UV exposure is the short
circuit current ( I sc ). Braisaz et al. (2014) proposed a model for short circuit degradation
due to UV exposure over time. They found that the degradation curve is not linear but
an exponentially decreasing curve. The short circuit is modeled as a function of UV as:
(11) UV model according to Braisaz et al. (2014)
I sc ( t) = I sc(0) − aDUV ( t) − b(1 − exp(−CDUV )( t)) (2.22)
Z t
DUV ( t) = E ( u) × 5.5% d u (2.23)
0
Here, DUV is the UV dose in in M J / m2 or kW h/ m2 , I sc is the short circuit current at
time (t), I sc (0) is the initial short circuit current and a, b and C are model coefficients.
The multiplication by 5.5% is due to the UV radiation between (280nm – 400nm) which
is approximately 5.5% of the total light spectrum E (u) Koehl et al. (2009).
(12) The Schwarzschild Law
k = A(I ) p (2.24)
The Schwarzschild law has been applied by Gu et al. (2015) to study the effect of intensity
and wavelength of spectral UV light on discoloration of laminated glass/EVA/PPE PV
modules. Here, k is a constant, I the intensity and p is the Schwarzschild coefficient.
Recommendation: When applying this expression in performance (power) prediction mod-
els where other loads are also applied, the parameter p must be calibrated according to
the knowledge of severity ranking Jordan et al. (2017a).

2.3.4 Degradation models for delamination, fatigue solder failure

and cell cracks
(13) Coffin-Manson’s equation
σ
N= (2.25)
(∆T )β1
The model is used to predict degradation modes caused by temperature cycling such as
encapsulant delamination, fatigue solder failure and cell cracks. According to Escobar and
Meeker (2007), the model describes the number of cycles to failure as (2.25), where ∆T
is the temperature range, σ and β1 are properties of the material and test setup. The
cycles-to-failure distribution for temperature cycling can also depend on the cycling rate
(e.g. due to heat buildup). An empirical extension of the Coffin-Manson relationship that
describes such dependencies is Escobar and Meeker (2007):
σ 1 ³ E × 11605 ´
a
N= exp (2.26)
(∆T )β1 ( f req)β2 T max (K )
where f req is the cycling frequency and E a is a quasi-activation energy.
(14) Crack propagation model
1
C a ( t) = C a ( t − 1) + ³ ´m (2.27)
x 125
Ta
2.3 A review of the available physical models 17

The model was suggested by Braisaz et al. (2014) and it was applied to simulate the
degradation of the short-circuit current I sc due to the expansion of cell cracks caused by
temperature. C a ( t) is the crack activation at time ( t), C a ( t − 1) the crack activation at
time ( t − 1), T a is the daily temperature amplitude, m a model parameter and x is the
number of thermal cycles. The crack activation/propagation model is dependent on the
daily temperature amplitude T a .

(15) Damage accumulation model

³ Q ´
D = C (∆T )n ( r (T ))m .exp − (2.28)
k B T max

The model was used by Bosco et al. (2016) in order to calculate the solder fatigue damage
in seven cities investigated in their study and compared to FEM simulated results. They
found out that the model fits well to the simulated calculations. In this equation, ∆T
is the mean daily maximum cell temperature change, T max the mean maximum daily
temperature, C a scaling constant, Q the activation energy, k B Boltzmann’s constant, r (T )
the number of times the temperature history increases or decreases across the reversal
temperature, T the period of a year and n & m are model constants similar to those in
the Coffin-Manson equation.

(16) Backsheet degradation rate model

T −T0
R D ≈ I X ( b + m × TOW ) × (T f ) 10 (2.29)

Here, I is the light intensity, X , b and m are fit parameters, TOW is the time of wetness,
T the temperature, T0 a reference temperature and T f is a multiplier for the increase in
degradation for a rise in temperature in 10 K steps.
The model is used to estimate a potential form of the degradation kinetics of the
backsheet. This model was applied by Kempe (2014) to model the uncertainty in a 25 year
equivalent test for module backside exposure to irradiance and temperatures in different
climatic zones.
Recommendation: As also mentioned by Kempe, the parameter that describes the effect
of time of wetness has very high uncertainties, we recommend careful comparison of the
relative change in degradation rate with changes in TOW. In case one wants to extract
thermal parameters such as activation energy, the multiplier term (T f ) can be replaced by
the Arrhenius term.

2.3.5 Physical models for combined degradation modes/stresses

Since degradation of PV modules in outdoor operation is influenced by multiple environ-
mental stresses, models based on multiple stresses are viable for outdoor service lifetime
prediction. In the literature, only a few authors Gaines et al. (1978); Bala Subramaniyan
et al. (2018) have proposed models based on the combination of several stresses.

(17) Model of Gaines

P max 1
= [1 − R D t] β (2.30)
P max(0)

R D = A f T f RH f M f G f ω (2.31)
18 2. Literature review

Gaines et al. (1978) proposed a model for power output degradation based on multiple ac-
celerated environmental stresses (2.30). Where R D is the degradation rate and the factors
f T , f RH , f M , f G and f ω are associated with a decrease in power output due to effects of
temperature (T ), relative humidity (RH ), mechanical stresses (from temperature differ-
ences) ( M ), gaseous concentration (G ) and the frequency of the temperature excursion (ω).
The mathematical form of each factor is formulated to represent the underlying physical
phenomena.

An Arrhenius form is used for temperature influenced effects:

³ B´
f T = exp − (2.32)
T
B denotes a constant parameter and T denotes temperature.

The effect of relative humidity f RH is represented by:

³ ³ 1 1 ´´C − DT
f RH = 1 + (RH )0 .exp C 0 − (2.33)
T C T0
the second term in the bracket corrects the relative humidity as a function of temperature,
given a specified relative humidity at T0 . C and D are constant parameters.

The mechanical/temperature excursion factor f M is represented by:

³ ³ ´´ ³ ³ ´´
h exp G 1 T1 − T1 + exp − G 2 T1 − T1 i
b b
fM = × exp( J ∆T ) (2.34)
D0
the first term in the bracket reflects the stresses arising from differences in expansion coef-
ficients of bonded materials. The constants G 1 , G 2 , D 0 and T b are chosen to represent the
estimated magnitudes of these fatigue effects. The factor exp( J ∆T ) estimates the magni-
tude of the temperature excursion ∆T , where J is a constant.

The effect of gaseous concentration f G is represented by:

h G iE − TF
fG = 1 + (2.35)
Go
here, E and F denote constant parameters and T is the temperature.

The frequency of the temperature excursion f ω is represented by:

h ω iP − T
Q

fω = 1 + (2.36)
ωo
ω is the frequency and P as well as Q are constant parameter. In a constant temperature
test, T is a constant and ω is taken to be zero. In the cyclic temperature tests, reciprocal
temperature is considered to be a sinusoidal function of time:
1
( t) = τ + ∆τ sin(ω t) (2.37)
T

1h 1 1 i
τ= − (2.38)
2 T min T max
2.4 Chapter Overview 19

T min and T max are the minimum and maximum temperatures associated with the temper-
ature cycles.
Recommendation: The model of Gaines presents the previous approach on multiple stress
modeling, however, the user should take caution that this model was developed and ap-
plied on PV modules that had a different construction from today’s modules. Therefore,
its application might need some modification to fit the current PV module construction
types.

(18) Degradation rate model of Subramaniyan

³ β1 ´
Rate(T, ∆T,UV , RH ) = β0 .exp − × (∆T dail y )β2 × (UVdail y )β3 × (RH dail y )β4
k B T max
(2.39)

Another model to calculate the degradation rate due to combined environmental stresses
has been proposed recently by Bala Subramaniyan et al. (2018). The model takes into
account the effect of both static and cyclic temperature, ultraviolet radiation and relative
humidity as (2.39). Where Rate(T, ∆T,UV , RH ) is the reaction rate, T max the daily max-
imum temperature of the module [K], ∆T dail y the daily cyclic temperature of the module
[K], UVdail y the daily daytime average irradiance [W/m2 ], RH dail y the daily average rel-
ative humidity [%] and k is the Boltzmann constant. The model parameters β0 , which
is the frequency factor [s−1 ]; β1 , the activation energy [eV]; β2 , the effect of cyclic tem-
perature; β3 , the effect of UV radiation and β4 , the effect of RH, can be estimated from
measured data through data fitting techniques. In their study, the model was calibrated
using outdoor measurements and then applied to predict the degradation rates in four
other climates.

2.4 Chapter Overview

Based on the reviewed literature, two main aspects/challenges have been identified as;

• Degradation indicator models are developed to fit specific degradation patterns based
on a specific degradation mode. A generalized model that can optimize the different
degradation patterns is still a challenge. Therefore, in this research we aim to propose
a model that can be generalized for different degradation patterns.

• Although several degradation rate models are developed for specific degradation
modes for indoor application, little advances are made to develop models for com-
bined climatic stresses that can be used for outdoor prediction. It is clearly seen
that the need for such a model dates back in the 1970’s however since then it has
been of recent that Bala Subramaniyan et al. (2018) added a contribution to this
effort. Therefore, further developments for a model that take into account multiple
environmental stress factors are proposed in this research.
20 2. Literature review
Chapter 3

Methodology

Table 3.1: Chapter 3 Nomenclatures

Symbols
ci clearness index
Ea activation energy [eV]
k degradation factor [1/year]
kB Boltzmann constant (8.62 × 10−5 eV/K)
P max power at maximum power point [W]
RD degradation rate [%/hour]
SD res residual standard deviation
T amb ambient temperature [0 C ]
T D ew dew point temperature [0 C ]
Tm module temperature [K]
Subscripts
cal calibration
h hydrolysis
p photo-degradation
t thermo-mechanical
T total
Abbreviations
∆T temperature cycles [K]
DH damp heat
FT failure time
GHI global horizontal irradiance
MSEP mean square error of prediction
NWP Numerical Weather Predictions
RUL remaining usefull lifetime
STC standard test conditions
t time
UV Ultraviolet [W /m2 ]
WS wind speed [m/s]
WVP saturated water vapour pressure

In this chapter, the methodologies for the proposed physical, data-driven and hybrid
models are described. The chapter is therefore, divided into three sections, 3.1, 3.2 and
22 3. Methodology

3.3. In section 3.1, the physical model is described, the experimental data used for model
calibration and validation as well as the methods used to process the climate data used
to simulate the global degradation maps are described. In section 3.2, the data-driven
approach is described as well as the experimental data used for model calibration and
validation. In section 3.3, we describe how the two methods are linked into a hybrid
model. For consistence and to allow the reader to follow, it is better to read section 3.1
with the corresponding results in section 4.1 and section 3.2 together with 4.2 respectively.

3.1 Methodology for the proposed physical model

In chapter 2 further development of a physical degradation rate model that take into ac-
count multiple climatic stress factors as well as a generalized degradation indicator model
were identified. In this research, we contribute to these challenges by proposing a new
model to quantify the impact of combined climatic stresses as well as a degradation indi-
cator model that can optimize several degradation patterns.

3.1.1 Combined climatic stresses degradation rate model

Photovoltaic modules are exposed to a variety of climatic loads during outdoor operation.
Over time, these loads trigger a number of degradation modes within the modules leading to
performance loss. In section, an approach to develop a combined climatic stress degradation
rate model is described.

A. Background and modelling assumptions

When developing a model for PV modules degradation prediction, three main aspects need
to be considered:
• Impact of PV materials variations. New materials are proposed frequently to improve
PV performance.

• The different operating climatic conditions. PV modules operate in different climates

for example in arid, maritime, tropical climates.

• The different PV technologies, for example: crystalline silicon, thin films, and differ-
ent module designs like bifacial, glass-glass or glass-backsheet.
However, a model that takes into account all the three aspects is rather complex and might
require an extensive experimental campaign which in turn makes it expensive. Moreover
the fact that, materials are changing frequently makes developing a generalized model a
challenging problem. In this regard, several assumptions and simplification have to be
used. In developing our model, the following assumptions are used.
• The degradation rates are assumed to be proportional to the applied stresses and to
have an Arrhenius temperature dependence. Therefore, the material dependence of
the model is evaluated through the activation energy.

• Three degradation processes are assumed to be necessary for service lifetime pre-
diction as; hydrolysis, photodegradation, and thermomechanical degradation. The
assumption is based on the ability to reproduce these degradation mechanism in the
3.1 Methodology for the proposed physical model 23

laboratory using accelerated ageing tests and that, the underlying physical and(or)
chemical degradation kinetics have been studied from different studies Zhu et al.
(2016); Kempe (2006); Sharma and Chandel (2013); Ndiaye et al. (2013); Gok et al.
(2017); Jordan and Kurtz (2013); Marín et al. (1996); Park et al. (2013); Wu et al.
(2014); Koehl et al. (2017) .
Empirical kinetics models to evaluate the degradation rate constant are proposed or
selected to describe the three degradation processes. These models are developed
depending on the applied climatic stresses as illustrated using the general reaction
equation below:

Stress1 + Stress2 + ...StressN → D e gradation precursor (3.1)

where Stress1, Stress2, ...StressN are the (climatic) degradation factors triggering
the degradation mechanism under evaluation. The kinetics of the dominating degra-
dation process is quantified as the degradation rate (R D ). The mathematical form of
the degradation rates of the three degradation processes is described in the following
sections.
• The three processes are described based on combinations of input climatic vari-
ables. The combinations are selected based on the current indoor testing procedures
Wohlgemuth (2012) which aim to reproduce degradation modes according to three
main processes. Therefore, the combination of the input stresses are assumed to be
responsible for triggering a specific reaction that might induce specific degradation
modes. Figure 3.1 summarizes the hypothesized degradation mechanisms that are
known to be induced by the applied loads.
• The effects of the applied loads is quantified on how much they affect the power
output of the module, hence power at maximum power point is used as a degradation
indicator in this research.

Figure 3.1: Schematic diagram of the modelling hypotheses. The combination of climatic
stress triggers the three main precursor processes and the triggered processes are linked to
different degradation modes
24 3. Methodology

B. Hydrolysis: Temperature and Relative humidity model

The Peck’s model described in chapter 2 equation (2.5) is selected to model the effect
of temperature and relative humidity based on the sensitivity analysis study in chapter 2,
figure 2.4. The model shows a good dependence of both temperature and relative humidity.
The models is hereby re-written as:
³ E ah ´
R Dh = A h .rh nef f .exp − (3.2)
kB Tm

100
rh e f f (%) = (3.3)
1 + 98.exp[−9.4 × (RH (%)/100)]
where R Dh is the degradation rate constant, k B is the Boltzmann constant (8.62 × 10−5 eV ),
T m is the module temperature (Kelvin), A h is the pre-exponential constant, rh e f f (%)
is the effective module relative humidity proposed by Koehl et al. (2012), n is a model
parameter that indicates the impact of relative humidity on power degradation. In this
context, E ah is defined as the activation energy for power degradation due to hydrolysis
process.

C. Photo-degradation: UV, temperature and relative humidity model

The formulation of photo-degradation has been developed to take into account the com-
bined effects of UV stress and moisture on PV modules. The baseline line assumption is
that, UV stress alone would influence some degradation modes but also some modes might
be influenced with a combined UV moisture stresses. For example, Ngo et al. (2016) has
demonstrated that, UV irradiation plays a significant role in generating acetic acid with the
presence of relative humidity that causes the power degradation of PV modules. Therefore
to take into account the effect of UV alone and the combined UV-moisture contributions
a model is proposed as:
³ E ap ´ ³ E ap ´
R D p = A h .UV X .rh nef f .exp − X
+ A h .UVdose .exp − (3.4)
kB Tm kB Tm
OR
³ E ap ´
R D p = A h .UV X (1 + rh nef f ).exp − (3.5)
kB Tm
where R D p is the rate due to photo-degradation processes, UV is the UV irradiance (W /m2 ),
X is a model parameter that indicates the impact of UV on power degradation and E ap is
the activation energy for power degradation due to photo-degradation process.

D. Thermo-mechanical degradation: T max, & ∆T

The model used to quantify the effects of thermo-mechanical stresses is the modified Coffin-
Mansion relationship that includes the effect of maximum temperature according to Esco-
bar and Meeker (2007). The modified model is re-written as;
³ E at ´
R Dt = A t .C N .(273 + ∆T )Θ .exp − (3.6)
k B T max
where ∆T = (T max − T min ) is the temperature difference (Kelvin), C N is the cycling rate,
T max and T min are the module maximum and minimum temperatures, E at the activation
energy of power degradation due to thermo-mechanical process.
3.1 Methodology for the proposed physical model 25

E. Combined (RH , T , UV T max & ∆T ) model

The transition from indoor degradation rate evaluation to outdoor is a challenging task
as the interaction of different stresses and processes leading to PV degradation is not well
known. The underlying assumption deployed in this research is that: some degradation
processes might lead to specific degradation modes independent of the others and that some
might have a synergistic nature as described in the schematic figure 3.2, which results in a
variety of degradation modes.

Figure 3.2: Schematic showing possible interactions of the three degradation processes

Hence, this assumption allows us to evaluate the total degradation rate as the sum of
both independent and dependent processes. The mathematical form of the total rate is
expressed as:

R DT = A N (R Dh + R D p + R Dt + R Dh .R D p + R Dh .R Dt + R D p .R D p + R Dh .R D p .R Dt ) (3.7)

OR
R DT = A N (1 + R Dh )(1 + R D p )(1 + R Dt ) − 1 (3.8)

OR
Y
n
R DT = A N . (1 + R D i ) − 1 (3.9)
i =1

where R DT (%/year) is the total degradation rate, R D i is the i th rate constant and n is the
total number of degradation processes. A N is the normalization constant of the physical
quantities, in this case it takes the units ( year −2 %).
26 3. Methodology

3.1.2 Degradation indicator model

As already mentioned in chapter 2, different electrical parameters such as: P max , I SC ,
R sh and R s are commonly modeled as degradation indicators to quantify the effects of
the applied stresses. In this study, P max is considered as the degradation indicator. The
reasons to use P max as an indicator are: it is easily calculated from the current-voltage (I-
U) curves unlike R sh and R s that are derived from fitting procedure of the illuminated I-U
curve with diode equations. Moreover, it’s a parameter needed for energy yield evaluation
and a metric used by manufacturers to give warranties.

A. Proposed power degradation model

Usually a linear-shaped power loss is assumed for outdoors degradation analysis and lifetime
predictions. However, as reported by Jordan et al. (2017b), non-linearity of power loss is
often observed in the field depending on the modules technologies or degradation modes.
In their study, the effect of the different degradation curves, observed (approximately)
in field performance on the levelized cost of energy (LCOE) was quantified using Monte
Carlo simulation. Köntges et al. (2014) also reported that, the loss in power can take
different shapes, for example the following categories: exponential-shaped, linear-shaped,
steps degradation and saturating power loss over time. A simplified non-linear power
degradation model that can optimize most these degradation shapes has been proposed as;
P ( t = t) ³ ³ Γ ´µ ´
= 1 − exp − (3.10)
P I nitial R D i .t
where P ( t = t) and P I nitial are the module output power at time t and the initial output
power, respectively. Γ is the power susceptibility, which is assumed to be a material
property, µ is the shape parameter and R D i is the degradation rate constant of degradation
process i or the total degradation rate.

B. Failure time function

The failure time (FT) is defined in this study as a 20% loss of the nominal power (common
manufacturer’s warranty). Using equations (3.8) and 3.10, the failure time can be derived
as:
Γ
FT = 1
(3.11)
R DT .(| log(0.2)|) µ

3.1.3 Experimental part

In order to calibrate and valid the models, two sets of experiments where carried out; first
using indoor accelerated ageing tests and second, using outdoor monitoring tests.

A. Experimental: Indoor accelerated aging

Distributed damp heat (DH), thermal cycling (TC) and combined DH-UV stress tests were
carried out at different test conditions as shown in 3.2. The condition are varied because
one set is used for model calibration and the other set is used for model validation. Table
3.3 shows the time steps and the measured power degradation at the different applied
test conditions. The tested modules are from the same manufacturer with p-type homo-
junction crystalline silicon (c-Si) cells, with a thermoplastic encapsulant material and no
aluminum layer as additional moisture barrier.
3.1 Methodology for the proposed physical model 27

Table 3.2: Experimental conditions for DH, combined UV/DH and TC tests
Test conditions
Experimental Test conditions for model calibration Test conditions for model validation

DH 750 C /85%RH (6500 hours) 850 C /85%RH (6500 hours)

180W / m2 /650 C /55%RH 180W / m2 /850 C /55%RH
UV/DH
(4000 hours) (4000 hours)
TC −40 C /400 C (3500 cycles)
0
−40 C /850 C (1200 cycles)
0

Table 3.3: Time steps and measured power degradation at different test conditions
DH: 750 C /85%RH
Time steps[hours] 0 2000 4000 5000 5500 6000 6500
Normalized power 1.000 0.991 0.967 0.993 0.986 0.990 0.987
DH: 850 C /85%RH
Time steps[hours] 0 500 1000 1500 2000 2500 3000 3500 4000
Normalized power 1.000 1.000 0.995 0.991 0.991 0.983 0.977 0.953 0.944
UV/DH: 180W / m2 /650 C /55%RH
Time steps[hours] 0 2000 2500 3000 3500 4000
Normalized power 1.00 0.976 0.972 0.964 0.953 0.945
UV/DH: 180W / m2 /850 C /55%RH
Time steps[hours] 0 2000 2500 3000 3500 4000
Normalized power 1.00 0.984 0.960 0.928 0.872 0.845
TC: −400 C /400 C
Cycle steps 0 2000 2010 2050 2100 2600 3000 3200 3400 3600
Normalized power 1.00 0.979 0.978 0.975 0.976 0.926 0.968 0.972 0.975 0.960
TC: −400 C /850 C
Cycle steps 0 200 400 600 800 1000 1200
Normalized power 1.00 0.965 0.957 0.941 0.937 0.920 0.888

B. Experimental: Outdoor monitoring

To clearly assess the effect of combined climatic stresses on performance degradation,

three identical experimental mono-crystalline silicon (mc-Si) modules are exposed and un-
der monitoring in three climatic zones, maritime in Gran Canaria, Spain; arid in Negev,
Israel; and alpine in Zugspitze, Germany (see Figure 3.3). At each test site, the electrical
performances and climatic conditions are under monitoring at high time-resolution. In
Gran Canaria, the tilt angle is 230 , and the azimuth angle is 1690 of PV modules. In
the Negev the tilt angle is 310 , and the azimuth angle 1800 for PV modules. The module
in Gran Canaria has been exposed for over 7 years and the ones in Negev as well as in
Zugspitze have been exposed over 5 years. A more detailed descriptions focusing on the
test sites and categorization of weathering stresses for PV modules in Koehl et al. (2011,
2018). Apart from the electrical performance measurements, the modules temperatures
are also recorded every 10 minutes. The sensors for measuring modules temperatures are
located under one of the central cells. They are Pt100 sensors, which are attached from the
back using adhesive aluminum tapes. Other meteorological data such as: relative humid-
ity, global irradiation, UV irradiance and wind speed are also under monitoring in all the
three zones with a one minute resolution. Figure 3.4 shows the temperature distribution
28 3. Methodology

Figure 3.3: The outdoor test facilities: Alpine: Zugspitze, Germany; Desert: Negev, Israel;
and Maritime: Gran Canaria, Spain.

in the three locations, the annual averages of UV dose and RH as well as the performance
measurements of (P max ).

C. Data processing of the three location

An irradiance filter between (800-1200 W/ m2 ) is used for all locations and the power is
corrected to standard test conditions (STC) of irradiance. The irradiance bin ensures that
only clear sky conditions were considered in order to have irradiance conditions near to
STC and to model a common situation for all the climates. The temperature filter applied
varied from location to location depending on the most frequent temperature that a module
experiences over its lifetime in each location (see figure 3.4. Forexample a temperature filter
between (300 C-350 C) is used for Gran Canaria, (350 C-450 C) for negev and (50 C-250 C) for
Zugspitze. To make sure that the power degradation observed for outdoor modules are
not due to soiling effects, periodic cleaning of the modules is done. To avoid seasonal
effects in climatic variables evaluation, five years data is averaged to calculate the models
input values (see table 3.4). Indeed, this ensures that, the values used in degradation
prediction correspond to what a module will experience during its lifetime. The mean
value of the module minimum and maximum temperature has been computed considering
upper and lower temperature bins as in figure 3.4 C. This also ensures that extreme values
corresponding to measurement anomalies are not used in degradation rates calculations.

3.1.4 Global climate data processing

The studies over large geographical regions can be made by processing global gridded data
estimated from NWP including satellite or reanalysis models Urraca et al. (2018). Even
3.1 Methodology for the proposed physical model 29

Figure 3.4: (A): Distribution of module temperature in the three climatic zones. (B): Total
UV dose and average annual relative humidity for the three zones measured for 5 years.
(C) : Ten minutes values of module temperature for 5 years. The dotted lines show the
mean maximum and minimum temperatures. (D) : Raw and filtered power measurements
in Gran Canaria, Spain. In cyan is the power raw data (before applying filters), in black is
the data after applying the irradiance and temperature filters and in red is the data after
STC correction.

Table 3.4: Summary of 5 years average climatic inputs used in simulation.

Input parameters
Location RH [%] T m [0 C ] UV [kWh/m2 ] T max [0 C ] T min [0 C ]

Negev 61.0 36.8 87.7 56.7 12.7

Canaria 68.0 30.6 101.0 43.6 19.6
Zugspitze 74.0 18.7 81.0 44.7 -2.30

though satellite-based estimations can be more accurate than the reanalysis-based ones,
the advantage of the second is the possibility to extract all the essential variables together
in the same data-set, without gaps and identical timestamps.
In this research, the ERA5 climate reanalysis data-set (C3S) (2017) was processed to
extract and model the climatic stresses necessary for the degradation rate evaluation. This
dataset provides a spatial resolution of 31 km and temporal resolution of hourly data from
1979. The estimated climate data is compared and validated with real ground measure-
30 3. Methodology

ments taken from the World Radiation Monitoring Centre - Baseline Surface Radiation
Network (WRMC-BSRN) Driemel et al. (2018) (the validation procedure is described in
the annex). Although not all variables required are directly available from the ERA5, we
calculate the missing local climate variables, WS, RH and UV irradiance from existing
models as described below. The derived maps are presented in the annex.

A. Evaluation of relative humidity

The relative humidity is not extracted directly from ERA5, so it is estimated using equa-
tions 3.12 and 3.13. The saturation water vapour pressure (WVP) over water and ice is
calculated using the Buck’s formula Buck (1981) from the dew point temperature (T D ew )
and ambient temperature
½ (T amb ):
³ ´³ ´
T T
WV P (T )[ k pa] = 0.61115.exp 23.036 333.70 . 279.82+T f or T < 0

³ T ´³ T ´
0.61115.exp 18.678 . f or T ≥ 0 (3.12)
234.84 257.14 + T

WV P (T D ew )
RH [%] = .100[%] (3.13)
WV P (T amb )

B. Evaluation of UV
Regarding the UV irradiation, although this variable is included in the ERA5 dataset,
it considers a wavelength-range up to 440 nm. In our case, the effective UV irradiance
covers up to 400 nm, since that is a typical average value where encapsulates change
the transmittance properties Miller et al. (2015). For this reason, we neglect the UV
irradiance given by ERA5, and model it using a formula proposed by Crommelynck and
Joukoff (1990) as expressed in equations 3.14-3.17. It is based on the clearness index ( c i )
and the global horizontal irradiance (GHI). The c i is calculated by dividing the GHI and
the top-of-atmosphere irradiance extracted from ERA5. Unfortunately, the lack of valid
measurements disallows us to do a benchmarking of UV models.

c i = max(0.1, min( c i , 0.7)) (3.14)

UVB = (1.897 − 0.860.c i ).1 e−3 .GH I (3.15)

UVA = (7.210 − 2.365.c i ).1 e−2 .GH I (3.16)

UV = UVA + UVB (3.17)

C. Evaluation of wind speed (WS)

The WS is a parameter of interest in this study because it is used to estimate the PV module
temperature (T m ) due to the related cooling effect of materials. The WS is calculated and
height corrected according to equation 3.18 and equation 3.19 Huld and Amillo (2015),
where û wind and v̂wind are the vector components of the wind, h ER A is the height from
ground which the wind is modelled in the ERA5 dataset, and h ER A is the assumed height
of the PV modules equal to 2 meters. The 2 meters height assumed are in accordance with
3.1 Methodology for the proposed physical model 31

the height of modelled ambient temperature (T amb ) and dew point temperature (T D ew )
given by ECMWF.
q
W S ER A = û2wind + v̂2wind (3.18)

³h ´
mod 0.2
WS = .W S ER A (3.19)
h ER A

D. Modelling module temperature

The module temperature is the most effective parameter in degradation rate evaluation,
since the rate models are built on the assumption that, the degradation follows an Arrhenius
temperature dependence . In most cases, the module temperature is estimated from,
ambient conditions of temperature, irradiance and wind speed using a number of different
models proposed by Kratochvil et al. (2004), Segado et al. (2015), Faiman (2008). In this
research the Faiman model (equation 3.20) is selected for module temperature evaluation
because it has been reported that, it provides a good accuracy under different climate
conditions for crystalline silicon PV modules Faiman (2008); Schwingshackl et al. (2013)
The model estimate the PV module temperature as a function of the GHI, WS and T amb
as:
GH I
T m = T amb + (3.20)
u 0 + u 1 .W S
where u0 is a coefficient describing the effect of the radiation on the module temperature
(W /0 Cm2 ) and u 1 describes the cooling by the wind (W s/0 Cm2 ). In this research, the values
of u0 = 26.9 (W /0 Cm2 ) and u1 = 6.2 (W s/0 Cm3 ) according to Huld and Amillo (2015).
The presented calculations consider an open-rack flat mounting configuration for the PV
modules at all locations. However, the installation over rooftops or any other surface and
the re-positioning can increase the thermal stress due to higher reception of photons from
the sun or more circulation of air.

3.1.5 Statistical error analysis methods

To analyse how well a set of data points fit with a given model, the residual standard
deviation as a measure of a goodness-of-fit is used. Given that,

R esidual = Ym − Y f (3.21)

The residual standard deviation (SD res ) is expressed as:

sP
(Ym − Y f )2
SD res = (3.22)
n−2

where Ym measured value, Y f fitted value and n number of data points.

To check the prediction accuracy and to evaluate the uncertainties of the model, the
percentage mean square error of prediction (MSEP) is used according to McKay et al.
(1999) as:
h ³ ´2 i
MSEP = 100 × V [P p ] + µ p p − µ pm (3.23)

Where V [P P ] is the variance of predicted power, µ p p and µ pm are the mean of predicted
and measured power respectively.
32 3. Methodology

3.2 Methodology for data-driven model

Different from the physical model presented in the previous section, here a data-driven
methodology for long-term degradation forecast is described. In this approach, the lifetime
forecast is based on the existing degradation history observed in the field and not on the
climatic stress factor. The schematic figure 3.5 summarises the methodology.

Figure 3.5: Illustration of the overall objective of the data-driven methodology. The major
steps are; data treatment, identification of degradation pattern and selection of the best
degradation model for long-term forecast

3.2.1 Background
The ever-growing secondary market of PV systems (that is, the transaction of solar plants
ownership) calls for reliable and high quality long-term PV degradation forecasts to mit-
igate the financial risks. Data-driven techniques that utilize monitored operational data
related to system’s performance are used in many fields, for example in the aircraft indus-
try Zhang et al. (2018); Listou Ellefsen et al. (2019) to forecast the future trend or the
remaining useful lifetime. They are normally applied to complex systems where developing
a physical model could be more complex and expensive.

For PV applications, a few authors have proposed data-driven prognosis models to

evaluate the remaining useful lifetime of PV modules. For example, Laayouj et.al Laayouj
et al. (2016) proposed a smart prognostic method for PV module health degradation and
remaining useful lifetime prediction. The model is based on two approaches; the on-line
diagnosis and data-driven prognosis. Also, Sheng et al Sheng et al. (2019) proposed an
autoregressive moving average model-filtered hidden Markov model to predict the residual
life for complex systems with multi-phase degradation. They applied the model to predict
3.2 Methodology for data-driven model 33

the residual life of a specific PV module system. Although both methods are reported
to provide good predictions, the biggest drawback of these methods is that their perfor-
mances are not rigorously validated or analysed with different degradation datasets. Both
methods are calibrated, validated and applied based on a single PV module or system
dataset. Moreover, for the later, the methodology was applied to simulated performance
measurements for its reliability evaluation and residual life prediction. PV modules can
exhibit different degradation scenarios Köntges et al. (2014) especially due to different
technologies, different failure modes as well as different operating local climates. There-
fore, it is very unlikely that a model calibrated and validated on a single dataset can be
generalized to apply to other degradation scenarios. In this regard, we propose a model
based on a rigorous analysis of degradation data of several PV modules as well as systems
of different technologies and installed in different locations. Hence the proposed model is
aimed to be generalized to the different degradation scenarios. Also related work is that
of Rizzo et.al Rizzo et al. (2018) where an algorithm for lifetime extrapolation, prediction,
and estimation was proposed. However, their algorithm is proposed only for emerging PV
technologies and for shorter time forecasts.
Generally, in spite of the recognized potential of empirical data-driven techniques for
time series forecast, limitations still exist for their application in long-term PV degradation
evaluation. Different factors such as outliers in the dataset, seasonal variations and many
other reducing factors (e.g. soiling) should be separated from long-term non reversible
degradation. The lack of a systematic and flexible approach to select parameters of these
techniques and their black-box character limit the understanding and control of their per-
formance. We address this issue by proposing a systematic and flexible approach with
adjustable model parameters to evaluate the degradation trend based on the nature of the
dataset under evaluation. The proposed method aims to evaluate the irreversible long-term
degradation of PV modules as well as systems. To achieve this we propose an iterative
algorithm for degradation trends evaluation that allows to separate seasonal variations and
other reversible performance reducing effects from irreversible degradation.
Another drawback of the available data-driven techniques is their accuracy when long-
term predictions are required after a short time interval and with limited data points.
For example, Taylor and Letham Taylor and Letham (2018) performed a comparison of
the forecasting accuracy of different statistical models at different time horizons. In their
study, most models displayed large uncertainties when applied after shorter time interval.
Indeed, the available techniques are based on fitting the available systems degradation
data using regression models and then applying a simple extrapolation to forecast the
lifetime. However, in practice, the system’s degradation history available may be short and
incomplete, and a simple extrapolation could lead to large uncertainties hence degrading
the reliability of the forecasts. To address this challenge, the proposed model is aimed at
improving the long-term forecasting accuracy using a small degradation history and few
data points. To achieve this, we propose a new concept using time dependent degradation
factor for degradation extrapolation instead of using a simple extrapolation for lifetime
forecast. To further improve the accuracy and to have a generalized model, the concept of
multiple degradation factor models dependent on degradation patterns is proposed.

3.2.2 Modelling assumptions

To begin with, the concept of the proposed data-driven model is based on the work re-
ported by Jordan et al Jordan et al. (2017), that associating specific degradation and
failure modes with specific time series behavior can aid in service life prediction. Based on
34 3. Methodology

this understanding, our model is based on analysing time series degradation patterns to
extracted different parameters (see Figure. 3.6).

Figure 3.6: Illustration of the different degradation pattern parameters extracted during
calibration and the remaining useful lifetime (RUL)

The extracted parameters are then used to formulate degradation rate functions. In
developing the model, we also use the following assumptions:

• Different degradation modes might cause differences in degradation patterns mean-

ing that it is unlikely that a single degradation function can represent all expected
degradation patterns. Therefore the approach depends on multi-degradation rate
functions.

• Some degradation modes might be triggered by other degradation modes and might
appear at certain stages of a module’s lifetime, meaning that using a constant degra-
dation rate extracted at a given stage of PV operation to represent the entire lifetime
could affect the forecast accuracy. Therefore, time dependent degradation rates are
proposed. Moreover, the usage of degradation factor is preferable to degradation rate
since we consider a non-linear degradation with a non-constant degradation rate.

A. "Time and degradation pattern" dependent models

Taking the aforementioned assumptions into consideration, we propose different degrada-

tion factor models dependent on time and degradation pattern. The degradation factor
models are expressed as:
³ ³ 1 ´ ´
y
k 1 = k cal 1 + A 1 .k cal .ln .τ11 .τ2x1 .ρ z1 .t (3.24)
∆P
³ ³1´ ´
k 2 = k cal 1 + A 2 .k cal . .ρ z1 .t (3.25)
τ1
3.2 Methodology for data-driven model 35

³ ³ 1 ´³1´ ´
k 3 = k cal 1 + A 3 .k cal .ln . .τ2x1 .ρ z1 .t (3.26)
∆ P τ1

Table 3.5: Definition of coefficients and parameters

Coefficients and parameters Definitions

t cal Time at calibration threshold

P cal Power at t cal
SD P cal
p
Standard deviation of calibrated data
τ1 = max( t cal ) -
³ ∆ t| = |´ln( t cal − t cal −1 )|
τ2 = | ln -
1
ρ = ln SD P cal -
−3/2
A 1 [ year ], A 2 [ year −1/2 ] and A 3 [ year −1/2 ] Proportionality constants
y1 , z1 and x1 Optimization parameters

It should be noted that, these models are purely empirical without any physical im-
plication, they are derived from degradation pattern parameters shown in figure 3.6 as
they describe well the specific degradation patterns. It is also important to note that the
model parameters are extracted after an initial 3% performance degradation threshold.
The 3% optimization threshold is calculated from the degradation trend that excludes
early degradation effects. The reasoning for using a 3% threshold are further described
and demonstrated in the results section.

B. Power degradation model

The power degradation function proposed in section 3.1, equation (3.10) is used to fit the
degradation data also in this part. However, since the evaluation is independent of material
properties evaluation or analysis, the model parameter Γ is fixed to one. Therefore, the
power degradation function can be re-written as:
P ( t = t) ³ ³ 1 ´µ ´
= 1 − exp − (3.27)
P N orm k cal t
where P N orm is the maximum power of the degradation trend and k cal is the degradation
factor [a−1 ] at calibration. In this case the degradation factor is preferred for use instead
of the degradation rate because of the non-linearity of the power degradation and non-
constant degradation.

