sensors

Review
Calibration Method for Particulate Matter Low-Cost Sensors
Used in Ambient Air Quality Monitoring and Research
Janani Venkatraman Jagatha 1, * , André Klausnitzer 1 , Miriam Chacón-Mateos 2 , Bernd Laquai 2 ,
Evert Nieuwkoop 3 , Peter van der Mark 3 , Ulrich Vogt 2 and Christoph Schneider 1

1 Geography Department, Humboldt-Universität zu Berlin, Unter den Linden 6, D-10099 Berlin, Germany;
[email protected] (A.K.); [email protected] (C.S.)
2 Department of Flue Gas Cleaning and Air Quality Control, Institute of Combustion and Power Plant
Technology (IFK), University of Stuttgart, Pfaffenwaldring 23, 70569 Stuttgart, Germany;
[email protected] (M.C.-M.); [email protected] (B.L.);
[email protected] (U.V.)
3 Netherlands Organisation for Applied Scientific Research, Anna van Buerenplein 1, 2595 DA The Hague,
The Netherlands; [email protected] (E.N.); [email protected] (P.v.d.M.)
* Correspondence: [email protected]; Tel.: +49-30-2093-6898

Abstract: Over the last decade, manufacturers have come forth with cost-effective sensors for mea-
suring ambient and indoor particulate matter concentration. What these sensors make up for in cost
efficiency, they lack in reliability of the measured data due to their sensitivities to temperature and
relative humidity. These weaknesses are especially evident when it comes to portable or mobile
measurement setups. In recent years many studies have been conducted to assess the possibili-





 ties and limitations of these sensors, however mostly restricted to stationary measurements. This
study reviews the published literature until 2020 on cost-effective sensors, summarizes the recom-
Citation: Venkatraman Jagatha, J.;
Klausnitzer, A.; Chacón-Mateos, M.;
mendations of experts in the field based on their experiences, and outlines the quantile-mapping
Laquai, B.; Nieuwkoop, E.; van der methodology to calibrate low-cost sensors in mobile applications. Compared to the commonly used
Mark, P.; Vogt, U.; Schneider, C. linear regression method, quantile mapping retains the spatial characteristics of the measurements,
Calibration Method for Particulate although a common correction factor cannot be determined. We conclude that quantile mapping can
Matter Low-Cost Sensors Used in be a useful calibration methodology for mobile measurements given a well-elaborated measurement
Ambient Air Quality Monitoring and plan assures providing the necessary data.
Research. Sensors 2021, 21, 3960.
https://doi.org/10.3390/s21123960 Keywords: air pollution; low-cost sensors; particulate matter; quantile mapping; mobile measurements

Academic Editor: Ole Hertel

Received: 23 April 2021

1. Introduction
Accepted: 6 June 2021
Published: 8 June 2021
Epidemiological studies reveal that there is concrete evidence of the connection be-
tween poor air quality due to particulate matter (PM) pollution and health [1–5]. The
Publisher’s Note: MDPI stays neutral
United Nations Sustainable Development Goal 11 includes clean air as a basic human
with regard to jurisdictional claims in
requirement for health and wellbeing and aims to reduce the per capita environmental
published maps and institutional affil- impact, especially in urban surroundings [6]. The World Health Organization (WHO)
iations. classifies outdoor air pollution as a leading environmental cause of death due to cancer [7].
Air quality (AQ) awareness has also been rapidly increasing among citizens over the last
decades [8]. Air quality monitoring stations (AQMS) run by institutions or governmental
agencies are usually point-based and location-specific. The spatial coverage of such mea-
Copyright: © 2021 by the authors.
surements is insufficient owing to the costs [9,10]. This fact, coupled with the advancements
Licensee MDPI, Basel, Switzerland.
in micro-sensing technology, contributes to a paradigm shift from conventional air quality
This article is an open access article
monitoring networks to rapid growth in air quality monitoring systems (AQMS) set up
distributed under the terms and either by private initiatives or public institutions using low-cost sensors (LCS) which com-
conditions of the Creative Commons plement existing air quality monitoring networks [11,12]. Purple-Air [13] LCS company for
Attribution (CC BY) license (https:// instance works together with the Environmental Protection Agency of the United States of
creativecommons.org/licenses/by/ America (EPA) in its projects [14]. In Germany, Breeze Technologies [15] works on projects
4.0/). together with different state governments within their projects.

Sensors 2021, 21, 3960. https://doi.org/10.3390/s21123960 https://www.mdpi.com/journal/sensors

Sensors 2021, 21, 3960 2 of 27

The EPA recognises the LCS for its versatility and has funded projects to assess the
quality of the LCS and its applications [16]. The draft roadmap for the next generation
of air quality monitoring from the EPA includes LCS [17]. The European Union’s science
hub, the European Commission (EC) followed suit and launched its projects to determine
the possibilities and limitations of LCS [18–20]. The Joint Research Centre (JRC) of the
EC has taken it upon itself to test LCS and classify them according to their probable
applications [21–23]. The JRC welcomes the frequent use of LCS by citizens, researchers,
and institutions despite their drawbacks. Sense-Box project [24], supported by the Federal
Ministry of Education and Research (BMBF), Germany, and the AirSensEUR project [25] of
the JRC, indicate the recognition of the LCS by government organisations. The European
Standardisation Organisation is working on a protocol [26,27] to evaluate LCS based on
common criteria [28].
Over the last decade, many manufacturers have come forth with LCS for PM (PM-LCS)
for measuring the ambient and indoor air quality. The costs range from EUR 15 to EUR 500
(see Section 2). PM-LCS are available as a unit produced by Original Equipment Manufac-
turers (OEM) and as a Sensor Box containing one or an ensemble of sensors produced by
the OEMs. In addition, there are ‘do-it-yourself’ sensors available for LCS enthusiasts to
custom design their sensor boxes. What these sensors make up for in cost, they lack in the
reliability of the measured data due to their straightforward technological concepts, their
sensitivities to temperature and relative humidity, and due to inappropriate usage when
the sensors are deployed for studies outside of the manufacture specifications, for example,
using an indoor sensor for outdoor measurements. In recent years a series of studies have
been conducted to assess the possibilities and limitations of these sensors [29–32].
Moreover, the sensor market is extremely volatile, with new manufacturers entering
the market and new sensors or upgrades of existing sensors hitting the market. This further
complicates studies employing LCS as some of the available reports on specific sensors are
becoming obsolete and new studies have to be carried out to assess the performance of
newly released ones.
This study aims to review the calibration of PM-LCS as PM is one of the major air
pollutants relevant in ambient air monitoring. It builds on a literature review and a wide
survey among experts in the field in Germany amended by interviews in other countries
of the European Union. Experts in the field were consulted for their experiences and
recommendations. We further report and discuss own experiments for the calibration of
mobile measurements with self-made PM-LCS measurement units. The sensors considered
for this study are the ones that are currently being mostly used by citizen science initiatives
and the scientific communities in Germany, but elsewhere as well and are defacto-standard
in many comparison studies [11,12,33].

2. State of Art in PM-LCS Correction Procedures

The LCS technology and its usage among individual laymen, for example, within
citizen science initiatives and researchers, are rapidly increasing. Since this is a relatively
new technology with a highly volatile market, a lot of information is still available as grey
literature and technical reports only [23]. The state of the art analysis of LCS in this chapter
is provided in the form of a short overview of LCS technologies, a literature review, and
from interviews with experienced and amateur researchers in the LCS area.

2.1. Principle Measurement Techniques for PM

Several methods, such as gravimetric sampling, photometry, and ß-radiation attenua-
tion, are used to determine the concentration of PM in the air [34,35]. However, photometry
using laser scattering is commonly used in most LCS due to its shorter response time and
lower power requirement.
In reference-grade PM instruments, as explained in Baumbach (1996) [34], a beam of
light is penetrated through an airflow channel (flue gas), and the extinction (logarithm of
the ratio of emitted light intensity to the attenuated light intensity) is measured. According
Sensors 2021, 21, 3960 3 of 27

to the Lambert-Beer law [36], extinction is directly proportional to the concentration of

interest in the flue gas. However, a light beam sent through a gas volume laden with
particles is not only attenuated but also scattered. This phenomenon of scattering of light
by particles is used in ambient air measurements. In light-scattered photometry, a beam of
light is sent by the emitter to a flicker mirror which alternatively generates a measuring
beam and a reference beam. The particles from the air sucked into the system scatters
the measuring beam which is detected by the photocell. The amount of scattered light is
measured in comparison to the reference beam which is indicative of the concentration of
the particles in the air.
Continuously measuring PM instruments, such as the Grimm 1.108/1.109 (henceforth
referred to as Grimm), for example, use the Mie-Scattering principle, in addition, to enable
particle sizing [37]. This is carried out in Grimm by using a semiconductor laser as a light
source and a measuring cell wherein the scattered light is led directly and via a mirror
onto a detector. The scattering light pulse of every single particle is counted along with
the intensity of the scattered light signal giving rise to both particle counts and particle
sizing, respectively. An accurate sampling volume of 1.2 L/min is used by Grimm to enable
precise and reproducible particle counting and particle sizing [37].
In PM-LCS such as the OPC series (OPC-N2, OPC-N3, and OPC-R1) from Alphasense
Ltd. [38]., Nova SDS011 (SDS011) [39], SPS30 from Sensiron GmbH [40] operate using
the laser scattering principle, similar to that of Grimm. Alphasense replaced the power-
ful pump and measurement cell in Grimm with a mircrofan leading to a virtual sensing
zone [38]. It measures the particles and creates a size-distribution. The mass concentration
is then obtained using an algorithm using the number concentration, refractive index,
particle density and a weighting-factor. Alphasense uses a particle density of 1.65 g/mL
but provides an option to alter the particle size density for each bin according to the need
of users and the refractive index of particles. SDS011 and SPS30 do not specify how a
“measurement cell” is created within their sensors. SDS011 and SPS30 use algorithms to
calculate the mass concentration for PM10 by using the particle number concentration mea-
sured, an assumed particle size distribution, and particle density. An in-depth discussion
of the measurement principles of most LCS is provided in Alfano et al. (2020) [41].
The information available from the datasheets of the most popular PM LCS sensors
are summarised in Tables 1 and 2. Table 1 lists the manufacturer, model number, the
dimensions of the sensor in mm, measurement principle used (laser scattering spectrometer,
LSS, or photometer), measurement and detection ranges, time resolution (T.R.), and the
approximate cost in Euro (€). Table 2 dives deeper into the information on the electrical
and performance characteristics of the sensors mentioned in Table 1 and lists the nominal
voltage (V) and maximum power consumption in Watts (W), operating temperature, and
relative humidity range under condensing or non-condensing (n.c.) conditions, uncertainty,
sensor life, availability of calibration, and the reaction time in seconds (s). Information on
the repeatability and drift was not available for any of the sensors. “NA” is used to indicate
when information was unavailable.
Sensors 2021, 21, 3960 4 of 27

Table 1. List of low-cost PM sensors with the name of the manufacturer, model number, dimensions in mm, measurement principle used, the measurement and detection ranges, time
resolution (T.R.), and the approximate cost.

Manufacturer Model Dimension Principle Measurement and Detection Range T.R. Cost
0.38–17 µm, 16 Channels (Number concentration), PM1,
OPC-N2 75 × 63.5 × 60 L.S.S. 1.4 s N.A.
PM2.5, PM10
0–2000 µg/m3
0.35–40 µm, 24 Channels (Number concentration), PM1,
Alphasense Ltd. (Great Britain) OPC-N3 75 × 63.5 × 60 L.S.S. 1s 415 €
PM2.5, PM10,
Temperature and RH
0.35–12.4 µm
OPC-R1 72 × 25.5 × 21.5 L.S.S. 16 Channels (Number concentration), PM1, PM2.5, PM10, 1s 210 €
Temperature and RH
DC1700 PM 0–106 Particle/cm3
Dylos Corp (USA) 17.8 × 11.4 × 7.6 L.S.S. 60 s 420 €
PM2.5/PM10 AQM >0.5 and >2.5 µm and PM2.5 and PM10 in µg/m3
0–1000 µg/m3
Honeywell (USA) HPMA115SO-XXX 36 × 43 × 24 L.S.S. PM2.5 in µg/m3 (PM10 in µg/m3 with N.A. 30 €
additional programming)
831 Aerosol Mass 0–1.000 µg/m3
Met One (USA) 159 × 92.2 × 50.8 Photometer 60 s 1700 €
Monitor >0.1 µm
SDS011 71 × 70 × 23 L.S.S. 0–999.9 µg/m3 0.3–10 µm 1s 30 €
Nova Fitness (China) SDS018 59 × 45 × 20 L.S.S. 0–999.9 µg/m3 0.3–10 µm 1s 30 €
SDS198 71 × 70 × 23 L.S.S. 0–20 mg/m3 1–100 µm 1s 30 €
0–500 µg/m3
PMS 1003 65 × 42 × 23 L.S.S. N.A. 15 €
0.3–1.0; 1.0–2.5; 2.5–10 µm in three channels
0–500 µg/m3
PMSA003 38 × 35 × 12 L.S.S. N.A. 20 €
Plantower 0.3–1.0; 1.0–2.5; 2.5–10 µm in three channels
(China)
PMS 3003 65 × 42 × 23 L.S.S. 0.3–1.0; 1.0–2.5; 2.5–10 µm in three channels N.A. 20 €
PMS 5003 N.A. L.S.S. N.A. N.A. 15 €
PMS 7003 N.A. L.S.S. N.A. N.A. 20 €
Sensors 2021, 21, 3960 5 of 27

Table 1. Cont.

Manufacturer Model Dimension Principle Measurement and Detection Range T.R. Cost

Samyoung PSML(LPO) N.A. Photometer 0–900 µg/m3 PM2.5 and PM1 1s N.A.
(South Korea) DSM501A N.A. Photometer >1 µm N.A. 17 €
1–1000 µg/m3
Sensiron
SPS30 40.6 × 40.6 × 12.2 L.S.S PM1, PM2.5, PM4, and PM10 (Mass) N.A. 40 €
(Switzerland)
PM0.5, PM1, PM2.5, PM4 and PM10 (Particle number)
GP2Y1010AU0F 46 × 30 × 17.6 Photometer N.A. N.A. 12 €
Sharp (Japan)
DN7C3CA006 51 × 53 × 40 Photometer 25–500 µg/m3 N.A. 22 €
PPD42NJ 59 × 45 × 22 Photometer >1 µm N.A. 25 €
PPD60PV-T2 88 × 60 × 20 Photometer >0.5 µm N.A. N.A.
Shinyei (China)
PPD20V 88 × 60 × 20 Photometer >1 µm N.A. N.A.
PPD71 34 × 30 × 28 Photometer >0.5 µm N.A. N.A.
Winsen
ZH03B 50 × 32.4 × 21 Photometer 0–1000 µg/m3 N.A. 32 €
(China)

Table 2. Electrical and performance characteristics of low-cost PM sensors with the name of the manufacturer, model number, operating temperature (T) and relative humidity (RH) range,
nominal voltage and maximum power consumption, uncertainty, sensor life, availability of calibration, and the reaction time in seconds (s); “n.c.” stands for “non condensing”.

Voltage and Power Integrated

Manufacturer Model T and RH Uncertainty Sensor Life Reaction Time
Consumption Calibration
−10–50 ◦ C 4.8–5.2 V
OPC-N2 N.A. N.A. No N.A.
0–95% (n.c.) 0.90 W

Alphasense Ltd. −10–50 ◦ C 4.8–5.2 V

OPC-N3 N.A. N.A. Yes N.A.
(Great Britain) 0–95% (n.c.) 0.90 W
−10–50 ◦ C 4.8–5.2 V
OPC-R1 N.A. N.A. Yes N.A.
0–95% (n.c.) 0.48 W
DC1700 PM
Dylos Corp (USA) N.A. 110 V or Battery N.A. N.A. Yes 6s
PM2.5/PM10 AQM
Sensors 2021, 21, 3960 6 of 27

Table 2. Cont.

Voltage and Power Integrated

Manufacturer Model T and RH Uncertainty Sensor Life Reaction Time
Consumption Calibration
±15 µg/m3 (0–100 µg/m3 )
−10–50 ◦ C 5 ± 0.2 V
Honeywell (USA) HPMA115SO-XXX ±15% (100–1000 µg/m3 ) 20,000 h No <6 s
0–95% (n.c.) 0.40 W
at 25 ± 5 ◦ C
831 Aerosol Mass 100–240 V to 8.4 V(AC/DC)
Met One (USA) 0–50 ◦ C N.A. N.A. Yes N.A.
Monitor Li-Battery, rechargeable
−20–60 ◦ C 5 ± 0.2 V Max. ±15% and ±10 µg/m3
SDS011 N.A. No <10 s
<70% 0.40 W at 25 ◦ C, 50% RH
−20–60 ◦ C 5 ± 0.2 V Max. ±15% and ±10 µg/m3
Nova Fitness (China) SDS018 N.A. No <10 s
<70% 0.35 W at 25 ◦ C, 50% RH
−20–60 ◦ C 5 ± 0.2 V Max. ±20% and ±30 µg/m3
SDS198 N.A. No <10 s
<70% 0.40 W at 25 ◦ C, 50% RH
PMS 1003 N.A. 5 ± 0.2 V N.A. N.A. No N.A.
−10–60 ◦ C ±10 µg/m3 (0–100 µg/m3 )
PMSA003 5 ± 0.5 V N.A. No 10 s
0–99% ±10% (100–500 µg/m3 )
Plantower
(China) PMS 3003 N.A. 5 ± 0.2 V N.A. N.A. No N.A.
PMS 5003 N.A. N.A. N.A. N.A. No N.A.
PMS 7003 N.A. N.A. N.A. N.A. No N.A.
−10–65 ◦C 5.0 ± 10% V
Samyoung PSML(LPO) N.A. N.A. No N.A.
<95% (n.c.) 0.43 W
(South Korea)
DSM501A −10–65 ◦ C 5 ± 0.5 V N.A. N.A. No N.A.
5 ± 0.5 V ±10 µg/m3 (0 to 100 µg/m3 )
Sensiron (Switzerland) SPS30 −10–60 ◦ C >8 Years Yes N.A.
0.30 W ±10% (100–1000 µg/m3 )
−10–60 ◦ C 5 ± 0.5 V
GP2Y1010AU0F N.A. N.A. No N.A.
10–90% 0.10 W
Sharp (Japan)
−10–60 ◦ C 5 ± 0.25 V
DN7C3CA006 N.A. N.A. No N.A.
10–90% 0.10 W
Sensors 2021, 21, 3960 7 of 27

Table 2. Cont.

Voltage and Power Integrated

Manufacturer Model T and RH Uncertainty Sensor Life Reaction Time
Consumption Calibration
0–45 ◦ C
PPD42NJ 5 ± 0.2 V N.A. N.A. No N.A.
<95% (n.c.)
0–45 ◦ C
PPD60PV-T2 5 ± 0.2 V N.A. N.A. No N.A.
Shinyei <95% (n.c.)
(China) 0–40 ◦ C
PPD20V 5 ± 0.2 V N.A. N.A. No N.A.
<95% (n.c.)
−10–60 ◦ C
PPD71 5 ± 10% V N.A. N.A. No N.A.
<95% (n.c.)
Winsen −10–50 ◦ C 5 ± 0.1 V
ZH03B N.A. 3 Years No N.A.
(China) 0–85% (n.c.) 0.60 W
Sensors 2021, 21, 3960 8 of 27

The SDS011 from Nova-Fitness Ltd., China, the PMS series from Plantower Technology,
China, and the OPC series from Alphasense Ltd., UK, are some of the most popular choices
of LCS. The SDS011 costs around EUR 30 without additional electronics to capture and
store data. It is one of the most used sensors in citizen science projects such as the Sensor
Community project in Europe [42]. The Plantower sensors are widely used in the USA in
research [43] and in citizen science initiatives such as the CityOS project [44]. The OPC
series, on the other hand, costs between EUR 300 and EUR 450 and is used more in research
work. The extra costs compared to, for example, the SDS01 sensor is compensated by the
ability of the OPC to provide a histogram of particle-sizes in 16 bins (OPC-N2) or 24 bins
(OPC-N3 and OPC-R1) and PM1 in addition to the overall mass distributions of PM10
and PM2.5. The SPS30 (Sensiron AG, Zürich, Switzerland) is gaining popularity but the
literature available on this sensor still is very limited.
The cost-effectiveness of LCS comes with its own disadvantages. The inherent lim-
itations of the PM sensors, when compared to expensive reference devices, introduce
variations in the measurements between the two devices. By construction, an expensive
standard device such as a Grimm Aerosol Spectrometer has an advantage over a LCS due
to the presence of a pump. Even though a ventilator is often present in LCS such as the
Alphasense family and SDS011 (Nova Fitness) sensors, the power of such a ventilator is
quite low as it produces a mere 300 mL/min or less sample flow rate compared to the
1.2 L/min sample flow rate of a Grimm 1.109/1.108. Conventional OPCs also have a
narrow air inlet that leads to the centre of a measurement chamber wherein the air sample
is illuminated with a laser source in a multiplex mode. This means that the laser intensity
is modulated, enabling the ability of the instrument to measure a wide range of particle
sizes [37]. The OPC-N2 has its patented system, wherein the expensive pump and narrow
inlet are replaced with a micro-fan which sucks in the air into an open scattering chamber,
wherein an elliptical mirror and a dual-element photodetector create a “virtual sensing
zone” where the laser light illuminates, scatters and is detected. The smaller sized particles
are calculated using a weighing to account for their underestimation in LCS [38].
LCS are generally not stand-alone instruments, which means that they need additional
electronics for power supply, configuration, and data storage. However, the OPC-N2 is
a stand-alone instrument with software included which runs on a Windows operating
system [45] and uses an internal SD card to store data. However, the instrument by itself is
not weather-proof nor does it come with temperature and RH sensors, a clock module, or a
global navigation satellite system (GNSS) receiver. Therefore, the OPC-N2 has to be set up
with additional microcomputers such as the Raspberry Pi [46] or Arduino [47] or with a
custom-made printed circuit board (PCB). However, the latest version, OPC-N3, comes
with built-in temperature and RH sensors.