3.2.3 Modeling approach

A. Failure time and remaining useful lifetime definitions
Chen Xiongzi et al Chen Xiongzi et al. (2011) defined the RUL of a system or a component
as the length from the current time to the end of its useful life. The question is how to define
the ”useful life”. In this study the useful life is defined as the non-reversible performance
loss such that the module or system power decreases by 20% of the ”maximum stable
power” measured in the field. The notion of a maximum stable power is introduced to
separate long term degradation from early stage degradation events such as light induced
degradation (LID) for p-type crystalline silicon modules Köntges et al. (2014) or light and
36 3. Methodology

elevated temperature induced degradation (LeTID for multicrystalline silicon and modules
with passivated emitter and rear cell (PERC)Kersten et al. (2017); Philipp et al. (2019). It
also helps to separate other reversible effects reducing module performance such as soiling
Zorrilla‐Casanova et al. (2013) and seasonal variations Cañete et al. (2014). Moreover,
due to these effects and variation of the outdoor conditions, the power printed on the PV
module label substantially deviates from the initial PV module power outdoors. From our
point of view, the maximum stable power can be easily compared among systems and is
describing the system performance well which is not necessarily the case for the nominal
power. The nominal power of a PV module/system may deviate substantially from the
real power since it is recorded under STC. These well defined conditions of 1000 W/m2
and 25◦ C are almost never prevalent as shown in the histogram in figure 3.7b.

Figure 3.7: A, Illustration of the stable power (red line) used in this study, the green doted
line is ”maximum stable power” (reference power) and the blue dotted is the failure time
threshold. B, Histogram of the data distribution throughout the 8.5 years

Although, the term failure time has already been defined in subsection 3.1.2 it is re-
defined here accordingly to the context. In this study, we therefore, define the FT as a
reduction in the module or system performance by 20% of the maximum stable power
(indicated by the green dotted line in figure 3.7). The 20% loss is purely arbitrary and it
is used in this case because of its consistence with the warranties given by manufacturers.
It might be interesting to predict the module or system performance for its entire lifetime
(until the module stops working), however, such a prediction could be more unreliable and
unrealistic since certain future events are unpredictable. Moreover, the longer a module
stays in the field the more sensitive it becomes to different degradation modes. That might
lead to a dramatic increase in the degradation rate which cannot be easily modeled. There-
fore, according to FT definition, the RUL is the length between the current time (CT) to
the FT and can be expressed as:

RU L = FT − CT (3.28)

1. Data filtering
The first filter applied to the input dataset is an irradiance filter. The upper limit for
irradiance was set and fixed at 1200 W/ m2 and the lower limit was chosen flexible between
600-900 W/ m2 depending on the location under evaluation. This clear sky filter is used as
a way to separate low performances due to low irradiance from degradation. The filtered
3.2 Methodology for data-driven model 37

power is then corrected to STC of irradiance by applying a linear correction of power to

1000 W/ m2 irradiance. The second step is to remove outliers. Because of the stochastic
behavior of outdoor conditions and anomalies in the measuring equipment, data measured
from outdoor PV modules as well as systems is usually subject to outliers. When not
treated well, Outliers might lead to large uncertainties, especially in degradation trend
evaluation. The outliers are removed by computing the standard deviation (SD) around
the mean value of the entire dataset, whereby all points that are out of the range of
( M ean ± 2 × SD ) are eliminated. For LID or LeTID sensitive systems, it is a crucial step
to analyse the degradation pattern during the first year of operation. These degradation
modes are characterized by strong non-linear trends in the initial phase which might either
stabilize gradually or undergo a regeneration phase Kersten et al. (2017); Philipp et al.
(2019). If the regeneration phase is detected, we recommend to eliminate the data until
the onset of this phase.

2. Data decomposition
Time series data is characterized by four major components; level, trend, seasonality, and
noise. There are two models that are used to model the effects of these components; additive
(linear) and multiplicative (non-linear) Brownlee (2017). In our case a multiplicative model
is used. The model suggests that the components are multiplied together as follows:

y( t) = Level × T rend × Seasonal it y × N oise (3.29)

This function is implemented and available for free use in the python library. The function
uses a moving average (MA) method to extract a degradation trend in the time series
data. The averaging depends on the required resolution (weekly, monthly or yearly) of
one’s interest and it determines the extracted degradation trend in this case.
In outdoor conditions seasonal and other different effects can reduce the performance
of a PV module or system. For non-reversible performance degradation evaluation it is
crucial to choose a good averaging temporal interval that eliminates seasonality and other
reversible effects. It has been reported that reversible effects such as inverter failures and
soiling of PV modules have a higher impact on performance loss rates than the actual per-
formance degradation which is non-reversible Klünter et al. (2019). These effects could be
module technology dependent, for example Virtuani et al. Virtuani et al. (2015) reported
that amorphous silicon (a-Si) systems are more affected by variations in the incident spec-
trum then crystalline silicon (c-Si) systems because of there narrower spectral sensitivity.
Therefore, it is important to separate these performance reducing effects from long term
degradation by choosing a good averaging temporal interval. The selection of a good tem-
poral interval could be very tedious if one has to do it manually. An iterative algorithm
has been implemented to assist in determining a proper interval for degradation trend
evaluation depending on the datasets being evaluated.

3. Iterative algorithm for degradation trend evaluation

One way to completely remove any variation in the degradation trend will be to apply a
strict constraint, which removes the difference between each n th and (n + 1) th value of the
trend that is greater or equal to zero, that is; trend [n − (n + 1)] ≥ 0. However, according
to the stochastic behavior of outdoor conditions, outdoor measurements usually contain
unavoidable outliers which can appear even after applying certain filters. This implies that
applying such a strict condition to outdoor datasets is quite unrealistic; in most cases this
38 3. Methodology

condition is not fulfilled or might require a considerable amount of computation time to

converge depending on the nature of the dataset. Therefore , a tolerance (tol) is introduced
and the condition is applied as:
h¯ max( dn) − min( dn) ¯i
¯ ¯
trend ¯ ¯ ≤ toll, f or : dn = n − ( n + 1) (3.30)
min( dn)
The process begins by initialising the temporal interval by: initial temporal interval
= 2% of the total length of the time series. Afterwards, the iterative loop is repeated
until the condition in 3.30 is fulfilled. The tolerance depends on the nature of the dataset
under investigation, for example it can be correlated with the resolution of the data or the
magnitude of outliers in the datasets. This makes the process of temporal interval selection
a quasi-automated process. By analyzing a number of datasets, a range of tol values can be
set granting the flexibility for application on broad datasets of the different distributions.

3.2.4 Experimental
Different data-sets for PV modules as well as systems that have been exposed for quiet a
long period of time with considerable degradation were used in this study. The data-sets
are from three different sources:

1. The first set of data are those of the TIcino SOlare (TISO)-10-kW PV plant in Lugano
(Switzerland) Virtuani et al. (2019); Annigoni et al. (2019). The TISO‐10 PV system
has been connected to the grid since 1982, and is the oldest installation of this kind in
Europe (We shall refer to them as ”TISO-Modules” here after). The performance (i.e.
current‐voltage curves) of 18 selected reference modules were measured at regular
intervals between 1982 and 2017. After 35 years in the field, these modules show a
degradation rate ranging from -0.2 to -0.7%/year considering a ± 3% measurement
uncertainty. In this study 10 of the 18 modules with a considerable degradation have
been used in module calibration and validation stages.

2. Another source of data is a PV plant installed at the airport of Bolzano/Italy (posi-

tion ca. 46.46 N, 11.33 E, elevation: 262 m) including 11 experimental PV systems,
which are in operation for 8.5 years and covering the time period from 01 February
2011 until 31 July 2019. They are referred to as ”Bolzano systems”. Most ma-
jor PV system technologies are included, namely one and three junction amorphous
silicon (a-Si), cadmium telluride (CdTe), CIGS, silicon solar cells made out of a
heterojunction with an intrinsic thin layer (HIT), mono-crystalline silicon (mc-Si),
poly-crystalline silicon (pc-Si) and poly-crystalline silicon string ribbon (ribbon). All
systems are part of one experimental PV plant, they are ground mounted with an
orientation of 8.5O west of south and a fixed tilt of 30O . The installed nominal power
for the systems range from 1 to 4 kWp per individual installation. According to a new
PV sensitive climate classification, proposed by Ascencio-Vázquez et al. Ascencio-
Vásquez et al. (2019), the climate in Bolzano is categorized as a temperature climate
with medium irradiation. The irradiance is measured with a Kipp & Zonen CMP11
pyranometer. Calibrations are performed in regular intervals and the measurement
uncertainty is between 2% and 4%. Additionally, climate data are taken from a
ground based meteo-station installed in close proximity to the test side.

3. Finally, data from the Desert Knowledge Australia Solar Centre (DKA Solar Cen-
tre) DKASC (2019), named hereafter as ”DKA systems”, are used. The data can
3.2 Methodology for data-driven model 39

be downloaded through 1 . The data sets used in this study are given as monthly
yield in kWh of three different systems; Kyocera-5.4kW-Poly-Si Dual (2008), eco-
Kinetics-26.5kW-mono-Si-Dual (2010) and trina-23.4kw-mono-dual (2009). For the
first system (Kyocera), data from 01 January 2009 to 01 December 2018 was used
in the analysis, for the eco-Kinetics system from 01 January 2011 to 01 December
2018 and for the Trina system, data from 01 January 2014 to 01 June 2019. More
information about these systems and datasets are available on the DKA Solar Cen-
tre website. It should be noted that the data plotted here have been normalized to
the maximum power and were subject to the outlier filters described in the previous
section.

3.2.5 Statistical errors analysis

The error measurement employed for the performance evaluation of the proposed forecast-
ing method is the root mean square error (RMSE). Given a measured (m) and a predicted
(p) value, respectively, for a given quantity and number of observations (n), the RMSE is
expressed as. s Pn
j =1 ( p i − m i )
2
RMSE = (3.31)
n
The metric deployed to compare the performance of the proposed method with other
methods is the relative difference. This is a relative comparison of the failure time fore-
casted after a 3% interval and on the full dataset. It is expressed as,
|FT3% − FT f ull |
R el di f f = (3.32)
max(|FT3% | or |FT f ull |)

where, R el di f f the relative difference is, FT3% and FT f ull are the failure time evaluated
after a 3% degradation and using the entire dataset respectively.

1
DKA data download: http://dkasolarcentre.com.au/locations/alice-springs/graphs?
sources=91
40 3. Methodology

3.3 Methodology of a hybrid model

The described physical models is very useful when one need to correlate the predicted
lifetime to the influencing climatic factors as well as dominating degradation mechanisms.
However, one draw back is on the accuracy when it comes to long-term degradation pre-
diction. And also, the described data-driven model could help to improve the long-term
prediction accuracy, however, it provides no information about the influencing degrada-
tion mechanisms. A model that can provide more accurate long-term predictions as well
as information of the dominating degradation mechanism is of great importance for finan-
cial evaluation as well as material development. In this regard, the two approaches are
combined into a hybrid model to achieve these two aspects. The combination of the two
approaches is illustrated in the schematic diagram (figure 3.8)

Figure 3.8: Flow chart illustrating the link between the physical and data-driven approach.
The red dashed square is the data-driven part, we implement an algorithm to identify the
degradation patterns and to select the appropriate rate function

Figure 3.8 illustrate the link between the physical and the data-driven approaches. The
fundamental assumption in this link is that, the module experience the same degradation
pattern in other geographical locations. This means that the degradation pattern param-
eters extracted in the training location can be applied in other locations to select the best
degradation function for long-term predictions. Therefore, the differences are within the
calculated degradation rate at calibration ( k cal ) which is dependent on the climatic stresses
of a given location.
Chapter 4

Results and Discussion

Table 4.1: Chapter 4 Nomenclatures

Symbols
ci clearness index [-]
Ea activation energy [eV]
k degradation factor [1/year]
FT3% Failure time calculated using -3% degradation [year]
FT f ull Failure time calculated complete dataset [year]
kB Boltzmann constant (8.62 × 10−5 eV/K)
P max power at maximum power point [W]
RD degradation rate [%/hour]
R el di f f relative difference
SD res residual standard deviation
Tm module temperature [K]
T mm measured module temperature [0 C ]
T mp modeled module temperature [0 C ]
(µ ) shape parameter [-]
Subscripts
cal calibration
h hydrolysis
p photo-degradation
t thermo-mechanical
T Total
Abbreviations
AH tropical with very high irradiance
AK tropical with high irradiance
BH desert with high irradiance
BK desert with very high irradiace
KGPV Köppen-Geiger-Photovoltaic
FP Facebook prophet
MSEP mean square error of prediction
PR performance ratio
RUL remaining usefull lifetime
STC standard test conditions
42 4. Results and Discussion

In this chapter, the results of the physical model, data-driven model as well as for the
hybrid model are presented and discussed respectively.

4.1 Results for the physical model

4.1.1 Properties of the proposed power degradation function

As already mentioned that, non-linearity of power loss is often observed in the field depend-
ing on the module technologies or degradation modes. And also that, the loss in power can
take different shapes, for example: exponential-shaped, linear-shaped, steps degradation
and saturating power loss over time. Moreover, since energy yield could highly depend on
how the performance degradation evolves over time (see figure 4.1b), it is very crucial to
choose a degradation model to take into account the different failure patterns. A simplified
non-linear power degradation model proposed in this research aims to take into considera-
tion these non-linearity dependencies of power degradation. To do this, a shape parameter
µ has been introduced to model all the degradation shapes as shown in figure 4.1a.

Figure 4.1: A) Optimization of power degradation shapes by altering the shape parameter
µ and B) Relative energy yield corresponding to the degradation shapes

Figure 4.1a shows the simulated degradation shapes 1-7 corresponding to different val-
ues of µ = 0.1, 0.2, 0.3, 0.4, 0.5, 0.8 & 1.0 respectively. On the one hand, when µ = 0.1
and µ = 0.2, one can achieve the commonly observed degradation shape in thin film tech-
nologies with a faster degradation at early stages of exposure and followed by stabilization
Jordan et al. (2017b). On the other hand, one can expect the degradation shape similar
to that when µ= 1.0 for glass-glass modules if we speculate that for this module design
there are less moisture pathways and moisture ingress compared to glass-backsheet mod-
ules. Therefore, the moisture induced degradation modes are slower at the earlier stages
of the module lifetime. However, as the breathable pathways and drying are also limited,
the moisture will accumulate over the years, leading to a dramatic increase in degradation
rate.
4.1 Results for the physical model 43

4.1.2 Calibration and validation of the specific degradation rate

models
Two basic approaches are applied for the calibration of the different degradation rate
models: (a) optimization of model performances and (b) through prior knowledge from
previous studies. Optimization of model performance, which compares measured and sim-
ulated data, is applied by the help of a built-in nonlinear least-squares solver in the GNU
Octave software. Prior knowledge, with the aid of sensitivity analysis, is used as a baseline
to select the initial fitting guesses and also as a confirmation that the extracted values are
in a meaningful range.
To derive the model parameters, equations 3.2, 3.5 and 3.6 are fitted on experimental
data from the indoor accelerated ageing tests at given specific test conditions as presented
in chapter 3 table 3.2. The extracted parameters are presented in table 4.2 together with
the percentage residual standard deviation (SD res ) (see equation 3.22). To further ensure
a better calibration procedure, in each case the distribution of the residuals is plotted and
analysed (to check if the residuals follow a normal distribution). The parameter Γ of the
power function in equation (3.10) has been normalized to one and the extracted shape
parameter (µ) is 0.7 for hydrolysis, 0.4 for photo- and thermo-mechanical degradation in
figure 4.2.
To validate the models, different sets of data measured at different test conditions (see
table 4.2 ) are used and by using the extracted parameters the simulations are compared
with the measured data points through evaluation of the mean square error of prediction
(MSEP, see equation 3.23) as also presented in table 4.2. The MSEP is also used as a metric
to correlate the uncertainties in model calibration with the predictions when compared
with the SD res . For example, a 0.5% SD res led to 0.025% MSEP for the hydrolyisis
model, a 1.65% SD res resulted into 0.216% MSEP for the thermo-mechanical model and
0.19% SD res led to 0.168% MSEP for the photo-degradation model. Although, there is a
correlation of the uncertainties due to model calibration as its evaluated in the thermo-
mechanical model, the uncertainties in experimental data sets used for validation can also
influence the evaluated mean square error therefore it’s also useful to plot and interpret the
results basing on the degradation trend. Evaluating a confidence interval of the prediction
can also help to analyze the model performance.

Table 4.2: Extracted parameters for the three rate models and the percentage residual
standard deviation (SDres) as well as the mean square error of prediction (MSEP)
Extracted models parameters
Degradation rate model Ai E a [eV] n X &Θ SD res MSEP

Hydrolysis: Eq(3.2) 6.11e4 0.91 1.90 - 0.5% 0.025%

Photodegradation: Eq(3.5) 1.20e-3 0.43 1.80 0.63 0.2% 0.168%
Thermo-mechanical: Eq(3.6) 9.10e-5 0.40 - 2.24 1.6% 0.216%

Figure 4.2 shows calibration and validation results for the hydrolysis, photo-degradation
and thermo-mechanical models (equations 3.2, 3.5 and 3.6) respectively. In black is the
measured power for DH 750 C/85RH; red is the respective model fit with a violet line at
5000 hours representing the optimized data points. The blue line is the predicted power
at DH 850 C/85RH, the light blue patch is the 95% prediction confidence interval and in
green the measured power at DH 850 C/85RH, used for model validation. The vertical
lines on measured data points indicate a 2.5% measurement uncertainty. The color usage
44 4. Results and Discussion

Figure 4.2: A: Hydrolysis model Eq(3.2) calibration and validation, the vertical violet line
represents the optimized data points; B: Residual distribution of the fitted data points
of the DH test; C: Photo-degradation model Eq(3.5) calibration and validation, and D:
Thermo-mechanical model Eq(3.6) calibration and validation. The blue patch is the 95%
confidence interval of prediction.

and explanation above are consistent for photo-degradation and thermo-mechanical graphs
using respective data sets. For all the models, the predictions are satisfactory and are within
a 95% confidence interval. The observable variations could be linked with the measurement
uncertainties.
It important to note that, when applying the physical models, the extracted parameters
are usually valid only for a particular module type. Meaning that the parameters have to
be evaluated when the module type or technology changes.

4.1.3 Calibration and validation of the combined degradation rate

model
The combined/total degradation rate (equation 3.8) is calibrated using outdoor measure-
ments of Gran Canaria. The dataset of Gran Canaria is selected over Negev and Zugspitze
because the module in Gran Canaria has been exposed longer compared to the ones of
Negev and Zugspitze, and moreover it shows a clear degradation trend. The extracted
model parameters are presented in table 4.3. On calibration, the residual standard devia-
tion is 2.34% and the derived parameters, Γ = 190 and µ = 0.19 for the power degradation
4.1 Results for the physical model 45

function in equation 3.10.

Table 4.3: Extracted model parameters of the individual rate models using the combined
rate model 3.8 and with Gran Canaria data
Extracted models parameters
Degradation rate model Ai E a [eV] n X &Θ

Hydrolysis: Eq(3.2) 4.91e7 0.74 1.90 -

Photodegradation: Eq(3.5) 71.83 0.45 1.90 0.63
Thermo-mechanical: Eq(3.6) 2.04 0.43 - 2.24

The model is validated by using the derived model parameters to simulate the degrada-
tion rates using the climatic datasets of Negev and Zugspitze. By comparing the simulated
degradation trend with the measured data (see figure 4.3), the accuracy of the model is
verified.

Figure 4.3: Right: Dotted black points are the measured data of Gran Canaria (calibrated
dataset); thick red line is the model fit; in green is the measured data of Negev, thick
blue line is the model prediction for Negev and the dashed lines indicate normalization to
the initial laboratory power values before outdoor exposure respectively. Left: Residual
distribution of the calibrated data points.

Figure 4.3 on the right shows the calibrated data of Gran Canaria and the predicted
degradation for Negev plotted together with the measured degradation. The alpine predic-
tions were left out to avoid too much information on the graph due to data fluctuations.
The outdoor predictions show a good agreement with the measured power degradation.

4.1.4 Degradation rates and lifetime prediction

Depending on the climate a module is installed in, different degradation modes might dom-
inate over the others. Using the proposed degradation models, equations (3.2), (3.5), (3.6),
(3.8) and the outdoor derived model parameters, it is possible to predict the dominating
degradation precursor and the total degradation rate as well as the failure time from equa-
tion (3.11) for any location with known climatic loads. In this section, annual degradation
rates of the mc-Si modules were predicted using input climatic loads of Zugspitze, Gran
Canaria and Negev as shown in table 4.4.
46 4. Results and Discussion

High thermo-mechanical degradation is predicted for Zugspitze in comparison to Gran

Canaria because of high module temperature variations in this climate zone. On the
contrary, small degradation due to hydrolysis is predicted in Zugspitze despite the high
levels of relative humidity. This could be explained by the low average module temperatures
experienced in this region, hence slowing hydrolysis processes and the absolute water vapor
concentration. In all cases, high rates are predicted in Negev. This can be explained
again by the higher temperatures in this zone that determines the reaction rates for other
degradation processes caused by other degradation factors such as hydrolysis by humidity
and photo-degradation by UV dose. The predicted failure time defined as a 20% loss in
power, show more severe degradation of maximum power output in arid climates where
temperatures are higher such as expected. This further confirms the previous studies
Annigoni et al. (2017) that temperature could be the primary accelerator of degradation.

Table 4.4: Predicted degradation rates of the mc-Si modules and failure time in the three
climatic zones with a 95% lower and upper confindence interval (CI).
Predicted rates and FT
Lower 95% Upper 95%
R Dh RD p R Dt R DT FT
Location CI CI
(%/year) (%/year) (%/year) (%/year) (years)
(years) (years)

Negev 0.169 0.216 0.225 0.74 21.4 16.7 27.1

Gran-Canaria 0.122 0.212 0.104 0.50 31.6 25.0 40.0
Zugspitze 0.043 0.103 0.129 0.3 52.8 42.7 65.0

4.1.5 Comparison of the proposed physical model with other phys-

ical model
We compare the predictions of the total degradation rate and failure time using the pro-
posed physical mode and the model proposed by Bala Subramaniyan et al. (2018) in equa-
tion (2.39). For meaningful comparison and to avoid any bias, both models are calibrated
using the same dataset (from Gran Canaria). Moreover, the model parameters are also
derived to be in range with those reported in their paper Bala Subramaniyan et al. (2018).
The simulated degradation rates and failure times are as presented in table 4.5. Note that
the values of Gran Canaria are perfectly the same because they are the reference calibrated
dataset for both models. The comparison is then made on the predictions of the other two
locations that is; of Negev and Zugspitze.
The evaluated rates are quite comparable for Negev but show very high difference for
Zugspitze. To further analyse the cause of this high discrepancy in the prediction, the
sensitivity analysis of the input variables is carried out. Temperature (T), relative humid-
ity (RH) and UV dose bins between (150 C -450 C ), (40%-100%) and (80-120kWh/a/m2 )
respectively are used to generate 500 combinations of T, RH and UV. The sensitivity of
the degradation rate with the climatic variables is as shown in figure 4.4.
As shown in figure 4.4, the model according to Bala Subramaniyan et al. (2018) has a
higher sensitivity to the input variable especially to the module temperature. At lower or
higher values of temperatures, the model overestimates or underestimates the degradation
rate respectively and this explains the observed variations in the predictions especially in
Zugsptize where the module temperatures are considerably low. The reason to this higher
4.1 Results for the physical model 47

Table 4.5: Comparison of the proposed model with the model according to Bala Subra-
maniyan et al. (2018)
Predicted rates and FT
R DT [% years] FT [% years] R DT [% years] FT [% years]
Location (Proposed) according to (Subramaniyan) according to
Eq(3.8) Eq(3.11 Eq(3.8) Eq(3.11)

Gran Canaria 0.50 31.6 0.50 31.6

Negev 0.74 21.4 0.80 20.0
Zugspitze 0.30 52.8 0.14 113.5

Figure 4.4: Sensitivity analysis of the proposed and the model according to Subramaniyan.
Note, each dot represents a value of relative humidity and increases to the bottom.

sensitivity on temperature could be attributed to the model assumption that, all the degra-
dation processes triggered by the combination of the four stresses have the same activation
energy. From a physical point of view, this is usually not the case as one can expect
different processes to have different kinetics and hence different activation energies. Also,
assuming a single activation energy could be too much simplification hence deteriorating
the model performance.

4.1.6 Global degradation rates mapping: A global PV degradation

risk analysis
The evaluation of the degradation rates dependent on specific degradation processes and
the total degradation rate is extend from the three available location to a global scale.
The aim is to create a global risk map of the three proposed mechanism and also the total
degradation rate. The climate datasets used are extracted, modelled and averaged from
the ERA5 reanalysis dataset for the years 2016, 2017 and 2018.
Figures 4.5, 4.6 4.7 and 4.8 show the worldwide degradation rates for hydrolysis, photo-
degradation, thermo-mechanical and total degradation respectively. The generated maps
are in accordance with the parameters from the studied mono-crystalline modules.
According to the recently proposed Köppen-Geiger-Photovoltaic (KGPV) climate scheme
Ascencio-Vásquez et al. (2019), hydrolysis-degradation presents the smallest contribution
in almost all the KGPV zones, but is considerable high for the tropical climates AH (trop-
48 4. Results and Discussion

Figure 4.5: Global mapping of hydrolysis degradation rate

Figure 4.6: Global mapping of photo-degradation rate

Figure 4.7: Global mapping of thermo-mechanical degradation rate

4.1 Results for the physical model 49

Figure 4.8: Global mapping of total degradation rate

Figure 4.9: European categorization of total degradation rate based on temperature, hu-
midity and UV irradiance for a mono-crystalline silicon PV module.

ical with very high irradiance) and AK (tropical with high irradiance), which zones are
related with high precipitation levels (humid areas) and temperature levels. This process
can provoke the moisture ingress leading to delamination of polymers or corrosion of solder
bonds Koehl et al. (2018).
Photo-degradation has the second-highest contribution to the total degradation rate.
This indicator combines the humidity, temperature and UV irradiance impacting the PV
module. The impact is similar to the hydrolysis-degradation but higher in terms of absolute
values due to the process triggered by UV irradiation. For desert areas, even though the
UV irradiation is high, the low humidity in the air decreases the damage of the PV cells
due to this mechanism Ngo et al. (2016). The photo-degradation is considerable high in
AH (tropical with very high irradiance) and AK (tropical with high irradiance) zones due
to the high climatic stresses of all variables (temperature, humidity and UV irradiation).
It is surprising that the highest degradation due to photo-degradation was not predicted
in the desert with high irradiance (BH) and desert with very high irradiace (BK) climates,
50 4. Results and Discussion

for example in the northern part of Africa where the UV is expected to be higher compared
to the tropical regions of central Africa. The reason to this can be attributed to the low
values of relative humidity in the dessert areas hence lowering the impact of the degradation
modes influenced by the combined UV-RH stresses.
Thermo-mechanical degradation exhibits the highest contribution to the total degra-
dation rate in almost all zones. This parameter is affected by the seasonal temperature
cycling (the difference between the maximal and minimal temperature of the year) and
also the annual average maximum temperature.

The total degradation rates calculated by the combination of the previous three degra-
dation mechanisms is also mapped. In accordance with the literature Omazic et al. (2019),
the highest degradation rate is identified in tropical areas (hot and humid). Interestingly,
the AK presents lower degradation than the AH, due to lower photo-degradation contribu-
tion (related to lower humidity). Globally the highest degradation rates (above 1.4 %/year)
are identified in locations next to the equator line. To facilitate the visualization and use
of degradation maps, we categorize the locations into bins of 0.2%/year ranging from 0 to
0.8%/year for Europe The categorized map is shown in figure 4.9 The total degradation
rates could reach 0.8%/year in the hottest areas of the south of Spain and Portugal for
Europe. In real operating conditions, external degradation factors, such as soiling might
increase the degradation rate if taken into account, but the degradation presented here
assumes only gradual and non-reversible degradation processes.

4.1.7 Uncertainties evaluation of the physical model

In order to be able to develop reliable predictive models, it makes it crucial to evaluate the
different sources of uncertainties that can deteriorate the model performance. Therefore,
in this section, the different sources of uncertainties are analyzed and discussed.

A. Uncertainties related to data quality

The quality of input data can highly affect the accuracy of the predictions. Here, the
impact of data outliers is analysed. The module power measurements were done every five
minutes, such high-resolution measurements and the frequently fluctuating environmental
conditions outdoors, lead to unavoidable noise even after applying filters and corrections.
By using an hourly moving average to minimize the noise in the datasets, the process of
model calibration and validation is repeated. The effect of the noise to the derived model
parameters as well as on failure time estimation is evaluated. Because the parameters: E a ,
n, X and of the degradation rates models are very sensitivity, they are assumed constant
in this analysis. Therefore, we evaluate the noise effect using the parameter Γ of the
power degradation function (3.10). In order to illustrate this effect, the simulations of the
failure time for the three location have been repeated using the hourly time-resolution and
compared with the five minutes resolution. The MSEP and relative difference in failure
time estimation with a five minutes and hourly resolution data are presented in table 4.10.
Figure 4.10 shows a plot of the simulated and measured data. Note, the dataset of Gran-
Canaria is still used as training or calibration dataset. The residual standard deviation
reduced to 2.04% and the derived model parameter reduced to Γ = 182.3.
Although the residual standard deviation improved from 2.23% for 5 minutes resolution
to 2.04% for hourly resolution, the improvement did not considerably led to reduction in
MSEP. However, it led to a 4.05% relative difference of the predicted failure time. This is
4.1 Results for the physical model 51

Table 4.6: Percentage mean error in prediction (MSEP) and relative difference of the
estimated FT for different data resolutions.
Uncertainties evaluation
Data Parameter FT Relative
Location MSEP [%]
Resolution Γ [ year ] difference

Negev 5 min 190.0 0.0230 21.4 4.05%

Hourly 182.3 0.0240 20.5
Zugspitze 5 min 190.0 0.0152 52.8 4.05%
Hourly 182.3 0.0150 50.7

Figure 4.10: Model calibration and validation with hourly resolution

consistent with the observations from indoor results that, the uncertainties in experimental
data sets used for validation can also influence the evaluated MSEP.

B. Uncertainties related to climatic variables evaluation

In most cases, the essential input climate variables (temperature, humidity and UV
irradiation) are evaluated from national or private weather services such as EAR5 used in
this study to evaluate the global degradation map. The key challenges with these databases
is that they are not developed for PV module or systems evaluation. In most cases the
required variables such as relative humidity and UV irradiation are not directly measured.
Therefore, they have to be evaluated from other variables such as precipitation and global
irradiation using specific mathematical models. Indeed, these models are associated with
their own uncertainties since they are difficult to calibrate and validate. Indeed, using
such calculated parameters poses a question on the absolute values accuracy of the predict
52 4. Results and Discussion

degradation rates or failure time. Moreover, parameters such as ambient temperatures

which is widely measured and available are not directly used as model inputs since they have
to be converted to module temperature for degradation evaluation. This also influence the
evaluated degradation rates and failure time. Since temperature is considered as the main
factor that influences the degradation mechanisms, here we demonstrate the uncertainties
involved in module temperature evaluation by comparing the predictions of the modeled
and measured module temperatures in the three locations.
As can be seen in table 4.7, the relative difference in measured (T mm ) and modeled
(T mp ) module temperatures can be as high as 47%. Also depending on the location where
the module is installed; this difference can lead to as much as 59.3% relative difference
between the degradation rate evaluated with the measured module temperatures (R DT mm )
and the degradation rate evaluated using modeled temperatures (R DT mp ). One important
observation is that, the uncertainties in rates or lifetime prediction due to the uncertain-
ties in module temperatures evaluation, are location dependent. For example, the modeled
module temperature showed high uncertainties for Zugspitze as compared to Gran Ca-
naria, on contrary high uncertainties in degradation rates are evaluated for Gran Canaria
compared to Zugspitze. This can be explained by the Arrhenius temperature dependence
in the degradation rate models. The effect can be clearly explained by the temperature
against the degradation rate plot Figure 4.11.

Table 4.7: Comparison of predicted rates and FT using measured and modeled module
temperature.
Predicted rates and FT
T mm T mp relative R DT mm R DT mp TFT mm TFT mp relative
Location
[0 C] [0 C] difference [0 C] [ 0 C] [ years] [ years] difference

Negev 36.63 25.35 44.47% 0.75 0.471 33.60 21.09 59.27%

Gran-Canaria 30.12 23.81 26.51% 0.48 0.345 45.80 33.0 38.79%
Zugspitze 4.98 3.65 36.38% 0.23 0.22 71.41 69.22 2.74%

At lower temperature, large variation of temperatures leads to small change in the

degradation rate and at high temperatures a small variation in temperature leads to a
relatively large change in the degradation rate. Hence, the uncertainties in module tem-
peratures evaluations have a more drastic effect on the accuracy of the predictions in
locations with high module operating temperatures.
4.2 Results for the data-driven model 53

Figure 4.11: Module temperatures against the degradation rate

4.2 Results for the data-driven model

4.2.1 "Time and degradation pattern" dependent models as well as
3% degradation threshold
To validate the assumptions used in this approach, here the pitfalls of using a constant
degradation rate for long-term degradation forecast are illustrated. Also, the dependence
of the model on time series degradation patterns is illustrated.
The power degradation function (equation 3.27) is calibrated at different performance
degradation intervals (at P = -1%, P = -2% and P = -3%). At each interval, a simple
extrapolation with a constant degradation rate at calibration (k cal ) is performed and failure
time evaluated. To analyse the dependence on degradation patterns, two of the TISO-
modules named TEA5 and TEA6 are used as shown in figure 4.12.
It is clearly visible that although the model fits the data very well at all calibration
intervals, the evaluated failure time is very different at each calibration interval. This effect
is observed for both modules TEA5 and TEA6. This result demonstrates three aspects:
one, a perfectly fitting model does not guarantee better forecast or prediction. Two, a
model that assumes a constant degradation factor/rate is not appropriate for long-term
PV performance degradation forecast. Three, the model forecast accuracy might highly
depend on a specific degradation pattern. This effect is visible in the two modules, for
example high discrepancies are visible for the TEA6 module compared to TEA5 module
even at similar calibration intervals. Therefore, based on these three observations, the
concept of ”time and degradation pattern” dependent degradation factor was introduced.
It is also visible that the forecasting accuracy improves as the calibration interval or per-
formance degradation increases, which seems like an obvious observation. However, since
the objective is to perform the prediction at the early stages of the module’s operation, a 3%
54 4. Results and Discussion

Figure 4.12: Fit and RUL prediction at different percentage power loss of two TISO mod-
ules, namely TEA5 and TEA6. The dotted lines represent forecasts at the three calibration
intervals

threshold was selected. This threshold was found optimal since it provided good forecasts
compared to values less than 3% loss. Moreover, another reason is that in situations where
the degradation trend is not monotonically decreasing as in many outdoor datasets, using
performance degradation of less than 3% might lead to misinterpretation of degradation
with seasonal variation or other performance reducing factors which are reversible. For
example, the TEA5-module in figure 4.12 shows that prediction after only 1% performance
degradation resulted in under-estimations compared to the measured degradation trend,
which is attributed to the variations in the degradation trend due to reversible effects.

4.2.2 Model calibration

The power degradation model 3.27 was fitted on measured datasets to extract the model
parameters ( k cal and µ). At the same time the degradation pattern parameters ∆P , ∆ t
and t cal are evaluated. During the calibration process, a total of 7 TISO-modules ( named
TEA1, TEA4, TEA6, TEA8, TEA10, TEB1, TEB2) and 2 DKA systems where optimized.
From these modules and systems, different failure patterns where extracted by altering the
tolerance (which changes the degradation trend). Despite the strong degradation and huge
variations in the data points observed for the DKA systems, they are selected for the
calibration process because the objective of proposed methodology is to be applicable on
a wide set of data distribution.
Figure 4.13 shows an example of one of the TISO-modules (TEA1) with annual mea-
surements for 35 years and one of the DKA systems (Kyocera system) with monthly average
values for 9 years, used for model calibration. The extracted parameters are used as the
basis to set logical conditions to select which degradation factor model represents better
a given degradation pattern. Over 10 logical conditions where achieved to represent the
different degradation patterns as presented in table 4.8.
Figure 4.14a shows the 7 TISO-modules used in this study for model calibration. Figure
4.14b is the corresponding change of the degradation factor over time of the 7 modules. It is
clearly visible that the rate of degradation factor variation is dependent on the degradation
patterns. This is evident for TEA1 module that displays a strong degradation but with a
low rate of change of the degradation factor over time. Furthermore, for some degradation
patterns like the one of TEA10, a constant degradation factor ( k = k cal ) was sufficient to
evaluate the long term degradation. Hence, for such a degradation pattern, it is enough to
4.2 Results for the data-driven model 55

Figure 4.13: TEA1 module (from the TISO-Modules) and Kyocera system (from DKA
systems) selected for calibration. In yellow are the measurements and red is the extracted
degradation trend and in thick blue line is the model prediction with the 95% confidence
interval. The horizontal green dotted line shows the -20% failure threshold of the degra-
dation trend

Table 4.8: Logical conditions for model selection and the corresponding extracted opti-
mization model parameters
Parameter Parameter Parameter
Logical conditions Model
y1 x1 z1

µ ≤ 0.45 & k cal ≤ 9.0 e − 4 k1 2.0 1.0 1.0

µ ≥ 0.45 & k cal ≥ 0.01 k1 1.0 0.1 1.0
µ ≤ 0.3 & ∆P ≥ 0.01 k1 1.0 0.1 1.0
µ ≥ 0.45 & ∆P ≥ 0.01 k1 1.0 1.5 1.0
µ ≥ 0.55 & k cal ≤ 0.005 k1 - 4.0 1.0
µ ≤ 0.4 & ∆P ≤ 0.01 & k cal ≤ 0.005 k1 - 4.5 1.0
µ ≤ 0.35 & ∆P ≥ 1.0 e − 4 & t cal ≤ 5 k2 - - 3.5
µ ≤ 0.45 & ∆P ≤ 1.0 e − 4 & k cal ≤ 0.009 k3 - 0.1 1.0
k cal ≥ 0.01 & ∆P ≤ 5.0 e − 5 or µ ≥ 0.45 k = k cal - - -
k cal ≤ 0.013 & ∆P ≤ 9.5 e − 6 & µ ≥ 0.43 k3 - 0.1 0.5
else: k1 0.0 2.5 1.0

forecast the lifetime by using only the simplified power degradation model e.g. equation
3.10.

4.2.3 Validation

The most important and challenging part of all predictive models is the validation process
of the models which usually requires different sets of measured data from the ones used
in the calibration process. In this study, PV modules and systems with a considerable
performance degradation were used to valid the performance of the model. An example is
shown in figure 4.15 for Trina and CIGS systems. By comparing the measured trend line
and the prediction line, the accuracy of the model was verified.
56 4. Results and Discussion

Figure 4.14: A, Different TISO-modules used in model calibration. B, Change of the

degradation factor over time of the 7 modules respectively

Figure 4.15: Model validation with experimental datasets of Trina (from DKA systems)
and CIGS (from Bolzano systems). In yellow are the measurements, in red is the extracted
degradation trend and in thick blue line is the model prediction with the 95% confidence
interval. The horizontal green dotted line shows the -20% failure threshold of the degra-
dation trend

4.2.4 Model application and comparison with statistical models

The performance of the proposed method is benchmarked against those of two statistical
forecasting methods, namely ARIMA and Facebook prophet (FP). The reasons to select
these methods is based on a study by Taylor and Letham Taylor and Letham (2018), where
the performance of the different statistical models in long-term forecast was compared. In
their study, FP was found to be outstanding in comparison with other models. The ARIMA
model was characterized with the highest uncertainties. We selected the best and the least
performing models to compare the accuracy of the proposed model and to investigate
whether the nature of the data under evaluation could affect the models’ performance.
Moreover, both methodologies are simple to apply and commonly used in different fields.
For example, the FP algorithm is implemented in common programming languages such
as python and R 1 . The methodology was created as a flexible time series model which is
configurable by non-experts. It is based on a decomposable time series model including
1
The code can be accessed at: http://https://facebook.github.io/prophet/
4.2 Results for the data-driven model 57

trend, seasonality, holidays (not important for our application) and an error term. The
model gives the opportunity to chose linear and logistic time series evolution, the second
being suited for the non-linear behavior of PV performance. Furthermore, among the
seasonality settings for different time resolutions, yearly seasonality was selected to detect
yearly variations in power production due to the seasons of a year. The ARIMA model has
free online packages Hyndman et al. (2018) and Hyndman and Khandakar (2008). In this
study an auto-arima function which uses the akaike information criterion (AIC) to get the
optimal model was used.

Table 4.9: Comparision of ARIMA, Facebook prophet and the proposed model at two
prediction intervals. where, mc1 is mc-Si1, mc2 is mc-Si2, 1j is 1j-a-Si1, 3j is 3j-a-Si1, pc1
is pc-Si1, pc2 is pc-Si2, pc3 is pc-Si3 and rib is ribbon1
System HIT1 mc1 mc2 pc1 pc2 pc3 1j 3j rib CdTe1 CIGS1
ARIMA
FT3% [years] 21.4 NA 18.8 10.1 14.2 18.2 16.2 NA 13.8 16.4 NA
RUL3% [years] 12.9 NA 10.2 1.6 5.7 9.7 7.8 NA 5.3 7.9 NA
FT f ull [years] 21.2 NA 14.5 NA 22.3 26.5 15.0 NA 21.8 12.5 4.0
RUL f ull [years] 12.8 NA 6.0 NA 13.8 18.0 6.5 NA 13.3 4.0 -4.5
Reld i f f [%] 0.9 - 22.8 NA 36.3 31.3 36.7 - 21.8 23.8 NA
Facebook prophet
FT3% [years] 34.6 37.8 13.8 9.8 9.5 15.3 15.1 21.0 9.4 14.9 NA
RUL3% [years] 26.1 29.3 5.3 1.3 1.0 6.8 6.6 12.5 0.9 6.4 NA
FT f ull [years] 16.7 37.8 14.7 33.1 24.7 18.7 14.9 21.0 24.5 12.8 5.6
RUL f ull [years] 8.2 29.8 6.2 24.7 16.2 10.2 6.4 12.5 16.0 4.2 -2.9
Reld i f f [%] 51.7 - 6.1 70.3 61.5 18.2 36.9 - 61.6 14.1 NA
Proposed
FT3% 16.5 82.5 17.2 17.8 25.2 25.4 18.7 25.2 25.2 15.8 5.8
RUL3% [years] 8.0 74.0 8.7 9.3 16.7 16.9 9.9 16.7 16.7 7.3 -2.7
FT f ull [years] 15.4 82.5 17.2 14.7 24.3 18.8 21.8 25.2 24.3 15.2 6.5
RUL f ull [years] 6.9 74.0 8.7 6.2 15.8 10.3 13.3 16.7 15.8 6.7 -2.0
Reld i f f [%] 6.7 - 0 17.4 3.6 25.9 14.2 - 3.6 3.8 10.8

The performance evaluation is carried out using the 11 Bolzano system datasets. The
comparison of the three models is twofold: first, the models are compared on the lifetime
forecast after a small performance degradation interval and second, on the consistence of the
forecast at two different forecasting intervals. In the first attempt, the same system input
data (normalized and STC irradiance corrected power time series with 3% performance
degradation as training data) were tested on ARIMA and FP models. Unfortunately, no
meaningful predictions were achieved because the number of calculated change points is
greater than the number of observations in the dataset. Instead, the monthly Performance
Ratio (PR) was used as the input parameter. The PR sets the actual yield of a PV system
in relation with the expected yield at STC conditions IEC61724-1 (2017). For model
comparison, the models are calibrated at two different intervals: first , at an interval
corresponding to the 3% performance degradation and second , using the entire datasets
corresponding to 8.5 years interval. The respective lifetime and RUL forecasts are shown
in table 4.9.
Considering a 3% calibration threshold, it is visible that there are high discrepancies
as well as unrealistic forecast scenarios (see figure 4.16) for both FP and ARIMA models.
58 4. Results and Discussion

In table 4.9 it is also observed that the failure times are either over- or under-estimated
for the ARIMA and Facebook models. The over- and under-estimations can be correlated
with the evaluated degradation trend which is influenced by the nature of the dataset as
well as the number of data-points. Considering the calibration using the entire dataset,
the FP and the proposed model show good agreement in the failure time forecast for most
of the systems. This excludes the mc-Si1 system that appeared to be very stable with a
performance degradation of less than 2% after 8.5 years. For this system, the proposed
model shows an over-estimation of the failure time because the evaluated degradation is too
small to achieve optimal calibrations. Comparing the variations in predictions at different
calibration intervals, it can be seen that the proposed model displays a good consistency
(on average 9.5% relative difference) compared to the other two models that displayed
unrealistic scenarios. Although the FP showed a good agreement with the proposed model
when calibrated using the entire dataset, it however displayed big variations in lifetime
forecast between the two calibration intervals. Overall, the ARIMA model does not perform
well in this study since it failed to converge for four systems. The model also tends to overfit
the data, hence making it more sensitive to reversible performance reducing events. This is
visible when looking at the pc-Si1 system (figure 4.17) where the model failed to converge
when calibrated on the entire dataset. This implies that such a model is not useful for
performance degradation forecast.

Figure 4.16: ARIMA, Facebook prophet and proposed methods when applied on a CIGS
system with fewer data-points using a 3% threshold. The horizontal dotted green line
shows the failure time threshold.

Figure 4.17: Forecast at 3% and entire data-set calibration thresholds using auto-ARIMA
model. The figure shows overfiting when calibration is done on the entire dataset.

Considering the calibration using the entire data-sets, the Facebook and the proposed
model showed a good agreement in the failure time forecast for most of the systems. This
4.2 Results for the data-driven model 59

excludes the mc-Si1 system that appeared to be very stable with a performance degrada-
tion of less than 2% after 8.5 years. For this system, the proposed model shows an over-
estimation of the failure time because the evaluated degradation is too small to achieve
optimal calibrations. Comparing the variations in predictions at different calibration in-
tervals, it can be seen that the proposed model displays a good consistence compared to
the other two models. Although the Facebook prophet showed a good agreement with
the proposed model when calibrated using the entire data-sets, it however displayed big
variations in lifetime forecast between the two calibration intervals. Overall, the ARIMA
model showed the under-performance in this study since for over four systems it failed to
converge. The model also tends to overfit the data, hence making it more sensitive with
performance reducing events that could be reversed. This can be clearly displayed in pc-
Si1 system (figure 4.17) where the model failed to converge when calibrated on the entire
dataset. This implies, that such a model is not sufficient for performance degradation
forecast.

4.2.5 Model limits and uncertainties

Although the proposed model has displayed a good performance on a number of PV mod-
ules and systems, it is however bound to some limits and uncertainties that could deterio-
rate the prediction accuracy. The model limits and uncertainties are identified as:

Figure 4.18: System showing a sudden drop in power (around the green box), in blue is
the model forecast. The horizontal green dotted line shows the -20% failure threshold of
the degradation trend

The model works well when the degradation is gradual and continuous, it cannot fore-
cast events that might lead to dramatic or sudden power losses, such as breakage, fire,
or catastrophic failures. For example, figure 4.18 shows one of the DKASC Alice Spring
systems; eco-Kinetics-26.5kW-mono-Si-Dual (2010), which experiences a sudden drop in
performance. According to the information from the DKA website, this sharp drop in
performance was attributed to one PV module in this PV array that was damaged during
a windstorm. Hence, the cause is identified as a partial failure of one array string in which
this damaged module is connected.
The model is based on degradation patterns; hence it is influenced by the extracted
degradation trends. The degradation trend is extracted from time series data by applying
60 4. Results and Discussion

the condition in equation 3.30. Changing the tolerance (tol) value affects the extracted
degradation trend as shown in figure 4.19a. Figure 4.19b shows how the calibration as well
as the forecast error varies with tol. The RMSE was calculated using equation 3.31 and is
converted to percentage.

Figure 4.19: A, Variation of the extracted degradation at different values of tol, the hor-
izontal dotted lines show the -20% failure threshold of the degradation trend. B, is the
variation of the fitting error (RMSE-fit) and forecast error (RMSE-pred) with different
values of tol

According to the evaluated datasets in this study, the value tol and the extracted
degradation trend highly depend on the resolution and the outliers in a given dataset. For
data sets with a monthly or yearly resolution, the values of tol ranged from 0.9 to 1.25
while datasets with high resolution of 15 minutes, the values were between 1.4 and 1.65.
The tolerance factor tol is very useful as it provides a parameter for optimization adapted
to the datasets variability unlike the black box automated algorithms.

4.2.6 Assessing the effects of PV modules long-term degradation

on lifetime energy yield
To mitigate the financial investment risks for PV systems stakeholders, it is a prerequisite
to reliably predict the long-term energy yield. Recently Reise et al. (2018) have reported
the different sources of uncertainties in PV systems yield predictions and assessments. A
big challenge was on how the degradation effects are taken into account during long-term
yield predictions. Indeed, the authors reported that, this is the most unexplored challenge
and varies from analysist to analysist. Therefore, in this section we show how the proposed
model could be applied to long-term yield predictions to lower the uncertainties associated
to long-term performance degradation and hence improve the long-term yield prediction
accuracy. Hence we point out that, the scope of the study is limited to analyzing the impact
(sensitivity) of degradation to long-term yield prediction and not to propose a model for
yield prediction.
The described method is applied to 7 of the TISO modules (TEA1, TEA4, TEA6,
TEA8, TEA10, TEB1, TEB2) to extract the degradation trends independent of measure-
ment uncertainties and to forecast the power degradation from a -5% to a 35 years time
frame (the current modules age). To expand our analysis, three degradation scenarios;
linear with a constant rate, non-linear (proposed but with a constant degradation rate) as
4.2 Results for the data-driven model 61

well as non-linear (proposed with a time dependent degradation rate) are assumed. The
comparison with a linear approximation is of ultimate importance because it is commonly
used during long-term yield evaluation according to Reise et al. (2018). Figure 4.20 shows
the results of two of the modules TEA1 and TEA8 that showed the best and the worst
forecasts when using the linear approximation.

Figure 4.20: (TEA1) and (TEA8) TISO modules that displayed lower and higher uncer-
tainties when long-term predictions are made with a linear approximation (black). The
green and blue curves show the forecasts with the non-linear degradation with time depen-
dent rate and with a constant rate respectively. In yellow are the measured data points
and in red is the degradation trend

It can be seen that the new approach (non-linaer with a time dependent degradation
rate) displays consistent results considering the measured data (real data). Since the in-
tended exercise is to assess how this will affect the long-term yield predictions, all the
approximations are correlated to the uncertainties in yield evaluations. This was achieved
by comparing the relative loss of lifetime energy yield of the measured degradation for the
35 years with the predicted ones as shown in table 4.10. It can be seen that using the pro-
posed long-term degradation method (non-linear with a time dependent rate) reduced the
prediction uncertainties to 37.6% on average. Moreover, it is seen that a linear approxima-
tion with a constant degradation rate is associated with the highest uncertainties of over
64.7 on average. Indeed, the dependence of the prediction uncertainties to the degradation
patterns is clearly visible in the two modules. For module TEA1 with a linear degradation
pattern the uncertainty is small compared to module TEA8 with an exponential degrada-
tion pattern. Moreover, the residual standard deviation calculated according to equation
3.22, further confirms these discrepancies. Meaning that the linear approximation cannot
reliably fit or describe the different degradation patterns observed in the field.
Table 4.11 shows the failure time (defined as the time needed for power to decrease
by -20% of the initial value). It is observed that using the proposed method (non-linear
approximation with a time dependent rate), consistent failure time forecasts are achieved.
On contrary, using a linear as well as a non-linear approximation but with a constant
rate, some unrealistic scenarios are predicted for some modules. Indeed these scenarios
are correlated with the degradation patterns. These calculations demonstrate that if these
scenarios are included in the lifetime yield predictions, they can lead to over estimation
of the lifetime yield. Hence the proposed model could provide a solution to improve the
lifetime yield accuracy.
62 4. Results and Discussion

Table 4.10: Comparison of evaluated loss in lifetime energy yield in 35 years using the 3
degradation approximation scenarios. Relative uncertainty is calculated according to the
measured yield loss
Module name TEA1 TEA4 TEA6 TEA8 TEA10 TEB1 TEB2
Measured
Loss of lifetime energy [years] -19.04 -7.86 -9.07 -6.58 -5.09 -8.87 6.97
Linear
Loss of lifetime energy [years] -15.70 -3.38 -0.46 +0.37 -0.29 -5.70 -3.57
Relative uncertainty [%] 17.5 57.0 94.9 105.7 94.3 35.7 48.8
Non-linear with a constant rate
Loss of lifetime energy [years] -12.74 -3.84 -1.38 -1.47 -1.21 -6.47 -4.93
Relative uncertainty [%] 33.1 51.1 84.7 78.6 76.2 27.1 29.3
Non-linear with a time dependent rate
Loss of lifetime energy [years] -39.7 -16.89 -4.25 -13.55 -2.46 -8.86 -6.81
Relative uncertainty [%] 52.0 53.4 52.8 51.4 51.6 0.1 2.3

Figure 4.21: Residual standard deviation of the 7 TISO modules fitted at -5% degradation
using linear and non-linear models respectively

4.2.7 Simplified User Interface (PVLife Toolbox)

After optimizing, testing and further validating our model under different data scenarios,
the described approach is embedded into a simplified user interface for PV lifetime forecast
called ‘PVLife toolbox’ as shown in figure 4.22. This simplified interface is aimed to be
applicable by any user, even without deep knowledge in data analysis. The toolbox has
for main buttons; the browse button that allows to enter the data path, the run button
4.2 Results for the data-driven model 63

Table 4.11: Comparison of evaluated loss in lifetime energy yield in 35 years using the 3
degradation approximation scenarios. Relative uncertainty is calculated according to the
measured yield loss
Module name TEA1 TEA4 TEA6 TEA8 TEA10 TEB1 TEB2
Linear
Failure time [years] 38.7 82.7 96.4 123.7 112.0 75.9 116.6
Non-linear with a constant factor
Failure time [years] 95.0 92.5 82.8 52.8 77.1 73.9 67.1
Non-linear with a time dependent factpr
Failure time [years] 29.8 33.1 38.0 33.4 44.1 38.8 39.9

Figure 4.22: PVLife Toolbox, on the left is the layout before simulation and on the right
is the layout after simulation

that starts the simulation, the clear button that deletes the current inputs-output to allow
entrance of new inputs and the quit button that exits the toolbox. The toolbox has been
created using the module Tkinter in python therefore it can run on the python interface.
64 4. Results and Discussion

4.3 Results for the hybrid model

4.3.1 Application of the hybrid model
After calibration and validation of both the physical and the data-driven methods, both
approaches are combined into a hybrid model as described in section 3.3. The hybrid model
is then applied to re-calculate the failure time of the three mono-crystalline modules. The
results are compared with the one from the physical model as in table 4.12. Figure 4.23
shows the evolution of the degradation trends for the physical as well as the hybrid model.
If we base our argument on the module installed on Zugspitze and on the current bill of
materials, we can conclude that the predicted failure time with the hybrid method is more
probable as compared to the failure time evaluated with only a physical model.

Figure 4.23: On the right: model and measured degradation trend until the time of expo-
sure. On the left: bold lines, measured trend, dotted lines model forecast for with a physical
model (using constant rate) and with a hybrid model (using time dependent rate). The
colors represent the three climatic locations.

Table 4.12: Comparison of the physical model and the hybrid models

Module Failure time [years] Relative

Location Hybrid model Physical model difference
Zugspitze 42.4 52.8 19.7 %
Gran Canaria 23.6 31.6 25.3 %
Negev 17.0 21.4 20.6 %
Chapter 5

Conclusion

When exposed in outdoor environment, PV modules as well as systems are affected by

continuous cycles of temperature, humidity, irradiation, mechanical stress, and soiling.
These environmental factors cause different degradation modes to take place within a PV
module and reduce the performance of the system. Therefore, it is necessary to develop
diagnostic techniques, lower the performance uncertainty, and predict the behavior of PV
systems with higher accuracy. In this research, two main modelling approaches to predict
the degradation rates, the failure time and the remaining useful lifetime of PV modules as
well as systems have been proposed.

On the one hand, a physical model for quantifying the impact of combined climatic stresses
on module maximum power output degradation is proposed. Degradation rate models are
proposed and validated with indoor measurements for specific degradation precursors. A
combined degradation rate model is developed. The model is calibrated and validated
with real field data sets using three identical mono-crystalline modules installed in three
climatic zones: maritime, arid and alpine. Severe degradation is predicted in the arid cli-
mate characterized by high temperatures. Using EAR5 reanalysis datasets to evaluate the
model’s input climatic variables, global risk maps of the specific degradation mechanisms
and the combined rate model are created. The maps are evaluated based on the studied
mono-crystalline modules, and they represent a correlation of different climatic zones to
the different degradation mechanisms. For-example, according to the the Köppen-Geiger
Photovoltaic (KGPV) climate classification, thermo-mechanical degradation is the harsh-
est for the studied PV module in mostly all climate zones, presenting the highest impact
in very high irradiation zones. Photo-degradation and hydrolysis-degradation show similar
global spatial distribution, but the first one is higher since it comprises also UV irradiation
as a degradation factor. Although, due to the high uncertainty in real degradation rate
of PV systems (solar resource, methodology of calculation, quality of operational data,
bill-of-materials of PV modules, etc.), the presented maps might not be representative of
the actual absolute degradation rates, however, they can be used as a guide to identify
possible risk areas in terms of climate stresses.

On the other hand, a data-driven approach is also proposed, the model is aimed to im-
prove the long-term degradation forecasting accuracy using a shorter degradation history.
In developing the model, multiple degradation factor models dependent on the degradation
patterns as well as time are proposed to describe the different degradation patterns. The
data-driven method is based on the monitored degradation and has been calibrated to pro-
vide reliable forecast using a short degradation history of only 3%. During the model devel-
66 5. Conclusion

opment, a focus is put on data treatment especially on extraction of the degradation trends
independent of any reversible performance reducing effect or measurement uncertainties.
Therefore, a computer aided approach is proposed to extract the degradation trends from
the outdoor data. The model has been validated using different PV module and system
datasets with observed long-term degradation. The performance of the proposed model
is benchmarked against two statistical methods, namely ARIMA and Facebook prophet,
using the time series of 11 experimental PV plants (with different PV technologies). The
proposed model displays outstanding performance when forecasts are made after a shorter
time compared to the two statistical models which displayed unrealistic forecasts. The
model also displays consistent results with a 9.5% average relative uncertainty of the eval-
uated failure time when forecasts are made at different intervals, which is not the case for
the two statistical models. Therefore, the obvious advantages of the proposed model over
other data-driven models are that, it is applicable after a small performance degradation
of only 3%, which usually can be observed after a short operation time and it is applicable
on fewer data points. Another advantage of the proposed approach is that, it is based
on a systematic approach for selecting the data and parameters of the models making it
applicable for degradation evaluation on a wide range of data distributions. Hence, the
models’ performance can be well understood and a correlation of different parameters can
be achieved, which is not the case for many empirical, data-driven techniques, especially
those with a black-box character. Moreover, the model is also applicable to all PV tech-
nologies.

On addition, the proposed data-driven model has been applied to assess the effects of
long-term degradation to lifetime yield prediction. The assessment has been carried out by
assuming three degradation scenarios; linear and non-linear with a constant degradation
rate as well as non-linear with a time dependent degradation rate (proposed). Using the
latter made it possible to lower the long-term yield prediction relative uncertainty to 37.6%
on average compared to 54.4% and 64.7% for a non-linear and linear with a constant degra-
dation rate respectively. Lastly, a ‘PVLife toolbox’ based on the data-driven algorithms
has been developed in this research with a simplified interface to be applicable by any one
even without deep knowledge in data analysis.

Finally, the two approach are combined into a hybrid model to achieve a good forecasting
accuracy by deploying data-driven algorithms and also to have a physical understanding
of the evaluated rates through physical models. The hybrid model has been benchmarked
with physical model to evaluate the lifetime of three aforementioned experimental mono-
crystalline modules and it proves to provide more consistent results. The proposed hybrid
model will aid to have more reliable long-term degradation forecast as well as having a
physical understanding of the dominating degradation mechanisms influencing the perfor-
mance degradation. We believe such a model is useful to calculate more reliable levelized
cost of energy and thus the economic viability of solar energy as well as to aid in PV
materials development that can withstand the different environmental conditions.
Bibliography

Annigoni, E., A. Virtuani, M. Cavvicio, G. Friesen, D. Chianese, and C. Ballif (2019). 35

years of photovoltaics: Analysis of the TISO-10-kW solar plant, lessons learnt in safety
and performance – Part 2. 38

Annigoni, E., A. Virtuani, F. Sculati-Meillaud, and C. Ballif (2017, June). Modeling

Potential-Induced Degradation (PID) of Field-Exposed Crystalline Silicon Solar PV
Modules: Focus on a Regeneration Term. In 2017 IEEE 44th Photovoltaic Specialist
Conference (PVSC), pp. 2794–2798. 14, 46

Ascencio-Vasquez, J. (2020). Modelling and worldwide assessment of performance and

aging of photovoltaic modules and systems. 75, 76

Ascencio-Vásquez, J., K. Brecl, and M. Topič (2019, October). Methodology of Köppen-

Geiger-Photovoltaic climate classification and implications to worldwide mapping of PV
system performance. Solar Energy 191, 672–685. 38, 47

Bala Subramaniyan, A., R. Pan, J. Kuitche, and G. TamizhMani (2018, September). Quan-
tification of Environmental Effects on PV Module Degradation: A Physics-Based Data-
Driven Modeling Method. IEEE Journal of Photovoltaics 8(5), 1289–1296. 1, 17, 19, 46,
47

Bosco, N., T. J. Silverman, and S. Kurtz (2016, July). Climate specific thermomechanical
fatigue of flat plate photovoltaic module solder joints. Microelectronics Reliability 62,
124–129. 17

Braisaz, B., C. Duchayne, M. Van Iseghem, and K. Radouane (2014, September). PV

Aging Model Applied to Several Meteorological Conditions. In 29th PVSEC Proceedings,
Amsterdam, pp. 2303–2309. 8, 14, 15, 16, 17

Brownlee, J. (2017, February). Introduction to Time Series Forecasting With Python: How
to Prepare Data and Develop Models to Predict the Future. Machine Learning Mastery.
37

Buck, A. L. (1981, December). New Equations for Computing Vapor Pressure and En-
hancement Factor. Journal of Applied Meteorology 20(12), 1527–1532. 30

(C3S) (2017). Era5: Fifth generation of ecmwf atmospheric reanalyses of the global
climate . copernicus climate change service climate data store. Technical report,
https://cds.climate.copernicus.eu/cdsapp, Paris. 29

Cañete, C., J. Carretero, and M. Sidrach-de Cardona (2014, February). Energy perfor-
mance of different photovoltaic module technologies under outdoor conditions. En-
ergy 65, 295–302. 36
68 BIBLIOGRAPHY

Chen Xiongzi, Yu Jinsong, Tang Diyin, and Wang Yingxun (2011, August). Remaining use-
ful life prognostic estimation for aircraft subsystems or components: A review. In IEEE
2011 10th International Conference on Electronic Measurement Instruments, Volume 2,
pp. 94–98. 6, 35

Crommelynck, D. and A. Joukoff (1990, January). A simple algorithm for the estimation
of the spectral radiation distribution on a horizontal surface, based on global radiation
measurements. Solar Energy 45(3), 131–137. 30

DKASC, A. S. (2019). DKASC, Alice Springs | DKA Solar Centre |,

http://dkasolarcentre.com.au/locations/alice-springs, 2019-11-08 . 38

Driemel, A., J. Augustine, K. Behrens, S. Colle, C. Cox, E. Cuevas-Agulló, F. M. Denn,

T. Duprat, M. Fukuda, H. Grobe, M. Haeffelin, G. Hodges, N. Hyett, O. Ijima, A. Kallis,
W. Knap, V. Kustov, C. N. Long, D. Longenecker, A. Lupi, M. Maturilli, M. Mimouni,
L. Ntsangwane, H. Ogihara, X. Olano, M. Olefs, M. Omori, L. Passamani, E. B. Pereira,
H. Schmithüsen, S. Schumacher, R. Sieger, J. Tamlyn, R. Vogt, L. Vuilleumier, X. Xia,
A. Ohmura, and G. König-Langlo (2018, August). Baseline Surface Radiation Network
(BSRN): structure and data description (1992–2017). Earth System Science Data 10(3),
1491–1501. 30

Dunn, L., M. Gostein, and B. Stueve (2013, February). Literature Review of the Effects
of UV Exposure on PV Modules. In NREL PV Module Reliability Workshop. 16

Escobar, L. A. and W. Q. Meeker (2007, August). A Review of Accelerated Test Models.

arXiv:0708.0369 [stat]. arXiv: 0708.0369. 11, 16, 24

Faiman, D. (2008). Assessing the outdoor operating temperature of photovoltaic modules.

Progress in Photovoltaics: Research and Applications 16(4), 307–315. 31

Gaines, G. B., R. E. Thomas, G. T. Noel, T. S. Shilliday, V. E. Wood, and D. C. Carmichael

(1978). Development of accelerated test design for service-life prediction of solar array
at Mead, Nebraska. Quarterly report. 17, 18

Gok, A., D. K. Ngendahimana, C. L. Fagerholm, R. H. French, J. Sun, and L. S. Bruckman

(2017, December). Predictive models of poly(ethylene-terephthalate) film degradation
under multi-factor accelerated weathering exposures. PLOS ONE 12(5), e0177614. 23

Gu, X., Y. Lyu, L.-C. Yu, C.-C. Lin, and D. Stanley (2015, December). Effect of Intensity
and Wavelength of Spectral UV Light on Discoloration of Laminated Glass/EVA/PPE
PV Module. In 3rd Atlas-NIST Workshop on PV Materials Durability. 16

Hacke, P., S. Spataru, S. Johnston, K. Terwilliger, K. VanSant, M. Kempe, J. Wohlgemuth,

S. Kurtz, A. Olsson, and M. Propst (2016, November). Elucidating PID Degradation
Mechanisms and In Situ Dark I–V Monitoring for Modeling Degradation Rate in CdTe
Thin-Film Modules. IEEE Journal of Photovoltaics 6(6), 1635–1640. 14

Hacke, P., S. Spataru, K. Terwilliger, G. Perrin, S. Glick, S. Kurtz, and J. Wohlgemuth

(2015, November). Accelerated Testing and Modeling of Potential-Induced Degrada-
tion as a Function of Temperature and Relative Humidity. IEEE Journal of Photo-
voltaics 5(6), 1549–1553. 14
BIBLIOGRAPHY 69

Halm, A., A. Schneider, V. D. Mihailetchi, L. J. Koduvelikulathu, L. M. Popescu, G. Gal-

biati, H. Chu, and R. Kopecek (2015, August). Potential-induced Degradation for En-
capsulated n-type IBC Solar Cells with Front Floating Emitter. Energy Procedia 77,
356–363. 14

Hattendorf, J., A. Blauaermel, D. Koshnicharov, M. C. Koekten, L. Wulff, W.-M. Gnehr,

R. Löw, and J. A. Esquivel (2012, October). Potential Induced Degradation in Mono-
Crystalline Silicon Based Modules: An Acceleration Model. In 27th European PVSEC
Proceedings, pp. 3405–3410. 14, 15

Huld, T. and A. M. G. Amillo (2015, June). Estimating PV Module Performance over

Large Geographical Regions: The Role of Irradiance, Air Temperature, Wind Speed and
Solar Spectrum. Energies 8(6), 5159–5181. 30, 31

Hyndman, R. J., G. Athanasopoulos, C. Bergmeir, G. Caceres, L. Chhay, M. O’Hara-Wild,

F. Petropoulos, S. Razbash, E. Wang, and F. Yasmeen (2018). forecast: Forecasting
Functions for Time Series and Linear Models version 8.10 from CRAN. 57

Hyndman, R. J. and Y. Khandakar (2008, July). Automatic Time Series Forecasting: The
forecast Package for R. Journal of Statistical Software 27(1), 1–22. 57

IEC61724-1 (2017). Photovoltaic system performance, Part 1: Monitoring. Geneva, CH:

International Electrotechnical Commission;2017. 57

Jordan, D. and S. Kurtz (2010, June). Analytical improvements in PV degradation rate

determination. In 2010 35th IEEE Photovoltaic Specialists Conference, pp. 002688–
002693. ISSN: 0160-8371, 0160-8371. 11

Jordan, D. C., C. Deline, M. G. Deceglie, A. Nag, G. M. Kimball, A. B. Shinn, J. J. John,

A. A. Alnuaimi, A. B. A. Elnosh, W. Luo, A. Jain, M. U. Saleh, H. von Korff, Y. Hu,
J.-N. Jaubert, and F. Mavromatakis (2020, January). Reducing Interanalyst Variability
in Photovoltaic Degradation Rate Assessments. IEEE Journal of Photovoltaics 10(1),
206–212. 1

Jordan, D. C. and S. R. Kurtz (2013). Photovoltaic Degradation Rates—an Analytical

Review. Progress in Photovoltaics: Research and Applications 21(1), 12–29. 5, 23

Jordan, D. C., T. J. Silverman, B. Sekulic, and S. R. Kurtz (2017). PV degradation

curves: non-linearities and failure modes. Progress in Photovoltaics: Research and
Applications 25(7), 583–591. 33

Jordan, D. C., T. J. Silverman, J. H. Wohlgemuth, S. R. Kurtz, and K. T. VanSant (2017a).

Photovoltaic failure and degradation modes. Progress in Photovoltaics: Research and
Applications 25, 318–326. 16

Jordan, D. C., T. J. Silverman, J. H. Wohlgemuth, S. R. Kurtz, and K. T. VanSant (2017b).

Photovoltaic failure and degradation modes. Progress in Photovoltaics: Research and
Applications 25(4), 318–326. 26, 42

Kempe, M. D. (2006, October). Modeling of rates of moisture ingress into photovoltaic

modules. Solar Energy Materials and Solar Cells 90(16), 2720–2738. 23
70 BIBLIOGRAPHY

Kempe, M. D. (2014, June). Evaluation of the uncertainty in accelerated stress testing.

In 2014 IEEE 40th Photovoltaic Specialist Conference (PVSC), pp. 2170–2175. ISSN:
0160-8371. 17

Kempe, M. D., G. Jorgensen, K. Terwilliger, T. J. McMahon, C. E. Kennedy, and T. T.

Borek (2007). Acetic Acid Production and Glass Transition Concerns with Ethylene-
Vinyl Acetate Used in Photovoltaic Devices. 8

Kersten, F., F. Fertig, K. Petter, B. Klöter, E. Herzog, M. Strobel, J. Heitmann, and J. W.

Müller (2017). System performance loss due to LeTID. 36, 37

Klünter, C., I. Rauschen, K. Reinartz, B. Müller, B. Farnung, and K. Kiefer (2019, Oc-
tober). Degradation in PV Power Plants: Theory and Practice. 36th European Photo-
voltaic Solar Energy Conference and Exhibition, 1331–1335. 37

Koehl, M., M. Heck, and S. Wiesmeier (2012, April). Modelling of conditions for accelerated
lifetime testing of Humidity impact on PV-modules based on monitoring of climatic data.
Solar Energy Materials and Solar Cells 99, 282–291. 24

Koehl, M., M. Heck, and S. Wiesmeier (2018). Categorization of weathering stresses for
photovoltaic modules. Energy Science & Engineering 6(2), 93–111. 27, 49

Koehl, M., M. Heck, S. Wiesmeier, and J. Wirth (2011, July). Modeling of the nominal
operating cell temperature based on outdoor weathering. Solar Energy Materials and
Solar Cells 95(7), 1638–1646. 27

Koehl, M., S. Hoffmann, and S. Wiesmeier (2017). Evaluation of damp-heat testing of

photovoltaic modules. Progress in Photovoltaics: Research and Applications 25(2), 175–
183. 23

Koehl, M., D. Philipp, N. Lenck, and M. Zundel (2009, August). Development and ap-
plication of a UV light source for PV-module testing. In Proceedings of SPIE - The
International Society for Optical Engineering 7412. 16

Kratochvil, J. A., W. E. Boyson, and D. L. King (2004). Photovoltaic array performance

model. 31

Köntges, M., S. Kurtz, C. E. Packard, U. Jahn, K. A. Berger, K. Kato, T. Friesen, H. Liu,

M. Van Iseghem, J. Wohlgemuth, D. Miller, M. Kempe, P. Hacke, F. Reil, N. Bogdanski,
W. Herrmann, C. Buerhop-Lutz, G. Razongles, and G. Friesen (2014). Review of Failures
of Photovoltaic Modules. Report, IEA International Energy Agency. 8, 14, 26, 33, 35

Köntges, M., G. Oreski, U. Jahn, M. Herz, P. Hacke, and K.-A. Weiß (2017). Assessment
of photovoltaic module failures in the field : International Energy Agency Photovoltaic
Power Systems Programme : IEA PVPS Task 13, Subtask 3 : report IEA-PVPS T13-
09:2017. Technical report, International Energy Agency, Paris. 1, 6

Laayouj, N., H. Jamouli, and M. E. Hail (2016, December). Intelligent Prognostic Frame-
work for Degradation Assessment and Remaining Useful Life Estimation of Photovoltaic
Module. Journal of Engineering and Technological Sciences 48(6), 772–795–795. 32
BIBLIOGRAPHY 71

Lausch, D., V. Naumann, A. Graff, A. Hähnel, O. Breitenstein, C. Hagendorf, and J. Bag-

dahn (2014, January). Sodium Outdiffusion from Stacking Faults as Root Cause for the
Recovery Process of Potential-induced Degradation (PID). Energy Procedia 55, 486–493.
14

Listou Ellefsen, A., E. Bjørlykhaug, V. Æsøy, S. Ushakov, and H. Zhang (2019, March).
Remaining useful life predictions for turbofan engine degradation using semi-supervised
deep architecture. Reliability Engineering & System Safety 183, 240–251. 32

Marín, M. L., A. Jiménez, J. López, and J. Vilaplana (1996, July). Thermal degradation
of ethylene (vinyl acetate). Journal of thermal analysis 47(1), 247–258. 23

Masuda, A., N. Uchiyama, and Y. Hara (2015, March). Degradation by acetic acid for crys-
talline Si photovoltaic modules. Japanese Journal of Applied Physics 54(4S), 04DR04.
8

McKay, M. D., J. D. Morrison, and S. C. Upton (1999, March). Evaluating prediction

uncertainty in simulation models. Computer Physics Communications 117(1), 44–51. 31

Meyers, B., M. Deceglie, C. Deline, and D. Jordan (2019, November). Signal Process-
ing on PV Time-Series Data: Robust Degradation Analysis without Physical Models.
arXiv:1907.09456 [eess]. arXiv: 1907.09456. 1, 2

Miller, D. C., E. Annigoni, A. Ballion, J. G. Bokria, L. S. Bruckman, D. M. Burns, Xinxin

Chen, L. Elliott, Jiangtao Feng, R. H. French, S. Fowler, Xiaohong Gu, P. L. Hacke,
C. C. Honeker, M. D. Kempe, H. Khonkar, M. Köhl, L.-E. Perret-Aebi, N. H. Phillips,
K. P. Scott, F. Sculati-Meillaud, T. Shioda, S. Suga, S. Watanabe, and J. H. Wohlgemuth
(2015, June). Degradation in PV encapsulation transmittance: An interlaboratory study
towards a climate-specific test. In 2015 IEEE 42nd Photovoltaic Specialist Conference
(PVSC), pp. 1–6. ISSN: null. 30

Moser, D., M. D. Buono, U. Jahn, M. Herz, M. Richter, and K. D. Brabandere (2017).