2.2. Literature Review on PM-LCS Calibration Studies

Lukeville (2019) [8] has undertaken a comprehensive overview of the basic principles
involved in the LCS measurement technology, ensuring their quality, reliability, and limi-
tations. She also provides information on how the air quality is measured in Europe and
how citizens can calibrate their sensors using existing official stations, and how they might
provide the data to the authorities for further use.
Due to the widespread and increasing popularity of LCS among scientists and citizens,
Aakash C. Rai et al. (2016) [21] and Aakash C. Rai and Prashant Kumar (2017) [22]
provide an overview of available stand-alone sensors, their technology, and the costs
involved. Updated and exhaustive information on not just stand-alone sensors but of
sensor systems, including black-box sensors, was carried out by the JRC [23]. This study
provides quantitative data on the performance of LCS against reference instruments. It
is concluded that the coefficient of determination (R2 ) metric, used by most studies that
evaluate a sensor’s capabilities, can be misleading on the quality of LCS. This is because
the R2 is overly dependent on a range of different reference measurement specifications
on the duration of the test, and the season and location of the test, making the changes in
Sensors 2021, 21, 3960 9 of 27

R2 not completely dependent on the LCS data quality or the calibration methods alone.
Due to these shortcomings, Karagulian et al. (2019) [23] state that the standardization of a
protocol for the evaluation of LCS has a high priority at an international level.
Kuula et al. (2020) [33] investigate the particle-size selectivity and its role in the
analysis of sources of errors in LCS. They report that six sensors, namely the Plantower
PMS5003, Nova SDS011, Sensiron SPS30, Sharp GP2Y1010AU0F, Shinyei PPD42NS, and
Omron B5W-LD0101, are compared against a Grimm 1.108 (2020) with a vibrating orifice
aerosol generator 3450 (VOAG, TSI Inc., Shoreview, MN, USA). The results show that none
of the sensors adhere to the detection ranges claimed by the manufactures. In comparison
with the Grimm 1.108, the sensors could achieve comparable data in one or two size bins
only, which is insufficient for a sensor to be able to provide reliable mass concentration data.
Sousan et al. (2016) [48] compared the Alphasense OPC-N2 to the Grimm PAS 1.108
with the SMPS C5.402 (Grimm Aerosol GmbH, Ainring, Germany) and APS 3321 (TSI Inc.,
Shoreview, MN, USA) as reference instruments. Salt, welding fume, and Arizona street dust
were used as input aerosols in an experimental setup, and the detection efficiency, response,
and precision of both number concentration and mass concentration were assessed. For
all the aerosols and PM metrics, the firmware-calculated mass concentrations had an R2
value of 0.97, whereas the number concentrations were found to be underestimated in the
lower particle-size range (salt and welding fumes) and overestimated for coarse particles
(Arizona street dust) when compared to reference instruments. The two OPCs, OPC-N2
and Grimm PAS 1.108 themselves were found to be consistent with each other.
Official air quality monitoring stations (AQMS) use accurate, but expensive devices.
This makes it difficult to set up multiple stations to allow higher spatial resolution. LCS
can come in handy in such situations by complementing AQMS. However, such setups
only return reliable measurements when the LCS is well calibrated and extensive post-
processing of the measured data is carried out. Di Antonio et al. (2018) [49] and Crilley et al.
(2018) [50] provide such a correction methodology for LCS. Di Antonio et al. (2018) [49] for
instance use the measured particle size distribution of the OPC-N2 sensor instead of the
mass concentration to derive a correction based on relative humidity (RH) for individual
particle sizes due to the hygroscopic properties of the dust particles. This is done by
using Koehler’s theory [51,52], see also Section 3.3, which can significantly improve sensor
performance and retain information on particle composition. The algorithm provided is
also flexible to changes in particle chemical composition and particle chemical speciation.
However, not all sensors in the market have the ability to provide particle size distribu-
tion. Most of the LCS widely used in citizen science projects output only the standard mass
concentration of PM of aerodynamic diameter 10 µm (PM10), 2.5 µm (PM2.5), and 1 µm
(PM1). Sensiron’s SPS30 produce the mass concentration of PM of aerodynamic diameter
4 µm (PM4) in addition to the standard mass concentrations. When an LCS measures only
the bulk PM it is difficult to implement a correction based on particle size. However, when
the information from all bins is available, as is the case of the OPC series, it is possible to
calculate the mass concentration using its own correction factors. Therefore, Crilley et al.
(2018) [50] have outlined a simple correction factor as reported in the following paragraph,
also based on the Koehler’s theory which is briefly described in Section 3.3. A number
of 14 instruments were used in their study. All 14 instruments were co-located and then
deployed. The instruments showed reasonable inter-unit precision and a reasonable agree-
ment to reference optical-particle counters, TEOM-FDS, Grimm PAS 1.108, and TSI 3330,
under low-to-normal RH. Under high ambient RH (>85%) a significant positive artefact was
detected, which reiterates the necessity to correct the measured data for ambient RH. To
correct for the ambient RH two aspects are to be noted here. First, the mass concentrations
as computed by the OPC-N2 using factory-set algorithms are ignored. They are instead
calculated from the particle size distribution (particle number concentration in different
size bins) data of the OPC-N2 and the reference instruments using a uniform particle
density of 1.65 g/mL (factory setting for OPC-N2). Second, applying this correction for
RH < 85% tends to overcorrect the data. Crilley et al. (2018) [50] also note that “all low-cost
Sensors 2021, 21, 3960 10 of 27

PM sensors will likely require calibration factors to obtain the dry particle weight unless
they actively dry the PM-containing air stream before it enters the device”. They also hint
at the use of heated inlets to reduce the RH in the air stream. Samad et al. (2021) [53]
investigated the use of a low-cost dryer for the OPC-N3, concluding that it can successfully
reduce the negative effects of the relative humidity on the PM results. However, this alters
the power requirements of the sensor with the consequence that either larger batteries are
needed or only shorter operation times can be accomplished for off-grid operations.
Crilley et al. (2020) [54] checked the validity of the calibration method described in
Crilley et al. (2018) [50] in four cities and on three different continents. They report that the
elevated particle mass concentration found in LCS is due to the bulk aerosol hygroscopicity
under different RH conditions. Crilley et al. (2020) [54] conclude that a factor based on
Koehler’s theory (k-factor) derived from in situ measurements, as they did, offers better
calibration and improves the performance of the OPC-N2. Nevertheless, in conditions
where in situ measurements are impractical, then Crilley et al. (2020) [54] suggest using a
“literature-based k-factor”.
Laquai and Saur (2017) [55] explain a calibration strategy for PM2.5 measured using
the SDS011 sensor and using the Grimm 1.108 as a reference instrument. Based on an
experimental setup, it is found that the PM10/PM2.5 ratio of the LCS gives an indication
of the particle mass distribution. With this information, a range of ratios between the LCS
sensor and Grimm for PM10 and PM2.5 is obtained in different particle spectra and a
correction algorithm is deduced.
Datta et al. (2020) [56] describe a calibration method using a gain-offset model and
linear regression for PM2.5 measurements. This study is an evaluation of a cluster of 32
LCS at one regulatory site. The evaluation was carried out using multiple linear regressions
with co-located data.
Zusman et al. (2020) [43] evaluated the performance characteristics of two LCS,
Plantower PMS A003 and Shinyei PPD42NS, in comparison to reference methods. They
developed a regional calibration model for seven metropolitan areas in the United States
of America. They observed occasional spikes of PM2.5 concentrations when the sensors
warmed up, which led to excluding the first 8 h of data after each deployment.
A common trend observed in all the literature is that they all agree that the LCS can
be a very important and useful commodity to complement existing AQMS, provided they
follow proper measurement practices, are compensated for effects of varying RH, and
apply data processing techniques. Nevertheless, the data processing in most of the LCS
relies on setting up a mathematical model to fit the data of the LCS to a reference device.
The regression model is usually the model of choice for LCS calibration [22,57]. However,
these studies focus on point-based stationary measurements, integrated over a longer
period of time (>=60 min). Mahajan and Kumar (2020) [58] observed that support vector
regression (SVR) appeared to be a promising approach to calibrate LCS when compared
to linear regression, artificial neural networks, and random forest regression. However,
the scope of their study is also limited to stationary measurements. The validity of such
models with respect to seasonal changes is not mentioned in these studies. Alfano et al.
(2020) [41] provide the most recent and extensive review of LCS and their calibration.
They emphasize the strong dependency of the performance accuracy of LCS on whether
the device is calibrated or not in the operative environment. Therefore, using co-located
calibrations to determine the accuracy of LCS on mobile platforms fails to account for the
micro-scale changes in the spatial characteristics that further affect the accuracy of LCS.

2.3. Interviews on Usage and Calibration of LCS

As a part of a project on developing a communication strategy for citizen science
projects and the general public on the usage of LCS for the German Environmental Agency
(Umweltbundesamt, UBA) a series of 18 interviews were conducted between October 2019
and August 2020 with researchers working with LCS. The list of participants can be found
in Appendix A. The following conclusions were drawn from the interviews:
Sensors 2021, 21, 3960 11 of 27

• Expert interviews show a lack of uniformity in the testing of sensors. New guidelines
are needed to make sensor testing procedures binding and comparable;
• When using sensors, it is important to be clear about what they are to be used for.
If the aim is to increase the environmental awareness of citizens or to test the air
quality (low pollution, high pollution) in a location, the quality of the data is sufficient.
Currently, the raw data of the sensors are not suitable for quantitative measurements
due to their poor reproducibility and stability characteristics;
• Many research groups have used the sensors without calibration. The number of
calibrations required during a measurement campaign is still unclear. Most research
groups carry out the calibrations in comparative measurements with standard mea-
suring instruments at the beginning, when the measurement campaign is short, and
additionally at the end in longer measurement campaigns;
• The data sheets provided by the manufacturers are partly insufficient. Therefore,
calibrations by the user are essential. In addition, each sensor must be calibrated
separately, since the characteristics of the sensors are individual even with sensors of
the same type;
• A big issue is that LCS are operated outside their specifications. Almost all require a
non-condensing environment. LCS are mostly sensors developed for indoor use. In
many cases these sensors are used for outdoor measurements, thus failing to provide
useable data;
• Single laboratory or co-location experiments are insufficient to determine the measured
values and characteristics of the sensors. If the sensors are to be used for mobile
measurements, stationary calibrations are insufficient. Furthermore, the age-related
drift of the sensors must be taken into account. The service life of the sensors is usually
less than specified by the manufacturer;
• A common platform for users of low-cost sensors for communication and exchange of
information and ideas is indispensable. The circle of users of such low-cost sensors
is constantly growing in private and commercial applications as well as in science
without proper assurance of quality and information regarding visualization and
interpretation of such measurements;
• Nevertheless, citizen scientists and the general public should be encouraged and
guided to work with LCS and the data acquired through them.

3. The URBMOBI 3.0 System for Mobile PM Measurements

3.1. Device Configuration of the URBMOBI 3.0 System
In order to have a sensor box on a mobile platform, a sensor ensemble was designed
and set up with a custom-made PCB called the URBMOBI 3.0, short for URban MOBIle
instrument in its development phase 3, building on earlier versions 1.0 and 2.0 [59]. URB-
MOBI 3.0 is equipped with a PM [38], nitrogen dioxide (NO2 ) [60], nitrogen oxide (NO) [61],
ozone (O3 ) [62], global radiation [63], and two temperature and RH [64] sensors (Figure 1).
 In order to have a sensor box on a mobile platform, a sensor ensemble was design
set up with a custom-made PCB called the URBMOBI 3.0, short for URban MOBIle instr
in its development phase 3, building on earlier versions 1.0 and 2.0 [59]. URBMOB
Sensors 2021, 21, 3960 equipped with a PM [38], nitrogen dioxide (NO2) [60], nitrogen oxide (NO) [61], ozo
12 of 27

[62], global radiation [63], and two temperature and RH [64] sensors (Figure 1).

Sensors 2021, 21, 3960 11 of 2

Figure 1. Printed circuit board (PCB), bottom-side (left) and top-side (right), of the URBMOBI 3.0
Figure 1. Printed
sensor-box. circuit
Photo: Evert board (PCB), bottom-side (left) and top-side (right), of the
Nieuwkoop. URBM
sensor-box. Photo: Evert Nieuwkoop.
TheURBMOBI
The URBMOBI 3.03.0
PCBPCB is equipped
is equipped with with
a GNSSa GNSS receiver
receiver and an and an SD
SD card card to stor
to store
dataoffline
data offline with
with a time
a time resolution
resolution ofFigure
of 2 s. 2 s. Figure 2 shows
2 shows theof
the setup setup of the URBMOBI
the URBMOBI sensor senso
ensembleininitsits
ensemble metal
metal box.
box. TheThe instrument
instrument is notiscompletely
not completely protected
protected against
against rain and rain an
spray water and in such cases, the measurements should be stopped.
spray water and in such cases, the measurements should be stopped.