Identification of technical risks in the photovoltaic value chain and quantification of the
economic impact. Progress in Photovoltaics: Research and Applications 25(7), 592–604.
6, 14

Munoz, M. A., M. C. Alonso-García, N. Vela, and F. Chenlo (2011, September). Early

degradation of silicon PV modules and guaranty conditions. Solar Energy 85(9), 2264–
2274. 16

Naumann, V., T. Geppert, S. Großer, D. Wichmann, H.-J. Krokoszinski, M. Werner, and

C. Hagendorf (2014, January). Potential-induced Degradation at Interdigitated Back
Contact Solar Cells. Energy Procedia 55, 498–503. 14

Ndiaye, A., A. Charki, A. Kobi, C. M. F. Kébé, P. A. Ndiaye, and V. Sambou (2013,

October). Degradations of silicon photovoltaic modules: A literature review. Solar
Energy 96, 140–151. 8, 23

Ngo, T., Y. Heta, T. Doi, and A. Masuda (2016, April). Effects of UV on power degradation
of photovoltaic modules in combined acceleration tests. Japanese Journal of Applied
Physics 55(5), 052301. 24, 49
72 BIBLIOGRAPHY

Omazic, A., G. Oreski, M. Halwachs, G. C. Eder, C. Hirschl, L. Neumaier, G. Pinter,

and M. Erceg (2019, April). Relation between degradation of polymeric components in
crystalline silicon PV module and climatic conditions: A literature review. Solar Energy
Materials and Solar Cells 192, 123–133. 50

Pan, R., J. Kuitche, and G. Tamizhmani (2011, January). Degradation analysis of solar
photovoltaic modules: Influence of environmental factor. In 2011 Proceedings - Annual
Reliability and Maintainability Symposium, pp. 1–5. ISSN: 0149-144X, 0149-144X. 8

Park, N., C. Han, and D. Kim (2013, December). Effect of moisture condensation on
long-term reliability of crystalline silicon photovoltaic modules. Microelectronics Relia-
bility 53(12), 1922–1926. 8, 23

Philipp, D., T. Geipel, P. Gebhardt, A. Schmid, T. Naeem, and E. Fokuhl (2019, October).
<span class=”highlight”>LeTID - </span>A<span class=”highlight”> Comparison of
Test Methods on Module Level</span>. 36th European Photovoltaic Solar Energy
Conference and Exhibition, 816–821. 36, 37

Phinikarides, A., N. Kindyni, G. Makrides, and G. E. Georghiou (2014, December). Review

of photovoltaic degradation rate methodologies. Renewable and Sustainable Energy
Reviews 40, 143–152. 2, 6

Reise, C., B. Müller, D. Moser, G. Belluardo, and P. Ingenhoven (2018, April). Task 13:
Uncertainties in PV System Yield Predictions and Assessments. 60, 61

Rizzo, A., A. Cester, M. V. Madsen, F. C. Krebs, and S. A. Gevorgyan (2018). A Novel

Algorithm for Lifetime Extrapolation, Prediction, and Estimation of Emerging PV Tech-
nologies. Small Methods 2(1), 1700285. 33

Schwingshackl, C., M. Petitta, J. E. Wagner, G. Belluardo, D. Moser, M. Castelli, M. Ze-

bisch, and A. Tetzlaff (2013, January). Wind Effect on PV Module Temperature: Anal-
ysis of Different Techniques for an Accurate Estimation. Energy Procedia 40, 77–86.
31

Segado, P. M., J. Carretero, and M. Sidrach‐de‐Cardona (2015). Models to predict the

operating temperature of different photovoltaic modules in outdoor conditions. Progress
in Photovoltaics: Research and Applications 23(10), 1267–1282. 31

Sharma, V. and S. S. Chandel (2013, November). Performance and degradation analysis for
long term reliability of solar photovoltaic systems: A review. Renewable and Sustainable
Energy Reviews 27, 753–767. 23

Sheng, Z., Q. Hu, J. Liu, and D. Yu (2019, January). Residual life prediction for complex
systems with multi-phase degradation by ARMA-filtered hidden Markov model. Quality
Technology & Quantitative Management 16(1), 19–35. 32

Swanson, R., M. Cudzinovic, D. DeCeuster, V. Desai, J. Jürgens, N. Kaminar, W. Mul-

ligan, L. Rodrigues-Barbarosa, D. Rose, D. Smith, A. Terao, and K. Wilson (2005,
January). The surface polarization effect in high-efficiency silicon solar cells. pp. 410.
14
BIBLIOGRAPHY 73

Taubitz, C., M. Kröber, M. Schütze, and M. B. Köntopp (2014, November). Potential

Induced Degradation: Model Calculations and Correlation between Laboratory Tests
and Outdoor Occurrence. In 29th European PVSEC Proceedings, pp. 2490–2494. 14,
15

Taylor, S. J. and B. Letham (2018, January). Forecasting at Scale. The American Statis-
tician 72(1), 37–45. 2, 33, 56

Urraca, R., T. Huld, F. J. Martinez-de Pison, and A. Sanz-Garcia (2018, August). Sources
of uncertainty in annual global horizontal irradiance data. Solar Energy 170, 873–884.
28

Virtuani, A., M. Caccivio, E. Annigoni, G. Friesen, D. Chianese, C. Ballif, and T. Sample

(2019). 35 years of photovoltaics: Analysis of the TISO-10-kW solar plant, lessons
learnt in safety and performance—Part 1. Progress in Photovoltaics: Research and
Applications 27(4), 328–339. 38

Virtuani, A., D. Strepparava, and G. Friesen (2015, October). A simple approach to model
the performance of photovoltaic solar modules in operation. Solar Energy 120, 439–449.
37

Whitfield, K., A. Salomon, S. Yang, and I. Suez (2012, June). Damp heat versus field
reliability for crystalline silicon. In 2012 38th IEEE Photovoltaic Specialists Conference,
pp. 001864–001870. ISSN: 0160-8371, 0160-8371. 8

Wohlgemuth, J. H. (2012). Standards for pv modules and components – recent devel-

opments and challenges. Presented at the 27th European Photovoltaic Solar Energy
Conference and Exhibition, Frankfurt, Germany. 1, 23

Wu, D., J. Zhu, T. R. Betts, and R. Gottschalg (2014). Degradation of interfacial adhesion
strength within photovoltaic mini-modules during damp-heat exposure. Progress in
Photovoltaics: Research and Applications 22(7), 796–809. 23

Zhang, S., Y. Zhang, and J. Zhu (2018, April). Residual life prediction based on dynamic
weighted Markov model and particle filtering. Journal of Intelligent Manufacturing 29(4),
753–761. 32

Zhu, J., M. Koehl, S. Hoffmann, K. A. Berger, S. Zamini, I. Bennett, E. Gerritsen, P. Mal-

branche, P. Pugliatti, A. D. Stefano, F. Aleo, D. Bertani, F. Paletta, F. Roca, G. Graditi,
M. Pellegrino, O. Zubillaga, F. J. C. Iranzo, A. Pozza, T. Sample, and R. Gottschalg
(2016). Changes of solar cell parameters during damp-heat exposure. Progress in Pho-
tovoltaics: Research and Applications 24(10), 1346–1358. 23

Zorrilla‐Casanova, J., M. Piliougine, J. Carretero, P. Bernaola‐Galván, P. Carpena,

L. Mora‐López, and M. Sidrach‐de‐Cardona (2013). Losses produced by soiling in the
incoming radiation to photovoltaic modules. Progress in Photovoltaics: Research and
Applications 21(4), 790–796. 36
74 BIBLIOGRAPHY
Mapping of global climate stress
factors

Figure 1: Global horizotal irradiation used to model UV. Since the ambient temperature is
propotional to the irradiance levels, the global distribution of ambient temperatures takes
the same trend. Figure reprinted from Ascencio-Vasquez (2020). The map is according to
EAR5 dataset and of the year 2018
76 Appendix . Mapping of global climate stress factors

Figure 2: Maximum module temperature distribution worldwide. The module temperature

is calculate from the ambient temperature using the Faiman model. Figure reprinted from
Ascencio-Vasquez (2020). The map is according to EAR5 dataset and of the year 2018

Figure 3: Distribution of relative humidity worldwide. The relative humidity is calcu-

lated from the ambient and the dew point temperatures using the Buck’s formula. Figure
reprinted from Ascencio-Vasquez (2020). The map is according to EAR5 dataset and of
the year 2018
Supporting articles

Paper 1. S. Lindig, I. Kaaya, K.A. Weiss, D. Moser, and M. Topic, “Review of statistical
and analytical degradation models for photovoltaic modules and systems as well
as related improvements,” IEEE J. Photovoltaic., vol. 8, no. 6, pp. 1773–1786,
Nov. 2018.
This paper is reprinted from the Open Access publication available at https://
ieeexplore.ieee.org/document/8472886. All rights remain with the authors
and editors.
Impact factor: 3.398

Paper 2. I. Kaaya, M. Koehl, A.P. Mehilli, S. de Cardona Mariano, & K.A. Weiss, ”Mod-
eling Outdoor Service Lifetime Prediction of PV Modules: Effects of Combined
Climatic Stressors on PV Module Power Degradation”. IEEE Journal of Photo-
voltaics,vol. 9, no. 4, pp. 1105–1112, May. 2019.
This paper is reprinted from the Open Access publication available at https:
//ieeexplore.ieee.org/document/8720037. All rights remain with the au-
thors and editors.
Impact factor: 3.398

Paper 3. J. A-Vásqueza, I. Kaaya, K. Brecla, K-A. Weiss and M.Topic, ”Global climate
data processing and mapping of degradation mechanisms and degradation rates
of PV modules”. energies MDPI ,vol. 12, no. 4749 Dec 2019.
This paper is reprinted from the Open Access publication available at https://
www.mdpi.com/1996-1073/12/24/4749/htm. All rights remain with the authors
and editors. Impact factor: 2.707
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE JOURNAL OF PHOTOVOLTAICS 1

Review of Statistical and Analytical Degradation

Models for Photovoltaic Modules and Systems
as Well as Related Improvements
Sascha Lindig , Ismail Kaaya, Karl-Anders Weiß, David Moser , and Marko Topic

Abstract—In this work, we investigate practical approaches of applications until the 1970’s and 1980’s. In the beginning of
available degradation models and their usage in photovoltaic (PV) the 1970’s, the dramatic price increase for fossil fuels and an
modules and systems. On the one hand, degradation prediction energy uncertainty because of the oil crisis raised the aware-
of models is described for the calculation of degradation at system
level where the degradation mode is unknown and hence the physics ness for a need of change in the energy supply [1]. Since then,
cannot be included by the use of analytical models. Several statis- the interest in renewable energies and solar energy in particular
tical models are thus described and applied for the calculation of has increased, which in turn led through scientific achievements
the performance loss using as case study two PV systems, installed to a steady reduction in installation costs and performance im-
in Bolzano/Italy. Namely, simple linear regression (SLR), classical provements of terrestrial PV systems. In the late 1990’s, the
seasonal-decomposition, seasonal- and trend-decomposition using
Loess (STL), Holt–Winters exponential smoothing and autoregres- first large-scale PV systems were installed [2]. Nowadays, PV
sive integrated moving average (ARIMA) are discussed. The per- module manufacturers guarantee a performance reduction of no
formance loss results show that SLR produces results with highest more than 20% within 25 years of operation at standard test
uncertainties. In comparison, STL and ARIMA perform with the conditions (STC) (modules tested indoor under TSTC = 25 ◦ C,
highest accuracy, whereby STL is favored because of its easier im- GSTC = 1000 W/m2 , AM 1.5) and even started to guarantee a
plementation. On the other hand, if monitoring data at PV module
level are available in controlled conditions, analytical models can be maximum degradation of 1%/year for the first ten years. Never-
applied. Several analytical models depending on different degrada- theless, the actual performance throughout the lifetime is quite
tions modes are thus discussed. A comparison study is carried out uncertain and unpredictable. Laboratory or field determination
for models proposed for corrosion. Although the results of the mod- of PV modules service life under real environmental conditions
els in question agree in explanation of experimental observations, requires an unacceptable length of time.
a big difference in degradation prediction was observed. Finally,
a model proposed for potential induced degradation was applied PV systems are affected by continuous cycles of temperature,
to simulate the degradation of PV systems maximum power in humidity, irradiation, mechanical stress, and soiling. These en-
three climatic zones: alpine (Zugspitze, Germany), maritime (Gran vironmental mechanisms cause different degradation modes to
Canaria, Spain), and arid (Negev, Israel). As expected, a more take place within a PV module and reduce the performance
severe degradation is predicted for arid climates. of the system. Therefore, it is necessary to develop diagnostic
Index Terms—Degradation models, performance loss, photo- techniques, lower the performance uncertainty, and predict the
voltaic (PV) modules, PV systems, service life prediction. behavior of PV systems with higher accuracy.
Commonly two approaches, statistical and analytical meth-
I. INTRODUCTION ods, are used for evaluating degradation rates of PV modules
and systems. This report describes quantitative degradation and
ECAUSE of high costs and limited efficiencies, photo-
B voltaic (PV) applications were exclusively used for space
service lifetime models currently used for PV modules and rec-
ommends further improvements. A review of available models
and improvements is crucial for accurate life-time calculations
Manuscript received June 1, 2018; revised July 27, 2018; accepted September of future energy PV systems. The first part of this work focuses
5, 2018. This work was supported by the European Union’s Horizon 2020 on available metrics of variables and the most common statis-
Program under GA. No. 721452—H2020-MSCA-ITN-2016. (Corresponding
author: Sascha Lindig.) tical models to retrieve the performance loss based on these
S. Lindig is with the Institute for Renewable Energy, EURAC Research - Viale metrics. The second part deals with analytical models, which
Druso 1, 39100 Bolzano, Italy, and also with the Faculty of Engineering, Univer- pinpoint specific degradation modes and their possible impact
sity of Ljubljana, 1000 Ljubljana, Slovenia (e-mail:,[email protected]).
I. Kaaya and K. A. Weiß are with the Fraunhofer Institute for Solar Energy on the performance of PV systems.
Systems, 79110 Freiburg, Germany (e-mail:, [email protected]; We believe that a more precise prediction of PV sys-
[email protected]). tem performance and the capability of linking performance
D. Moser is with the Institute for Renewable Energy, EURAC Research -
Viale Druso 1, 39100 Bolzano, Italy (e-mail:,[email protected]). losses to relevant degradation modes will increase public trust
M. Topic is with the Faculty of Engineering, University of Ljubljana, 1000 in solar energy. Additionally, it will help stakeholders such
Ljubljana, Slovenia (e-mail:,[email protected]). as investors, PV plant owners, operation and maintenance,
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. and insurance companies as well as other parties involved
Digital Object Identifier 10.1109/JPHOTOV.2018.2870532 to favor more beneficial and accurate business models and
2156-3381 © 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution
requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE JOURNAL OF PHOTOVOLTAICS

to more efficiently operate and maintain PV systems in the Models for degradation are generally either data-driven
future. or derived from physical principles via stochastic processes.
Although data-driven models are more commonly applied to
II. PV MODULES DEGRADATION MODES analyze degradation data, viewing degradation through stochas-
tic processes helps researchers to theoretically characterize the
Degradation modes are effects that irreversibly degrade the
degradation process. Therefore, a coupling of both models could
performance of a PV module/system or may cause safety prob-
enhance the knowledge of what is happening in PV systems.
lems [3]. A great number of different degradation modes are
Data-driven models help to examine the overall performance
observed in PV modules, both under outdoor operation and in-
loss of a system over time and by using analytical models
door testing. The most commonly observed degradation modes
conclusions of what triggered these losses might be derived.
include [3]: light induced degradation (LID), solder fatigue
failure, silver grid finger delamination, bypass diode failure,
delamination, cell cracks, corrosion, polymeric discoloration, A. Data-Driven Models
ultraviolet (UV) degradation of the cell, polymeric mechani- Data-driven models are often empirical employed to estimate
cal failure, and potential induced degradation (PID). Each of degradation rates based on statistical analysis of given data sets.
these degradation modes has different causes and is triggered The goal of the statistical analysis is to calculate the trend of
by different stress factors. Apart from the modes listed, differ- the PV performance time-series and to translate the slope of the
ent technical risks, which affect the PV performance and the trend to an annual loss rate, in units of %/year [7]. Although these
resulting costs, were found by Moser et al. [4]: glass breakage, models can provide consistent performance loss rates (PLRs),
snail track, defective backsheet, hotspot, soiling, overheating, which are useful for data extrapolation and service life predic-
and failure junction box. An occurring degradation mode can tions, they do not directly provide evidence for the degradation
have an increasing impact on the PV performance over time. It modes taking place in the module. Other effects such as diffuse
develops either in isolation or in combination with other degra- soiling, snow, shading or module mismatch have also a direct
dation modes or technical risks and might lead to the failure of impact on the performance trend. Therefore, it is more accurate
a PV module. The term failure for electrotechnical devices is to talk about a PLR rather than a degradation rate.
defined as “the termination of the ability of an item to perform
a required function” [5]. While this definition serves as a clear
B. Analytical Models
guideline for most devices, the failure of a PV component is
somewhat more complex. For example, although a PV module Analytical models are based on the physical/chemical theo-
can still be technically usable, its power output might be too low ries of a specific degradation mode. These models represent the
to verify the continuation of its operation from an economical mechanism involved in complex physical/chemical processes.
point of view. Within the scope of this work, a failure is defined For well-known PV module degradation modes, several analyt-
as the necessity to replace a PV component, because of its ulti- ical models to forecast PV module degradation are available.
mate, economic or safety-related failure. A clear understanding All these models are based on the principle of understanding
of the definition of a degradation mode is also still a challenge the underlying process, but they are still only heuristic models,
and stress should be put on common nomenclature to define which do not include the influence of material parameters.
the same degradation mode with the same terminology. For the In the following chapters, we discuss the most commonly
moment, accelerated aging tests are being utilized in the study used performance loss models.
of some of these modes. However, there is no proof/evidence
that the results from these tests reflect what exactly happens to IV. PERFORMANCE LOSS MODELS
the modules in outdoor conditions.
Before applying any statistical model, the observed data are
generally treated using filtering techniques depending on factors
III. DEGRADATION MODELING like irradiance or standard deviation ranges and subsequently
Degradation models are used to relate a test item’s estimated averaged or added up over certain time periods. This step is
failure time with the wear and tear during its usage period. performed as data preparation to minimize outliers and noise
The failure time is defined as the end of the lifetime of a PV and to remove values corresponding to inhomogeneous irradi-
component because of its failure. Degradation models help to ance conditions on the irradiance sensor and the PV system [8].
quantify the performance loss PV modules and systems are Afterwards, a performance metric can be applied to a pretreated
experiencing under operation. Degradation in PV systems is the data set and the PLR is calculated by using statistical methods.
reduction in efficiency with which a PV system is converting These steps are necessary to minimize seasonal oscillations and
light of the sun into electricity over time [6]. This appears at to eliminate outliers resulting in the reduction of the overall un-
all levels of a PV system, be it at cell, module or system level. certainty in the estimation of the PLR. In the following, a short
To model PV module degradation modes, the knowledge of overview of the most common performance metrics as well as
internal loads like temperature, chemical conditions, irradiance, statistical methods is presented. In Section IV-C, a comparison
and mechanical loads in/on the PV module is required. One very of the statistical models in question is performed on a case study
important part is to convert external loads to internal loads of with data of two PV systems installed at the airport of Bolzano
the module. in Italy.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LINDIG et al.: REVIEW OF STATISTICAL AND ANALYTICAL DEGRADATION MODELS FOR PHOTOVOLTAIC MODULES AND SYSTEMS 3

A. Performance Metrics Tmod − TSTC are considered. An average performance model for
each PV type is considered and the k-coefficients are calculated
Statistical performance loss models need to be applied on cer-
tain PV system performance-rating parameters. The parameters using data from different modules of the same PV technology
[10], [14]. This model creates a matrix instead of a single well-
are expressed through performance metrics, which are mea-
defined value for the maximum power point.
sured or calculated in a specified interval. Performance metrics
are ideal to compare the performance of different systems in Within the PVs for utility scale and applications project, an-
other widely used model, PVUSA, was developed [15], [16].
different climates. They can be categorized into three different
groups. These are: 1) electrical parameters directly taken from While calculating corrected power values, it is assumed that
I–V curves recorded either outdoor or indoor and corrected to the PV system current primarily depends on irradiance and the
voltage on module temperature Tmod . Tmod in turn is strongly de-
STC, 2) empirical metrics such as PVUSA [9], the 6-k-values
performance model (applied in the PVGIS online tool) [10] or pendent on ambient temperature, irradiance, and wind speed. A
the Sandia models [11], and 3) normalized and/or corrected met- regression of the systems maximum power output is performed
against PVUSA test conditions (PTC) by
rics such as performance ratio [7]. Great care has to be taken
when selecting performance metrics. The choice as well as pos- PMPP = GPOA (A + BGPOA + CTam + DuW )
sible corrections such as corrected power for temperature and
irradiance [12] will influence the results. Therefore, the out- (GPTC = 1000 W/m2 , TPTCA M = 20 ◦ C, uW = 1 m/s). (2)
come of a certain performance loss model applied on a specific First, measurements at high irradiance values (G ≥ 800 W/m2 )
performance metric needs to be evaluated and put into context in the plane of array (POA) are selected and fitted to calculate
to understand the validity of the results. monthly values for the coefficients A, B, C, and D, applying
1) Evaluation of I–V Curves: Electrical parameters of the multivariate regression. Afterwards, the coefficients are used to
I–V curve include power, voltage, and current at the maximum receive monthly ratings at PTCs (substituting meteorological
power point, the open circuit voltage, and the short circuit cur- data values). It should be noted that this methodology is op-
rent. With these parameters, it is possible to calculate the fill timized for crystalline silicon PV. An adapted version of the
factor. A PV systems performance loss is observable when com- equation including another coefficient E was developed to con-
paring the values of periodically performed measurements of sider thin-film technologies [7], [9].
systems in operation [13]. If an unexpected decline in one of 3) Normalized and Corrected Metrics: Normalized and/or
the parameters appears, the affected modules can be examined corrected metric parameters are useful when comparing dif-
indoors or outdoors, for example, with electroluminescence or ferent PV technologies in different climates. Here, PV system
thermal imaging cameras. For detailed characterization, indoor performance data are either normalized to comparable, unit-free
measurements can lead to the most accurate results. However, metrics or corrected in respect to outdoor conditions. One of the
removing PV modules from the field is time consuming with most commonly used metrics is the performance ratio (PR),
the possibility of damaging the modules during transport and which is an adequate indicator for the quality of a PV installa-
handling. These considerations need to be taken into account tion. The PR is calculated by dividing the final (or array) yield
to decide which strategy to follow depending on the extent and Yf (a) (depending if ac- or dc-power is evaluated) with the ref-
the complexity of the detected problem and which stakeholders erence yield Yref [17]. The yields are ratios of measured values
are involved. For example, for insurance claims, outdoor mea- of power or irradiance with values obtained under STC
surement may be sufficient while certified indoor measurements
could be required for PV module warranty claims. Yf PAC /PSTC
PR = = (3)
2) Empirical Metrics: The empirical metrics presented are Yref GPOA /GSTC
models, which aim to obtain performance data while taking Ya PDC /PSTC
into account the dependence between the PV system output and PRDC = = . (4)
Yref GPOA /GSTC
prevailing outdoor conditions [9]. The idea is to receive PV
system performance parameter like the efficiency or the maxi- When studying the PV performance, it is advisable to use dc-
mum power through the application of formulas, which consist related performance metrics in order to eliminate possible influ-
of empirical coefficients and weather data. Two widely used ences because of inverter degradation or misbehavior.
models are the 6-k-values performance model and PVUSA. A promising correction method, presented by Belluardo et al.
The 6-k-values performance model describes system perfor- [18], evaluates the irradiance and temperature corrected power
mance through the relative efficiency ηREL correlated to STC as under STC conditions as follows:
a function of in-plane irradiance GPOA and module temperature GSTC 1
PT ,G corr = Pmax (5)
Tmod [14] G 1 + γ(Tmod − TSTC ).
ηREL (G , T  ) = 1 + k1 ln(G ) + k2 ln(G )2 + k3 T  Here, γ is the temperature coefficient of the PV systems power
at STC, which is stated on the datasheet. Since γ is retrieved
+ k4 T  ln(G ) + k5 T  ln(G )2 + k6 T 2 . (1)
at 1000 W/m2 and highly temperature dependent, a preliminary
Equation (1) has to be fitted to experimental data to obtain the data filtering, similar to the filter applied in PVUSA, should be
empirical coefficients k1 − k6 . Hereby, the normalized in-plane performed to assure the accuracy of the temperature coefficient
irradiance G = G/GSTC and the normalized temperature T  = in use.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE JOURNAL OF PHOTOVOLTAICS

Methods, which correct absolute values like power, voltage equations below
or current, can be additionally normalized by dividing the cor-
rected value by the nominal installed value under STC. This step ŷ = Tt + St + et , ŷ = Tt × St × et . (7)
simplifies a possible comparison between different PV systems. Here, T is the trend, S the seasonality, and e the remaining part
A clear advantage of these rating techniques is the possibility to of the data [7].
evaluate the performance loss in any desired time resolution. 3) HW Seasonal model: The HW seasonal model contains
a forecast equation and three smoothing equations as shown
B. Purely Statistical Methods below
Statistical analysis methods are used to retrieve trends of per- ŷt+1|t = lt + bt + st−S +1 (8)
formance time-series. These time-series are some sort of perfor-
mance metrics, which are discussed in Section IV-A. The slope lt = A(yt − st−S ) + (1 − A)(lt−1 + bt−1 ) (9)
of a trend function can be interpreted as the PLR. It is possible bt = B(lt − lt−1 ) + (1 − B)bt−1 (10)
to accumulate these ratings for any given time resolution into an
easily comparable annual aging value. The difficulty is to find st = C(yt − lt−1 − bt−1 ) + (1 − C)st−S . (11)
a good estimation of the PLR since the application of a certain Here, lt is the level, bt the slope, and st the seasonal component.
statistical method on a performance metric and a defined filter A, B, and C are smoothing parameters. If monthly data are
determines the result significantly. Statistical analysis methods evaluated, the period of seasonality, S, equals to the value of 12.
can be divided into model-based methods like linear regression, The HW model is either additive or multiplicative, depending
classical seasonal decomposition (CSD), Holt–Winters (HW) on the seasonal behavior. In case of evaluating a PV systems
exponential smoothing or autoregressive integrated moving av- performance, the additive method should be selected because
erage (ARIMA) and nonmodel-based methods such as seasonal the seasonal variations are approximately constant throughout
and trend decomposition using Loess (STL). In the following, the series. The seasonal component is then computed in absolute
these commonly used methods are described. terms and has a mean of around zero. The level equation (9) is a
1) Simple Linear Regression (SLR): Performance metrics of weighted average between the seasonally adjusted observation
any kind are most commonly applied on linear regression be- (yt − st−S ) and the nonseasonal one-step-ahead forecast (lt−1 +
cause of the straight-forward approach. The fitted trend line is bt−1 ). The slope is a weighted average of the level at time t
given by minus the level at t − 1, and the trend at t − 1. The selection
of smoothing parameters determines how fast the exponential
ŷ = at + b. (6)
weights decline over the past observations. The HW method can
Hereby, a represents the gradient and b is the intercept with the be especially useful for computing the future behavior of a PV
y-axis. The SLR-algorithm uses the method of least squares. The system [7], [20], [21].
idea of this method is to sum up squared values of the difference 4) Autoregressive Integrated Moving Average: ARIMA is a
between trend line and actual measurement points and to find model, which can contain several methods in a multiplicative
the minimum value for this sum. Squares are used to add up only way and can be described as ARIMA (p, d, q) (P, D, Q). Here,
positive numbers and to put more weight on more widely scat- p is the auto-regressive, d the differencing, and q the moving
tered residuals. This method overemphasizes outliers as well as average order as well as P is the seasonal autoregressive, D
seasonal variations and can result in large uncertainties. Because the seasonal differencing, and Q the seasonal moving average
of that, performance metrics, which reduce seasonal oscillation order. Due to the flexibility of the model, seasonal variations,
should be applied if the SLR-algorithm is used. errors, outliers, and level shifts can be addressed in a proper
2) Classical Seasonal Decomposition: Another commonly way. ARIMA is applied using the following [7]:
used statistical model is CSD. By using CSD, the seasonality φ(T )φS (T S ) d D S
S yt = φ(T )φS (T )et (12)
and a certain irregular component are separated from a set of
measured time-series data to receive a clear trend over time. T is the delay operator, φ(T ) = (1 − φ1 T − · · · − φp T p ) is an
This technique helps to get a fast idea of a performance loss autoregressive polynomial in T of degree P, φ(T S ) is an au-
of the system in question. The trend is obtained by applying a toregressive polynomial in T S of degree PS , φ(T ) a mov-
centered moving smoothing on a time-series with a certain sea- ing average polynomial in T of degree q, and φS (T S ) is a
sonal period m. When using monthly data, the seasonal period is moving average polynomial of degree QS in T S . Apart from
usually set to 12. Here, the first value is computed by averaging that, d = (1 − T )D is a nonseasonal differencing operator
over the first 12 months. Due to the 12-month centered moving and D S = (1 − T ) is a seasonal differencing operator and
S d

average, 6 months at the beginning and 6 months at the end of grasps nonstationarity in the relevant location in consecutive
the observation period are not included in the computation. To periods [22].
calculate the seasonality, the trend is subtracted from the mea- The stationarity of the time-series determines the optimal
sured data and each month throughout the years of surveillance ARIMA model; a transformation using differencing to achieve
is averaged. What remains at the end is an irregular component stationarity might be indispensable. Stationarity is described by
[19]. Depending on the stability of the seasonal component, a constant mean and variance, resulting in a nonexisting trend
an additive or a multiplicative model is used as shown in the and the graph seems more like white noise. There are different
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LINDIG et al.: REVIEW OF STATISTICAL AND ANALYTICAL DEGRADATION MODELS FOR PHOTOVOLTAIC MODULES AND SYSTEMS 5

ways to difference a time-series, the simplest and most common is done by applying local regression on a data window with a
way being first-order differencing [23] certain width. The regression curve is fitted to the data within
the window. The closer the points are to the center of the win-
ŷ = yt − yt−1 . (13)
dow (higher weight), the greater is the impact on the regression
Here, the differenced value is the change between two consecu- line calculation. The weight is reduced on those points that are
tive values of the original time-series. The resulting time-series furthest from the regression curve. The whole step of regression
has T-1 values. Seasonal or second-order differencing are further and weighting is repeated several times to receive a point on the
examples of how to create stationarity within the time-series in Loess-curve, which is at the center of the window. By moving the
question. window across the data, the complete Loess curve is computed.
The heart of the ARIMA model is the application of autore- What follows is that each point of the Loess curve is the intersect
gression. To perform an autoregression, the desired variable is of a regression curve and the center of the respective window.
computed by applying a linear combination of past values of the
variable. The general form of an autoregressive model of order C. Comparison of Statistical Models
p is
The statistical models presented are applied on the uncor-
ŷ = c + φ1 yt−1 + φ2 yt−2 + · · · + φp yt−p + et (14) rected performance ratio data sets of a mono-crystalline (mc-
Si) and an amorphous silicon (a-Si) system using the software
where c is a constant and et is the remainder.
R. The mc-Si system contains 14 PV modules and has a rated
The moving average model used within ARIMA has a dif-
power of 1960 Watt-peak (Wp). The second installation includes
ferent purpose than the one for CSD. Here, the moving average
12 amorphous silicon modules with a power of 1200 Wp. The
uses past forecast errors in a model similar to a regression. The
observed data have a resolution of 15 min and were averaged
aim of the moving average model is to predict a forecast instead
over whole months. The systems in question were installed in
of smoothing the trend cycle of past values [21].
Bolzano/Italy in August 2010 and are evaluated for seven years
While using the software environment R, the function seas, a
from March 2011 until February 2018 in order to eliminate
function within the R-package “seasonal” which automatically
initial degradation effects such as the Staebler–Wronski effect
performs seasonal adjustments, automatically calculates the op-
or short-term LID. The monitored data are prefiltered to ex-
timal ARIMA (p, d, q)(P, D, Q) variables to apply on the data
clude data with performance ratio values below 1% and above
set [24]. For the application of ARIMA within Section IV-C, the
200% and a POA irradiation of less than 50 W/m2 and more
extracted parameters for both systems are (0, 1, 1) (0, 1, 1).
than 1500 W/m2 . This was done to remove extreme outliers
5) Seasonal–Trend Decomposition Using LOESS: STL is a
and measurement errors. The simplicity of the filter was chosen
continuation of CSD and Loess is a method to estimate nonlinear
to intensify possible deviations among the models. The irradi-
relationships. The centered moving average is replaced by a
ance data are recorded with a pyranometer. For each model,
locally weighted regression to extract the trend [25]. Because
the relative annual PLRs, the corresponding uncertainty and the
of that, the estimates become more robust and are less affected
intercept with the y-axis are given. The PLR of the more sophis-
by missing data and outliers. Similar to CSD, STL decomposes
ticated models are calculated by applying a linear regression to
a seasonal time-series into three components (trend, seasonal,
the respective trend, which was extracted through the statistical
remainder) and is described by
model. To receive the yearly relative PLR and the corresponding
Yt = Tt + St + Rt . (15) uncertainty, the following formulas are used [27]:
STL contains an inner and an outer loop. Every time a run within 12a
PLR = (16)
the inner loop is performed, the seasonal and trend components b
are updated. The number of runs within the inner loop are mostly  
  2  2
equal to 1 or 2. The outer loop includes an inner loop followed  12 −12a
uPLR = × u2a + × u2b (17)
by a calculation of robustness weights. This calculation serves b b2
as an input for the following inner loop to decrease the im-
pact of transient, abnormal behavior on the trend and seasonal where a and b are the fitting coefficients of the linear regres-
parts [26]. sion, u2a,b the variances of these fitting coefficients, and uPLR
To better grasp the idea of Loess, the method is explained the standard deviation of the PLR. This uncertainty calculation
when applied within the software R. Here, two parameters have corresponds to a confidence interval of 68%.
to be chosen, the trend window and the seasonal window. The Two definitions of the PLR can be found in the literature,
seasonal window is either periodic or the span of the Loess win- in relative terms as PLR = 12a/b or absolute with PLR = 12a
dow for seasonal extraction. The smaller the values, the faster [18]. Here, the relative PLR was chosen because it makes it eas-
the trend and seasonal components can change. A high value ier to generalize the findings to the energy yield of the array using
for the seasonal window forces the seasonal part to be peri- the initial yield of the plant. The results of these calculations are
odic, in this case just the means for the monthly values are used less aimed to deliver the best possible combination of filtering
(seasonal component for January is mean of all January val- techniques, performance metrics, and statistical models but are
ues). After calculating the seasonally adjusted data, (measured intended to provide a direct comparison between the presented
data minus seasonality) the trend is Loess-smoothened. This analysis methods. While the uncertainties of the resulting PLRs
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE JOURNAL OF PHOTOVOLTAICS

Fig. 1. Comparison of statistical models on PR-data, circles represent PLR including uncertainties (primary axis), triangles represent initial PR (secondary axis),
on the left is mc-Si system, on the right is a-Si system.

TABLE I
COMPARISON OF STATISTICAL MODELS

were used to rate the statistical models, the remainder, where

applicable, serves as a validation of the parameter fit. The re-
mainder should have Gaussian white noise properties, such as
being uncorrelated and normally distributed.
In Fig. 1, the relative PLRs, uncertainties and initial PR values
are given. The initial value is the intercept with the y-axis.
The stated uncertainty is the uncertainty of the PLR against the
extracted trends, computed by the individual statistical methods.
It can be seen that the performance of the mc-Si system
degrades 0.5%–0.6% per year, the one of a-Si close to 1.8% per
year. Because of cabling and other system losses the initial PR
value is below the theoretical value under STC. As expected, Fig. 2. Trend-lines of CSD- (red-straight) and STL-model (blue-dotted) of
linear regression shows in both cases the highest uncertainty. unfiltered monthly PR-data of a mc-Si system.
For all other cases, a trend was first filtered from the data set
on which a linear regression was performed. This step leads to
an outcome with higher certainties. While all more advanced into account and because of this, the decrease is stronger over
methods show similar results with regards to the uncertainty of time. This in turn results additionally in an overestimation of
the mc-Si system, STL and ARIMA outperform the others when the initial PR value.
applied to the a-Si system. Table I summarizes the models in respect to different char-
In case of CSD, the loss ratio is for both systems higher acteristics of time-series. In general, SLR might serve as a first
in comparison and probably overestimated. In Section IV-B2, it indication of a PLR determination but is not suited for accurate
was mentioned that, when using CSD, the first and last months of calculations due to its simplistic approach. It has been shown
the data set are lost due to the applied centered moving average. that the more sophisticated selected analysis methods perform
This is visible in Fig. 2. Here, the extracted trend-lines of the PR very similar for crystalline systems, both in the estimation of
of the mc-Si system using CSD and STL are shown. Within the the initial PR and the PLR. An exception hereby is the usage
first six months of observation, the trend of the PR has a roughly of CSD for short time-series. The exclusion of the first and last
stable value. When applying CSD, this time period is not taken observations can falsify the final PLR. When a thin-film system
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LINDIG et al.: REVIEW OF STATISTICAL AND ANALYTICAL DEGRADATION MODELS FOR PHOTOVOLTAIC MODULES AND SYSTEMS 7

is subject of the calculations, STL and ARIMA show the best (t = 0) before aging and at time (t = t) after aging or in the
results. field.
When applying HW, ARIMA or STL, it is very important to
perform crucial modeling steps with great care to receive the A. Degradation Models for Corrosion
optimal results. In case of ARIMA, the time-series has to be
Corrosion is one of the most occurring degradation modes
transformed in order to reach stationarity. Since PV time-series
in PV modules [30]. Corrosion is caused by the presence of
are highly seasonal, a seasonal differentiation is essential. For
high temperature and high humidity in the module. Humidity
the same reason, the seasonal window parameter within STL
can enter the module through the backsheet or the layers of
has to be set as periodic and the additive method is preferred
the encapsulant and spread into the module [31], weakening
when using the HW model.
the adhesive bonds between the interfaces. One hypothesis is
Filtering is an integral part when computing PLR values.
that humidity, which catalyzes corrosive processes, leads to a
If performance metrics are corrected using temperature coeffi-
formation of acetic acid through the hydrolysis of vinyl-acetate
cients, which are retrieved at a POA irradiance of 1000 W/m2 ,
monomers present in the EVA [32]–[34]. Corrosion attacks the
an appropriate POA irradiance should be selected. Within this
metallic connections of PV cells and results in a loss of adhesive
work, the preliminary filter are treating outliers and values cor-
strength between the cells and the metallic frame, as well as an
responding to measurement errors not sufficiently. This was
increased leakage current and therefore a loss in performance
done on purpose to amplify the impact of outliers and measure-
[32]. Empirical models based on power at maximum power
ment inaccuracies on the final results. SLR and CSD treat all
point and series resistance as degradation indicators have been
values with similar weights and are therefore strongly affected
proposed to model corrosion according to [35]–[37]. The models
by outliers. HW’s weighted average, STL’s locally weighted
are as follows.
regression, and the combination of similar techniques within
1) The Model of Pan [35]
ARIMA are well suited for outlier handling.
Another statistical method worth mentioning is the year-on- Pmax
= exp(−RD tβ ). (18)
year model developed by Sunpower [28] and later improved by Pmax(0)
NREL [29]. It is implemented within the Python RdTools for
2) Pmax and Rs Models According to Braisaz [36]
the analysis of PV data. This method has a complete different
approach as the here discussed models, as it is using a loss rate 1 − exp(−B)
Pmax = (19)
distribution instead of one single value. The gradient between 1 + exp(RD t − B)
two related data points in consecutive years (hour, day, week,
Rs = Rs(0) + exp(RD t − B) (20)
month) determine a single PLR. The median of this gradient,
the gradients of all remaining data points of that two years, where Pmax and Rs are the output power and series resistance
and all following years determine the final performance loss at time (t), Pmax(0) and Rs(0) are the power output and series
per year. The power loss rates are computed using a 100% per- resistance at time (t = 0), β is the experimental parameter, B
formance baseline value. This approach is excluded within this is a coefficient to be defined, and RD is the degradation rate
study. A comparison is difficult to perform because the initial determined according to (21), (22), and (23).
value is preset and, in contrast with the performed computations, 1) Models for Degradation Rate (RD ) Calculation: Kinetic
the data aggregation is done in an irradiance-weighted manner. models are developed on the primary assumption that the rate
Nevertheless, this approach is of special interest if high quality of degradation is proportional to the concentration of water in
field irradiance data are not available because it can compute PV modules and that the rate constant has a Arrhenius temper-
the PR based on a modeled clear-sky irradiance. ature dependence. Three models according to [19] and [38] are
proposed, namely the Peck’s model, the Eyring model, and the
exponential model.
V. MODELING SPECIFIC DEGRADATION MODES
Peck’s model:
USING ANALYTICAL MODELS  
Ea
In order to explain experimental observations of different RD .Peck = A exp − RH n . (21)
kB T
degradation modes, analytical models are developed based on
the physical/chemical theories of the degradation mode in ques- Eyring model:
tion. These models are environmental stress oriented. The hy-  
−Ea b
potheses of a particular degradation mode are built depending on RD .Eyring = A exp − . (22)
kB T RH
specific environmental stresses applied, and on the assumption
that the kinetics of a specific degradation mode are influenced by Exponential model:
one dominating process. Electrical parameters such as power at  
−Ea
maximum power point (Pmax ), short circuit current (ISC ), shunt RD .Exp = A exp . exp(m × RH). (23)
kB T
(Rsh ), and series resistance (Rs ) are commonly modeled as
degradation indicators. Hence, the environmental stresses and Here, Ea is the activation energy of the degradation process
their interactions with the PV module components are assessed [eV], T the module temperature [K], kB is the Boltzmann con-
based on the reduction of the initial electrical parameter at time stant (8.62 × 10−5 eV/K), and RH is the relative humidity [%].
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE JOURNAL OF PHOTOVOLTAICS

TABLE II
COMPARISON OF MODELS FOR DEGRADATION RATE

Fig. 3. Module 1. Maximum power degradation models of Pan (P) and Braisaz
A, n, b, and m are model parameters. RD [%/h] is the inverse of (B), simulated for three climates: Zugspitze (green), Gran Canaria (blue), and
the mean time to failure at a given condition. In order to obtain Negev (purple).
A, Ea , n, b, and m in (21), (22), and (23), the equations can be
fitted to experimental data or represented on a logarithmic scale
by a straight line, using the following equations: mean values of temperature and relative humidity were used.
  In both cases, the Peck’s model was used for degradation rate
Ea
ln(RD .Peck ) = ln(A) − + n × ln(RH) (24) calculation and both models were fitted to indoor data to extract
kB T the model parameters.
 
Ea b Both models show that temperature is the most relevant fac-
ln(RD .Eyring ) = ln(A) − − (25) tor that influences the PV degradation process, visible by the
kB T RH
  power losses depicted in Fig. 3. This can be seen by a small loss
Ea through lower degradation in power for the module installed at
ln(RD .Exp ) = ln(A) − + m × RH. (26)
kB T Zugspitze, where the climate is characterized by low tempera-
tures and high levels of relative humidity. However, the models
A plot of ln(RD ) versus 1/T [K] gives an Arrhenius plot with
completely differ in degradation predictions. According to the
a slope Ea /kB and an intercept ln(A).
simulations, the model of Pan converges at a relatively fast rate
2) Comparison of Corrosion Models: The models of Peck,
compared with the one of Braisaz. This could be due to the in-
Eyring, and exponential were applied to fit indoor data sets
fluence of the model coefficients such as B and β. Moreover, the
(damp heat at 85C/85% RH), of two c-Si modules, with module
time parameter for the Pan model follows a power law, hence β
1 showing a good performance stability compared with module
might be an accelerating factor.
2. The models were compared based on the extracted parame-
ters (see Table II) as well as the deviation from the fitted data
B. Degradation Models for PID
points. All models are consistent concerning the influence of
the activation energy on the degradation rate and that the pre- PID has been observed in all PV technologies and in almost
dicted values are in the range of literature values for polymeric all operating climates. It does not occur so frequently, but if
materials, which usually range between 0.6 and 2.0 eV [39]. it does, its effect can lead to a severe performance loss within
Nevertheless, the Eyring model shows a significant difference a short period [4], [41]–[44]. In general terms, PID is caused
in the extracted activation energy in comparison to that of Pecks by the difference in potential between the cells and the sup-
and the exponential model. It also has the highest percentage port structure of the module. This difference drives a leakage
deviation of fitted data points in both cases. Therefore, from this current that can lead to power degradation. Different types of
study, we can conclude that the Peck’s model has a better fit PID occur depending on the module technology. For crystalline
compared with the other models. silicon PV, two degradation modes can be identified, PID-p (for
Degradation models are utilized for the simulation of power polarization or passivation) and PID-s (for shunting). PID-p is
output degradation for module 1 to predict its performance a temporary and reversible degradation of the passivation layer,
in three climatic zones: alpine (Zugspitze, Germany), mar- which reduces the performance due to a surface recombination
itime (Gran Canaria, Spain), and arid (Negev, Israel), assuming increase [45]–[47]. PID-s is because of a leakage current in-
the degradation is because of corrosion according to Pan and volving an ionic flow of Na+ from the glass, encapsulant or cell
Braisaz. The module temperature is given for a standard c-Si surface into the cell, diffusing into the silicon stacking faults and
PV module type glass-backsheet construction installed at the shunting the cell [48]. The sodium incorporation in the Si sur-
three test sites and relative humidity was calculated from ambi- face degrades primarily the FF, the Voc , and finally the Isc . The
ent conditions according to [40]. For all the simulations, annual relevant stress factors for PID-s include [42]; high temperature,
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LINDIG et al.: REVIEW OF STATISTICAL AND ANALYTICAL DEGRADATION MODELS FOR PHOTOVOLTAIC MODULES AND SYSTEMS 9

TABLE III
PIDHACKE MODEL

relative humidity, system voltage, light, bias-junction potential,

and injected carriers. The models proposed for PID degradation
according to [36], [43], [44], [49]–[51] are as follows.
1) Pidhacke Model: A parabolic model was proposed by
Hacke et al. [44] to fit the power degradation of c-Si modules
 
Pmax Ea
= 1 − A exp − RH n × t2 . (27)
Pmax(0) kB T
Fig. 4. Simulated normalized maximum power degradation because of PID
The constants A and n are determined by fitting the equation in three climatic zones; Freiburg (green), Gran Canaria (blue) and Negev (red).
to experimental results. The parameters have to be determined
for each module type. This parabolic model is applicable to the
beginning of the degradation phases of PID-s, as it can fit the parameters for the module power. The model is written as
beginning of a sigmoid and does not describe the stabilization
Pmax (U, T, RH, t) = Pmax(0) (1 − P (t)) (30)
phase of the sigmoidal curve. Annigoni et al. [49] used the

indoor data to determine coefficients of the model for distinct 1 − exp τt1
aging contributions (temperature, relative humidity, and time) P (t) = P∞
;
and then applied the model including a voltage term in (28) to 1 − exp t−t τ2
0

outdoor PID degradation for different climates

  −2
U − U0
  P (U ) = 1 + exp (31)
Pmax Ea Φ
= 1 − A exp − RH × t × U.
n 2
(28)
Pmax(0) kB T
t0 = a × b × tˆ0 ; τ1 (T ) = b2 × τˆ1 ; τ2 = τˆ2
(32)
In this study, a similar approach is adopted to extract the model  
coefficients in Table III and to simulate maximum power degra- H0 T − T0
a(H) = ; b(T ) = exp . (33)
dation because of PID in three climatic zones; alpine (Zugspitze, H φ
Germany), maritime (Gran Canaria, Spain), and arid (Negev,
Israel). In all simulations, a constant voltage of 500 V was as- The model includes six adaptation parameters: tˆ0 , U0 , τˆ1 ,
sumed. The simulation results, shown in Fig. 4, were consistent τˆ2 , Φ, and φ. H0 and T0 are scaling parameters. The function
with the ones of Annigoni, saying that a more severe degradation P (t) describes the power loss caused by degradation. P∞ (U )
is predicted for arid climates. is its limit for t → ∞, and a(H) and b(T ) are the acceleration
2) Exponential Model: The model was applied by Hacke functions of relative humidity and temperature. For T = 0 and
et al. [43] to predict PID occurrence in thin-film modules in H = 0, they are equal to 1, therefore τˆ1 , τˆ2 are the time constants
the field using accelerated tests. Considering shunting, which is under these conditions. τ2 remains constant for a given module.
the PID mode that occurs first, an exponential model based on To determine the model’s parameters, the power degradation
module temperature T and relative humidity RH was found to is measured as a function of time with the system voltage U
fit well the PID rate for multiple stress levels of a CdTe module as parameters and a fixed humidity H as well as temperature
in chamber tests. The power model is of the form T . The saturating power P∞ is extracted by fitting P (t) to the
measured data.
    4) Taubitz Model: Taubitz et al. [51] proposed a regener-
d Pmax Ea
= 1 − Af (U ) exp − RH n × t (29) ation model for shunt resistance evolution over time because
dt Pmax(0) kB T of PID degradation. The shunt resistance was modeled in three
phases: shunting phase, regeneration phase, and transition phase
where f (U ) expresses the voltage dependency. as follows:
3) Model of Hattendorf: The model of Hattendorf et al. [50] Shunting phase:
is based on a matrix of indoor experiments where modules are  
exposed to varying voltage, module temperature, and ambient −t
Rsh (t) = aS exp . (34)
humidity. The conditions are varied to determine the model bS (t)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE JOURNAL OF PHOTOVOLTAICS

Regeneration phase: 1) Schwarzschild Law: The Schwarzschild Law has been

  applied by Gu et al. [55] to study the effect of intensity and
t wavelength of spectral UV light on discoloration of laminated
Rsh (t) = CR + aR exp . (35)
bR (t) glass/EVA/PPE PV modules. The law is a function of intensity as
Transition phase: k = A(I)p . (41)
2
Rsh (t) = aT (T )(t + bT (T )) + CT (36) Here, k is a constant, I the intensity, and p is the Schwarzschild
coefficient.
where aS , bS , CT , bT , aT (T ), bS (T ), bT (T ), CR , and aR are
Recommendation: When applying this expression in perfor-
constants and have to be determined for a specific module type.
mance (power) prediction models where other loads are also
Some of them are dependent on the module temperature T . The
applied, the parameter p must be calibrated according to the
constants are determined by measuring the times tS , tT , and tR
knowledge of severity ranking [56].
for reaching certain target values.
5) PID Model According to Braisaz et al. [36]: The model
D. Degradation Models for Delamination, Fatigue Solder
is based on shunt resistance Rsh degradation as an indicator, as
Failure, and Cell Cracks
it is the most important parameter for PID. The evolution of Rsh
as a function of voltage, temperature, and relative humidity was 1) Coffin–Manson’s Equation: The model is used to predict
modeled as degradation modes caused by temperature cycling such as en-
Rsh(0) capsulant delamination, fatigue solder failure, and cell cracks.
Rsh (t) = (37) According to Escobar and Meeker [38], the model describes the
1 + aRD t
  number of cycles to failure as
B Ea σ
RD = A × U exp − N= (42)
1 + exp(−C(RH) + D) kB T (ΔT )β 1
(38)
where ΔT is the temperature range and σ and β1 are properties
where Rsh is the shunt resistance at time (t), Rsh(0) is the initial of the material and test setup. The cycles-to-failure distribution
shunt resistance, A, B, C, and D are model coefficients, U is for temperature cycling can also depend on the cycling rate
the applied voltage, and RD is the degradation rate. (e.g., because of heat buildup). An empirical extension of the
Recommendation: The model contains many coefficients Coffin–Manson relationship that describes such dependencies
whose usage is not clearly described, such coefficients might is [38]
affect the physical interpretation of the results when used in-  
σ 1 Ea × 11605
appropriately. Therefore, we recommend that one should have N= exp (43)
a prior knowledge of the impact of the parameters on which (ΔT )β 1 (freq)β 2 Tmax (K)
the coefficients are being applied to the degradation process in where freq is the cycling frequency and Ea is a quasi-activation
question. energy.
2) Crack Propagation Model: The model was suggested by
C. Models for UV Degradation Braisaz et al. [36] and it was applied to simulate the degradation
of the short-circuit current Isc due to the expansion of cell cracks
UV light exposure has been reported to cause PV module
caused by temperature. The model takes the form
degradation in a number of ways. Exemplary, it could result in
discoloration of the encapsulant material [52] or delamination 1
Ca (t) = Ca (t − 1) +
m . (44)
at the glass encapsulant or cell encapsulant interface [53]. The x 125
Ta
parameter most impacted by UV exposure is the short circuit
current (Isc ). Braisaz et al. [36] proposed a model for short cir- Ca (t) is the crack activation at time (t), Ca (t − 1) the crack
cuit degradation because of UV exposure over time. They found activation at time (t − 1), Ta is the daily temperature amplitude,
that the degradation curve is not linear but an exponentially de- m a model parameter, and x is the number of thermal cycles.
creasing curve. The short circuit is modeled as a function of The crack activation/propagation model is dependent on the
UV as daily temperature amplitude Ta .
3) Damage Accumulation Model: The model was used by
Isc (t) = Isc(0) − aDUV (t) − b(1 − exp(−CDUV )(t)) (39)
Bosco et al. [57] in order to calculate the solder fatigue damage
 t
in seven cities investigated in their study and compared with
DUV (t) = E(u) × 5.5% du. (40) FEM simulated results. They found out that the model fits well
0
to the simulated calculations. The model is written as
Here, DUV is the UV dose in MJ/m2 or kWh/m2 , Isc is the  
Q
short circuit current at time (t), Isc (0) is the initial short circuit D = C(ΔT ) (r(T )) exp −
n m
. (45)
kB Tmax
current, and a, b, and C are model coefficients. The multipli-
cation by 5.5% is because the UV radiation (280–400 nm) is In this equation, ΔT is the mean daily maximum cell temper-
approximately 5.5% of the total light spectrum E(u) [54]. ature change, Tmax the mean maximum daily temperature, C a
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LINDIG et al.: REVIEW OF STATISTICAL AND ANALYTICAL DEGRADATION MODELS FOR PHOTOVOLTAIC MODULES AND SYSTEMS 11

scaling constant, Q the activation energy, kB Boltzmann’s con- cal/temperature excursion factor fM is represented by
stant, r(T ) the number of times the temperature history increases ⎡

⎤
or decreases across the reversal temperature, T the period of a exp G1 T1 − T1b + exp −G2 T1 − T1b
year, and n and m are model constants similar to those in the fM = ⎣ ⎦
D0
Coffin–Manson equation.
4) Backsheet Degradation Model: The model is used to es- × exp (JΔT ). (51)
timate a potential form of the degradation kinetics of the back-
sheet. This kinetic model was applied by Kempe [58] to model The first term in the bracket reflects the stresses arising from dif-
the uncertainty in a 25-year equivalent test for module backside ferences in expansion coefficients of bonded materials. The con-
exposure to irradiance and temperatures in different climatic stants G1 , G2 , D0 , and Tb are chosen to represent the estimated
zones. The degradation rate model is written as magnitudes of these fatigue effects. The factor exp(JΔT ) esti-
mates the effect of the magnitude of the temperature excursion
T −T 0
RD ≈ I X (b + m × TOW) × (Tf ) 10 . (46) ΔT , where J is a constant
 G E − FT
Here, I is the light intensity, X, b, and m are fit parameters, fG = 1 + . (52)
TOW is the time of wetness, T the temperature, T0 a reference Go
temperature, and Tf is a multiplier for the increase in degrada- Here, E and F denote constant parameters and T is the temper-
tion for a rise in temperature in 10 K steps. ature. The frequency of the temperature excursion fω is repre-
Recommendation: As also mentioned by Kempe, the param- sented by
eter that describes the effect of time of wetness has very high  ω P − T
Q

uncertainties, we recommend careful comparison of the relative fω = 1 + . (53)

ωo
change in degradation rate with changes in TOW. In case one
wants to extract thermal parameters such as activation energy, ω is the frequency and P as well as Q are constant parameters.
the multiplier term (Tf ) can be replaced by the Arrhenius term. In a constant temperature test, T is a constant and ω is taken to
be zero. In the cyclic temperature tests, reciprocal temperature
E. Model Based on Multiple Stresses is considered to be a sinusoidal function of time
1
Since degradation of PV modules in outdoor operation is in- (t) = τ + Δτ sin(ωt) (54)
T
fluenced by multiple environmental stresses, models based on
multiple stresses are viable for outdoor service lifetime predic- 1 1 1 
τ= − . (55)
tion. 2 Tmin Tmax
1) Model of Gaines: Gaines et al. [37] proposed a model for Tmin and Tmax are the minimum and maximum temperatures
power output degradation based on multiple accelerated envi- associated with the temperature cycles.
ronmental stresses. The model suggested is Recommendation: The model of Gaines presents the previous
Pmax 1
approach on multiple stress modeling, however, the user should
= [1 − RD t] β (47) take caution that this model was developed and applied on PV
Pmax(0)
modules that had a different construction from today’s modules.
RD = AfT fR H fM fG fω (48) Therefore, its application might need some modification to fit
the current PV module construction types.
where RD is the degradation rate and the factors fT , fR H , fM , 2) Model of Subramaniyan: Another model to calculate the
fG , and fω are associated with a decrease in power output due degradation rate because of combined environmental stresses
to effects of temperature T , relative humidity RH, mechanical has been proposed recently by Subramaniyan et al. [59]. The
stresses M (due to temperature differences), gaseous concen- model takes into account the effect of both static and cyclic
tration G, and the frequency of the temperature excursion. The temperature, UV radiation and relative humidity as
mathematical form of each factor is formulated to represent the  
underlying physical phenomena. An Arrhenius form is used for β1
Rate(T, ΔT, UV, RH) = β0 exp −
  kB Tmax
B
fT = exp − . (49) × (ΔTdaily )β 2 × (U Vdaily )β 3 × (RHdaily )β 4 (56)
T
where Rate(T, ΔT, U V, RH) is the reaction rate, Tmax the daily
B denotes a constant parameter and T denotes temperature. The
maximum temperature of the module [K], ΔTdaily the daily
effect of relative humidity fRH is represented by
cyclic temperature of the module [K], UVdaily the daily daytime
  C − DT average irradiance [W/m2 ], RHdaily the daily average relative
1 1
fRH = 1 + (RH)0 exp C0 − . (50) humidity [%], and k is the Boltzmann constant. The model
TC T0
parameters β0 , which is the frequency factor [s−1 ], β1 , the acti-
The second term in the bracket corrects the relative humid- vation energy [eV], β2 , the effect of cyclic temperature, β3 , the
ity as a function of temperature, given a specified relative hu- effect of UV radiation and β4 , the effect of RH, can be estimated
midity at T0 . C and D are constant parameters. The mechani- from measured data through data fitting techniques. In the scope
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE JOURNAL OF PHOTOVOLTAICS

of this paper, the model is presented but has not been applied or more advanced models. In case of ARIMA, the time-series has
tested in any way due to a lack of combined stress data. A more to be stationary. The model parameters need to be chosen based
detailed description and application of the model can be found on the time-series behavior. That means that PV-related param-
in [59]. eters, like the varying temperature dependency of different PV
technologies, or the prevailing weather conditions have to be
taken into account.
VI. DISCUSSION AND CONCLUSION Unfortunately, it is not possible to determine if a (or which)
In this paper, several PV system performance loss method- degradation mode occurs on the basis of calculated PLRs. Apart
ologies are reviewed. Hereby, statistical and analytical models from degradation modes, other factors such as shading or soiling
are taken into account. might be a reason for a reduced performance. Because of that,
First, a discussion about statistical models to determine the it is important to not only study the data of a PV system, but
PLR of PV systems from available outdoor data is presented. also to undertake regularly visual and electrical inspections and
The performance loss trend is retrieved by applying filters, connect the findings with the calculated PLR. An idea of how
performance metrics, and statistical models on data sets. By to isolate the occurrence of degradation modes within a module
performing detailed in-depth performance studies, it might be under surveillance might be the application of the presented
possible to gain a greater understanding about the root causes models on the short-circuit current, the open-circuit voltage and
of the decrease in the power output of PV systems over time. the fill factor. Hence, it might be possible, together with the
Especially when considering current and voltage behavior, spe- inclusion of results from the studies of accelerated tests, to find
cific degradation modes could be identified and at a later stage patterns in the trends of these values.
verified by visual inspection techniques. In the second part of this work, several analytical models
As a first measure, appropriate filtering techniques have to for specific degradation modes, which trigger the aforemen-
be applied on the data set in question. The choice of the filter tioned performance losses to a large extent, are further studied.
will strongly depend on the performance metric and/or statis- The models proposed for corrosion and PID were implemented
tical model and, in case of an inappropriate filter window, will to simulate the maximum power degradation in three climatic
falsify the final outcome. On the other hand, pretreatment of zones. A strong influence of the climate is evident in all the sim-
the data set is necessary to eliminate outliers, noise, and mini- ulation results where, as expected, a more severe degradation is
mize seasonal oscillation. Before deciding which performance predicted for arid climates. The key observations in this study
metric or statistical model to use, the PV system technology, the are as follows.
length of the observed period, climatic conditions, and mounting 1) Although these models give a preliminary approximation
system (rack, tracker) should be taken into consideration. Pre- of the time-evolution of power performance, they do not
vailing seasonality, temperature/irradiance dependency of the provide any information on the physical processes taking
I–V curve parameters and noncorrelated outliers (data errors, place within the module.
shading effects, etc.) will increase the uncertainty of the re- 2) The models are developed based on numerous assump-
sults and influence the final PLR. The final aim is to receive tions and simplifications, moreover the hypotheses of a
a clear performance trend. In Section IV-C, the discussed sta- particular degradation mode are modeled depending on
tistical models have been applied on monitored field data of environmental stress factors and do not take into consid-
one monocrystalline silicon and one amorphous silicon system eration the influence of material parameters.
to retrieve long-term performance trends. Thereby, the applica- 3) None of the models is universal, that is, they can well
tion of SLR resulted in performance ratings with the greatest describe the degradation of a specific type of a PV tech-
uncertainties in comparison. The usage of CSD produced per- nology and fail on the other. Therefore, it is necessary to
formance rates with low uncertainties but due to the elimination be certain that a chosen degradation model is valid for a
of the first and last months of monitored data through the cen- specific application.
tered moving average, this technique is not recommended for 4) The analytical models are developed and validated based
data sets, which just consider short time periods. It was seen on indoor data from accelerated tests. Although some au-
that CSD overestimated the performance loss of both systems. thors went on to simulate outdoor conditions based on
The remaining techniques, namely HW exponential smoothing, indoor observations, a big challenge remains of how to
ARIMA, and STL, are performed on a similar high level of ac- interpret the results for multiple environmental stresses
curacy and the results are almost identical. HW experiences a using indoor data.
slightly higher uncertainty when applied to the amorphous sili- According to these observations, we recommend further
con PV system. It seems that ARIMA and STL are better suited developments for models that take into account both material
for noncrystalline PV systems due to their favorable treatment and multiple environmental stress factors. The development of
of the temperature behavior of the system in question. These such models need to be related to indoor as well as outdoor
three techniques exclude the seasonal part in time-series of PV observations.
performance metrics, which is an important modeling step in or-
der to receive a clear performance trend. Nevertheless, it has to
be stressed that the application of statistical models, especially ACKNOWLEDGMENT
ARIMA, has to be performed with great care and that it is not a This project has received funding from the European Union’s
rudimentary exercise to retrieve accurate model parameters for Horizon 2020 research and innovation programme in the frame-
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

LINDIG et al.: REVIEW OF STATISTICAL AND ANALYTICAL DEGRADATION MODELS FOR PHOTOVOLTAIC MODULES AND SYSTEMS 13

work of the project “SolarTrain” under the Marie Sklodowska- [24] C. Sax, Seasonal: R Interface to X-13-ARIMA-SEATS, R package version
Curie GA No 721452. 1.6.1. 2017. [Online]. Available: https://CRAN.R-project.org/package=
seasonal
[25] A. Phinikarides, G. Makrides, and G. E. Georghiou, “Comparison of
REFERENCES analysis method for the calculation of degradation rates of different pho-
tovoltaic technologies,” in Proc. 27th EU PVSEC Proc., 2013, pp. 3211–
[1] C. S. Solanki, Solar Photovoltaics - Fundamentals, Technologies and 3215.
Applications vol. 2. New Delhi, India: PHI Learning Private Limited, [26] R. B. Cleveland, W. S. Cleveland, J. E. McRae, and I. Terpenning, “STL:
2012. A seasonal-trend decomposition procedure based on loess,” J. Official
[2] I. M. Peters, H. Liu, T. Reindl, and T. Buonassisi, “Global prediction Statist., vol. 6, no. 1, pp. 3–33, 1990.
of photovoltaic field performance differences using open-source satellite [27] Evaluation of Measurement Data—Guide to the Expression of Uncertainty
data,” Joule, vol. 2, no. 2, pp. 307–322, Feb. 2018. in Measurement,”International Bureau of Weights and Measures, BIPM-
[3] M. Kontges et al., Assessment of Photovoltaic Module Failures in the JCGM-100:2008, Sevres, FR, Guide, 2008.
Field. Report IEA PVPS T13-09:2017, Subtask 3, 2017. [28] E. Hasselbrink et al., “Validation of the PVLife model using 3 million
[4] D. Moser et al., “Identification of technical risks in the photovoltaic value module-years of live site data,” in Proc. IEEE 39th Photovolt. Spec. Conf.,
chain and quantification of the economic impact,” Prog. Photovolt.: Res. Jun. 2013, pp. 0007–0012.
Appl., vol. 25, no. 7, pp. 592–604, 2017. [29] D. C. Jordan, C. Deline, S. R. Kurtz, G. M. Kimball, and M. Anderson,
[5] International Electrotechnical Vocabulary. Chapter 191: Dependability “Robust PV degradation methodology and application,” IEEE J. Photo-
and Quality of Service, IEC60050-191, International Electrotechnical volt., vol. 8, no. 2, pp. 525–531, Mar. 2018.
Comission, Geneva, CH, Standard, 1990. [30] A. Ndiaye et al., “Degradations of silicon photovoltaic modules: A liter-
[6] D. C. Jordan and S. R. Kurtz, “Photovoltaic degradation rates - An an- ature review,” Solar Energy, vol. 96, pp. 140–151, Oct. 2013.
alytical Review: Photovoltaic degradation rates,” Prog. Photovolt.: Res. [31] N. Park, C. Han, and D. Kim, “Effect of moisture condensation on long-
Appl., vol. 21, no. 1, pp. 12–29, Jan. 2013. term reliability of crystalline silicon photovoltaic modules,” Microelec-
[7] A. Phinikarides, N. Kindyni, G. Makrides, and G. E. Georghiou, “Review tron. Rel., vol. 53, no. 12, pp. 1922–1926, Dec. 2013.
of photovoltaic degradation rate methodologies,” Renew. Sustain. Energy [32] M. D. Kempe et al., “Acetic acid production and glass transition concerns
Reviews, vol. 40, pp. 143–152, Dec. 2014. with ethylene-vinyl acetate used in photovoltaic devices,” Solar Energy
[8] D. Moser, M. Pichler, and M. Nikolaeva, “Filtering procedures for reliable Mater. Solar Cells, vol. 91, no. 4, pp. 315–329, Feb. 2007.
outdoor temperature coefficients in different photovoltaic technologies | [33] K. Whitfield, A. Salomon, S. Yang, and I. Suez, “Damp heat versus field
journal of solar energy engineering | ASME DC,” J. Solar Energy Eng., reliability for crystalline silicon,” in Proc. 38th IEEE Photovolt. Spec.
vol. 136, no. 2, pp. 021006-1–021006-10, 2013. Conf., Jun. 2012, pp. 1864–1870.
[9] C. M. Whitaker et al., “Application and validation of a new PV per- [34] A. Masuda, N. Uchiyama, and Y. Hara, “Degradation by acetic acid for
formance characterization method,” in Proc. Conf. Record 26th IEEE crystalline Si photovoltaic modules,” Jpn. J. Appl. Phys., vol. 54, no. 4S,
Photovolt. Spec. Conf., Sep. 1997, pp. 1253–1256. pp. 04DR04-1–04DR04-6, Mar. 2015.
[10] G. H. Yordanov, “Relative efficiency revealed: Equations for k1-k6 of the [35] R. Pan, J. Kuitche, and G. Tamizhmani, “Degradation analysis of solar
PVGIS model,” in Proc. IEEE 40th Photovolt. Specialist Conf., Jun. 2014, photovoltaic modules: Influence of environmental factor,” in Proc. Annu.
pp. 1393–1398. Rel. Maintainability Symp., Jan. 2011, pp. 1–5.
[11] D. L. King, J. A. Kratochvil, and W. E. Boyson, “Photovoltaic array [36] B. Braisaz, C. Duchayne, M. Van Iseghem, and K. Radouane, “PV aging
performance model,” Tech. Rep. SAND2004-3535, 919131, Sandia Nat. model applied to several meteorological conditions,” in Proc. 29th PVSEC
Lab., Albuquerque, NM, USA, Aug. 2004. Proc., Amsterdam, The Netherlands, Sep. 2014, pp. 2303–2309.
[12] E. Skoplaki and J. A. Palyvos, “On the temperature dependence [37] G. B. Gaines et al., “Development of an accelerated test design for
of photovoltaic module electrical performance: A review of effi- predicting the service life of the solar array at mead, nebraska,” Quarterly
ciency/power correlations,” Solar Energy, vol. 83, no. 5, pp. 614–624, May Techn. Rep., NASA, USA, Feb. 1979.
2009. [38] L. A. Escobar and W. Q. Meeker, “A review of accelerated test models,”
[13] D. C. Jordan, J. H. Wohlgemuth, and S. R. Kurtz, “Technology and climate Statist. Sci., vol. 21, no. 4, pp. 552–577, Nov. 2006.
trends in PV module degradation,” in Proc. 27th EU PVSEC Proc., 2012, [39] R. Laronde, A. Charki, D. Bigaud, and P. Excoffier, “Reliability evaluation
pp. 3118–3124. of a photovoltaic module using accelerated degradation model,” in Proc.
[14] T. Huld et al., “A power-rating model for crystalline silicon PV mod- SPIE, 2011, vol. 8112.
ules,” Solar Energy Mater. Solar Cells, vol. 95, no. 12, pp. 3359–3369, [40] M. Koehl, M. Heck, and S. Wiesmeier, “Modelling of conditions for
Dec. 2011. accelerated lifetime testing of Humidity impact on PV-modules based on
[15] C. Jennings, “PV module performance at PG&E,” in Proc. Conf. Rec. 20th monitoring of climatic data,” Solar Energy Mater. Solar Cells, vol. 99,
IEEE Photovolt. Spec. Conf., Sep. 1988, pp. 1225–1229. pp. 282–291, Apr. 2012.
[16] C. Whitaker, T. Townsend, and H. Wenger, “Effects of irradiance and [41] J. Berghold et al., “Potential-induced degradation (PID) and its correlation
other factors on PV temperature coefficients,” in Proc. Conf. Rec. 22nd with experience in the field,” Photovolt. Int., vol. 19, pp. 85–92, 2013.
IEEE Photovolt. Spec. Conf., Oct. 1991, pp. 608–613. [42] M. Kontges et al., “Review of failures of photovoltaic modules,” Report
[17] Photovoltaic System Performance Monitoring—Guidelines for Measure- IEA PVPS T13-01:2014, Subtask 3.2, 2014.
ment, Data Exchange and Analysis, IEC61724:1998, International Elec- [43] P. Hacke et al., “Elucidating PID degradation mechanisms and in-situ dark
trotechnical Comission, Geneva, CH, Standard, 1998. I-V monitoring for modeling degradation rate in CdTe thin-film modules,”
[18] G. Belluardo et al., “Novel method for the improvement in the evaluation IEEE J. Photovolt., vol. 6, no. 6, pp. 1635–1640, Nov. 2016.
of outdoor performance loss rate in different PV technologies and com- [44] P. Hacke et al., “Accelerated testing and modeling of potential-induced
parison with two other methods,” Solar Energy, vol. 117, pp. 139–152, degradation as a function of temperature and relative humidity,” IEEE J.
Jul. 2015. Photovolt., vol. 5, no. 6, pp. 1549–1553, Nov. 2015.
[19] D. C. Jordan and S. R. Kurtz, “Analytical improvements in PV degradation [45] V. Naumann et al., “Potential-induced degradation at interdigitated back
rate determination,” in Proc. 35th IEEE Photovolt. Spec. Conf., Jun. 2010, contact solar cells,” Energy Procedia, vol. 55, pp. 498–503, Jan. 2014.
pp. 688–693. [46] A. Halm et al., “Potential-induced degradation for encapsulated n-type
[20] C. C. Holt, “Forecasting seasonals and trends by exponentially weighted IBC solar cells with front floating emitter,” Energy Procedia, vol. 77,
moving averages,” Int. J. Forecasting, vol. 20, no. 1, pp. 5–10, Jan. pp. 356–363, Aug. 2015.
2004. [47] R. Swanson et al., “The surface polarization effect in high-efficient silicon
[21] R. J. Hyndman and G. Athanasopoulos, Forecasting: Principles and Prac- solar cells,” in Proc. 15th Int. PVSEC Proc., Jan. 2005.
tice. Melbourne, Australia: OTexts, 2013. [48] D. Lausch et al., “Sodium outdiffusion from stacking faults as root cause
[22] A. Phinikarides, G. Makrides, N. Kindyni, A. Kyprianou, and G. E. for the recovery process of potential-induced degradation (PID),” Energy
Georghiou, “ARIMA modeling of the performance of different pho- Procedia, vol. 55, pp. 486–493, Jan. 2014.
tovoltaic technologies,” in Proc. 39th IEEE Photovolt. Spec. Conf., [49] E. Annigoni et al., “Modeling potential-induced degradation (PID) in
Jun. 2013, pp. 0797–0801. crystalline silicon solar cells: From accelerated-aging laboratory testing
[23] S. G. Makridakis and S. C. Wheelwright, Forecasting, 3rd ed. New York, to outdoor prediction,” in Proc. 32nd EU PVSEC Proc., Munich, Germany,
NY, USA: Wiley, Jan. 1998. Jun. 2016, pp. 1558–1563.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