Figure2.2.URBMOBI
Figure URBMOBI 3.03.0 device
device for mobile
for mobile air quality
air quality and meteorology
and meteorology with for
with sensors sensors for fine dus
fine dust
(OPC-N2), gases (NO
(OPC-N2), gases (NO2 , NO, and O3 ), air temperature, and humidity (SHT35), and global radiation radiatio
2 , NO, and O 3 ), air temperature, and humidity (SHT35), and global
(EKOML-01);
(EKO ML-01); Photo:
Photo: Janani
Janani Venkatraman
Venkatraman Jagatha.
Jagatha.

3.2.
3.2.Calibration
Calibrationof of
thethe
URBMOBI 3.0 System
URBMOBI in a Stationary
3.0 System Setup Setup
in a Stationary
As a first step, the LCS is calibrated against a reference device, Grimm 1.108, in
As a first step, the LCS is calibrated against a reference device, Grimm 1.108, in
a stationary setup. The setup and procedure were followed according to Laquai et al.
stationary
(2020) setup.3The
[65]. Figure setup
shows andsetup
the test procedure
using awere followed
particle according
generator. to Laquai
The particle et al. (2020
generator
is[65]. Figure
a simple 3 shows
Zarges™ box the test
acting as setup using
a particle a particle
chamber. generator.
The top portion ofThe particle
the box generator is
is provided
simple
with ZargesTM box
a loudspeaker andacting
a smoke asgenerator.
a particleFlour
chamber.
is usedThe
as atop portion
source of the box
of particles is provide
greater
with2.5
than a loudspeaker and a smoke
µm and is dispensed generator.
with the Flour
help of the is used asThe
loudspeaker. a source
smokeofgenerator
particles greate
isthan
used2.5
to µm
dispense particles
and is with less
dispensed withthan
the 2.5 µmofaerodynamic
help diameter.
the loudspeaker. TheBoth
smokeparticle
generator i
dispensers are connected and operated via external circuitry. The bottom part
used to dispense particles with less than 2.5 µm aerodynamic diameter. Both particle of the box is dis
pensers are connected and operated via external circuitry. The bottom part of the box i
fitted with a Grimm 1108 device as the reference and space to place the LCS to be tested
The entire experiment is carried out under constant temperature (≈20 °C) and RH (≈50%
conditions. The results of the experiment are presented in Figure 4.
Sensors 2021, 21, 3960 13 of 27

fitted with a Grimm 1108 device as the reference and space to place the LCS to be tested.
Sensors 2021, 21, 3960 The entire experiment is carried out under constant temperature (≈20 ◦ C) and RH (≈1250%)
of 25
conditions. The results of the experiment are presented in Figure 4.

Figure3.3.Self-built
Figure Self-builtparticle
particlegenerator
generatorasasmeasuring
measuringsetup
setupfor
forcomparison
comparisonmeasurements
measurementsbetween
betweenaa
reference instrument (Grimm 1.108, bottom right) and URBMOBI 3.0 (bottom/middle left);
reference instrument (Grimm 1.108, bottom right) and URBMOBI 3.0 (bottom/middle left); Photo:
Photo: Bernd Laquai.
Bernd Laquai.

Fromthe
From thetests
testsininthe
theparticle
particlegenerator,
generator,ititisisobserved
observedthat
thatfor
forparticles
particlesabove
above2.52.5µm
µm
aerodynamic diameter, the URBMOBI 3.0—OPC-N2 and the reference
aerodynamic diameter, the URBMOBI 3.0—OPC-N2 and the reference device are com- device are compa-
rable. However,
parable. However, forfor
smaller
smallerparticles (<2.5
particles µm µm
(<2.5 aerodynamic
aerodynamic diameter) the URBMOBI
diameter) the URBMOBI 3.0—
OPC-N2 tendstends
3.0—OPC-N2 to overestimate partialpartial
to overestimate mass concentration when compared
mass concentration to the reference
when compared to the
device. This
reference means
device. that
This a size-dependent
means compensation
that a size-dependent function is function
compensation necessaryisto calibrate
necessary
thecalibrate
to OPC-N2the against
OPC-N2 the against
referencethedevice, assuming
reference device,the reference
assuming thedevice to bedevice
reference more toaccu-
be
rate in an absolute sense. The compensation function is generated by assuming
more accurate in an absolute sense. The compensation function is generated by assuming a a linear
relationship
linear between
relationship the URBMOBI
between 3.0—OPC-N2
the URBMOBI 3.0—OPC-N2 (LCS)(LCS)
and the
andreference device:
the reference device:
𝑃𝑀2.5 = =
PM2.5 a ∗𝑎 PM2.5
∗ 𝑃𝑀2.5 ++C,𝐶, (1)
(1)
Re f LCS

where 𝑃𝑀2.5 is the PM2.5 concentration of the reference device (Grimm 1.108),
where PM2.5Re f is the PM2.5 concentration of the reference device (Grimm 1.108), PM2.5 LCS
𝑃𝑀2.5 is the PM2.5 concentration of the URBMOBI 3.0—OPC-N2 (LCS), “a” denotes
is the PM2.5 concentration of the URBMOBI 3.0—OPC-N2 (LCS), “a” denotes slope, and “C”
slope, and “C” the constant in the calibration equation. The assumption of a linear rela-
the constant in the calibration equation. The assumption of a linear relationship is because
tionship is because constant
constant temperature and RHtemperature
conditions areandmaintained.
RH conditions Theare maintained.
result The result of
of the compensation
the compensation
function is providedfunction is 4c.
in Figure provided in Figure
It is apparent that4c.after
It isapplying
apparentthis
thatcorrection
after applying this
the OPC-
correction the OPC-N2 in the URBMOBI 3.0 ensemble is well suited
N2 in the URBMOBI 3.0 ensemble is well suited for both PM10 and PM2.5 measurements for both PM10 and
PM2.5such
under measurements under conditions.
dry and constant such dry and constant conditions.
However, the goal for the URBMOBI 3.0 instrument is the implementation of an LCS
to measure the ambient particulate matter concentration. Therefore, a second test with the
LCS setup was carried out outdoors with a Grimm 1.109 as a reference device. The test
was carried out in the Adlershof suburb of the Berlin metropolitan area between 2020–11–
12 22:00:00 UTC and 2020–11–15 12:00:00 UTC at an altitude of 35 m, away from the road,
on the building of the Geography Department of Humboldt-Universität zu Berlin. The air
inlet of the reference device was at a height of 1.5 m and the air inlet of the LCS was at a
height of 1.25 m above the floor of the roof. During the measurement period, the ambient
temperature (T) ranged between 8.5 °C to 18 °C and the relative humidity (RH) between
55% to 80%. It can be seen in Figure 5a that the sensor clearly overestimates PM10 concen-
trations as compared to the data measured using the Grimm 1.109 device. The gain of the
LCS increases as the RH increases (Figure 5). Furthermore, Figure 5(a-ii) and Figure 5(c-
i), both have the same coefficient of determination (𝑅 ) value, even though Figure 5(a-ii)
 bles. Comparing Figure 5(a-ii) and Figure 5(c-i), it is clear that a simple integration over
longer periods is insufficient. It is important to consider the effects of the temperature and
RH to explain and calibrate LCS.

Sensors 2021, 21, 3960

𝑃𝑀10 = 𝑎 ∗ 𝑃𝑀10 + 𝑏 ∗ 𝑇 + 𝑐 ∗ 𝑅𝐻 + 𝐶, 14 of 27 (2)

where a, b, and c are the slopes of PM10LCS, T, and RH respectively.

Figure 4. Results of the calibration tests recorded by the URBMOBI 3.0 (metal housing) and the reference instrument
Figure 4. Results of the calibration tests recorded by the URBMOBI 3.0 (metal housing) and the reference instrument
(Grimm 1.108) inside the particle generator with a time integral of 6 s; (a) PM10 scatter plot between Grimm 1.108 raw data
(Grimm 1.108) inside the particle generator with a time integral of 6 s; (a) PM10 scatter plot between Grimm 1.108 raw
and URBMOBI 3.0; (b) PM2.5 scatter plot between Grimm 1.108 and URBMOBI 3.0 raw data; (c) PM2.5 scatter plot between
data and URBMOBI 3.0; (b) PM2.5 scatter plot between Grimm 1.108 and URBMOBI 3.0 raw data; (c) PM2.5 scatter plot
Grimm 1.108 raw data and URBMOBI 3.0 data corrected using the compensation function.
between Grimm 1.108 raw data and URBMOBI 3.0 data corrected using the compensation function.
However, the goal for the URBMOBI 3.0 instrument is the implementation of an LCS
to It is observed
measure that PM10
the ambient also needs
particulate matter to be correctedTherefore,
concentration. under ambient
a secondconditions,
test with theas op-
posed
LCS to justwas
setup PM2.5 as out
carried found in thewith
outdoors laboratory
a Grimm experiment using device.
1.109 as a reference constant
Theair
testconditions
was
carried out in the Adlershof suburb of the Berlin metropolitan area between
and a particle generator. This means that the URBMOBI 3.0 instrument is in principle well 2020–11–12
22:00:00
suited UTC andmeasurements,
for ambient 2020–11–15 12:00:00 UTC at
provided an altitude
a careful andof 35 m, away from
well-designed the road,
cleaning and pro-
on the building of the Geography Department of Humboldt-Universität zu Berlin. The
cessing of the data is carried out. Single values are unreliable, especially when a high tem-
air inlet of the reference device was at a height of 1.5 m and the air inlet of the LCS was
poral resolution is used. Therefore, it is highly recommended to make temporal integrals
at a height of 1.25 m above the floor of the roof. During the measurement period, the
over at least
ambient 60 s which
temperature (T)isranged
also evident
betweenfrom8.5 ◦ Cthe comparison
to 18 ◦ C and the of Figure
relative 5(c-i) and
humidity Figure
(RH)
5(c-ii).
between 55% to 80%. It can be seen in Figure 5a that the sensor clearly overestimates
PM10 concentrations as compared to the data measured using the Grimm 1.109 device.
The gain of the LCS increases as the RH increases (Figure 5). Furthermore, Figure 5a–ii
and Figure 5c–i, both have the same coefficient of determination (R2 ) value, even though
Figure 5a–ii integrates the data over time (60 s) while Figure 5c–i keeps the 6-s interval
but performs a multilinear regression using the temperature and relative humidity as
explaining variables. Comparing Figure 5a–ii and Figure 5c–i, it is clear that a simple
integration over longer periods is insufficient. It is important to consider the effects of the
temperature and RH to explain and calibrate LCS.

PM10Re f = a ∗ PM10 LCS + b ∗ T + c ∗ RH + C, (2)

where a, b, and c are the slopes of PM10LCS , T, and RH respectively.