14 IEEE JOURNAL OF PHOTOVOLTAICS

[50] J. Hattendorf et al., “Potential induced degradation in mono-crystalline David Moser received the B.Sc. and M.Sc. degrees
silicon based modules: An acceleration model,” in Proc. 27th Eur. PVSEC from the University of Trento, Italy, in 2003 and 2006,
Proc., Oct. 2012, pp. 3405–3410. respectively, the Ph.D. degree in physics from the Sal-
[51] C. Taubitz, M. Krber, M. Schtze, and M. B. Koentopp, “Potential induced ford University, Salford, U.K., in 2010.
degradation: Model calculations and correlation between laboratory tests He coordinates the activities of the research group
and outdoor occurrence,” in Proc. 29th Eur. PVSEC Proc., Nov. 2014, Photovoltaic Systems, Institute for Renewable En-
pp. 2490–2494. ergy, EURAC, Bolzano, Italy. His work focuses on
[52] L. Dunn, M. Gostein, and B. Stueve, “Literature review of the effects of characterizing indoor and outdoor behavior, perfor-
UV exposure on PV modules,” in NREL PV Module Reliab. Workshop, mance, and reliability of photovoltaic (PV) modules
Feb. 2013. and systems, building integration of PV systems, and
[53] M. A. Munoz, M. C. Alonso-Garcia, and Vela, “Early degradation of monitoring of outdoor PV plants. He is also active in
silicon PV modules and guaranty conditions,” Solar Energy, vol. 85, no. 9, PV potential studies on a regional scale and member of the Board of Directors
pp. 2264–2274, 2011. of The Association of European Renewable Energy Research Centres.
[54] M. Koehl, D. Philipp, N. Lenck, and M. Zundel, “Development and ap-
plication of a UV light source for PV-module testing,” in Proc. SPIE - Int
Soc. for Opt. Eng. 7412, Aug. 2009, pp. 7412021–7412027.
[55] X. Gu, Y. Lyu, L.-C. Yu, C.-C. Lin, and D. Stanley, “Effect of inten-
sity and wavelength of spectral UV light on discoloration of laminated
Glass/EVA/PPE PV module,” in 3rd Atlas-NIST Workshop PV Mater.
Durability, Dec. 2015.
[56] D. C. Jordan, T. J. Silverman, J. H. Wohlgemuth, S. R. Kurtz, and K. T.
VanSant, “Photovoltaic failure and degradation modes,” Progress Photo-
voltaics: Res. Appl., vol. 25, pp. 318–326, 2017.
[57] N. Bosco, T. J. Silverman, and S. Kurtz, “Climate specific thermomechan- Karl-Anders Weiß received the Diploma degree
ical fatigue of flat plate photovoltaic module solder joints,” Microelectron. in physics and economics and the Ph.D. degree in
Rel., vol. 62, pp. 124–129, Jul. 2016. physics from the University of Ulm, Germany, in
[58] M. D. Kempe, “Evaluation of the uncertainty in accelerated stress testing,” 2005 and 2014, respectively.
in Proc. IEEE 40th Photovolt. Spec. Conf., Jun. 2014, pp. 2170–2175. He is Head of the Service Life Analysis Group,
[59] A. B. Subramaniyan, R. Pan, J. Kuitche, and G. TamizhMani, “Quantifica- Fraunhofer Institute for Solar Energy Systems,
tion of environmental effects on PV module degradation: A physics-based Freiburg, Germany. His areas of interest include
data-driven modeling method,” IEEE J. Photovolt., vol. 8, no. 5, pp. 1289– degradation of materials in solar applications, accel-
1296, Sep. 2018. erated testing of components and materials of solar
systems, numerical simulations, methods to analyze
degradation of polymers, nondestructive analytical
Sascha Lindig received the B.Eng. degree in pho- methods climatic loads and service life prediction.
tovoltaic and semiconductor technologies from the
University of Applied Science Jena, Jena, Germany,
in 2011, and the M.Eng. degree in environmental and
energy engineering from the Leipzig University of
Applied Sciences, Leipzig, Germany, in 2014. He is
currently working toward the Ph.D. degree in elec-
trical engineering with the University of Ljubljana,
Ljubljana, Slovenia, in collaboration with EURAC
Research, Bolzano, Italy.
He is working in the frame of the Marie
Sklodowska Curie SOLAR-TRAIN project on statistical performance loss mod- Marko Topic received the Ph.D. degree from the
els of PV systems, degradation patterns and solar economics. University of Ljublana, Slovenia, in 1996.
He is the Head of Laboratory of Photovoltaics
and Optoelectronics, University of Ljubljana, Ljubl-
jana, Slovenia, acts as the Chairman of the European
Ismail Kaaya was born in Uganda. He received the
Technology and Innovation Platform Photovoltaics
B.S. degree in applied physics from the International
(ETIP-PV) since 2016 and previously as the Chair-
University of Africa, Khartoum, Sudan, in 2013, the
man of European Photovoltaic Technology Platform
Postgraduate Diploma degree in condensed matter (since 2014). He is currently a full Professor with the
physics from the Abdus-Salam International Center
Faculty of Electrical Engineering at the University
for Theoretical Physics, Trieste, Italy, in 2015, and the
of Ljubljana, and he has a very broad research ex-
M.S. degree in renewable energy science and technol-
perience in photovoltaics, thin-film semiconductor materials, electron devices,
ogy from Ecole Polytechnique, Palaiseau, France, in optoelectronics, electronic circuits, and reliability engineering.
2016. He is currently working toward the Ph.D. de-
Dr. Topic is a member of the Slovenian Academy of Engineering and has
gree in photovoltaics with Fraunhofer Institute for
received several prestigious awards, including the “Zoisova nagrada” in 2008
Solar Energy Systems, Freiburg, Germany.
(the highest award of the Republic of Slovenia for Scientific and Research
His research interests includes advanced characterization of hydrogenated
Achievements).
amorphous silicon, modeling spatial inhomogenities/(microscopic defects) in
solar cells, and development of service life models for PV modules.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE JOURNAL OF PHOTOVOLTAICS 1

Modeling Outdoor Service Lifetime Prediction of PV

Modules: Effects of Combined Climatic Stressors
on PV Module Power Degradation
Ismail Kaaya , Michael Koehl, Amantin Panos Mehilli, Sidrach de Cardona Mariano, and Karl Anders Weiss

Abstract—Photovoltaic modules are exposed to a variety of

climatic loads during outdoor operation. Over time, these loads
trigger a number of degradation modes within the modules leading
to performance loss. This paper quantifies the impact of combined
climatic loads on the module’s maximum power output using a
mathematical approach. Three degradation precursor reactions,
namely, hydrolysis, photodegradation, and thermomechanical
degradation, are assumed to be necessary for service lifetime pre-
diction. For each reaction, an empirical kinetics model is proposed
and validated with indoor test measurements. A generalized model
to quantify the effects of combined climatic loads is proposed. The
generalized model is calibrated and validated using outdoor test
measurements. The model is then applied to predict the annual
degradation rates and a 20% performance loss of three identical
monocrystalline modules installed in three benchmarking cli- Fig. 1. Schematic diagram of the modeling hypotheses.
mates: maritime (Gran Canaria, Spain), arid (Negev, Israel), and
alpine (Zugspitze, Germany) using real monitored meteorological
data. A degradation of 0.74%/year corresponding to 21.4 years et al. [2] have proposed a model to link the module degradation
operation time was predicted as the highest for an arid environ- path and environmental variables. According to our previous
ment, compared with 0.50%/year and 0.3%/year degradation for publication [3], the current state-of-the-art degradation models
maritime and alpine environments, respectively. The proposed available for PV modules and systems were reviewed. Among
models will find applications in outdoor predictions as well as in
the combined stress accelerated tests to develop test designs. the key observations was that most of the degradation models
are developed for specific degradation modes using controlled
Index Terms—Climatic zones and service lifetime prediction, test conditions and are validated based on indoor measurements
degradation model, degradation modes, photovoltaic (PV) module.
from accelerated tests. Since a PV module in outdoor operation
experiences numerous climatic loads, which, in turn, might lead
I. INTRODUCTION to different degradation modes, combined stress models are a
prerequisite to estimate PV module degradation.
HE search for a combined stress model for photovoltaic
T (PV) module lifetime prediction dates back to the 1970s,
for example, Gaines et al. [1] proposed a quantitative model for
The main motivation of this paper is to bridge the gap toward
service lifetime prediction of PV modules in outdoor operation.
A combined stress model is proposed based on the physics of
accelerated testing using multiple environmental stresses, which
failure. The approach deployed in this paper is based on ana-
was used to develop the test design. Recently, Subramaniyan
lyzing and modeling degradation modes under various climatic
stresses. The input stresses are assumed to be responsible for
Manuscript received January 31, 2019; revised March 21, 2019 and April 26, triggering a specific reaction that might induce specific degra-
2019; accepted May 3, 2019. This work was supported in the framework of dation modes (see Fig. 1). The effect of the applied stresses has
the European Union’s Horizon 2020 Marie Skłodowska-Curie Actions Program
(SOLAR-TRAIN project) under Grant 721452. (Corresponding author: Ismail been evaluated with experimental data using accelerated age-
Kaaya.) ing tests. We assume that it is a crucial step to first evaluate the
I. Kaaya is with the Fraunhofer Institute of Solar Energy, 79110 Freiburg, effects of different stresses in controlled conditions using accel-
Germany, and also with the University of Malaga, 29016 Malaga, Spain
(e-mail: [email protected]). erated tests. This is needed to correlate the power degradation
M. Koehl, A. P. Mehilli, and K. A. Weiss are with the Fraunhofer Insti- to specific degradation modes using different characterization
tute of Solar Energy, 79110 Freiburg, Germany (e-mail: Michael.koehl@ise. methods. It also helps to understand the physics of failure of the
fraunhofer.de; [email protected]; karl-anders.weiss@
ise.fraunhofer.de). different degradation modes.
S. de Cardona Mariano is with the University of Malaga, 29016 Malaga, Three main aspects need to be considered for a modeling
Spain (e-mail: [email protected]). approach applicable for PV service lifetime prediction.
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. 1) Impact of PV material variations: New materials are pro-
Digital Object Identifier 10.1109/JPHOTOV.2019.2916197 posed frequently to improve PV performance.
2156-3381 © 2019 EU
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE JOURNAL OF PHOTOVOLTAICS

2) Different operating climatic conditions: PV modules op- be to use the induction and the onset of the degradation phase
erate in different climates, for example, in arid, maritime, as it might be a good representation of natural degradation.
tropical climates.
3) Different PV technologies, for example, crystalline sili- A. Module Output Power Degradation Function
con (c-Si), thin films, and different module designs like Since the module power at the maximum power point (PMPP )
bifacial, glass–glass, or glass–backsheet. is a quantity commonly used by manufacturers to set warranties,
All these factors might lead to different degradation modes, it has been selected for use as a degradation indicator in this pa-
rates, and performance degradation. To our knowledge, it is a per. Hence, in this context, degradation is defined as the gradual
rather complex challenge to have a single model taking into deterioration in the module PMPP over time. The effects of ap-
account all the above aspects. In this paper, a simplified approach plied climatic stresses are then quantified by how much they lead
is proposed, as a first approximation to take into account some to a reduction in the initial power over time. The module output
of the above aspects in a single model. power as a function of time is proposed as
  μ 
PMPP (t) B
II. POWER DEGRADATION FUNCTION AND DEGRADATION = 1 − exp − (1)
PMPP (0) ki t
RATE MODELS
where PMPP (t) and PMPP (0) are the module output power at
The fact that the module is exposed to several stress factors time t and the initial output power, respectively. B is the power
at the same time, and that outdoor conditions have a stochastic susceptibility, which is assumed to be a material property, μ is
nature, makes it difficult to model outdoor conditions. A model the shape parameter, and ki is the degradation rate constant of
that includes all the above dependences requires a huge number degradation process i.
of unknown coefficients to be evaluated. Some of the required
coefficients might require experimental procedures to be evalu- B. Degradation Rate Models for Controlled Indoor Conditions
ated, making it time consuming and expensive.
One way to overcome these obstacles is to use an approach that Empirical kinetics models to evaluate the degradation rate
minimizes the number of coefficients to be evaluated. Assump- constant ki are proposed depending on the applied stresses as
tions should be made based on the dominating loads, processes, presented in the following general reaction equation:
impacts, or mechanisms for degradation evaluation. Constant or Stress A + Stress B + . . . Stress N → Degradation
average quantities might be used to compute the average rates, in
order to overcome the stochastic behavior of input parameters. precursor. (2)
Using accelerated tests to quantify the impact of applied loads In this paper, three main degradation precursor reactions are
on power degradation and to evaluate the sensitivity of model assumed to be hydrolysis, photodegradation, and thermome-
parameters for varying controlled test conditions helps to start chanical degradation. Depending on the applied stresses, a reac-
with modeling before investigating outdoor conditions. This also tion constant is evaluated to quantify the impact of the applied
helps to relate the performance losses to specific degradation stresses on power degradation over a specified time period.
modes. However, one of the major pitfalls of accelerated testing 1) Hydrolysis-Driven Degradation Due to Temperature and
is that it may focus on one degradation mode while masking oth- Relative Humidity:
ers. The masked degradation modes could be the first to show  
n EH
up or dominate in the field under different operating conditions kH = AH .rheff .exp − (3)
[4]. Therefore, the underlying physics and chemistry related to kB .Tm
the degradation must be known to avoid these obstacles. The where kH is the rate constant, kB is the Boltzmann constant
assumptions and hypotheses used in this paper are based on the (8.62 × 10−5 ), Tm is the module temperature (Kelvin), AH is
prior knowledge and studies [5]–[9], [11]–[15], which deal with the pre-exponential constant, rheff (%) is the effective module
the underlying physical and or chemical degradation kinetics. relative humidity (RH) proposed by Koehl et al. [16], and n is
Fig. 1 summarizes the hypothesized degradation mechanisms a model parameter that indicates the impact of RH on power
that are known to be induced by the applied loads. degradation. In this context, EH is defined as the activation en-
The models should be calibrated using nonextreme test con- ergy for power degradation. Equation (3) is the commonly used
ditions in order to obtain a good correlation of the indoor accel- Peck’s model, which evaluates the degradation impact due to
erated conditions to outdoor conditions. As already mentioned, RH and temperature [17]. The Peck’s model was selected based
very extreme conditions might generate new degradation modes on the study of our previous article [3], since the model showed
that might never happen outdoors [4]; using these conditions for the best performance in DH result calibration.
calibration of the model might result in high uncertainties. More- 2) Photodegradation Due to UV Dose, Temperature, and Rel-
over, some of the degradation patterns observed in accelerated ative Humidity:
tests are unlikely to occur in natural aging leading to misin-  
EP
terpretations. For example, during an extended damp heat (DH) kP = AP .(UVdose )X . (1 + rhneff ) .exp − (4)
test, three phases of power degradation were described by Koehl kB .Tm
et al. [15] as induction, degradation, and saturation phases. One where kP is the rate constant, UVdose is the integral UV dose
way to use such a degradation pattern for outdoor ageing would (kW/m2 ), EP is the activation energy for power degradation due
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

KAAYA et al.: MODELING OUTDOOR SERVICE LIFETIME PREDICTION OF PV MODULES 3

TABLE I
EXPERIMENTAL (EXP) CONDITIONS FOR DH COMBINED UV/DH
AND TC TESTS

Fig. 2. Schematic diagram showing rates in controlled indoor conditions and

the synergistic nature of outdoor conditions.

the failure time can be calculated using the following relation:

to photoreaction, and X is a model parameter that indicates the
impact of UV dose on power degradation. B
tf = 1 . (8)
3) Thermomechanical Degradation Due to Temperature kT × [|log (0.2)|] µ
Cycles: The most commonly used model for thermal cycling
(TC) is the Coffin–Manson relationship. According to Escobar III. MODEL CALIBRATION
and Meeker [17], the effect of temperature cycling can depend
Two basic approaches were applied for the calibration of
largely on the maximum temperature TU . The cycle-to-failure
the different degradation rate models: 1) optimization of model
distribution for temperature cycling can also depend on the cy-
performances; and 2) through prior knowledge from previous
cling rate (e.g., due to heat buildup). Therefore, a modified form
studies. Optimization of model performance, which compares
of the Coffin–Manson relationship including this effect is
  measured and simulated data, was applied with the help of a
θ ET m built-in nonlinear least-squares solver in the GNU Octave soft-
kT m = AT m .(ΔT ) .CN .exp − (5)
kB .TU ware. Prior knowledge, with the aid of sensitivity analysis, was
used as a baseline to select the initial fitting guesses and also as
where ΔT = (TU − TL ) is the temperature difference (Kelvin),
a confirmation that the extracted values are in meaningful range.
CN is the cycling rate, TU and TL are the module upper and lower
limit temperatures, and ET m is the activation energy of power
A. Experimental Indoor
degradation.
Distributed DH, TC, and combined DH-UV stress tests were
C. Degradation Rate Model for Outdoor Conditions carried out at different test conditions under the framework of
the SOPHIA project at different test laboratories. Table I shows
The transition from indoor degradation rate evaluation to out-
the different test conditions used in this paper for model cali-
door is a challenging task due to not yet knowing enough how
bration and validation. The studied modules are from the same
different stresses, degradation processes, and the induced degra-
manufacturer. They are P-type homojunction c-Si with thermo-
dation interact.
plastic encapsulant material and without an aluminium layer as
The underlying assumption deployed in this paper is that some
additional moisture barrier.
degradation processes might lead to specific degradation modes
independent of the others, and that some might have a synergistic
nature (see Fig. 2), which results in a variety of degradation B. Experimental Outdoor
modes. Hence, this assumption allows us to evaluate the total Three identical monocrystalline silicon (mc-Si) modules are
degradation rate as the sum of both independent and dependent under monitoring in three climatic zones: maritime (in Gran
processes. The mathematical form of the total rate is expressed Canaria, Spain), arid (in Negev, Israel), and alpine (in UFS
as Zugspitze, Germany). In Gran Canaria, the tilt angle is 23°, and
the azimuth angle 169° for the PV module. In Negev, the tilt
kT = AN . (1 + kh ) (1 + kp ) (1 + kT m ) − 1 (6)
angle is 31°, and the azimuth angle 180° for PV modules. The
n
 module in Gran Canaria has been exposed for over seven years,
kT = AN . (1 + ki ) − 1 (7) and the ones in Negev and Zugspitze have been exposed over
i=1
five years outdoors.
where kT (%/year) is the total degradation rate (%/year), ki Apart from the performance measurements, the modules tem-
is the ith rate constant, and n is the total number of degrada- peratures are also recorded every 10 min. The sensors for mea-
tion processes. AN is the normalization constant of the physical suring module temperatures are located under one of the central
quantities; in this case, it takes the units (year−2 /%). cells. They are Pt100 sensors, which are attached from the back
Using (1) and (7) and defining failure time (tf ) as a 20% loss using adhesive aluminum tapes. Other meteorological data such
in maximum power output (common manufacturer’s warranty), as RH, global irradiation, UV irradiance, and wind speed are
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE JOURNAL OF PHOTOVOLTAICS

Fig. 3. Distribution of module temperature in the three climatic zones.

Fig. 5. Ten minutes values of module temperature for five years. The dotted
lines show the mean maximum and minimum temperatures.

requires a huge amount of data and experimental campaigns.

Although the model parameters are technology or module spe-
cific, the formulations presented in this paper are flexible and
can be adapted for application in other modules designs and
technologies.
On the one hand, extracting model parameters for each eval-
uated PV module type makes degradation prediction more com-
plex and rather expensive and time consuming. In this case, a
model parameter B in (1), the so-called power susceptibility, is
introduced. It could allow simulating a percentage increase in
performance due to an improvement in materials of different PV
modules when other model parameters are kept constant.
Fig. 4. Total UV dose and average annual RH for the three zones measured
for five years. On the other hand, we assume that the power degradation
shapes could be linked to module technologies or module de-
TABLE II sign. For example, a faster degradation at early stages of expo-
SUMMARY OF FIVE YEARS AVERAGE CLIMATIC INPUTS USED IN SIMULATION
sure and followed by stabilization is commonly observed in thin-
film modules [18]. The other speculation is made on the module
design, for example, glass–glass modules have less moisture
pathways and moisture ingress compared with glass–backsheet
modules. This means that moisture-induced degradation modes
are slower at the earlier stages of the module lifetime. How-
ever, as the breathable pathways and drying are also limited, the
moisture will accumulate over the years, leading to a dramatic
increase in the degradation rate. For this case, one could expect
also under monitoring in all the three zones with a 1-min reso-
a degradation shape like that in Fig. 6 when μ = 0.96.
lution. The annual averages of UV dose and RH are as shown in
The proposed power degradation function includes a shape pa-
Fig. 4.
rameter μ, making it possible to optimize all the possible degra-
Five years of measured datasets were used to evaluate the
dation shapes. Fig. 6 shows degradation shapes optimization by
averages (see Table II), to ensure that the values used in degra-
changing the value of μ.
dation prediction correspond to what a module will experience
Moreover, it is very crucial to optimize the power degrada-
during its lifetime. The mean value of the module minimum and
tion shapes, since they are linked to energy yield evaluation as
maximum temperature has been computed considering upper
the yield corresponds to the integral of power with respect to
and lower temperature bins as in Fig. 5.
operation time, hence the area under the curves in Fig. 6.
IV. RESULTS AND DISCUSSION
B. Sensitivity Analysis of Input Loads
A. Model Properties A sensitivity analysis of climatic input loads of temperature,
It is rather complex to develop a generalized model that cov- RH, and UV dose on the failure time was carried out. Tempera-
ers all these requirements for taking into account innovations ture (T), RH, and UV dose bins between (15–55 °C), (40–100%),
in materials and designs. Moreover, validation of such a model and (80–100 kW/h/m2 ), respectively, were used to generate 500
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

KAAYA et al.: MODELING OUTDOOR SERVICE LIFETIME PREDICTION OF PV MODULES 5

Fig. 8. Schematic diagram of models’ calibration and validation procedure

using for both indoor and outdoor datasets.

TABLE III
EXTRACTED MODEL PARAMETERS FOR INDOOR MODULES AND THE
PERCENTAGE DEVIATION (DEV) FROM THE FITTED DATA

Fig. 6. Optimization of power degradation shapes by altering the shape

parameter µ.

Fig. 9. Hydrolysis model (3) calibration and validation. The blue patch is the
95% confidence interval of prediction; the violet line represents the optimized
Fig. 7. Sensitivity analysis of T, UV, and RH. The arrow indicates increasing data points.
RH represented by the dots.

combinations of T, RH, and UV. The variation of failure time to

each combination set is shown in Fig. 7.
According to this analysis, the model is more sensitive to
temperature as compared with RH and UV dose; this can be
explained by the Arrhenius temperature dependence in the rate
models. Moreover, during the sensitivity analysis, a threshold
of RH (highlighted in blue) was observed. This was linked to
the Peck’s model for hydrolysis. Simulations using the peck’s
model only to evaluate the failure time confirmed this tendency
at humidity levels above 80%. The model is still applicable,
since these conditions are not usually the case outdoors, and in
this paper, the varying input load was temperature for indoor
conditions. Fig. 10. Photodegradation model (4) calibration and validation. The blue patch
is the 95% confidence interval of prediction.
C. Model Validation
As for any predictive model, the crucial part is to validate for specific test conditions. The extracted rate models param-
the model. There are several ways of model validation. In this eters are presented in Table III. The parameter B in (1) was
paper, the approach used is verification with real experimental normalized to one, and the extracted shape parameter μ was 0.7
measurements. Fig. 8 shows the steps used for validation of the for Fig. 9 and 0.4 for Figs. 10 and 11.
models. It should be noted that the extracted parameters are valid only
1) Model Validation With Indoor Datasets: The proposed for a particular module type. The parameters have to be evalu-
models (3)–(5) were calibrated using indoor experimental data ated for each different module separately. The models were then
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE JOURNAL OF PHOTOVOLTAICS

TABLE IV
EXTRACTED MODEL PARAMETERS FOR OUTDOOR MODULES

Fig. 11. Thermomechanical model (5) calibration and validation. The blue
patch is the 95% confidence interval of prediction.

applied to predict the degradation when the test conditions are

varied. The model predictions are compared with experimental
outcomes for validation.
Fig. 9 shows calibration and validation results for the hy-
drolysis model (3). In black is the measured power for DH
75 °C/85RH; red is the respective model fit with a violet line Fig. 12. Combined model (7) calibration and validation with 5-min resolution.
at 5000 h representing the optimized data points. The blue line
is the predicted power at DH 85 °C/85RH, the light blue patch
is the 95% prediction confidence interval; green represents the and Zugspitze, and moreover, it shows a clear degradation
measured power at DH 85°C/85RH, used for model validation. trend. A filter at module temperature (30–35 °C) and irradiance
The vertical lines on measured data points indicate a 2.5% mea- (800–1200 W/m2 ) was used because it is the most frequent tem-
surement uncertainty. The color usage and explanation above perature that a module experiences over its lifetime in this region
are consistent with Figs. 10 and 11 for photodegradation (4) and (see Fig. 3). The irradiance bin ensures that only clear sky con-
thermomechanical (5) models using respective datasets. For all ditions were considered in order to have irradiance conditions
the models, the predictions are satisfactory and are within a 95% near to standard test conditions (STC) and to model a common
confidence interval. The observable variations could be linked situation for all the climates. The power was corrected to STC of
with the measurement uncertainties. irradiance. To make sure that the power degradation observed for
To correlate the uncertainties in model calibration with the outdoor modules is not due to soiling effects, periodic cleaning
predictions, the percentage mean square error of prediction of the modules is done.
(MSEP) [19] was evaluated as The extracted rate model parameters are presented in Table IV.
  2  The percentage deviation of the fitted data points was 2.34%
MSEP = 100 × V [Pp ] + μPp − μPm (9) and the derived parameters were B = 190 and μ = 0.19 for the
power function (1).
where Pp and Pm are the predicted and measured power, respec-
Fig. 12 shows calibration and validation results. The black
tively, V is the variance of the predicted power, and μPp and μPm
dots are the measured power for Gran Canaria; the red line is
are the mean of predicted and measured powers, respectively.
the respective model fit. The blue line is the predicted power
A 0.5% deviation led to 0.025% MSEP for the hydrolysis
for Negev, and the measured power for Negev is in green. The
model, 1.65% deviation resulted into 0.216% MSEP for the ther-
dotted lines indicate normalization to the initial laboratory power
momechanical model, and 0.19% deviation led to 0.168% MSEP
values before outdoor exposure. The alpine predictions were
for the photodegradation model. Although there is a correlation
left out to avoid too much information on the graph due to data
of the uncertainties due to model calibration as it is evaluated
fluctuations. The outdoor predictions show a good agreement
in the thermomechanical model, the uncertainties in experimen-
with the measured power degradation.
tal datasets used for validation can also influence the evaluated
mean square error; therefore, it is also useful to evaluate the
confidence interval of the prediction. D. Degradation Rates and Lifetime Prediction
2) Model Validation With Outdoor Datasets: The power Depending on the climate a module is installed in, different
degradation model (1) with ki = kT [total degradation rate (6)] degradation modes might dominate over the others. Using the
was calibrated using the monitored dataset of Gran Canaria. proposed degradation models (3)–(5) and (7) and the outdoor
The dataset from Gran Canaria was selected over Negev and derived model parameters, it is possible to predict the dominating
Zugspitze because the module in Gran Canaria has been ex- degradation precursor and the total degradation rate, as well as
posed for quite a long time compared with the ones in Negev the failure time (8) for any location with known climatic loads.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

KAAYA et al.: MODELING OUTDOOR SERVICE LIFETIME PREDICTION OF PV MODULES 7

TABLE V
PREDICTED DEGRADATION RATES USING (3)–(5) AND (7) OF THE mc-Si MODULES AND FAILURE TIME (8) IN THE THREE CLIMATIC ZONES

In this paper, annual degradation rates of the mc-Si modules

were predicted using input climatic loads of Zugspitze, Gran
Canaria, and Negev, as shown in Table V.
High thermomechanical degradation is predicted for
Zugspitze in comparison with Gran Canaria because of high
module temperature variations in this climate zone. On the
contrary, small degradation due to hydrolysis is predicted in
Zugspitze despite the high levels of RH. This could be explained
by the low average module temperatures experienced in this re-
gion, hence slowing hydrolysis processes and the absolute wa-
ter vapor concentration. In all cases, high rates are predicted in
Negev. This can be explained again by the higher temperatures in
this zone that determines the reaction rates for other degradation
processes caused by other degradation factors such as hydrolysis
by humidity and photodegradation by UV dose. The predicted Fig. 13. Combined model (7) calibration and validation with hourly resolution.
failure time defined as a 20% loss in power, as expected, shows
more severe degradation of the maximum power output in arid TABLE VI
PERCENTAGE MEAN ERROR IN PREDICTION (MSEP) AND RELATIVE
climates where temperatures are higher. This further confirms DIFFERENCE (REL-DIFF) OF THE ESTIMATED FAILURE TIME (tf )
the previous studies [20] that temperature could be the primary FOR DIFFERENT DATA RESOLUTIONS (DATA RES)
accelerator of degradation.

E. Evaluation of Uncertainties Due to Data Quality

The module power measurements were done every 5 min;
such high-resolution measurements and the frequently fluctuat-
ing environmental conditions outdoors lead to unavoidable noise
even after applying filters and corrections. By using a moving
average of 1 h to minimize the noise in the datasets, the process
V. CONCLUSION
of model calibration and validation was repeated. The effect of
the noise to the derived model parameters as well as on failure A degradation model for quantifying the impact of combined
time estimation was evaluated. Because of the high sensitivity climatic stresses on module maximum power output degrada-
of the models parameters, Ea , n, X, and θ, they were assumed tion has been introduced. Degradation rate models have been
constant, and the variations in calibration and lifetime estima- proposed and validated with indoor measurements for specific
tions due to the noise effect are evaluated using the parameter degradation precursors. A combined degradation rate model has
B of the power degradation function (1). been proposed and validated with real field datasets. The model
Fig. 13 shows the calibration and validation results for has been applied for the evaluation of the degradation rates and
hourly resolution. The deviation of the fitted data points re- the prediction of losses in the power output of monocrystalline
duced to 2.04%, and the derived model parameter reduced to modules installed in three climatic zones: maritime, arid, and
B = 182.3 alpine. A stronger degradation has been predicted in an arid
Table VI shows the MSEP and relative difference in failure climate, which could be explained by a higher mean module
time estimation with 5-min data and hourly averaged data. Al- temperature, as well as high temperature variations in this zone.
though the fitting deviation improved from 2.23% to 2.04% us- In addition, a correlation of dominating degradation pre-
ing hourly resolution, the improvement did not considerably led cursors to the operating climate has been analyzed for the
to reduction in MSEP, but the effect is visible in failure time es- three climates. Thermomechanically induced modes dominate
timation with a relative difference of 4.05%. This is consistent in Zugspitze and Negev due to high temperature variations in
with the observations from indoor results that the uncertainties these zones. In Gran Canaria, photodegradation dominates due
in experimental datasets used for validation can also influence to high UV values and relatively high average module tempera-
the evaluated MSEP. tures experienced in this region.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE JOURNAL OF PHOTOVOLTAICS

The proposed power degradation model presented in this pa- [8] A. Ndiaye et al., “Degradations of silicon photovoltaic modules: A litera-
per has been calibrated and validated using specific modules ture review,” Sol. Energy, vol. 96, pp. 140–151, Oct. 2013.
[9] A. Gok et al., “Predictive models of poly(ethylene-terephthalate) film
design and technology. However, the formulations are flexible degradation under multi-factor accelerated weathering exposures,” PLoS
and could be applied to other modules designs and technologies. ONE, vol. 12, no. 5, May 2017, Art. no. e0177614.
Apart from the uncertainties due to models’ derived parame- [10] J. Zhu et al., “Changes of solar cell parameters during damp-heat ex-
posure,” Prog. Photovolt. Res. Appl., vol. 24, no. 10, pp. 1346–1358,
ters and input datasets, some degradation modes might be out- Oct. 2016.
comes of other degradation modes and might appear at certain [11] D. C. Jordan, T. J. Silverman, J. H. Wohlgemuth, S. R. Kurtz, and K. T.
stages of a module’s lifetime. This makes the predictions more VanSant, “Photovoltaic failure and degradation modes,” Prog. Photovolt.
Res. Appl., vol. 25, no. 4, pp. 318–326, 2017.
complex using only analytical models like the one described in [12] M. L. Marín, A. Jiménez, J. López, and J. Vilaplana, “Thermal degradation
this paper, hence impacting the model accuracy especially for of ethylene (vinyl acetate),” J. Thermal Anal., vol. 47, no. 1, pp. 247–258,
longer time predictions. One way to solve this problem could Jul. 1996.
[13] N. Park, W. Oh, and D. H. Kim, “Effect of Temperature and humidity on
be to apply a combination of analytical models with data-driven the degradation rate of multicrystalline silicon photovoltaic module,” Int.
models or apply computer algorithms embedded with analyti- J. Photoenergy, vol. 2013, Dec. 2013, Art. no. 925280.
cal models to determine the best solution for predicting outdoor [14] D. Wu, J. Zhu, T. R. Betts, and R. Gottschalg, “Degradation of interfa-
cial adhesion strength within photovoltaic mini-modules during damp-heat
service lifetime. exposure,” Prog. Photovolt. Res. Appl., vol. 22, no. 7, pp. 796–809, Jul.
2014.
REFERENCES [15] M. Koehl, S. Hoffmann, and S. Wiesmeier, “Evaluation of damp-heat
testing of photovoltaic modules,” Prog. Photovolt. Res. Appl., vol. 25,
[1] G. B. Gaines, R. E. Thomas, G. T. Noel, T. S. Shilliday, and V. E. Wood, no. 2, pp. 175–183, Feb. 2017.
“Development of an accelerated test design for predicting the service life [16] M. Koehl, M. Heck, and S. Wiesmeier, “Modelling of conditions for accel-
of the solar array at Mead, Nebraska,” Jet Propulsion Lab, Pasadena, CA, erated lifetime testing of humidity impact on PV-modules based on moni-
USA, Tech. Rep. DOE/JPL-954328-79/11, Jun. 1979. toring of climatic data,” Sol. Energy Mater. Sol. Cells, vol. 99, pp. 282–291,
[2] A. B. Subramaniyan, R. Pan, J. Kuitche, and G. TamizhMani, “Quantifica- Apr. 2012.
tion of environmental effects on PV module degradation: A physics-based [17] L. A. Escobar and W. Q. Meeker, “A review of accelerated test models,”
data-driven modeling method,” IEEE J. Photovolt., vol. 8, no. 5, pp. 1289– Statist. Sci., vol. 21, no. 4, pp. 552–577, Nov. 2006.
1296, Sep. 2018. [18] D. C. Jordan, T. J. Silverman, B. Sekulic, and S. R. Kurtz, “PV degradation
[3] S. Lindig, I. Kaaya, K. Weiß, D. Moser, and M. Topic, “Review of statistical curves: Non-linearities and failure modes,” Prog. Photovolt. Res. Appl.,
and analytical degradation models for photovoltaic modules and systems as vol. 25, no. 7, pp. 583–591, Jul. 2017.
well as related improvements,” IEEE J. Photovolt., vol. 8, no. 6, pp. 1773– [19] M. D. McKay, “Evaluating prediction uncertainty,” Nucl. Regulatory
1786, Nov. 2018. Commission, Rockville, MD, USA, Rep. NUREG/CR-6311, 1995.
[4] W. Q. Meeker and L. A. Escobar, “Pitfalls of accelerated testing,” IEEE [20] E. Annigoni et al., “Modeling potential-induced degradation (PID) in crys-
Trans. Rel., vol. 47, no. 2, pp. 114–118, Jun. 1998. talline silicon solar cells: From accelerated-aging laboratory testing to out-
[5] J. Zhu, D. Montiel-Chicharro, T. R. Betts, and R. Gottschalg, “Correlation door prediction,” in Proc. 32nd Eur. Photovolt. Sol. Energy Conf. Exhib.,
of degree of EVA crosslinking with formation and discharge of acetic acid 2016, pp. 1558–1563.
in PV modules,” in Proc. 33rd Eur. Photovolt. Sol. Energy Conf. Exhib.,
2017, pp. 1795–1798.
[6] M. Kempe, “Modeling of rates of moisture ingress into photovoltaic mod-
ules,” Sol. Energy Mater. Sol. Cells, vol. 90, pp. 2720–2738, Oct. 2006.
[7] V. Sharma and S. S. Chandel, “Performance and degradation analysis for
long term reliability of solar photovoltaic systems: A review,” Renew.
Sustain. Energy Rev., vol. 27, pp. 753–767, Nov. 2013. Authors’ photographs and biographies not available at the time of publication.
 energies
Article
Global Climate Data Processing and Mapping of
Degradation Mechanisms and Degradation Rates of
PV Modules
Julián Ascencio-Vásquez 1, * , Ismail Kaaya 2,3 , Kristijan Brecl 1 , Karl-Anders Weiss 2 and
Marko Topič 1
1 Faculty of Electrical Engineering, University of Ljubljana, Tržaška cesta 25, 1000 Ljubljana, Slovenia;
[email protected] (K.B.); [email protected] (M.T.)
2 Fraunhofer Institute of Solar Energy, Heidenhofstr. 2, 79110 Freiburg im Breisgau, Germany;
[email protected] (I.K.); [email protected]; (K.-A.W.)
3 School of Industrial Engineering, University of Malaga, 29016 Malaga, Andalucia, Spain
* Correspondence: [email protected]




Received: 17 October 2019; Accepted: 7 December 2019; Published: 12 December 2019


Abstract: Photovoltaic (PV) systems are the cheapest source of electricity in sunny locations and
nearly all European countries. However, the fast deployment of PV systems around the world is
bringing uncertainty to the PV community in terms of the reliability and long-term performance of
PV modules under different climatic stresses, such as irradiation, temperature changes, and humidity.
Methodologies and models to estimate the annual degradation rates of PV modules have been studied
in the past, yet, an evaluation of the issue at global scale has not been addressed so far. Hereby, we
process the ERA5 climate re-analysis dataset to extract and model the climatic stresses necessary for
the calculation of degradation rates. These stresses are then applied to evaluate three degradation
mechanisms (hydrolysis-degradation, thermomechanical-degradation, and photo- degradation) and
the total degradation rate of PV modules due to the combination of temperature, humidity, and
ultraviolet irradiation. Further on, spatial distribution of the degradation rates worldwide is computed
and discussed proving direct correlation with the Köppen-Geiger-Photovoltaic climate zones, showing
that the typical value considered for the degradation rate on PV design and manufacturer warranties
(i.e., 0.5%/a) can vary ± 0.3%/a in the temperate zones of Europe and rise up to 1.5%/a globally. The
mapping of degradation mechanisms and total degradation rates is provided for a monocrystalline
silicon PV module. Additionally, we analyze the temporal evolution of degradation rates, where a
global degradation rate is introduced and its dependence on global ambient temperature demonstrated.
Finally, the categorization of degradation rates is made for Europe and worldwide to facilitate the
understanding of the climatic stresses.

Keywords: photovoltaic; degradation; PV systems; climate zones; climate data

1. Introduction
The current worldwide energy transition from conventional to renewable energy sources already
proves and even more forecasts the expansion of photovoltaic systems to new and diverse locations.
On a global scale, we are rapidly approaching the so-called “Terawatt-scale Photovoltaics” era, which
indicates the surpass of 1 TW of installed capacity of photovoltaic (PV) systems and highlights an
important milestone for the PV industry [1]. This milestone is also boosted by the decrease in prices
of the PV technologies, which are already the cheapest source of electricity in many countries of
Europe [2].

Energies 2019, 12, 4749; doi:10.3390/en12244749 www.mdpi.com/journal/energies

Energies 2019, 12, 4749 2 of 16

However, this diversification is presenting some constraints in terms of financial and technical
risks, due to the unclear response of PV modules and materials under different climate conditions. The
financial risks are related to the estimation of the long-term energy yield, mainly due to degradation rate
calculations and solar resource variability, as represented by calculations of Levelized-Cost-Of-Electricity
(LCOE) for PV systems. The technical risks are related to the failure modes prone to occur for different
Bill-of-Materials (BOM) and Balance-of-System (BOS) under different climatic stresses which will vary
from location to location. For example, while in deserts the risk of sand-storms is high, in highlands wind
gusts and snow loads can damage the PV installations. Important findings and calculation methods
about the climatic triggers of degradation modes under different climate zones have been presented
in [3–11]. Irradiation, temperature, and humidity have been identified as the main degradation
precursors, leading to different degradation mechanisms depending on the stress level in each climate
zone. In qualitative terms, some generalizations can be made by climate: in tropical climates, the
combination of high humidity and high temperature is stated as being the harshest for PV reliability,
where the PV modules are prone to delamination, corrosion or Potential-Induced-Degradation (PID).
Desert and steppe climates stimulate degradation modes, such as encapsulant discoloration, backsheet
chalking or delamination. Temperate, cold and polar climates typically present low degradation due
to the low climatic stress, but modules are prone to fast power drops due to hail, storms or snow loads
inducing cell cracks, glass cracks and interconnect breakage. In quantitative terms, Jordan et al. have
been the pioneers in the degradation rate assessment under different climate conditions [4], but the
complexity of the issue does not yet allow the conclusion of a generalized estimation model per climate
zone. The synergy of climate stressors makes the estimation and prediction of the degradation rates
(i.e., expressed as a reduction of the power per year, %/a) not trivial, given that not only different
module types react differently regarding the climate but also that material interactions will behave
differently [6].
Studies at a global scale present the mapping of PV performance indicators (e.g., performance
ratio and energy yield) using different Geographic Information Systems (GIS) datasets and
approaches [12–14]. Regarding the long-term operation of PV systems, usually this issue is simplified
by considering a typical value (i.e., 0.5%/a for crystalline PV modules), also stated on manufacturer
performance warranties. However, it is known that the impact of climate on material properties and
energy production will vary over time and location. Interesting approaches to calculate the degradation
rates have been proposed in references [10,15]. Both models consider as main climate degradation
factors the average temperatures, temperature changes, humidity, and ultraviolet (UV) irradiation,
which is in line with indoor testing and standards for PV reliability. However, those models have not
been applied at global scale and a PV degradation mapping has not been published so far.
In this paper, the ERA5 climate reanalysis dataset [16] is processed to extract and further to
model the essential climate variables for the study of PV degradation (temperature, humidity, and UV
irradiation). The estimated climate data is compared and validated with real ground-measurements
taken from different sources [17–19]. Then, the module temperature and degradation rates are
estimated globally. Using the Köppen-Geiger-Photovoltaic (KGPV) climate classification [12], the
spatial distribution of degradation rates in view of climate zones is evaluated. Then, the temporal
evolution of degradation rates compared with the global increase in ambient temperature is presented.
The global mapping of degradation mechanisms and total degradation rates are provided, together
with a categorization of degradation rates in Europe and worldwide.

2. Climate Data Processing

The studies over large geographical regions can be made by processing GIS data estimated from
Numerical Weather Predictions (NWP) including satellite or reanalysis models [20]. Even though
satellite-based estimations can be more accurate than the reanalysis-based ones, the advantage of the
second is the possibility to extract all the essential variables together in the same dataset, without gaps
and identical timestamps.
Energies 2019, 12, 4749 3 of 16

Energies
One 2020,
of the 13, x FOR
latest PEER REVIEW
datasets released is the “ERA5” created by the European Centre for Medium-Range 3 of 17

Weather Forecasts (ECMWF). In comparison with previous datasets (e.g., ERA-Interim), ERA5 provides
One of the latest datasets released is the “ERA5” created by the European Centre for Medium-
a spatial resolution of 31 km and temporal resolution of hourly data from 1979 [16]. Also in the
Range Weather Forecasts (ECMWF). In comparison with previous datasets (e.g., ERA-Interim), ERA5
literature, the improvement
provides on the
a spatial resolution accuracy
of 31 km and of the global
temporal horizontal
resolution irradiance
of hourly (GHI)
data from 1979is[16].
reported
Also in[20].
Although not all variables required are directly available from the ERA5, we calculate
the literature, the improvement on the accuracy of the global horizontal irradiance (𝐺𝐻𝐼) is reportedthe missing
local climate
[20]. variables, wind speed (WS), relative humidity (RH), and UV irradiance (UV), using
existing models.
Although not all variables required are directly available from the ERA5, we calculate the
missing local climate variables, wind speed (𝑊𝑆), relative humidity (𝑅𝐻), and UV irradiance (𝑈𝑉),
2.1. Local Climate
using Variables
existing models.

While high
2.1. Local windVariables
Climate speed (WS) can increase mechanical loads on the PV installation [21] and be a
trigger of further degradation processes, we use it only to estimate the PV module temperature (Tmod )
While high wind speed (𝑊𝑆) can increase mechanical loads on the PV installation [21] and be a
due to the related cooling effect on materials. The WS is calculated and height-corrected according to
trigger of further degradation processes, we use it only to estimate the PV module temperature (𝑇𝑚𝑜𝑑 )
Equations (1) and (2) [22,23], where uwind and vwind are the vector components of the wind, hERA is the
due to the related cooling effect on materials. The 𝑊𝑆 is calculated and height-corrected according
heighttofrom ground(1)which
Equations the[22,23],
and (2) wind where
is modelled
𝑢̅𝑤𝑖𝑛𝑑 in
andthe𝑣̅ERA5are dataset, hmod is the assumed
andcomponents
the vector height
of the wind,
𝑤𝑖𝑛𝑑
of the PV
ℎ𝐸𝑅𝐴modules equalfrom
is the height to 2 ground
m. Thewhich
2 m height assumed
the wind is in accordance
is modelled in the ERA5with the height
dataset, of modelled
and ℎ𝑚𝑜𝑑 is the
ambient temperature
assumed (Tthe
height of amb )
PVand dew
modules point
equaltemperature
to 2 m. The 2 (T
m ) given
height
dew by the
assumed isECMWF.
in accordance with the
height of modelled ambient temperature (𝑇𝑎𝑚𝑏
q) and dew point temperature (𝑇𝑑𝑒𝑤 ) given by the
ECMWF. WSERA = u2wind + v2wind (1)
2 2
𝑊𝑆𝐸𝑅𝐴 = √𝑢̅ 𝑤𝑖𝑛𝑑
!0.2 + 𝑣̅𝑤𝑖𝑛𝑑 (1)
h
WS = mod ·WSERA (2)
hERA
ℎ 0.2
𝑚𝑜𝑑
𝑊𝑆 = ( 𝐸𝑅𝐴) · 𝑊𝑆 (2)
The relative humidity (RH) is also not extractedℎ𝐸𝑅𝐴 directly from ERA5, so it is estimated using
Equations The
(3) and
relative humidity (𝑅𝐻) is also not extracted directly from over
(4). The saturation water vapor pressure (WVP) ERA5,water and
so it is ice is calculated
estimated using
using Buck’s Formula [24,25] from the dew point temperature (T
Equations (3) and (4). The saturation water vapor pressure (𝑊𝑉𝑃) ) and ambient temperature (Tamb ).
Dew over water and ice is calculated
using Buck’s Formula [24,25] from
 the dew point temperature (𝑇𝐷𝑒𝑤 ) and ambient temperature (𝑇𝑎𝑚𝑏 ).
(23.036 − T )·( 279.82 T
+T𝑇 )
 0.61115·exp (23.036333.70

 𝑇 for T < 0
WVP(T )[kPa] =  0.61115 · 𝑒𝑥𝑝 −
T 333.70 )·(279.82+𝑇
T
(18.678− 234.84 )·( 257.14+T )
)
for 𝑇 < 0 (3)
𝑊𝑉𝑃(𝑇)[𝑘𝑃𝑎] =0.61121·exp
{ for T ≥ 0 (3)


𝑇 𝑇
0.61121 · 𝑒𝑥𝑝(18.678−234.84)·(257.14+𝑇) for 𝑇 ≥ 0
WVP(Tdew )
RH [%] = 𝑊𝑉𝑃(𝑇𝑑𝑒𝑤 )·100[%] (4)
𝑅𝐻[%] WVP
= (Tamb ) · 100[%] (4)
𝑊𝑉𝑃(𝑇𝑎𝑚𝑏 )
UV irradiance given in the ERA5 dataset covers a wavelength range up to 440 nm. However,
UV irradiance given in the ERA5 dataset covers a wavelength range up to 440 nm. However,
this variable is usually
this variable referred
is usually for for
referred wavelengths
wavelengthsbelow
below400400 nm. Thisdistinction
nm. This distinctioncancan cause
cause large
large
differences of theof
differences estimated UV irradiation.
the estimated UV irradiation.ForFor
example,
example,measurements
measurementsin inLjubljana, Slovenialead
Ljubljana, Slovenia lead to
an overestimation of 45% of
to an overestimation of45%
the of
energy in theinUV
the energy the part (see(see
UV part Figure 1).1).
Figure

(a) (b)

1. (a)1.Spectral
FigureFigure irradiance
(a) Spectral irradiancefrom
from280
280 nm to440
nm to 440nmnm under
under a day
a day of cloudy
of cloudy sky conditions.
sky conditions. (b)
(b) Spectral irradiance
Spectral irradiancefrom
from280
280 nm to 440
nm to 440nm
nmunder
undera day
a day
of of clear
clear skysky conditions.
conditions.

For this Forreason,

this reason, we neglect
we neglect UVUV
thethe irradiancedata
irradiance datagiven
given by
by ERA5,
ERA5,and andmodel
model it up to 400
it up nm nm
to 400
using ausing
methoda method
proposedproposed
in ref. in ref.
[26] and[26] and expressed
expressed in Equations
in Equations (5)–(8),(5)–(8),
whichwhich is based
is based on theon the
clearness
index (kt ) and the global horizontal irradiance (GHI). The kt is calculated by dividing the GHI and the
Energies 2019, 12, 4749 4 of 16

Energies 2020, 13, x FOR PEER REVIEW 4 of 17

top-of-atmosphere irradiance extracted from ERA5. Unfortunately, the lack of valid measurements
clearness index (𝑘𝑡 ) and the global horizontal irradiance (𝐺𝐻𝐼). The 𝑘𝑡 is calculated by dividing the
disallows that we validate the model worldwide.
𝐺𝐻𝐼 and the top-of-atmosphere irradiance extracted from ERA5. Unfortunately, the lack of valid
measurements disallows that we validate
∗ the model worldwide.
kt = max(0.1, min(kt , 0.7)) (5)
𝑘𝑡∗ = 𝑚𝑎𝑥(0.1, 𝑚𝑖𝑛(𝑘
 𝑡 , 0.7)) (5)
UVB = 1.897 − 0.860 kt∗ ·1e−3 ·GHI (6)
𝑈𝑉𝐵 =(1.897 − 0.860 𝑘𝑡∗) · 1𝑒 −3 · 𝐺𝐻𝐼 (6)
UVA = 7.210 − 2.365 kt∗ ·1e−2 ·GHI (7)
𝑈𝑉𝐴 = (7.210 − 2.365 𝑘𝑡∗ ) · 1𝑒 −2 · 𝐺𝐻𝐼 (7)
UV = UVA + UVB (8)
𝑈𝑉 = 𝑈𝑉𝐴 + 𝑈𝑉𝐵 (8)
The GIS data extracted and computed is compared with 15 ground meteorological stations involved
in the World TheRadiation
GIS data Monitoring
extracted andCenter
computed is compared
- Baseline Surfacewith 15 ground
Radiation meteorological
Network (WRMC-BSRN)stations [17],
involved in the World Radiation Monitoring Center - Baseline Surface Radiation Network (WRMC-
one station in Alice Springs, Australia provided by the Desert Knowledge Australia (DKA) Solar
BSRN) [17], one station in Alice Springs, Australia provided by the Desert Knowledge Australia
Centre [18] and one station in the Atacama Desert provided by Universidad de Chile [19]. The station
(DKA) Solar Centre [18] and one station in the Atacama Desert provided by Universidad de Chile
locations
[19].are
Theshown
stationinlocations
Figure 2are and moreindetails
shown Figureare presented
2 and in Table
more details are A1 in the Appendix
presented in Table A1A. inAt
thethese
stations, GHI,
Appendix A.T , and RH are measured. The comparison is carried out on a daily
ambAt these stations, 𝐺𝐻𝐼, 𝑇𝑎𝑚𝑏 , and 𝑅𝐻 are measured. The comparison is carried out on average resolution
for thea daily
yearsaverage
2016, 2017, and for
resolution 2018,
theexcept for the
years 2016, 2017,Chilean
and 2018, station
exceptwhere
for the the timestation
Chilean framewhere
ranges thefrom
2010 to 2014.
time Weranges
frame use the inverse
from 2010 todistance
2014. Weweighting (IDW)
use the inverse as interpolation
distance method
weighting (IDW) [27] to calculate
as interpolation
method [27]
the estimated to calculate
values the estimated
at specific location values
reducing at specific location reducing
the geographical the geographical
mismatch mismatch and
between measured
between
computed values. measured and computed values.
The time-series
The time-series for each
for each location
location are are compared
compared by by
thethe cumulativedistribution
cumulative distributionfunction
function (CDF)
(CDF) and
and the coefficient-of-determination (R 2). In Figure 3, locations with high and low accuracy are
the coefficient-of-determination (R2 ). In Figure 3, locations with high and low accuracy are presented,
presented, while the rest of the locations are presented in Table A2 in the Appendix A. For example,
while the rest of the locations are presented in Table A2 in the Appendix A. For example, on one hand,
on one hand, data in Cabauw (The Netherlands) show almost perfect fitting for the three variables
data in Cabauw
(𝑇𝑎𝑚𝑏 , 𝐺𝐻𝐼, (The Netherlands)
and 𝑅𝐻), evidencingshow almost
that ERA5 canperfect
be usedfitting for thedata
as synthetic three variables
in some (TambOn
locations. , GHI,
the and
RH), evidencing
other hand, in Howrah (India), the accuracy of 𝐺𝐻𝐼 and 𝑅𝐻 is low but could be improved with a in
that ERA5 can be used as synthetic data in some locations. On the other hand,
Howrah (India), thealgorithm,
post-processing of GHI
accuracy e.g., and RH is low but could be improved with a post-processing
ref. [28].
algorithm,An e.g.,excellent
ref. [28].agreement of 𝑇𝑎𝑚𝑏 was noticed possibly because of the high number of
observational
An data in ERA5.
excellent agreement of T𝐺𝐻𝐼
amb was 𝑅𝐻 also possibly
and noticed show good performance
because of theeven
highthough
number they
of are purely
observational
data inmodelled.
ERA5. GHI and RH also show good performance even though they are purely modelled.

2. Location
FigureFigure of meteorological
2. Location stations
of meteorological usedused
stations to compare the GHI,
to compare Tamb ,𝑇and
the 𝐺𝐻𝐼, 𝑎𝑚𝑏 ,RH
andwith
𝑅𝐻 the modelled
with the
modelled
data from ERA5 data from ERA5
reanalysis reanalysis dataset.
dataset.
Energies 2019, 12, 4749 5 of 16
Energies 2020, 13, x FOR PEER REVIEW 5 of 17

R2=0.89 R2=0.99 R2=0.82

R2=0.55 R2=0.97 R2=0.65

Figure 3. Cumulative distribution functions (CDFs) and statistical indicators for comparison of
ground-measurements
Figure 3. Cumulative and modelled functions
distribution data from(CDFs)
the ERA5
and reanalysis dataset and
statistical indicators forthe inverse distance
comparison of
ground-measurements
weighting and modelled
(IDW) interpolation. data from
We present the ERA5
a high reanalysis
accuracy and lowdataset and thecase,
accuracy inverse distance(CAB,
Cabauw
weighting
Lopik, Utrecht,(IDW) interpolation. We
The Netherlands) andpresent
Howraha high accuracy
(HOW, Westand low accuracy
Bengal, India). case,
The Cabauw (CAB,of the
comparison
Lopik, Utrecht, The Netherlands) and Howrah (HOW, West
other 15 locations is presented in Table A2 in the Appendix A. Bengal, India). The comparison of the
other 15 locations is presented in Table A2 in the Appendix A.
2.2. Operating Conditions of PV Modules
2.2. Operating Conditions of PV Modules
One of the most important climatic loads to analyze the degradation is the PV module temperature
One of the most important climatic loads to analyze the degradation is the PV module
(Tmod ). Typically this variable is estimated by using the Ross model [29], which is a function of the
temperature (𝑇𝑚𝑜𝑑 ). Typically this variable is estimated by using the Ross model [29], which is a
ambient temperature, irradiance and the Ross coefficient (kRoss ) as presented in Equation (9). However,
function of the ambient temperature, irradiance and the Ross coefficient (𝑘𝑅𝑜𝑠𝑠 ) as presented in
a higher accuracy
Equation of Tmod a estimations
(9). However, higher accuracy under different
of 𝑇𝑚𝑜𝑑 climate
estimations conditions
under different for crystalline
climate conditionssilicon
for PV
modules can besilicon
crystalline achieved by using
PV modules canthe Faiman model
be achieved by using(see
the Equation
Faiman model(10))(see
[22,30]. In Figure
Equation 4, a visual
(10)) [22,30].
comparison
In Figure of 4,
thea estimated Tmod is illustrated.
visual comparison The main
of the estimated 𝑇𝑚𝑜𝑑 differences areThe
is illustrated. presented in areas with
main differences are high
wind speed where the cooling effect will be taken into account when using the Faiman model.
presented in areas with high wind speed where the cooling effect will be taken into account when
using
Further the calculations
Faiman model. in this paper are based on the Faiman model to estimate the PV module
Further
as acalculations
function ofinthethis Tpaper are based on the Faiman model to estimate the PV module
amb , GHI, and WS, considering u0 equal to 26.9 W/m / C and u1
temperature 2 ◦

equaltemperature
to 6.2 Ws/m as3a/◦function of the 𝑇are
C [23], which , 𝐺𝐻𝐼, and
𝑎𝑚𝑏typical 𝑊𝑆, reported
values consideringfor𝑢c-Si
0 equal
PV to 26.9 W/min/°C
modules
2
theand 𝑢1
open-rack
equal to 6.2 Ws/m3/°C [23], which are typical values reported for c-Si PV modules in the open-rack
mounting configuration.
mounting configuration.
Tmod = Tamb + kRoss ·GHI (9)
𝑇𝑚𝑜𝑑 = 𝑇𝑎𝑚𝑏 + 𝑘𝑅𝑜𝑠𝑠 ∙ 𝐺𝐻𝐼 (9)
GHI
Tmod = Tamb + (10)
u0 +𝐺𝐻𝐼
u1 ·WS
Energies 2020, 13, x FOR PEER REVIEW 𝑇𝑚𝑜𝑑 = 𝑇𝑎𝑚𝑏 + 6 of(10)
17
𝑢0 + 𝑢1 · 𝑊𝑆

Figure 4. (a) Maximal module temperature using the Ross model. (b) Maximal module temperature
Figure 4. (a) Maximal module temperature using the Ross model. (b) Maximal module temperature
using the Faiman model. (c) Absolute difference between the estimated Tmod using Faiman and Ross
using the Faiman model. (c) Absolute difference between the estimated 𝑇𝑚𝑜𝑑 using Faiman and Ross
models. Annual average data from 2016 to 2018.
models. Annual average data from 2016 to 2018.

3. Spatial Distribution of PV Degradation Rates

Degradation mechanisms will be triggered not only from one individual degradation factor but
due to a combination of them. In ref. [10], three degradation processes are defined and empirically
expressed due to the combination of climate degradation factors: Hydrolysis-degradation ( 𝑘𝐻 )
related to the effect of temperature and humidity, photo-degradation (𝑘𝑃 ) depends on temperature,
Energies 2019, 12, 4749 6 of 16

3. Spatial Distribution of PV Degradation Rates

Degradation mechanisms will be triggered not only from one individual degradation factor but
due to a combination of them. In ref. [10], three degradation processes are defined and empirically
expressed due to the combination of climate degradation factors: Hydrolysis-degradation (kH ) related
to the effect of temperature and humidity, photo-degradation (kP ) depends on temperature, humidity
and UV irradiance, and thermo-mechanical-degradation (kTm ) due to high temperature and temperature
differences. The models are presented in Equations (11) to (14):

EH
 
kH = AH ·rhnef f · exp − , (11)
kB ·Tm

EP
   
kP = AP ·(UVdose )X · 1 + rhnef f · exp − , (12)
kB ·Tm
!
ETm
kTm = ATm ·(∆T )θ ·CN · exp − , (13)
kB ·TU
n
Y
kT = AN (1 + ki ) − 1, (14)
i=1

where the parameters are:

kH : Hydrolysis degradation rate (%) rhe f f : effective relative humidity
kP : Photo-degradation rate (%) UVdose : integral UV dose (kW/m2 )
kTm : Thermomechanical degradation rate (%) Tm : average module temperature
kT : Total degradation rate (%) TU : upper module temperature
kB : Boltzmann constant (8.62 × 10−5 ) TL : lower module temperature
EH : Activation Energy for Hydrolysis degradation (eV) ∆T = TU − TL : temperature difference
EP : Activation Energy for Photo-degradation (eV) n: model parameter that indicates the impact of
RH on power degradation.
ETm : Activation Energy for Thermomechanical
X: model parameter that indicates the impact of
degradation (eV)
UV dose on power degradation.
AH : Pre-exponential constant for Hydrolysis degradation θ: model parameter that indicates the impact of
∆T on power degradation.
AP : Pre-exponential constant for Photo-degradation CN : Cycling rate
ATm : Pre-exponential constant for Thermomechanical
AN : normalization constant of the physical
degradation
quantities (a−2 /%)
The degradation mechanisms are quantified by using the fitting coefficients published in
ref. [10] for a high-performance monocrystalline silicon PV module installed in the open-rack
mounting configuration.
Further on, the calculated degradation rates can be directly related to climate zones. To simplify
the spatial distribution analysis of degradation rates, we use the Köppen-Geiger-Photovoltaic (KGPV)
scheme proposed in ref. [12] to compare the climate zones in terms of annual degradation rates, as
shown in Figures 5 and 6. Each KGPV climate zone is defined by two letters, the first one implies the
relation of temperature and precipitation (TP-zones) and the second is related to the irradiation level,
as H-zones. The definition of each letter is as follows:
Temperature-Precipitation (TP) Zones Irradiation (H) Zones
A: Tropical climate K: Very high irradiation
B: Desert climate H: High irradiation
C: Steppe climate M: Medium irradiation
E: Temperate climate L: Low irradiation
D: Cold climate
F: Polar climate
Energies 2019, 12, 4749 7 of 16

Hydrolysis-degradation presents the smallest contribution in almost all the KGPV zones, but is
considerably higher for the tropical climates (AH and AK), which zones are related to high precipitation
levels (humid areas) and temperature levels. This process can provoke moisture ingress leading to
delamination of polymers or corrosion of solder bonds [31].
Photo-degradation has the second-highest contribution to the total degradation rate. This indicator
combines the humidity, temperature, and UV irradiance impacting the PV module. The impact is
similar to hydrolysis-degradation but higher in terms of absolute values due to the process triggered
by UV irradiation. For desert areas, even though the UV irradiation is high, the low humidity in the
air decreases the estimated damage of the PV cells due to this mechanism. The photo-degradation
is considerable high in tropical zones (AH and AK) due to the high climatic stresses of all variables
(temperature, humidity, and UV irradiation).
Thermo-mechanical degradation exhibits the highest contribution to the total degradation rate
in all zones, except in the AH zone, where the temperature variations are minimal. This parameter
is affected by seasonal temperature cycling (the difference between the maximum and minimum
temperature of the year) and also the annual average maximum ambient temperature.
The total degradation rates calculated by the combination of the previous three degradation
mechanisms Equation (13) are also evaluated per each KGPV climate zone. In accordance with the
literature [6], the highest degradation rate is identified in tropical areas (hot and humid). Interestingly,
the AK (tropical with very high irradiation) presents lower degradation than the AH (tropical with
high irradiation), due to lower photo-degradation contribution (related to lower humidity).
The steppe climate has higher humidity than the deserts and therefore, the total degradation rate
is increased by higher hydrolysis-degradation and photo-degradation. As expected CK (steppe with
very high irradiation) might be more stressful for PV modules than CH (steppe with high irradiation),
due to higher UV irradiation.
Temperate climates (DM and DH) result in average degradation rates of 0.42%/a and 0.58%/a,
respectively. Those climate zones are predominant in Europe, and their values are in accordance with
the typical 0.5%/a degradation rate assumed along within the PV community. However, the spread of
Energies 2020,
the values can 13, x FORaround
range PEER REVIEW
± 0.3%/a in the temperature zones across Europe. 8 of 17
Energies 2020, 13, x FOR PEER REVIEW 8 of 17

Figure 5.Figure
Figure
5. distribution
Spatial Spatial distribution of the degradation
of the degradation
5. Spatial distribution
mechanisms
mechanisms
of the degradation
in view
in view
mechanisms
of Köppen-Geiger-Photovoltaic
of the
in view
the Köppen-Geiger-
of the Köppen-Geiger-
Photovoltaic (KGPV) climate zones.
(KGPV) climate zones. Photovoltaic (KGPV) climate zones.

Figure 6. Spatial
Figure distribution
6. Spatial distributionofofthe
thetotal
total degradation rates
degradation rates inin view
view of the
of the KGPV
KGPV climate
climate zones.
zones. The The
Figure 6. Spatial distribution of the total degradation rates in view of the KGPV climate zones. The
average
average of total
of total degradationrate
degradation rateper
perclimate
climate zone
zoneisisindicated
indicatedbelow each
below label
each in %/a.
label in %/a.
average of total degradation rate per climate zone is indicated below each label in %/a.
Temperate climates (DM and DH) result in average degradation rates of 0.42%/a and 0.58%/a,
Temperate climates (DM and DH) result in average degradation rates of 0.42%/a and 0.58%/a,
respectively. Those climate zones are predominant in Europe, and their values are in accordance with
respectively. Those climate zones are predominant in Europe, and their values are in accordance with
the typical 0.5%/a degradation rate assumed along within the PV community. However, the spread
the typical 0.5%/a degradation rate assumed along within the PV community. However, the spread
of the values can range around ± 0.3%/a in the temperature zones across Europe.
of the values can range around ± 0.3%/a in the temperature zones across Europe.
Cold and polar areas present average degradation rates around or below 0.2%/a. In real
Cold and polar areas present average degradation rates around or below 0.2%/a. In real
operating conditions, we expect higher values due to external degradation factors, such as snow
operating conditions, we expect higher values due to external degradation factors, such as snow
accumulation over the PV systems or mechanical loads due to wind gusts which are not included in
Energies 2019, 12, 4749 8 of 16

Cold and polar areas present average degradation rates around or below 0.2%/a. In real operating
conditions, we expect higher values due to external degradation factors, such as snow accumulation
over the PV systems or mechanical loads due to wind gusts which are not included in the calculations
as yet.
Desert areas exhibit similar average degradation rates than temperate areas. In absolute climate
values, desert areas are hotter and dryer than temperate areas. However, in the relative contribution
of degradation mechanisms, the high thermomechanical stress together with low hydrolysis and
photo-degradation makes the average similar to the moderate climate stressors of the temperate
climate. In real operating conditions, external degradation factors, such as soiling might increase the
degradation rate if taken into account, but the degradation presented here assumes only gradual and
non-reversible degradation processes.

4. Mapping of Degradation Mechanisms and Total Degradation Rate

The worldwide mapping of the degradation mechanisms (hydrolysis-degradation,
photo-degradation and thermomechanical-degradation) are presented in Figure A1 in the Appendix B.
As mentioned before, the simulations are based on the degradation model developed in ref. [10], and
coefficients fitted for a specific PV module (high-performance mono-crystalline silicon PV module).
The climate datasets used were extracted, modelled and averaged from the ERA5 reanalysis dataset for
the years 2016, 2017, and 2018.
Figure A2 in the Appendix B shows the calculated worldwide degradation rate combining the
three degradation mechanisms based on the main climate degradation factors.
To facilitate the visualization and possible use of our degradation maps, we categorize the locations
into bins of 0.2%/a ranging from 0% to 0.8%/a for Europe and ranging from 0% to 1.4%/a around the
Energies 2020, 13, x FOR PEER REVIEW 9 of 17
World. The categorized maps are shown in Figures 7 and A3, respectively. The total degradation rates
could Europe,
reach 0.8%/a in the hottest
and globally areas
the highest of the south
degradation of (above
rates Spain and Portugal
1.4%/a) for Europe,
are identified and globally
in locations next to the
highest
thedegradation
equator line.rates (above 1.4%/a) are identified in locations next to the equator line.

7. European
FigureFigure categorization
7. European of of
categorization total degradation
total degradation rate basedon
rate based ontemperature,
temperature, humidity,
humidity, andand
UV UV
irradiance for a for
irradiance specific monocrystalline
a specific silicon
monocrystalline siliconPV
PVmodule.
module.

5. Uncertainty overover
5. Uncertainty Time-Temporal Evolution
Time-Temporal Evolution
The temporal
The temporalevolution of the
evolution climate
of the climatealso
alsomight
might be
be a factor
factortotoconsider
consider while
while estimating
estimating the the
degradation rates of PV modules. In ref. [12] the decrease of the Performance Ratio due
degradation rates of PV modules. In ref. [12] the decrease of the Performance Ratio due to climate to climate
changechange effect
effect is is already
already reported.
reported. Hereby,
Hereby, the annual
the annual degradation
degradation ratesrates are simulated
are simulated for PV
for the themodule
PV
module described previously, using historical data from ERA5. For convenience
described previously, using historical data from ERA5. For convenience in terms of computational in terms of
computational resources, degradation rates based on hourly data are calculated only for the years
1980, 1990, 2000, 2010, 2016, 2017, and 2018. The ambient temperature is extracted for every year from
1979 to 2018.
In Figure 8, the evolution of the ambient temperature and degradation rates is shown. The global
calculations represent the average values for land-surface between the Latitudes −60° and 60°. We
notice that an increase in the ambient temperature also increases the degradation of the PV modules
Energies 2019, 12, 4749 9 of 16

resources, degradation rates based on hourly data are calculated only for the years 1980, 1990, 2000,
2010, 2016, 2017, and 2018. The ambient temperature is extracted for every year from 1979 to 2018.
In Figure 8, the evolution of the ambient temperature and degradation rates is shown. The global
calculations represent the average values for land-surface between the Latitudes −60◦ and 60◦ . We
notice that an increase in the ambient temperature also increases the degradation of the PV modules
overEnergies
time. 2020, 13, x FOR PEER REVIEW 10 of 17

Figure 8. Temporal evolution of the global ambient temperature and the total degradation rates. The
Figure 8. Temporal evolution of the global ambient temperature and the total degradation rates. The
calculations are are
limited to to
the land-surface −60 ◦ ◦
calculations limited the land-surface between thelatitudes
between the latitudes−60° andand
60°.60 .

6. Discussions
The correlation with the KGPV climate zones shows a good agreement with the literature.
Tropical
The used climates are again
degradation presented
model as thetakes
currently harshest
intoforaccount
PV modules
pure due to thedegradation
climate high humidity and and
factors
high
it does nottemperature.
include temporal Desert and steppe climates
or external degradationpresent high degradation
factors or failure modes rates duesuchtoashigh daily
light-induced
temperature changes, even excluding soiling. Under no consideration of wind and snow loads, the
degradation (LID), light and elevated temperature degradation (LeTID), potential-induced-degradation
temperate, cold and polar climates present low degradation rates.
(PID) or mechanical damage due to wind or snow loads. Neither the underperformance due to soiling
Even though the results are coherent, some climate and degradation modelling topics need
or snow covering
further is considered.
investigation: (1) lack Also,
of UVthe final results
irradiation of total degradation
measurements spread around ratestherepresent
world does thenot
value at
steady
allow a representative validation of models, (2) moisture ingress and related triggering of to a
conditions (for example, after LID). The calculations presented in this paper are related
specific monocrystalline
degradation processes, silicon
such PVas module,
corrosionand the values could
or delamination, needbetodifferent for devices
be understood for with different
different
BOMinteractions
but we expect the trendand
of materials, to remain the same for all
(3) the understanding of crystalline
degradationsilicon technologies.
mechanisms under high UV
irradiance and extremely
The correlation with thelow KGPVhumidity
climateexposure.
zones shows a good agreement with the literature. Tropical
climates Despite
are again the presented
current limitations, the average
as the harshest fordegradation
PV modules ratesdue
calculated
to the agree
high with the typical
humidity and high
temperature. Desert and steppe climates present high degradation rates due to high daily tropical
values used in manufacturer warranties and developers, showing that in the worst case (e.g., temperature
climates) an average degradation of 1%/a can offer more than 20 years of performance above 80%,
changes, even excluding soiling. Under no consideration of wind and snow loads, the temperate, cold
and the global average degradation rate of 0.5%/a could promise a lifetime operation of 40 years.
and polar climates present low degradation rates.
Additionally, the mapping and results presented can give a new perspective of the climatic stresses
Even though
worldwide when the results are
considering thecoherent,
installationsome
of PVclimate
modules and degradation
at different modelling
locations. Also, the topics
high need
further investigation: (1) lack of UV irradiation measurements spread
resolution of the GIS data, including ambient temperature, UV irradiation and humidity, together around the world does not
allowwith
a representative
the module operatingvalidation of models,helps
temperature, (2) moisture
to identify ingress and
relevant relatedfor
variables triggering
PV systems of degradation
where
processes,
measured such as corrosion
data or delamination,
is not available. The degradation needmechanisms
to be understood
and the for totaldifferent interactions
degradation rates of
calculated for the land-surface are available as Supplementary Materials
materials, and (3) the understanding of degradation mechanisms under high UV irradiance and attached to this article.
extremely low humidity exposure.
7. Conclusions
Despite the current limitations, the average degradation rates calculated agree with the typical
values usedIn this paper, we usedwarranties
in manufacturer the ERA5 climate reanalysis dataset
and developers, for the
showing thatmodelling
in the worst of PVcase
degradation
(e.g., tropical
mechanisms and total degradation rates worldwide. We demonstrated
climates) an average degradation of 1%/a can offer more than 20 years of performance that by extracting the ambient
above 80%,
and and dew point
the global averagetemperature,
degradation the relative
rate ofhumidity
0.5%/a couldcan be estimated
promise at any location.
a lifetime operation Also, byyears.
of 40
extracting the global and top-of-atmosphere irradiances, the ultraviolet irradiance was estimated.
Additionally, the mapping and results presented can give a new perspective of the climatic stresses
Then, the extracted and modelled variables (temperature, irradiance, and humidity) were combined
worldwide when considering the installation of PV modules at different locations. Also, the high
to compute the degradation of PV modules in terms of photo-degradation, hydrolysis-degradation,
resolution of the GIS data, including ambient
and thermo-mechanical-degradation. temperature,
The quantification of UV
eachirradiation
degradationand humidity,
mechanism together
allowed us with
to estimate the total degradation rate globally for a specific monocrystalline silicon PV module.
To validate the estimation of climate variables, we extracted the time series of 17 ground
measurement stations from the World Radiation Monitoring Center - Baseline Surface Radiation
Energies 2019, 12, 4749 10 of 16

the module operating temperature, helps to identify relevant variables for PV systems where measured
data is not available. The degradation mechanisms and the total degradation rates calculated for the
land-surface are available as Supplementary Materials attached to this article.