It is observed that PM10 also needs to be corrected under ambient conditions, as
opposed to just PM2.5 as found in the laboratory experiment using constant air conditions
and a particle generator. This means that the URBMOBI 3.0 instrument is in principle
well suited for ambient measurements, provided a careful and well-designed cleaning
and processing of the data is carried out. Single values are unreliable, especially when a
high temporal resolution is used. Therefore, it is highly recommended to make temporal
integrals over at least 60 s which is also evident from the comparison of Figures 5c–i
and 5c–ii.
Sensors 2021,21,
Sensors2021, 21,3960
3960 15
14 of
of 27
25

Figure 5. URBMOBI 3.0 (LCS) vs. Grimm 1.109 (reference device) PM10 comparison at Berlin-
Figure 5. URBMOBI
Adlershof; (a) scatter3.0
plot(LCS) vs. Grimm
between 1.1091.109
the Grimm (reference
raw data device)
andPM10
URBMOBIcomparison
3.0 rawatdata
Berlin-Ad-
with a
lershof; (a) scatter plot between the Grimm 1.109 raw data and URBMOBI 3.0 raw data
time integral of 6 (i) and 60 (ii) seconds; (b) scatter plot between the Grimm 1. 109 raw data with a time
and
integral of 6 (i) and 60 (ii) seconds; (b) scatter plot between the Grimm 1. 109 raw data and UR-
URBMOBI 3.0 data corrected with linear regression with a time integral of 6 and 60 s (c) scatter plot
BMOBI 3.0 data corrected with linear regression with a time integral of 6 and 60 s (c) scatter plot
between the Grimm 1.109 raw data and URBMOBI 3.0 data corrected with multi-linear regression,
between the Grimm 1.109 raw data and URBMOBI 3.0 data corrected with multi-linear regression,
using
using temperature
temperature and and RH
RH as
as explaining
explaining variables,
variables, with
with aa time
timeintegral
integralof
of66and
and6060s.s.
3.3. Calibration of the URBMOBI 3.0 in a Mobile Setup
3.3. Calibration of the URBMOBI 3.0 in a Mobile Setup
The URBMOBI 3.0 is envisioned to be deployed for mobile applications. In addition to
The URBMOBI
hygroscopic 3.0 LCS
effects, the is envisioned to be deployed
has to be proofed for mobile
for sensitivities of applications. In addition
the photodetector, angle
to the
of hygroscopic effects,defects,
laser, electronic the LCS has to be proofed
or vibrations causing afor sensitivities
systematic error.ofTo
thecompensate
photodetector,
and
angle offor
correct thethe
laser, electronicarising
uncertainties defects,from
or vibrations causing
these factors and toa systematic
evaluate the error. To compen-
performance of
sateURBMOBI
the and correct3.0 forinthe uncertainties
a mobile setup, arising
a Grimm from these
1.109 factors and
instrument was tocarried
evaluate the perfor-
additionally
mance
as of the URBMOBI
a reference device, on 3.0 in a mobile setup,
a predetermined a Grimm
bicycle route. 1.109 instrument was
The measurements werecarried ad-
carried
ditionally
out along anas 18a reference
km route, device,
covering ondifferent
a predetermined
local-climate bicycle
zonesroute.
(LCZ)The[66]measurements
and land-use
were carried
classes (LUC) out along anin18the
as classified kmCorine
route,land
covering different(CLC)
classification local-climate
[67] in thezones (LCZ) [66]
North-Western
part of Berlin, Germany. Each measurement round took approximately 1.5 h to [67]
and land-use classes (LUC) as classified in the Corine land classification (CLC) in the
complete
North-Western part of Berlin, Germany. Each measurement round took
with the bicycle ridden at a mean speed of 15 km/h. Figure 6 shows the route across the approximately 1.5
h to complete with
various LCZ and LUC. the bicycle ridden at a mean speed of 15 km/h. Figure 6 shows the
route across the various LCZ and LUC.
Sensors 2021, 21, 3960 15 of 25
Sensors 2021, 21, 3960 16 of 27

Figure 6. Measurement route at Berlin-Hermsdorf across different local climate zones and land
Figure 6. Measurement route at Berlin-Hermsdorf across different local climate zones and land use
use classes.
classes.
The following steps were performed for calibration (Figure 7):
The following steps were performed for calibration (Figure 7):
1. The data sets were checked for outliers and inconsistencies due to manual or electrical
1. Theerrors.
data sets
Thewere checked
first and for 1%
the last outliers and of
quantile inconsistencies
the URBMOBIdue 3.0 to manual
data or electri- as
are considered
cal outliers
errors. The first and
and removed. the last 1% quantile of the URBMOBI 3.0 data are considered
2. as outliers and
Low performance removed.of LCS due to RH is an issue repeatedly discussed in different
2. Low studies on LCS.ofTo
performance LCS due to RHfor
compensate is an
theissue repeatedly
effect of aerosoldiscussed in different
hygroscopicity, stud-
the method
ies described
on LCS. To compensate for the effect of aerosol hygroscopicity, the
by Crilley et al. (2018) [50] wherein a correction factor “C”, derived based method de-
scribed by Crilley et al. (2018) [50] wherein a correction factor “C”, derived
on the Köhler’s theory [51], is used. Crilley et al. (2018) [50] state that for a situation based on
thewith
Köhler’s theory [51], is used. Crilley et al. (2018) [50] state that for a
RH < 60% a calibration against suitable reference instruments is sufficient. In the situation with
RHexperiments
< 60% a calibration againstRH
we conducted, suitable
ranged reference
from 50% instruments is sufficient.
to 85%. Therefore, it wasIndecided
the ex- to
periments
use the we conducted,
correction factor RH ranged
based from 50%theory
on Köhler’s to 85%. forTherefore,
the entireitdataset.
was decided to
The value
useofthe correction factor based on Köhler’s theory for the entire dataset.
k is assumed to be 0.4 since the measurements were carried out in an urban area The value of
ĸ issimilar
assumed to be
to that 0.4 study
of the since the measurements
conducted in Crilley were
et al.carried
(2020) out
[54].inThe
an URBMOBI
urban area3.0
similar to that of the study conducted in Crilley et al. (2020) [54].
data is corrected for relative humidity using the following Equations (3) and (4). The URBMOBI 3.0
data is corrected for relative humidity using the following Equations (3) and (4).
k
ĸ ρp
C1
𝐶= =+1 + , 1
, (3)(3)
−1 + aw

𝑎 is the
where, where, aw measured
is the measured relative humidity
relative over 100,
humidity overĸ 100,
is equal to 0.4,toand
k is equal 0.4,density of
and density
particle of ) is set (ρ
(𝜌particle top )1.65
is setg/mL. The
to 1.65 C-factor
g/mL. The is then applied
C-factor to the measurement
is then applied data
to the measurement
using: data using:
PMraw
= =
𝑃𝑀PMcorr ,
C
, (4)(4)

3. 3. TheThe difference
difference between
between thethe medians
medians of the
of the URBMOBI
URBMOBI 3.03.0 (Uc)
(Uc) data
data setset corrected
corrected forfor
thethe influence
influence of humidity
of humidity andand
thethe Grimm
Grimm 1.109
1.109 (G)(G) data
data setset is subtracted
is subtracted from
from thethe
URBMOBI
URBMOBI 3.0 3.0 to bring
to bring thethe measurements
measurements into
into thethe same
same range
range as as
thethe Grimm
Grimm 1.109
1.109
data and then labeled with the subscript
data and then labeled with the subscript “s”. “s”.
4. 4. TwoTwo models,
models, linear
linear regression
regression (lm)
(lm) andand quantile
quantile mapping
mapping (qm)
(qm) areare tested
tested forfor calibra-
calibra-
tion. Each model uses two approaches. The first approach uses 100%
tion. Each model uses two approaches. The first approach uses 100% of the Grimm of the Grimm
1.109 concurrent dataset to calibrate the URBMOBI 3.0 data (G~Uc). The second
Sensors 2021, 21, 3960 16 of 25

Sensors 2021, 21, 3960 17 of 27

1.109 concurrent dataset to calibrate the URBMOBI 3.0 data (G~Uc). The second ap-
proach limits the derivation of calibration parameters to 20% of the common data,
10% at the beginning and 10% at the end, to check whether the statistical quantities
approach
found inlimits the derivation
this way can be usedof calibration parameters
to reliably to 80%
adjust the 20% of the common
original data, the
data during
10% at the
mobile beginning and
measurement 10% at parallel
without the end, reference.
to check whether the statistical quantities
5. found
As anin this way step,
additional can beoutliers
used tothat
reliably
mightadjust
have the 80%
been original
missed data“1”
in step during the
are identified
mobile measurement without parallel reference.
after step “4” as outliers in a boxplot. These values are removed and the Grimm 1.109
5. As
andanURBMOBI
additional step, outliers that might
3.0 (RH-corrected) have beenagain
is correlated missed(G~Wo).
in step “1” are identified
after step “4” as outliers in a boxplot. These values are removed and the Grimm 1.109
6. The accuracy of the corrected URBMOBI 3.0 data (Uc) is checked. Accuracy (A) in
and URBMOBI 3.0 (RH-corrected) is correlated again (G~Wo).
this case is the percentage of data points that are within ±10% of the Grimm 1.109
6. The accuracy of the corrected URBMOBI 3.0 data (Uc) is checked. Accuracy (A) in
datacase
this point at the
is the same measurement
percentage of data pointstimethatafter calibration.
are within ±10% of the Grimm 1.109
data point at the same measurement time after calibration.
Figure 7 shows the analysis and comparison of quantile mapping (Figure 7a–f) and
Figure 7 shows the analysis and comparison of quantile mapping (Figure 7a–f) and
linear model (Figure 7g–l) on a bicycle-based mobile measurement carried out in Berlin-
linear model (Figure 7g–l) on a bicycle-based mobile measurement carried out in Berlin-
Hermsdorf.AAdescription
Hermsdorf. description of each
of each plotplot in the
in the factsheet
factsheet is given
is given below:
below:

Figure7.7.Factsheet
Figure Factsheetforfor PM10
PM10 calibration
calibration of URBMOBI
of URBMOBI 3.0 with
3.0 with GRIMM GRIMM
1.109 1.109
as the as the reference;
reference; an an
exampleofofaameasurement
example measurementround roundatatBerlin-Hermsdorf.
Berlin-Hermsdorf.TheTheUBRMOBI
UBRMOBI3.03.0data
dataobtained
obtained after cali-
after
bration using
calibration usingthe
the quantile mappingmethod
quantile mapping method has
has anan accuracy
accuracy of 36.7%
of 36.7% inexample.
in this this example.
DetailsDetails
for for
subfigures are provided in the
subfigures are provided in the text.text.
Sensors 2021, 21, 3960 18 of 27

Plots (a) to (f) in Figure 7 provide corrections based on quantile mapping:

• Time series in 30 s interval (preprocessed): Observation (obs), is the URBMOBI 3.0
(U) data (shown in dark orange). Reference (ref) is the Grimm data (shown in black).
“mO” denotes the offset between the median of Grimm and URBMOBI 3.0. “sqm” is
the corrected URBMOBI 3.0 time series after applying the median offset and before
quantile mapping (shown in red). “sqm20” is the same as sqm but using only 20% of
the Grimm device data (first and last 10%; this procedure reduces the time resolution;
shown in blue) (Figure 7a).
• Distribution of Grimm, URBMOBI 3.0, and corrected URBMOBI 3.0 data shown in
Figure 7a as boxplots (Figure 7b).
• Distribution of Grimm, URBMOBI 3.0, and corrected URBMOBI 3.0 data as seen in
Figure 7a (Figure 7c).
• Comparison of methods used for URBMOBI 3.0 correction shown in Figure 7a, includ-
ing a regression line based on simple linear regression. “obs” refers to the correlation
between original URBMOBI 3.0 and Grimm data. “sqm” is the correlation between
corrected URBMOBI 3.0 data using median offset before quantile mapping and the
Grimm data (shown in dark orange). “sqm20” is same as sqm but with 20% of the
Grimm data (first and last 10%; fewer data points) as reference. r2 is the correlation
coefficient of the correlation between Grimm and the corrected URBMOBI 3.0 data. r2p
denotes predicted r2 based on the same data (Figure 7d).
• Correlation between sqm and Grimm (same as sqm in Figure 7d). “A” provides the
accuracy of corrected URBMOBI 3.0 data. It is defined as the percentage of data points
that are within the range of ±10% of the Grimm data point at the same measurement
timestamp (Figure 7e).
• Correlation between sqm20 and Grimm (same as sqm in d) (Figure 7f).
The plots (g) to (l) in Figure 7 detail corrections based on multi-linear regression:
• Time series in preprocessed 30s interval: Observations (obs, URBMOBI 3.0) are shown
in dark orange. Reference (ref, Grimm) is shown in black. “lm” refers to the corrected
URBMOBI 3.0 data based on multi-linear regression (lm(G~U + RH + T)) wherein
the intercepts and coefficients are gathered using a 5-min mean of URBMOBI 3.0
and Grimm data. Intercepts and coefficients are applied to the original URBMOBI
3.0 in the 30 s interval are shown in blue. For URBMOBI 3.0 data that was already
RH-corrected with the C-factor RH was not considered for the multi-linear regression.
“wo” is the same as lm, but with the outliers removed before calculating the intercepts
and coefficients over a 5-min mean of both URMOBI 3.0 and Grimm data. Outliers are
based on 30 s interval data: >1.5·Inter quartile range (IQR) as shown in light green
(Figure 7g).
• Distribution of Grimm, URBMOBI 3.0, and corrected URBMOBI 3.0 data as boxplots
as shown in Figure 7g (Figure 7h).
• Distribution of Grimm, URBMOBI 3.0, and corrected URBMOBI 3.0 data shown in
Figure 7g (Figure 7i).
• Comparison of methods used for URBMOBI 3.0 correction shown in Figure 7g in-
cluding a regression line based on simple linear regression: obs–correlation between
URBMOBI 3.0 and Grimm data (shown in dark orange); lm–correlation between
Grimm and corrected URBMOBI 3.0 based on 5-min means and Grimm data (shown
in blue); wo–correlation between Grimm and corrected URBMOBI 3.0 data without
outliers (shown in green). For a description of r2 and RMSE see Figure 7d with x as
either obs, lm, or wo. r2 and r2p are not the correlation coefficients of the multi-linear
model which was used for corrections). The accuracy (A) for correction without
outliers was calculated using the same method as described for Figure 7e (Figure 7j).
• Same as Figure 7j, but zoomed in for lm. Accuracy of correction with outliers was
calculated using the same method as described for Figure 7e (Figure 7k).
Sensors 2021, 21, 3960 19 of 27

• Correlation between wo and Grimm-without-outliers (woGrimm). Accuracy of cor-

rection without outliers compared to Grimm without outliers was calculated using
the same method as described for Figure 7e (Figure 7l).
Although the linear regression model (lm) seems to look better (Figure 7a,g), it is
also apparent in comparing the two figures that the spatial and temporal variability is lost
in linear regression. Each measurement round is individually calibrated and a factsheet
similar to Figure 7 is generated. Similar patterns are observed for most of the measurement
rounds. Therefore, the quantile assignment is considered as the superior method. Figure 8
summarises all the rounds and generates a boxplot for each of the correction models
considered for PM10, PM2.5 and PM1. It also shows the summary of the comparison
between the different models and their respective coefficients of determination:

n ei2
r 2 = 1 − ∑ i =1 2
, (5)
∑in=1 (Ui − U )

where the predicted coefficient of determination (r2p ), and the normalized mean square of
deviation (nRMSE) were:

RMSEi
nRMSEi = − RMSEmin , (6)
RMSEmax

for all measurement rounds at the Hermsdorf site. The r2 measures how well the model
explains the given data and is dependent on the number of independent, explaining
variables. When the number of independent variables and polynomial terms increases, it
customizes itself to fit the peculiarities and random noise in the sample instead of reflecting
the entire population. It also does not predict what would happen to the chosen model
when it is used to calculate a different data set. The r2p on the other hand, provides a good
fit for the given data. Additionally, it can determine how well a regression model can
make predictions [68,69]. This combined with nRMSE is used to assess the best model for
calibrating LCS data on a mobile platform. In this respect, higher r2p and/or a lower nRMSE
indicate a better model. Figure 8 shows that the correction procedures work in similar
ways for PM10, PM2.5 and PM1. Quantile mapping is significantly better than all other
correction models for all three pollutants. The “sqm20” and lm methods produce better
results for PM1 than for PM10 and PM2.5. It can also be concluded that it is important to
not only assess mobile data with respect to its statistics (Figure 8) but also to check the time
series (Figures 7 and 9) in order to account for possible impacts of spatial characteristics.
It is important to note that each round in the Hermsdorf measurement campaign has
been calibrated and corrected individually. This in itself speaks for the inherent difficulties
in calibrating any data acquired with a LCS on a mobile platform. Although the qm
method seems to be a good option for calibrating LCS on a mobile platform, it does not
work similarly for all the measurement rounds. This can be seen in Figure 9, wherein the
URBMOBI 3.0 data is corrected to within the range of the Grimm data but the accuracy is
0% (Figure 9e).
 Sensors 2021, 21, 3960 20 of 27
Sensors 2021, 21, 3960 19 of 25

Figure 8.
Figure Summary of
8. Summary of different
different correction
correctionmethods,
methods,namely
namelycoefficients
coefficientsofofdetermination
determination(r2(r²),
), the
the
predicted coefficient of determination (r 2 ), and the normalized mean square of deviation (nRMSE)
predicted determination (𝑟p ), and the normalized mean square of deviation (nRMSE)
for RH
for RH corrected PM10, PM2.5PM2.5 and
and PM1,
PM1,for forthe
themeasurement
measurementcampaign
campaignininBerlin-Hermsdorf.
Berlin-Hermsdorf. TheThe
Accuracy
Accuracy (A) is also provided
provided forfor the
the correlations
correlationsbetween
betweenthe
theGrimm
Grimm1.109
1.109and
andURMBOBI
URMBOBI3.0 3.0raw
raw
data
data (G~Uc),
(G~Uc), Grimm
Grimm 1.109 and and URBMOBI
URBMOBI 3.0 3.0corrected
correcteddata
datausing
usingquantile
quantilemethod
methodwith
with100%
100% data
data
(G~sqm), Grimm1.109
(G~sqm), Grimm 1.109vs. vs. URBMOBI
URBMOBI 3.0 corrected
3.0 corrected using method
using quantile quantilewith
method with
20% data 20% data
(G~sqm20),
(G~sqm20), Grimm vs.3.0
Grimm vs. URBMOBI URBMOBI
corrected3.0 corrected
with with linear
linear regression regression
(G~lm), (G~lm),1.109
and Grimm and Grimm 1.109 and
and URBMOBI
URBMOBI 3.0 with all outliers removed (G~wo).
3.0 with all outliers removed (G~wo).
Sensors 2021,21,
Sensors2021, 21,3960
3960 21 of
2027
of 25

Figure9.9.Factsheet
Figure Factsheetfor
forPM10
PM10calibration
calibrationofofURBMOBI
URBMOBI 3.03.0 with
with Grimm
Grimm 1.109
1.109 as as
thethe reference;
reference; an an
example of a measurement round at Berlin-Hermsdorf with an accuracy of zero percent
example of a measurement round at Berlin-Hermsdorf with an accuracy of zero percent calculated calculated
after the calibration using the Quantile mapping method. Details for the subfigures are provided in
after the calibration using the Quantile mapping method. Details for the subfigures are provided in
the text under Section 3.3 and are the same as in Figure 7.
the text under Section 3.3 and are the same as in Figure 7.

4.4.Summary
Summaryand andConclusions
conclusion
Despitetheir
Despite theirshortcomings,
shortcomings,LCSLCShavehavebecome
becomeanan alternative
alternative forfor those
those whowho cannot
cannot
affordexpensive
afford expensivedevices
devices or/andfor
or/and forthose
thosewho
who want
want toto expand
expand AQMS
AQMS networks.
networks. The The field
field
ofofLCS
LCSisisone
onewith
withmany
manyopen
openquestions,
questions, especially
especially concerning
concerning thethe behaviour
behaviour of sensors
of sensors
under
underchanges
changesinintemperature
temperatureandandrelative
relativehumidity,
humidity,and
and aging. ToTo
aging. increase
increasethethe
environ-
environ-
mental awareness of people or to provide qualitative analysis of the AQ at a high
mental awareness of people or to provide qualitative analysis of the AQ at a high spatial spatial
resolution,
resolution,the
thedata
dataprovided
providedbybymost
mostofof thethe
LCS areare
LCS sufficient. However,
sufficient. However, quantitative
quantitative
measurements
measurementsusingusingOEMs
OEMsare
arenot
notsuitable
suitablewithout
withoutat at
least additional
least additional temperature
temperature andand
Sensors 2021, 21, 3960 22 of 27

humidity measurements and bias corrections as well as cross-calibration with existing

reference stations using the conventional established equipment of AQMS networks.
It appears that laboratory or co-location experiments alone might be insufficient to
determine the accuracy of measured values and the reliability and characteristics of the
sensors. Therefore, calibration of the sensors on a mobile platform has to be performed
using at least partly parallel measurements with quality assured standard devices. At this
stage, it is not recommended to only use LCS in mobile setups, especially for quantitative
measurements. However, if and when used, stationary calibrations are insufficient due to
rapid changes in environmental conditions such as temperature and humidity. However,
using quantile mapping wherein 20% of the data is used to correct the rest of the data set
can come in handy to avoid relying on a reference-grade device for the entire measurement
time. Setting up two or more reference stations or using existing AQMS stations along the
measurement route and measuring next to them for a couple of minutes, and using this as
a “20%” data to calibrate the LCS, could provide the same effect.
In addition, the effect of wind speed and the inaccuracies resulting from vibrations
need to be further investigated. Moreover, as the expert interviews have revealed, age-
related drift of the sensors and their service life must be accounted for.
Datasheets provided by the manufacturers are quite often insufficient to assess the
characteristics of the sensors. A variety of studies from independent researchers, scientists,
and public authorities have developed testing and calibration methods to compensate for
that. However, the lack of uniformity in the testing of sensors partly invalidates the results
or makes it hard to work out comparisons between studies and across different sensors.
The number of calibrations to be carried out during a measurement campaign is also not
easy to assess. This calls for establishing a seal of approval based on standardised testing
procedures, and the development and establishment of general guidelines for the use of
such sensors in AQ networks both for general monitoring and scientific measurement
campaigns. Further, a common platform for users of LCS for communication, information,
and idea exchange is indispensable as the circle of users is constantly growing in private,
commercial, and scientific applications.

Author Contributions: Conceptualization, J.V.J. and C.S.; State of art, J.V.J. and M.C.-M.; URBMOBI
3.0 system for mobile PM measurements, J.V.J., A.K., B.L., E.N., P.v.d.M., U.V., C.S.; formal analysis,
J.V.J., A.K., B.L. and C.S.; resources, J.V.J., B.L., U.V., C.S.; data curation, J.V.J. and A.K.; writing—
original draft preparation, J.V.J.; writing—review and editing, J.V.J., A.K., M.C.-M., C.S.; visualization,
J.V.J., A.K., and C.S.; supervision, C.S.; project administration, C.S.; funding acquisition, C.S. All
authors have read and agreed to the published version of the manuscript.
Funding: “This research was funded by the Federal Ministry of Research and Education, Germany
(BMBF) as a part of the “Urban Climate Under Change [UC]2 ” project within its sub-projects with
grant numbers 01LP1602B and 01LP1912B and the German Environment Agency (Umweltbundesamt,
UBA), Grant No. 3718-51-240-0.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author.
Acknowledgments: We gratefully acknowledge the assistance of Klaus Hartmann, Sabine Fritz, Ines
Langer and Sebastian Schubert for the contribution to both the logistics and data processing. We
would also like to thank the two anonymous reviewers and the editors, who substantially helped to
improve the study. We acknowledge support by the German Research Foundation (DFG) and the
Open Access Publication Fund of Humboldt-Universität zu Berlin.
Conflicts of Interest: The authors declare no conflict of interest.
Sensors 2021, 21, 3960 23 of 27

Appendix A

Table A1. Experts contacted as a part of a project on developing a communication strategy for citizen science projects and the general public on the usage of LCS for the German
Environmental Agency (Umweltbundesamt, UBA).

Name Organisation City, Country Project Sensor

Uhoo,
Saúl García Instituto de Salud Carlos III Madrid, Spain ICARUS
ioTech-portable PM sensor
Department of Environmental Sciences, Uhoo,
David Kocman Ljubljana, Slovenia ICARUS
Institut “Jozes Stefan” ioTech-portable PM sensor
Uhoo,
Ondřej Mikeš RECETOX, Masaryk University Brno, Czech Republic ICARUS
ioTech-portable PM sensor
Feasibility study, for the use of sensors in the Sensors from Alphasense Ltd.,
Bernd Laquai University of Stuttgart Stuttgart, Germany
context of an epidemiological study. SDS011
Andreas Madsasck Ok Lab Stuttgart, Germany SENSOR.COMMUNITY SDS011
Norwegian Institute for Air Research iFLINK European Citizen Science Association
Núria Castell Oslo, Norway Alphasense Ltd., SDS011
(NILU) Working Group
The state of Hennef, The state of Moers, and own
Robert Heinecke Breeze Technology Hamburg, Germany Purchased from different companies
Projects
National Institute for Public Health and
Hester Volten the Environment, Ministry of Health, Netherlands Multiple Citizen Science Projects SDS011, Alphasense Ltd.
Welfare and Sport.
Helge Simon & Lorenz Harr Johannes Gutenberg–Universität Mainz Mainz, Germany Own Project Alphasense Ltd.
European Commission Joint Research
Michel Gerboles & Annette Borowiak Ispra, Italy AirSensEUR, AQUILA Different sensor boxes
Centre
Institut für Energie and Umwelttechnik
Christof Asbach Duisburg, Germany Own Project SDS011, sensors from Alphasense Ltd.
(IUTA) e.V.
Sensoren zur Messung von Aerosolen and
SDS011, Alphasense Ltd. (NO2-B43F,
Dieter Klemp & Robert Wagner Forschungszentrums Jülich Juelich, Germany reaktiven Gasen and Analyse ihrer Auswirkung
NO-B4, OX-B431, CO-B4)
auf die Gesandheit (SMARAGD)
EEA report on Assessing air quality through
Anke Lükewille European Environmental Agency (EEA) Copenhagen, Denmark
citizen science
Air quality awareness project that
Panagiota Syropoulou DRAXIS, hackAIR Greece hackAIR
eventually leads to the use of SDS011.
Sensors 2021, 21, 3960 24 of 27

Table A1. Cont.

Name Organisation City, Country Project Sensor

Institute for Advanced Sustainability
Erika von Schneidemesser Potsdam, Germany Urban Climate Under Change [UC]2 Earth Sense (Zephyr)
Studies (IASS)
Andreas Phillip Universität Augsburg Augsburg, Germany Smart AQnet SDS011, Alphasense Ltd.
EUCENTRE, Department of Risk Uhoo,
Alberto Gotti Pavia, Italy ICARUS
Scenarios ioTech-portable PM sensor
Uhoo,
Danielle Vienneau Swiss TPH Basel, Switzerland ICARUS
ioTech-portable PM sensor
Markus Pesch Grimm Aerosol Technik GmbH Muldestausee, Germany Smart AQnet Alphasense OPC N3
Stefan Hogekamp PALAS Karlsruhe, Germany Own Project Own technology
M. M. Prada Bosch Renningen, Germany Own Project Different sensors
Sensors 2021, 21, 3960 25 of 27

References
1. Mehta, A.J.; Zanobetti, A.; Koutrakis, P.; Mittleman, M.A.; Sparrow, D.; Vokonas, P.; Schwartz, J. Associations between Short-term
Changes in Air Pollution and Correlates of Arterial Stiffness: The Veterans Affairs Normative Aging Study, 2007–2011. Am. J.
Epidemiol. 2013, 179, 192–199. [CrossRef]
2. Hemmingsen, J.G.; Rissler, J.; Lykkesfeldt, J.; Sallsten, G.; Kristiansen, J.; Møller, P.; Loft, S. Controlled exposure to particulate
matter from urban street air is associated with decreased vasodilation and heart rate variability in overweight and older adults.
Part. Fibre Toxicol. 2015, 12. [CrossRef]
3. Lee, B.-J.; Kim, B.; Lee, K. Air Pollution Exposure and Cardiovascular Disease. Toxicol. Res. 2014, 30, 71–75. [CrossRef] [PubMed]
4. Münzel, T.; Gori, T.; Al-Kindi, S.; Deanfield, J.; Lelieveld, J.; Daiber, A.; Rajagopalan, S. Effects of gaseous and solid constituents of
air pollution on endothelial function. Eur. Heart J. 2018, 39, 3543–3550. [CrossRef]
5. Weichenthal, S. Selected physiological effects of ultrafine particles in acute cardiovascular morbidity. Environ. Res. 2012, 115,
26–36. [CrossRef] [PubMed]
6. United Nations. Transforming Our World: The 2030 Agenda for Sustainable Development Goals. Available online: https:
//sustainabledevelopment.un.org/post2015/transformingourworld (accessed on 29 January 2020).
7. World Health Organisation. Outdoor Air Pollution a Leading Environmental Cause of Cancer Deaths. Available online:
http://www.euro.who.int/en/health-topics/environment-and-health/urban-health/news/news/2013/10/outdoor-air-poll
ution-a-leading-environmental-cause-of-cancer-deaths (accessed on 29 January 2020).
8. Lukeville, A. Assessing Air Quality through Citizen Science; Publications Office of the European Union: Luxembourg, 2019; ISBN
1977-B449.
9. Federal Ministry of Justice and Consumer Protection, Germany. Neununddreißigste Verordnung zur Durchführung des Bundes-
Immissionsschutzgesetzes (Verordnung über Luftqualitätsstandards und Emissionshöchstmengen—39. BImSchV) Anlage 5 (zu
den §§ 14 und 15) Kriterien für die Festlegung der Mindestzahl der Probenahmestellen für ortsfeste Messungen der Werte für
Schwefeldioxid, Stickstoffdioxid und Stickstoffoxide, Partikel (PM10, PM2,5), Blei, Benzol und Kohlenmonoxid in der Luft. In
Anlage 5 39. BimSchV—Einzelnorm; 2010. Available online: https://www.gesetze-im-internet.de/bimschv_39/anlage_5.html
(accessed on 8 February 2020).
10. Federal Ministry of Justice and Consumer Protection, Germany. Neununddreißigste Verordnung zur Durchführung des Bundes-
Immissionsschutzgesetzes (Verordnung über Luftqualitätsstandards und Emissionshöchstmengen—39. BImSchV) Anlage 6
(zu den §§ 1, 16 und 19) Referenzmethoden für die Beurteilung der Konzentrationen von Schwefeldioxid, Stickstoffdioxid und
Stickstoffoxiden, Partikeln (PM10 und PM2,5), Blei, Benzol, Kohlenmonoxid und Ozon. In Anlage 6 39. BimSchV—Einzelnorm;
2010. Available online: https://www.gesetze-im-internet.de/bimschv_39/anlage_6.html (accessed on 8 February 2020).
11. Adamiec, E.; Dajda, J.; Gruszecka-Kosowska, A.; Helios-Rybicka, E.; Kisiel-Dorohinicki, M.; Klimek, R.; Pałka, D.; Was, ˛ J. Using
Medium-Cost Sensors to Estimate Air Quality in Remote Locations. Case Study of Niedzica, Southern Poland. Atmosphere 2019,
10, 393. [CrossRef]
12. Wesseling, J.; de Ruiter, H.; Blokhuis, C.; Drukker, D.; Weijers, E.; Volten, H.; Vonk, J.; Gast, L.; Voogt, M.; Zandveld, P.; et al.
Development and Implementation of a Platform for Public Information on Air Quality, Sensor Measurements, and Citizen Science.
Atmosphere 2019, 10, 445. [CrossRef]
13. PurpleAir. PurpleAir|Real Time Air Quality Monitoring. Available online: https://www2.purpleair.com/ (accessed on 1
February 2021).
14. Johnson, K.; Holder, A.; Frederick, S.; Clements, A. PurpleAir PM2.5 U.S. Correction and Performance during Smoke Events
4/2020. In Proceedings of the International Smoke Symposium, Raleigh, NC, USA, 20–24 April 2020.
15. Breeze Technologies. Hochlokale Luftqualitätsdaten für eine lebenswertere Umwelt: Breeze Technologies. Available online:
https://www.breeze-technologies.de/de/ (accessed on 18 February 2021).
16. United States Environmental Protection Agency. Performance Evaluation of Low-Cost Air Quality Sensors|US EPA. Available online:
https://www.epa.gov/research-fellowships/performance-evaluation-low-cost-air-quality-sensors (accessed on 8 February 2020).
17. United States Environmental Protection Agency. DRAFT Roadmap for Next Generation Air Monitoring. 2013. Available online:
https://www.epa.gov/sites/production/files/2014--09/documents/roadmap-20130308.pdf (accessed on 8 February 2020).
18. Gerboles, M.; Spinelle, L.; Signorini, M. AirSensEUR: An Open Data/Software/Hardware Multi-Sensor Platform for Air Quality
Monitoring. Part A: Sensor Shield; Publications Office of the European Union: Luxembourg, 2015; Available online: https:
//publications.jrc.ec.europa.eu/repository/handle/JRC97581 (accessed on 15 March 2019).
19. European Commission; DRAXIS. hackAIR. Available online: https://www.hackair.eu/ (accessed on 10 February 2020).
20. Draxis Environmental SA. Emission: Integrated Platform for the More Efficient Monitoring of Air Pollution with the Use of IoT
Network. Available online: https://draxis.gr/projects/emission (accessed on 10 February 2020).
21. Rai, C.A.; Kumar, P.; Pilla, F.; Skouloudis, A.; Camprodon, G. Summary of Air Quality Sensors and Recommendations for Applica-
tion. iSCAPE—Improving the Smart Control of Air Pollution in Europe. 2016. Available online: https://www.iscapeproject.eu/w
p-content/uploads/2018/12/Resubmitted-D1.5-Summary-of-air-quality-sensors-and-recommendations-for-application.pdf (ac-
cessed on 2 June 2021).
22. Rai, C.A.; Kumar, P.; Pilla, F. Summary of Air Quality Sensors and Recommendations for Application. 2017. Available
online: https://www.iscapeproject.eu/wp-content/uploads/2017/09/iSCAPE_D1.5_Summary-of-air-quality-sensors-and-r
ecommendations-for-application.pdf (accessed on 2 June 2021).
Sensors 2021, 21, 3960 26 of 27