7. Conclusions
In this paper, we used the ERA5 climate reanalysis dataset for the modelling of PV degradation
mechanisms and total degradation rates worldwide. We demonstrated that by extracting the ambient
and dew point temperature, the relative humidity can be estimated at any location. Also, by
extracting the global and top-of-atmosphere irradiances, the ultraviolet irradiance was estimated.
Then, the extracted and modelled variables (temperature, irradiance, and humidity) were combined to
compute the degradation of PV modules in terms of photo-degradation, hydrolysis-degradation, and
thermo-mechanical-degradation. The quantification of each degradation mechanism allowed us to
estimate the total degradation rate globally for a specific monocrystalline silicon PV module.
To validate the estimation of climate variables, we extracted the time series of 17 ground
measurement stations from the World Radiation Monitoring Center - Baseline Surface Radiation
Network (WRMC-BSRN), Desert Knowledge Australia (DKA) Solar Centre, and Universidad de
Chile which include ambient temperature, global horizontal irradiance and relative humidity. The
cumulative distribution function and coefficient-of-determination (R2 ) were calculated to prove the
good agreement between the ground measurements and the modelled data from ERA5 interpolated
using the Inverse-Distance-Weighting method. For the validation, we considered daily average values
from 2016 to 2018.
In terms of global spatial distribution, we found a clear correlation between the
Köppen-Geiger-Photovoltaic (KGPV) climate classification and the estimated degradation rates.
In the temperate zones of Europe, the average degradation rate is in accordance with the typical
degradation rate of 0.5%/a considered widely by the PV community, however, this value can vary
around ± 0.3%/a for a specific year.
From our calculations, thermomechanical degradation is the harshest for the studied PV module
in nearly all climate zones, presenting the highest impact in very high irradiation zones, such as CK,
BK, and AK. Photo-degradation and hydrolysis-degradation show similar global spatial distribution,
but the former is higher since it also comprises UV irradiation as a degradation factor.
The temporal evolution of the degradation rates is directly correlated with the global ambient
temperature and it is evidence that climate change could impact the long-term performance of
PV systems.
We developed new maps for degradation mechanisms and the total degradation rate for the
studied PV module which can be directly integrated as a new layer over the KGPV climate classification
map and provide a rapid understanding of performance and degradation globally. However, due
to the high uncertainty in the real degradation rate of PV systems (solar resource, methodology of
calculation, quality of operational data, bill-of-materials of PV modules, etc.), the maps are presented as
a guide to identify possible risk areas in terms of climate stress but not to give quantified degradation
rates for specific locations for any PV module or PV system.

Supplementary Materials: The datasets generated in this article are available online as Mendeley data at
https://data.mendeley.com/datasets/3nt652dwwx/1.
Author Contributions: J.A.-V., I.K. worked on the data processing, simulations and analyses. K.B., K.-A.W.,
and M.T. discussed the results and helped to improve the schemes. All the authors discussed the results and
contributed, read and commented on the manuscript.
Funding: This project received funding from the European Union’s H2020 programme SOLAR-TRAIN under
grant agreement No 721452 and research programme P2-0197 funded by the Slovenia Research Agency.
Acknowledgments: The authors would like to thank the organizations ECMWF, DKA Solar Centre, Universidad
de Chile and WRMC-BSRN for providing climate datasets.
Conflicts of Interest: The authors declare no conflict of interest.
Energies 2019, 12, 4749 11 of 16

Appendix A Meteorological Stations and Data Validation

Table A1. Details of meteorological stations used for the validation of GHI, Tamb , and RH.

Elevation
Label Location Latitude Longitude Available Time Frame
[m]
ALI Alice Springs, Australia −23.76 133.88 560 2008-09 to 2019-11
CAB Cabauw, Netherlands 52.0 4.9 0 2005-02 to 2019-05
CAR Carpentras, France 44.1 5.1 100 1996-09 to 2018-12
CNR Pamplona, Spain 42.8 −1.6 471 2009-07 to 2019-02
E13 Oklahoma, USA 36.6 −97.5 318 1994-01 to 2017-05
FUA Fukuoka, Japan 33.6 130.4 3 2010-04 to 2019-04
GOB Gobabeb, Namibia −23.6 15.0 407 2012-05 to 2019-05
GUR Gurgaon, India 28.4 77.2 259 2014-07 to 2019-01
HOW Howrah, India 22.6 88.3 51 2014-10 to 2019-01
LIN
Energies Lindenberg, Germany 52.2 14.1 125 1994-10 to 2017-01
Energies 2020,
Energies 2020, 13,
2020, 13, xxx FOR
13, FOR PEER
FOR PEER REVIEW
PEER REVIEW
REVIEW
12
12 of
12 of 17
of 17
17
LRC2020,
Energies 13, x FORVirginia,
PEER REVIEW USA 37.1 −76.4 3 2014-12
12 of to
17 2019-05
Energies
Energies 2020,
2020, 13,
13, xx FOR
FOR PEER
PEER REVIEW
REVIEW 12
12 of
of 17
17
PALLRC Palaiseau, France 48.7 2.2 1562014-12 to 2019 2003-06 to 2019-02
LRC
LRC
Virginia,
Virginia,
USA
Virginia, USA
USA 37.1
37.1
37.1
−76.4
−76.4
−76.4
33 2014--12
12 to 2019---05
to 2019 05
PAY LRC
PAL Payerne, Switzerland
Virginia,
Palaiseau, USA
France 37.1
48.746.8 −76.4
2.2 6.9 33
156 491 2014
2014 - 12 to
2003-06 to2019
2019-02
05
1992-10
--05 to 2019-03
LRC
PAL
LRC
PAL Virginia,
Palaiseau,
Virginia,
Palaiseau, USA
France
USA
France 37.1
48.7
37.1
48.7 −76.4
2.2
−76.4
2.2 33
156
156 2014
2014 -
- 12
2003-06
12 to
to
to 2019
2019-02
2019
2003-06toto2019-02-05
2019-0205
SAP PAL
PAY Sapporo,
Palaiseau,
Payerne, Japan
France
Switzerland 48.7
46.843.1 2.2
6.9 141.3 156
491 17
2003-06 2010-04
1992-10 to 2019-03 to 2019-04
PAY
PAL
PAL
PAY Payerne,
Palaiseau,
Palaiseau,
Payerne, Switzerland
France
France
Switzerland 46.8
48.7
48.7
46.8 6.9
2.2
2.2
6.9 491
156
156
491 1992-10to
2003-06
2003-06
1992-10 to2019-02
to
to 2019-03
2019-02
2019-03
PAY
SAP San Pedro
Payerne,
Sapporo, de Atacama,
Switzerland
Japan 46.8
43.1 6.9
141.3 491
17 1992-10
2010-04 toto2019-03
2019-04
SPAPAYSAP
PAY
SAP Sapporo,
Payerne,
Payerne, Japan
Switzerland
Switzerland
Sapporo, Japan −22.98
43.1
46.8
46.8
43.1 141.3
6.9
6.9
141.3 −68.16 49117
491
17 2390 2010-04
1992-10
1992-10
2010-04to to
to
to 2019-04
2009-05 to 2014-06
2019-03
2019-03
to2019-04
2019-04
SAP San Pedro Chile
Sapporo, deJapan
Atacama, 43.1 141.3 17 2010-04
SAP SanSapporo,
Pedro dedeJapan
Atacama, 43.1 141.3 17 2010-04 to 2019-04
TATSAPSPA
SPA San
SPA
San Sapporo,
Pedro
Tateno,
Pedro
Japan
Atacama,
Japan
de Atacama,
43.1
−22.98
−22.98
−22.9836.1 141.3
−68.16
−68.16 140.1
−68.16
17
2390
2390
2390
2010-04
252009-05 to
2009-05 to
2009-05
2019-04
to 2014-06
2014-06
1996-02 to 2019-03
SPA San
San Pedro
Pedro Chile
Chile
de
de Atacama,
ChileAtacama, −22.98 −68.16 2390 2009-05 toto2014-06
2014-06
TIRSPA
SPA Tiruvallur, India
ChileJapan −22.98
−22.98 13.1 −68.16
−68.16 80.0 2390
2390 362009-05
2009-05 to 2014-06
2014-08 to 2019-01
TAT
TAT
TAT
Tateno,
Tateno,
Chile Japan
ChileJapan
Tateno,
36.1
36.1
36.1
140.1
140.1
140.1
25
25
25 1996-02to
1996-02
1996-02 to2014-06
to
to
2019-03
2019-03
2019-03
TAT
TIR
TIR Tateno, Japan
Tiruvallur,
Tiruvallur, India
India 36.1
13.1
13.1 140.1
80.0
80.0 25
36
36 1996-02
2014-08 to
2014-08to to2019-03
2019-01
to2019-03
2019-01
TAT
TAT
TIR Tateno,
Tateno, Japan
Tiruvallur,Japan
India 36.1
36.1
13.1 140.1
140.1
80.0 25
25
36 1996-02
1996-02
2014-08 to 2019-03
TIR Tiruvallur, India 13.1 80.0 36 2014-08 toto2019-01
2019-01
TIR
Table Tiruvallur,
TIRA2. Cumulative India
Tiruvallur,Distribution
India 13.1
Function
13.1 (CDF)80.0
for GHI, Tamb36
80.0 and RH to
36 2014-08
compare
2014-08 to
to 2019-01
ground stations
2019-01
Table
Table
from A2.
A2.
BSRN, Cumulative
Cumulative
DKA Solar Distribution
Distribution
Centre and Function
Function (CDF)
(CDF)
Universidad defor
for 𝐺𝐻𝐼
𝐺𝐻𝐼
Chile 𝑇𝑇𝑎𝑚𝑏
,,with and
and
modelled𝑅𝐻
𝑅𝐻 to
to
and compare
compare ground
ground
interpolated data from
TableA2.
Table A2.Cumulative
CumulativeDistribution
DistributionFunction
Function(CDF)
(CDF)for 𝐺𝐻𝐼, , 𝑇𝑇𝑎𝑚𝑏 and
for 𝐺𝐻𝐼 𝑎𝑚𝑏 𝑅𝐻 totocompare
and 𝑅𝐻 comparegroundground
stations
Table A2.
stations
Table from
A2.from BSRN,
Cumulative
BSRN,
Cumulative DKA
DKA Solar
Distribution
Solar
DistributionCentre
Centre and
Function
and
Function Universidad
(CDF) for
Universidad
(CDF) for de
𝐺𝐻𝐼
de
𝐺𝐻𝐼 Chile
, 𝑇
Chile
, 𝑇
𝑎𝑚𝑏with
and
with
and modelled
𝑅𝐻 to
modelled
𝑅𝐻 to and
compare
and
compareinterpolated
ground
interpolated
ground
ERA5
stations
climate
stations from
reanalysis.
fromBSRN,
BSRN, DKASolar
DKA SolarCentre
Centreand
andUniversidad
Universidadde deChile
Chile𝑎𝑚𝑏
𝑎𝑚𝑏with modelled and interpolated
with modelled and interpolated
data
data from
stations from
from
stations ERA5
ERA5
from climate
BSRN, DKA
climate
BSRN, DKA reanalysis.
Solar
Solar Centre
Centre and
reanalysis. and Universidad
Universidad de de Chile
Chile withwith modelled
modelled and
and interpolated
interpolated
data from ERA5 climate reanalysis.
data from ERA5 climate reanalysis.
data
data from
from ERA5
ERA5 climate reanalysis.
climateCumulative
CumulativeDistribution
reanalysis. Functions
Distribution Functions
Cumulative Distribution
Distribution Functions
Functions
Var. R22
Var.
Var. R R2
R2
Cumulative Var.
Cumulative Distribution Functions Var. R22
Cumulative
Cumulative Distribution
Distribution Functions
Functions Var.
Var. 0.76 R
R2
𝐺𝐻𝐼GHI0.76
𝐺𝐻𝐼
𝐺𝐻𝐼
0.76 0.76
𝑇𝑇
𝐺𝐻𝐼 0.76
0.98
0.98
𝐺𝐻𝐼 T
𝑎𝑚𝑏
𝑎𝑚𝑏
𝑇𝑅𝐻 0.76
𝑎𝑚𝑏amb0.98
0.76 0.98
𝑇𝐺𝐻𝐼𝑅𝐻 0.980.91
0.91
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 RH 0.98
0.91
0.98 0.91
𝑅𝐻 0.91
𝑎𝑚𝑏
𝑅𝐻
𝑅𝐻 0.910.91
𝐺𝐻𝐼GHI0.89
𝐺𝐻𝐼
𝐺𝐻𝐼
0.89
0.89 0.89
𝑇𝑇
𝐺𝐻𝐼𝑎𝑚𝑏 0.89
0.98
0.98
𝐺𝐻𝐼 𝑎𝑚𝑏
𝑇𝑅𝐻 0.89
T 0.98
0.98
0.89 0.98
𝑇𝐺𝐻𝐼𝑅𝐻amb
𝑎𝑚𝑏
0.82
0.82
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 RH 0.98
0.82
0.98 0.82
𝑅𝐻 0.82
𝑎𝑚𝑏
𝑅𝐻
𝑅𝐻 0.820.82
𝐺𝐻𝐼
𝐺𝐻𝐼 0.78
0.78
𝐺𝐻𝐼
GHI 0.78 0.78
𝑇𝑇
𝐺𝐻𝐼𝑎𝑚𝑏 0.78
0.98
0.98
𝐺𝐻𝐼 𝑎𝑚𝑏
𝑇𝑅𝐻 0.78
0.98
0.78
𝑇𝐺𝐻𝐼𝑅𝐻Tamb
𝑎𝑚𝑏 0.98
0.64
0.64 0.98
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 0.98
0.64
0.98
𝑅𝐻 𝑎𝑚𝑏 0.64
RH 0.64
𝑅𝐻
𝑅𝐻 0.640.64
𝐺𝐻𝐼
𝐺𝐻𝐼 0.81
0.81
0.81
𝐺𝐻𝐼 0.81
𝑇𝑇
𝐺𝐻𝐼 GHI
𝑎𝑚𝑏 0.99
0.99
0.81 0.81
𝐺𝐻𝐼
𝐺𝐻𝐼 𝑎𝑚𝑏
𝑇𝑅𝐻
𝑎𝑚𝑏 0.99
0.81
𝑇𝑎𝑚𝑏𝑅𝐻T 0.99
0.85
0.85 0.99
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻 0.99
amb0.85
0.99
𝑅𝐻 𝑎𝑚𝑏 0.85
𝑅𝐻
𝑅𝐻RH0.85
0.85 0.85
𝐺𝐻𝐼
𝐺𝐻𝐼 0.82
0.82
0.82
𝐺𝐻𝐼 0.82
𝑇𝑇
𝐺𝐻𝐼𝑎𝑚𝑏 0.99
0.99
𝐺𝐻𝐼
𝑇𝑅𝐻
𝐺𝐻𝐼
𝑇𝑎𝑚𝑏
GHI
𝑎𝑚𝑏
𝑎𝑚𝑏
0.82
0.99
0.82 0.82
𝑅𝐻 0.99
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻
0.80
0.80
0.99
𝑅𝐻 Tamb
𝑎𝑚𝑏 0.80
0.99
0.80 0.99
𝑅𝐻
𝑅𝐻RH0.80
0.80 0.80
𝐺𝐻𝐼 0.65
𝐺𝐻𝐼
𝐺𝐻𝐼
0.65
0.65
𝑇𝑇
𝐺𝐻𝐼𝑎𝑚𝑏 0.65
0.72
0.72
𝐺𝐻𝐼 𝑎𝑚𝑏 0.65
𝑇𝑅𝐻 0.72
0.65
𝐺𝐻𝐼
𝑇𝑎𝑚𝑏 GHI
𝑎𝑚𝑏 0.72
0.71 0.65
𝑅𝐻
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻 0.71
0.72
𝑅𝐻 0.71
0.72
T 0.71
𝑎𝑚𝑏 0.72
𝑅𝐻 amb
𝑅𝐻 0.71
0.71
RH 0.71
𝐺𝐻𝐼 0.63
𝐺𝐻𝐼
𝐺𝐻𝐼
0.63
0.63
𝑇𝑇𝑎𝑚𝑏
𝐺𝐻𝐼 0.63
0.98
0.98
𝐺𝐻𝐼 𝑎𝑚𝑏 0.63
𝑇𝑅𝐻 0.98
0.63
𝐺𝐻𝐼
𝑇𝑎𝑚𝑏 𝑎𝑚𝑏
𝑅𝐻 0.98
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻
0.82
0.82
0.98
𝑅𝐻 0.82
0.98
𝑎𝑚𝑏 0.82
𝑅𝐻
𝑅𝐻 0.820.82
𝐺𝐻𝐼 0.90
𝐺𝐻𝐼
𝐺𝐻𝐼
0.90
0.90
𝑇𝑇
𝐺𝐻𝐼𝑎𝑚𝑏 0.90
0.99
0.99
𝐺𝐻𝐼
𝑇 𝑎𝑚𝑏 0.90
0.99
0.90
𝑇𝐺𝐻𝐼 𝑎𝑚𝑏
𝑅𝐻
𝑅𝐻 0.990.93
0.93
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 0.99
0.93
0.99
𝑅𝐻 𝑎𝑚𝑏 0.93
𝑅𝐻
𝑅𝐻 0.930.93
𝐺𝐻𝐼 0.89
𝐺𝐻𝐼
𝐺𝐻𝐼
0.89
0.89
𝑇𝑇
𝐺𝐻𝐼𝑎𝑚𝑏 0.89
0.99
0.99
𝐺𝐻𝐼
𝑇 𝑎𝑚𝑏 0.89
0.99
0.89
𝑇𝐺𝐻𝐼 𝑎𝑚𝑏
𝑅𝐻
𝑅𝐻 0.990.78
0.78
𝑇𝑇𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 0.99
0.78
0.99
 𝐺𝐻𝐼 0.81
0.81
𝐺𝐻𝐼 0.81
𝑇𝐺𝐻𝐼
𝐺𝐻𝐼 0.81
0.99
𝑇 𝑎𝑚𝑏
𝐺𝐻𝐼
𝑇 𝑎𝑚𝑏 0.99
0.81
0.99
𝑇𝑅𝐻
𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 0.99
0.85
0.85
𝑇𝑅𝐻
𝑎𝑚𝑏 0.99
0.85
𝑅𝐻 0.85
𝑅𝐻 0.85

𝐺𝐻𝐼 0.82
0.82
𝐺𝐻𝐼 0.82
𝐺𝐻𝐼
Energies 2019, 12, 4749 𝑇
𝐺𝐻𝐼
𝑇 𝑎𝑚𝑏 0.82
0.99
0.99 12 of 16
𝑇 𝑎𝑚𝑏 0.82
𝐺𝐻𝐼 0.99
𝑇𝑅𝐻
𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 0.99
0.80
0.80
𝑇𝑅𝐻
𝑎𝑚𝑏 0.99
0.80
𝑅𝐻 0.80
𝑅𝐻 0.80

Table A2. Cont. 𝐺𝐻𝐼 0.65

0.65
𝐺𝐻𝐼 0.65
𝑇𝐺𝐻𝐼
𝐺𝐻𝐼 0.65
0.72
𝑇
𝑇
𝑎𝑚𝑏 0.72
𝑎𝑚𝑏 0.65
𝐺𝐻𝐼 0.72
𝑇𝑅𝐻
𝑎𝑚𝑏 0.72
Cumulative Distribution Functions 𝑅𝐻
𝑎𝑚𝑏
𝑇𝑅𝐻
𝑎𝑚𝑏
0.71
Var.0.72
0.71
0.71 R2
𝑅𝐻 0.71
𝑅𝐻 0.71

𝐺𝐻𝐼
𝐺𝐻𝐼 0.63
GHI0.63
0.63 0.63
𝑇𝐺𝐻𝐼
𝐺𝐻𝐼 0.63
0.98
𝑇
𝐺𝐻𝐼
𝑇 T
𝑎𝑚𝑏
𝑎𝑚𝑏 0.98
0.63
amb
0.98 0.98
𝑇𝑅𝐻
𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 0.98
0.82
0.82
𝑇𝑅𝐻
𝑎𝑚𝑏RH0.98
0.82 0.82
𝑅𝐻 0.82
𝑅𝐻 0.82

𝐺𝐻𝐼 0.90
0.90
𝐺𝐻𝐼
𝐺𝐻𝐼GHI0.90
0.90 0.90
𝑇
𝐺𝐻𝐼
𝑇 𝑎𝑚𝑏 0.99
0.99
𝑎𝑚𝑏 0.90
𝐺𝐻𝐼
𝑇
𝑇𝑅𝐻 T 0.99
𝑎𝑚𝑏 0.99
𝑎𝑚𝑏 amb
𝑅𝐻 0.93 0.99
𝑇𝑅𝐻
𝑎𝑚𝑏 0.93
0.99
0.93
𝑅𝐻 RH
𝑅𝐻 0.930.93 0.93

𝐺𝐻𝐼 0.89
0.89
𝐺𝐻𝐼 0.89
𝑇𝐺𝐻𝐼
𝐺𝐻𝐼 GHI0.89
0.99 0.89
𝑇 𝑎𝑚𝑏
𝐺𝐻𝐼
𝑎𝑚𝑏
𝑇𝑅𝐻 0.99
0.89
0.99
𝑇𝑅𝐻 Tamb0.99
𝑎𝑚𝑏
𝑎𝑚𝑏
𝑇𝑅𝐻
0.99
0.78
0.78 0.99
𝑎𝑚𝑏
𝑅𝐻 0.78
𝑅𝐻RH0.78
0.78 0.78
𝐺𝐻𝐼 0.87
0.87
𝐺𝐻𝐼 0.87
𝑇𝐺𝐻𝐼 0.87
𝐺𝐻𝐼
𝑇 GHI0.87
𝑎𝑚𝑏
𝐺𝐻𝐼
𝑎𝑚𝑏
0.99
0.99 0.87
𝑇𝑅𝐻
𝑇 𝑎𝑚𝑏 0.99
0.99
𝑅𝐻
𝑎𝑚𝑏
𝑇𝑅𝐻 0.83
Tamb0.99
0.83 0.99
𝑎𝑚𝑏
𝑅𝐻 0.830.83
𝑅𝐻RH0.83 0.83
𝐺𝐻𝐼 0.79
0.79
𝐺𝐻𝐼 0.79
𝐺𝐻𝐼
𝑇
𝐺𝐻𝐼
𝑇 𝑎𝑚𝑏 0.79
0.98
0.98
𝐺𝐻𝐼
𝑇 GHI
𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏
0.79
0.98 0.79
𝑇𝑅𝐻
𝑎𝑚𝑏 0.98
0.47
0.47
𝑇𝑅𝐻
𝑎𝑚𝑏T 0.98
0.47 0.98
Energies2020,
2020,13,
13,xxFOR
FORPEER
PEERREVIEW
REVIEW 𝑅𝐻 amb
0.47
13 of 17
Energies 𝑅𝐻 13
Energies 2020, 13, x FOR PEER REVIEW RH13ofof17
0.47
17 0.47
Energies 2020, 13, x FOR PEER REVIEW 13 of 17

𝐺𝐻𝐼
𝐺𝐻𝐼 0.72
GHI0.72 0.72
𝑇𝐺𝐻𝐼 0.72
0.99
𝑇𝐺𝐻𝐼T
𝑎𝑚𝑏
𝑎𝑚𝑏 0.72
0.99 0.99
𝑇𝑅𝐻
𝑎𝑚𝑏
amb0.99
𝑇𝑅𝐻 0.62
0.99
𝑅𝐻RH0.62
𝑎𝑚𝑏
0.62 0.62
𝑅𝐻 0.62

𝐺𝐻𝐼 0.79
𝐺𝐻𝐼GHI0.79
0.79 0.79
𝑇𝐺𝐻𝐼
𝑇𝐺𝐻𝐼 0.84
0.79
𝑎𝑚𝑏 0.84
T
𝑎𝑚𝑏
𝑇𝑅𝐻
𝑎𝑚𝑏 0.84
amb
0.75
0.84
𝑇𝑅𝐻 0.84
𝑅𝐻 0.75
𝑎𝑚𝑏
𝑅𝐻RH0.75
0.75 0.75

0.86
𝐺𝐻𝐼 0.86
𝐺𝐻𝐼
𝑇𝐺𝐻𝐼 0.86
GHI0.86
0.99 0.86
𝑇𝐺𝐻𝐼
𝑎𝑚𝑏 0.99
𝑎𝑚𝑏
𝑇𝑅𝐻
𝑎𝑚𝑏 0.99
T 0.86
0.86 0.99
𝑇𝑅𝐻 0.99
𝑅𝐻 amb
𝑎𝑚𝑏
0.86
𝑅𝐻RH0.86 0.86

0.49
𝐺𝐻𝐼 0.49
𝐺𝐻𝐼
𝑇𝐺𝐻𝐼 0.49
0.96
𝑇𝐺𝐻𝐼GHI
𝑎𝑚𝑏 0.49
0.96 0.49
𝑇𝑎𝑚𝑏
𝑅𝐻
𝑎𝑚𝑏 0.96
0.75
𝑅𝐻
𝑇𝑎𝑚𝑏 0.96
Tamb0.75 0.96
𝑅𝐻 0.75
𝑅𝐻 0.75
RH 0.75
Energies 2019, 12, 4749 13 of 16

Energies 2020, 13, x FOR PEER REVIEW 14 of 17

Appendix B Mapping of Degradation Mechanisms and Total Degradation Rate
Appendix B. Mapping of Degradation Mechanisms and Total Degradation Rate

Figure A1. Global mapping of degradation mechanisms. (a) Hydrolysis-degradation, (b) photo-
Figure A1. Global
degradation mapping
and (c) of degradation
thermomechanical mechanisms. top Hydrolysis-degradation, middle
degradation.
photo-degradation and bottom thermomechanical degradation.
Energies 2019, 12, 4749 14 of 16
Energies
Energies 2020,
2020, 13,
13, xx FOR
FOR PEER
PEER REVIEW
REVIEW 15
15 of
of 17
17

Figure
Figure A2. A2.
Figure A2. Total
Total
Total degradation
degradation
degradation rates for
foraaaspecific
ratesfor
rates specific
specific mono-crystalline
mono-crystalline silicon
mono-crystalline silicon PV module
PVPV
silicon module using
using
module the
the Kaaya
using Kaaya
the Kaaya
model.
model. Climate
Climate data
data from
from ERA5
ERA5 for
for the
the average
average between
between 2016
2016 and
and 2018.
2018.
model. Climate data from ERA5 for the average between 2016 and 2018.

Figure
Figure A3. A3.
Figure A3. Categorization
ofof
Categorization
Categorization of total degradation
degradation rates
totaldegradation
total rates for
rates foraaaspecific
for specific mono-crystalline
mono-crystalline
specific silicon
silicon
mono-crystalline PV
PV module
siliconmodule
PV module
usingusing
using the
the Kaaya
the KaayaKaaya model.
model.
model. Climate
Climate
Climate data from
datafrom
data ERA5
ERA5 for
fromERA5 for the
for theaverage
the average
average between
between 2016
2016
between and 2018.
andand
2016 2018.
2018.

References
References
References
1.
1. Haegel,
Haegel, N.M.;
N.M.; Atwater,
Atwater, H.;
H.; Barnes,
Barnes, T.;
T.; Breyer,
Breyer, C.;
C.; Burrell,
Burrell, A.;
A.; Chiang,
Chiang, Y.-M.;
Y.-M.; De
De Wolf,
Wolf, S.;
S.; Dimmler,
Dimmler, B.;
B.; Feldman,
Feldman,
1. Haegel, N.M.; Atwater, H.; Barnes, T.; Breyer, C.; Burrell, A.; Chiang, Y.-M.; De Wolf, S.; Dimmler, B.;
D.;
D.; Glunz,
Glunz, S.;
S.; et
et al.
al. Terawatt-scale
Terawatt-scale photovoltaics:
photovoltaics: Transform
Transform global global energy.
energy. Science
Science 2019,
2019, 364,
364, 836–838.
836–838.
Feldman, D.; Glunz, S.; et al. Terawatt-scale photovoltaics: Transform global energy. Science 2019, 364,
2.
2. Vartiainen,
Vartiainen, E.;
E.; Masson,
Masson, G.; G.; Breyer,
Breyer, C.;
C.; Moser,
Moser, D.;
D.; Román
Román Medina,
Medina, E. E. Impact
Impact of of weighted
weighted average
average cost
cost of
of
836–838. [CrossRef]
capital,
capital, capital
[PubMed] and other
capital expenditure,
expenditure, and other parameters
parameters on on future
future utility‐scale
utility‐scale PVPV levelised
levelised cost
cost of
of electricity.
electricity.
2. Vartiainen,
Prog. E.; Masson,
Prog. Photovolt.
Photovolt. Res. G.; 2019,
Res. Appl.
Appl. Breyer, C.; Moser, D.; Román Medina, E. Impact of weighted average cost of
2019, doi:10.1002/pip.3189.
doi:10.1002/pip.3189.
capital,
3.
3. Jordan, D.C.; Wohlgemuth, J.H.; Kurtz,parameters
capital
Jordan, D.C.;expenditure,
Wohlgemuth, and
J.H.; other
Kurtz, S.R.
S.R. Technology
Technologyon future
and utility-scale
and Climate
Climate Trends
Trends inPVPV
in levelised
PV Module cost of electricity.
Module Degradation.
Degradation.
Prog.InPhotovolt. Res. Appl.
the 2019.
27th [CrossRef]
In Proceedings of the 27th European Photovoltaic Solar Energy Conference and Exhibition, Frankfurt,
Proceedings of European Photovoltaic Solar Energy Conference and Exhibition, Frankfurt,
3. Germany,
Germany,
Jordan, 24–28
24–28 September
September
D.C.; Wohlgemuth, 2012.
2012.
J.H.; Kurtz, S.R. Technology and Climate Trends in PV Module Degradation.
4. Jordan,
Jordan, D.C.;
4. Proceedings
In D.C.;ofKurtz,
Kurtz,
the 27thS.R.; VanSant,
VanSant, K.;
S.R.;European K.; Newmiller, J.J. Compendium
Newmiller, Solar
Photovoltaic Compendium of
of photovoltaic
Energy Conference and degradation
photovoltaic Exhibition, rates:
degradation rates:
Frankfurt,
Photovoltaic
Photovoltaic degradation
degradation
Germany, 24–28 September 2012. rates.
rates. Prog.
Prog. Photovolt.
Photovolt. Res.
Res. Appl.
Appl. 2016,
2016, 24,
24, 978–989.
978–989.
5.
5. Jordan,
Jordan, D.C.; Silverman,
Silverman, T.J.;T.J.; Wohlgemuth,
Wohlgemuth, J.H.; J.H.; Kurtz,
Kurtz, S.R.;S.R.; VanSant,
VanSant, K.T. Photovoltaic
Photovoltaic failure
failure and
4. Jordan, D.C.;D.C.;
Kurtz, S.R.; VanSant, K.; Newmiller, J. Compendium ofK.T.
photovoltaic degradation and
rates:
degradation
degradation modes:
modes: PV PV failure
failure and
and degradation
degradation modes.
modes. Prog.Prog. Photovolt.
Photovolt. Res.
Res. Appl.
Appl. 2017,
2017, 25,
25, 318–326.
318–326.
Photovoltaic degradation rates. Prog. Photovolt. Res. Appl. 2016, 24, 978–989. [CrossRef]
6.
6. Omazic,
Omazic, A.;
A.; Oreski,
Oreski, G.;G.; Halwachs,
Halwachs, M.; M.; Eder,
Eder, G.C.;
G.C.; Hirschl,
Hirschl, C.; C.; Neumaier,
Neumaier, L.; L.; Pinter,
Pinter, G.;
G.; Erceg,
Erceg, M.
M. Relation
Relation
5. Jordan, D.C.; Silverman, T.J.; Wohlgemuth, J.H.; Kurtz, S.R.; VanSant, K.T. Photovoltaic failure and degradation
between
between degradation
degradation of of polymeric
polymeric components
components in in crystalline
crystalline silicon
silicon PV
PV module
module andand climatic
climatic conditions:
conditions: AA
modes: PV failure
literature
literature review.
review. and degradation
Sol.
Sol. Energy Mater.modes.
Energy Mater. Sol. Prog.
Sol. Cells
Cells 2019,Photovolt.
2019, 192,
192, 123–133. Res. Appl. 2017, 25, 318–326. [CrossRef]
123–133.
6. Omazic, A.; Oreski, G.; Halwachs, M.; Eder, G.C.; Hirschl, C.; Neumaier, L.; Pinter, G.; Erceg, M. Relation
between degradation of polymeric components in crystalline silicon PV module and climatic conditions:
A literature review. Sol. Energy Mater. Sol. Cells 2019, 192, 123–133. [CrossRef]
Energies 2019, 12, 4749 15 of 16

7. Halwachs, M.; Neumaier, L.; Vollert, N.; Maul, L.; Dimitriadis, S.; Voronko, Y.; Eder, G.C.; Omazic, A.;
Mühleisen, W.; Hirschl, C.; et al. Statistical evaluation of PV system performance and failure data among
different climate zones. Renew. Energy 2019, 139, 1040–1060. [CrossRef]
8. Eder, G.C.; Voronko, Y.; Dimitriadis, S.; Knöbl, K.; Újvári, G.; Berger, K.A.; Halwachs, M.; Neumaier, L.;
Hirschl, C. Climate specific accelerated ageing tests and evaluation of ageing induced electrical, physical,
and chemical changes. Prog. Photovolt. Res. Appl. 2018, 1–16. [CrossRef]
9. Dubey, R.; Chattopadhyay, S.; Kuthanazhi, V.; Kottantharayil, A.; Singh Solanki, C.; Arora, B.M.;
Narasimhan, K.L.; Vasi, J.; Bora, B.; Singh, Y.K.; et al. Comprehensive study of performance degradation of
field-mounted photovoltaic modules in India. Energy Sci. Eng. 2017, 5, 51–64. [CrossRef]
10. Kaaya, I.; Koehl, M.; Mehilli, A.P.; de Cardona Mariano, S.; Weiss, K.A. Modeling Outdoor Service Lifetime
Prediction of PV Modules: Effects of Combined Climatic Stressors on PV Module Power Degradation.
IEEE J. Photovolt. 2019, 9, 1105–1112. [CrossRef]
11. Lindig, S.; Kaaya, I.; Weis, K.-A.; Moser, D.; Topic, M. Review of Statistical and Analytical Degradation
Models for Photovoltaic Modules and Systems as Well as Related Improvements. IEEE J. Photovolt. 2018, 8,
1773–1786. [CrossRef]
12. Ascencio-Vásquez, J.; Brecl, K.; Topič, M. Methodology of Köppen-Geiger-Photovoltaic climate classification
and implications to worldwide mapping of PV system performance. Sol. Energy 2019, 191, 672–685.
[CrossRef]
13. Peters, I.M.; Liu, H.; Reindl, T.; Buonassisi, T. Global Prediction of Photovoltaic Field Performance Differences
Using Open-Source Satellite Data. Joule 2018, 2, 307–322. [CrossRef]
14. Victoria, M.; Andresen, G.B. Using validated reanalysis data to investigate the impact of the PV system
configurations at high penetration levels in European countries. Prog. Photovolt. Res. Appl. 2019, 27, 576–592.
[CrossRef]
15. Bala Subramaniyan, A.; Pan, R.; Kuitche, J.; TamizhMani, G. Quantification of Environmental Effects on PV
Module Degradation: A Physics-Based Data-Driven Modeling Method. IEEE J. Photovolt. 2018, 8, 1289–1296.
[CrossRef]
16. Copernicus Climate Change Service (C3S) ERA5: Fifth Generation of ECMWF Atmospheric Reanalyses of
the Global Climate. Copernicus Climate Change Service Climate Data Store (CDS) 2017. Available online:
https://cds.climate.copernicus.eu/cdsapp#!/home (accessed on 13 October 2019).
17. Driemel, A.; Augustine, J.; Behrens, K.; Colle, S.; Cox, C.; Cuevas-Agulló, E.; Denn, F.M.; Duprat, T.;
Fukuda, M.; Grobe, H.; et al. Baseline Surface Radiation Network (BSRN): Structure and data description
(1992–2017). Earth Syst. Sci. Data 2018, 10, 1491–1501. [CrossRef]
18. Desert Knowledge Australia Centre Download Data. Location: Alice Springs. Available online: http:
//dkasolarcentre.com.au/historical-data/download (accessed on 25 November 2019).
19. Molina, A.; Falvey, M.; Rondanelli, R. A solar radiation database for Chile. Sci. Rep. 2017, 7, 14823. [CrossRef]
20. Urraca, R.; Huld, T.; Martinez-de-Pison, F.J.; Sanz-Garcia, A. Sources of uncertainty in annual global horizontal
irradiance data. Sol. Energy 2018, 170, 873–884. [CrossRef]
21. Camus, C.; Offermann, P.; Weissmann, M.; Buerhop, C.; Hauch, J.; Brabec, C.J. Site-specific assessment
of mechanical loads on photovoltaic modules from meteorological reanalysis data. Sol. Energy 2019, 188,
1134–1145. [CrossRef]
22. Schwingshackl, C.; Petitta, M.; Wagner, J.E.; Belluardo, G.; Moser, D.; Castelli, M.; Zebisch, M.; Tetzlaff, A.
Wind Effect on PV Module Temperature: Analysis of Different Techniques for an Accurate Estimation.
Energy Procedia 2013, 40, 77–86. [CrossRef]
23. Huld, T.; Amillo, A. Estimating PV Module Performance over Large Geographical Regions: The Role of
Irradiance, Air Temperature, Wind Speed and Solar Spectrum. Energies 2015, 8, 5159–5181. [CrossRef]
24. Buck, A.L. New Equations for Computing Vapor Pressure and Enhancement Factor. J. Appl. Meteorol. 1981,
20, 1527–1532. [CrossRef]
25. Buck Research Instruments, LLC. CR-1A User’s Manual; Buck Research Instruments, LLC: Aurora, CO, USA,
2012.
26. Wald, L. A Simple Algorithm for the Computation of the Spectral Distribution of the Solar Irradiance at Surface;
Research Report; Mines ParisTech: Paris, France, 2018.
27. Kuo, P.-H.; Chen, H.-C.; Huang, C.-J. Solar Radiation Estimation Algorithm and Field Verification in Taiwan.
Energies 2018, 11, 1374. [CrossRef]
Energies 2019, 12, 4749 16 of 16

28. Yang, D. Post-processing of NWP forecasts using ground or satellite-derived data through kernel conditional
density estimation. J. Renew. Sustain. Energy 2019, 11, 026101. [CrossRef]
29. Skoplaki, E.; Palyvos, J.A. Operating temperature of photovoltaic modules: A survey of pertinent correlations.
Renew. Energy 2009, 34, 23–29. [CrossRef]
30. Faiman, D. Assessing the outdoor operating temperature of photovoltaic modules. Prog. Photovolt. Res. Appl.
2008, 16, 307–315. [CrossRef]
31. Koehl, M.; Heck, M.; Wiesmeier, S. Categorization of weathering stresses for photovoltaic modules.
Energy Sci. Eng. 2018, 6, 93–111. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).