23. Karagulian, F.; Barbiere, M.; Kotsev, A.; Spinelle, L.; Gerboles, M.; Lagler, F.; Redon, N.; Crunaire, S.; Borowiak, A. Review of the
Performance of Low-Cost Sensors for Air Quality Monitoring. Atmosphere 2019, 10, 506. [CrossRef]
24. senseBox. Available online: https://sensebox.de/en/ (accessed on 18 February 2021).
25. Air Quality Monitoring Open Framework—AirSensEUR. Available online: https://airsenseur.org/website/airsenseur-air-qualit
y-monitoring-open-framework/ (accessed on 18 February 2021).
26. Schuetze, A.; Hertel, O.; Karatzas, K.; Tiebe, C.; Conrad, T.; Gerboles, M. Workshop: Setting Standards for Low-Cost Air Quality
Sensors, 11 April 2019, Berlin—LMT. Available online: https://www.lmt.uni-saarland.de/index.php/de/aktuelles/43-aktuelles-
weiterbildung/1245-setting-standards (accessed on 24 December 2020).
27. Gerboles, M. The Road to Developing Performance Standards for Low Cost Sensors in Europe. Available online: https:
//www.epa.gov/sites/production/files/2020--02/documents/session_01_b_gerboles.pdf (accessed on 23 August 2019).
28. European Commission. Measuring Air Pollution with Low-Cost Sensors: Thoughts on the Quality of Data Measured by Sensors.
2017. Available online: https://ec.europa.eu/environment/air/pdf/Brochure%20lower-cost%20sensors.pdf (accessed on 2 June
2021).
29. Spinelle, L.; Gerboles, M.; Villani, M.G.; Aleixandre, M.; Bonavitacola, F. Field calibration of a cluster of low-cost available sensors
for air quality monitoring. Part A: Ozone and nitrogen dioxide. Sens. Actuators B Chem. 2015, 215, 249–257. [CrossRef]
30. Johnson, N.E.; Bonczak, B.; Kontokosta, C.E. Using a gradient boosting model to improve the performance of low-cost aerosol
monitors in a dense, heterogeneous urban environment. Atmos. Environ. 2018, 184, 9–16. [CrossRef]
31. de Vito, S.; Esposito, E.; Castell, N.; Schneider, P.; Bartonova, A. On the robustness of field calibration for smart air quality
monitors. Sens. Actuators B Chem. 2020, 310, 127869. [CrossRef]
32. Kuula, J. Opportunities and Limitations of Aerosol Sensors to Urban Air Quality Monitoring. Ph.D. Thesis, Finnish Meteorological
Institute, Helsinki, Finland, 2020. ISBN 9789523361171.
33. Kuula, J.; Mäkelä, T.; Aurela, M.; Teinilä, K.; Varjonen, S.; González, Ó.; Timonen, H. Laboratory evaluation of particle-size
selectivity of optical low-cost particulate matter sensors. Atmos. Meas. Tech. 2020, 13, 2413–2423. [CrossRef]
34. Baumbach, G. Air Quality Control; Springer: Berlin/Heidelberg, Germany, 1996; ISBN 978–3–642–79003–4.
35. Emeis, S. Measurement Methods in Atmospheric Sciences: In Situ and Remote; Borntraeger: Stuttgart, Germany, 2010; ISBN
9783443010669.
36. Chemistry Dictionary. Beer-Lambert Law|History, Definition & Example Calculation. Available online: https://chemdictionary
.org/beer-lambert-law/ (accessed on 1 February 2021).
37. Grimm Aerosol Technik GmbH. Portable Laser Aerosol Spectrometer and Dust Monitor_Model_1108–1109. Available online:
http://www.wmo-gaw-wcc-aerosol-physics.org/files/opc-grimm-model--1.108-and-1.109.pdf (accessed on 13 June 2016).
38. Alphasense OPC-N2. Available online: https://www.manualslib.com/manual/1540841/Alphasense-Opc-N2.html (accessed on
5 August 2017).
39. Nova Fitness Co., Ltd. SDS011 Laser PM2.5 Sensor Specification. Available online: http://www.inovafitness.com/en/a/chanpi
nzhongxin/95.html (accessed on 2 June 2021).
40. Sensiron. Sensors Specification Statement: How to Understand Specifications of Sensiron Particulate Matter Sensors. Available
online: https://www.sensirion.com/en/download-center/particulate-matter-sensors-pm/particulate-matter-sensor-sps30/
(accessed on 5 June 2020).
41. Alfano, B.; Barretta, L.; Del Giudice, A.; de Vito, S.; Di Francia, G.; Esposito, E.; Formisano, F.; Massera, E.; Miglietta, M.L.;
Polichetti, T. A Review of Low-Cost Particulate Matter Sensors from the Developers’ Perspectives. Sensors 2020, 20, 6819.
[CrossRef]
42. Sensor Community. Build Your Own Sensor and Join the Worldwide Civic Tech Network. Available online: https://sensor.com
munity/en/ (accessed on 24 March 2021).
43. Zusman, M.; Schumacher, C.S.; Gassett, A.J.; Spalt, E.W.; Austin, E.; Larson, T.V.; Carvlin, G.; Seto, E.; Kaufman, J.D.; Sheppard, L.
Calibration of low-cost particulate matter sensors: Model development for a multi-city epidemiological study. Environ. Int. 2020,
134, 105329. [CrossRef] [PubMed]
44. CityOS Air: Community Driven Air Monitoring Network. Available online: https://cityos.io/air (accessed on 1 June 2021).
45. Microsoft Corporation. Available online: https://www.microsoft.com/de-de/ (accessed on 20 October 2020).
46. Raspberry Pi Foundation. Available online: https://www.raspberrypi.org/ (accessed on 20 October 2020).
47. Arduino IDE. Available online: https://www.arduino.cc/ (accessed on 20 October 2020).
48. Sousan, S.; Koehler, K.; Hallett, L.; Peters, T.M. Evaluation of the Alphasense Optical Particle Counter (OPC-N2) and the Grimm
Portable Aerosol Spectrometer (PAS-1.108). Aerosol. Sci. Technol. 2016, 50, 1352–1365. [CrossRef]
49. Di Antonio, A.; Popoola, O.A.M.; Ouyang, B.; Saffell, J.; Jones, R.L. Developing a Relative Humidity Correction for Low-Cost
Sensors Measuring Ambient Particulate Matter. Sensors 2018, 18, 2790. [CrossRef]
50. Crilley, L.R.; Shaw, M.; Pound, R.; Kramer, L.J.; Price, R.; Young, S.; Lewis, A.C.; Pope, F.D. Evaluation of a low-cost optical
particle counter (Alphasense OPC-N2) for ambient air monitoring. Atmos. Meas. Tech. Discuss. 2018, 1–24. [CrossRef]
51. Köhler, H. The nucleus in and the growth of Hygroscopic droplets. Trans. Faraday Soc. 1936, 32, 1152–1161. [CrossRef]
52. Khvorostyanov, V.I.; Curry, J.A. Refinements to the Köhler’s theory of aerosol equilibrium radii, size spectra, and droplet
activation: Effects of humidity and insoluble fraction. J. Geophys. Res. 2007, 112. [CrossRef]
Sensors 2021, 21, 3960 27 of 27

53. Samad, A.; Melchor Mimiaga, F.E.; Laquai, B.; Vogt, U. Investigating a Low-Cost Dryer Designed for Low-Cost PM Sensors
Measuring Ambient Air Quality. Sensors 2021, 21, 804. [CrossRef]
54. Crilley, L.R.; Singh, A.; Kramer, L.J.; Shaw, M.D.; Alam, M.S.; Apte, J.S.; Bloss, W.J.; Hildebrandt Ruiz, L.; Fu, P.; Fu, W.; et al.
Effect of aerosol composition on the performance of low-cost optical particle counter correction factors. Atmos. Meas. Tech. 2020,
13, 1181–1193. [CrossRef]
55. Laquai, B.; Saur, A. Development of a Calibration Methodology for the SDS011 Low-Cost PM-Sensor with Respect to Professional
Reference Instrumentation 2017. Available online: https://www.researchgate.net/publication/322628807_Development_of_a_Ca
libration_Methodology_for_the_SDS011_Low-Cost_PM-Sensor_with_respect_to_Professional_Reference_Instrumentation (ac-
cessed on 2 June 2021).
56. Datta, A.; Saha, A.; Zamora, M.L.; Buehler, C.; Hao, L.; Xiong, F.; Gentner, D.R.; Koehler, K. Statistical field calibration of a
low-cost PM2.5 monitoring network in Baltimore. Atmos. Environ. 2020, 117761. [CrossRef]
57. Tanzer, R.; Malings, C.; Hauryliuk, A.; Subramanian, R.; Presto, A.A. Demonstration of a Low-Cost Multi-Pollutant Network to
Quantify Intra-Urban Spatial Variations in Air Pollutant Source Impacts and to Evaluate Environmental Justice. Int. J. Environ.
Res. Public Health 2019, 16, 2523. [CrossRef] [PubMed]
58. Mahajan, S.; Kumar, P. Evaluation of low-cost sensors for quantitative personal exposure monitoring. Sustain. Cities Soc. 2020, 57,
102076. [CrossRef]
59. Seidel, J.; Ketzler, G.; Bechtel, B.; Thies, B.; Philipp, A.; Böhner, J.; Egli, S.; Eisele, M.; Herma, F.; Langkamp, T.; et al. Mobile
measurement techniques for local and micro-scale studies in urban and topo-climatology. Erde 2016, 147, 15–39. [CrossRef]
60. Alphasense NO2-A43BF. Available online: http://www.alphasense.com/WEB1213/wp-content/uploads/2018/12/NO2A43F.
pdf (accessed on 19 March 2019).
61. Alphasense NO-A4. Available online: http://www.alphasense.com/WEB1213/wp-content/uploads/2016/03/NO-A4.pdf
(accessed on 19 March 2019).
62. Alphasense OX-A431. Available online: http://www.alphasense.com/WEB1213/wp-content/uploads/2018/12/OXA431.pdf
(accessed on 21 March 2019).
63. EKO Instruments B.V. ML-01 Si-Pyranometer. Available online: https://eko-eu.com/products/solar-energy/si-pyranometers
/mL-01-si-pyranometer (accessed on 2 June 2021).
64. Sensiron SHT 35. Available online: https://www.sensirion.com/fileadmin/user_upload/customers/sensirion/Dokumente/2
_Humidity_Sensors/Datasheets/Sensirion_Humidity_Sensors_SHT3x_Datasheet_digital.pdf (accessed on 2 June 2021).
65. Laquai, B.; Chourdakis, I.; Chacon, M.; Samad, A.; Solis, G.; Vogt, U. A Lower-Cost PM/NO2 Air Quality Measurement Unit with
a Low-Cost Air Dryer for Stationary Outdoor Use. 2020. Available online: http://www.opengeiger.de/Feinstaub/SMboxPaper
V1_4_060220_1058.pdf (accessed on 2 June 2021).
66. Fenner, D.; Meier, F.; Bechtel, B.; Otto, M.; Scherer, D. Intra and inter ‘local climate zone’ variability of air temperature as observed
by crowdsourced citizen weather stations in Berlin, Germany. Meteorol. Z. 2017, 26, 525–547. [CrossRef]
67. European Environment Agency. CORINE Land Cover. Available online: https://www.eea.europa.eu/publications/COR0-land
cover (accessed on 18 February 2021).
68. Frost, J. How to Interpret Adjusted R-Squared and Predicted R-Squared in Regression. Available online: https://statisticsbyjim.
com/regression/interpret-adjusted-r-squared-predicted-r-squared-regression/ (accessed on 18 February 2021).
69. Hopper, T. Can We Do Better than R-Squared? Available online: https://tomhopper.me/2014/05/16/can-we-do-better-than-r-
squared/ (accessed on 18 February 2021).