A Sensitivity Analysis of the Weather Research and Forecasting

(WRF) Model for Finding Wind Resource Potential

Rovick P. Tarife, Romeo C. Crampatanta, Jr., and William F. dela Cerna, Jr.

numerical prediction model in wind resource has been a

Abstract— Due to the unsustainable use of fossil fuels, the use practice over the years. These models take into consideration
of renewable energy such as wind energy takes the spotlight for solving physical governing motions of the atmosphere with
power generation. However, assessment for finding a good wind varying degree of complexity in hopes to replicate the
resource potential poses a tough challenge for the researchers
and modelers. Weather Research and Forecasting Model can
behavior and interaction of the earth and the atmosphere. The
identify a wind resource potential and it uses a numerical model includes parameterization schemes for solar and
representation based on the current weather measurement. A infrared radiation, cloud microphysics and convection
sensitivity analysis is carried out with the selection of the month (cumulus clouds), a soil and more.
of January and July as representative for the dry and wet
season in the Philippines. Two sets of parameters were tested Most studies about WRF Model did not take into
and for the month of January Pleim-Xiu-ACM2-Pleim-Xiu
exhibited better results. For the month of July,
consideration the effects of longer simulation although
MM5-YSU-Noah showed better results. Based on the outcome seasonal simulation is a practice already. Several studies
of the whole simulation, the area of Tubod, Lanao del Norte conducted sensitivity analysis to know the behavior of the
showed a good wind resource potential subjected to further model in response to the domain being simulated. A
studies. sensitivity analysis can give to the modeler vital information
about the use and influence of several model parameters
Index Terms— About four key words or phrases in [3]-[10].
alphabetical order, separated by commas.

This study makes use of the ARW core of WRF 3.6.1 and
I. INTRODUCTION focuses on simulating near-surface wind. A sensitivity
analysis of several physical schemes was developed for the
The importance of renewable energy nowadays is area of Mindanao, one of the major islands in the Philippines.
indisputable. Thus, renewable energy takes a significant role This study also aims to determine which combinations of
in power generation. This is to address the effects that the model parameters provides the most accurate results that will
conventional way of power generation brought to earth. The aid in the identification of the area where wind project is
limited accessibility of the natural resources cannot be viable.
promptly replaced by the normal means on a level of its
utilization. With this, nonrenewable energy sources such as
petroleum gas, oil and coal are ought to be not a practical II. METHODOLOGY
means because their formation takes billions of years.
Moreover, the excessive production of greenhouse gases A. Datasets
which can lead to major climate change. By utilizing The datasets of static fields (topography, land use, land water
sustainable power source, our decrease in reliance to masks, land cover classification, albedo) utilized in this study
petroleum products and gas reserves can help on enhancing were obtained and interpolated from the National Research
the security and sustainability. It will also help in mitigating for Atmospheric Research (NCAR) data base. This data
the impact of climate change. Of all the renewable sources, serves as input data to the WRF model for initial and
wind energy has been in a spotlight. boundary condition. This is driven by National Centers for
Environmental Prediction (NCEP) Final Operational Global
In the Philippines, rapid development of the usage of wind Analysis, with 1° of spatial resolution (both in latitude and
energy as a primary source of electricity is undeniable [1]. longitude) and 6 h of temporal sampling. The actual field data
With that, the challenge of evaluating the wind resource of a from Philippine Atmospheric, Geophysical and Astronomical
certain domain is proven to be a difficult task for the Services Administration (PAGASA) EL Salvador was also
developers and the researchers. The use of mesoscale utilized for validation purposes.

Manuscript received October 14, 2019 B. Model Configuration

R. P. Tarife is with the Electrical Engineering and Department,
MSU-Iligan Institute of Technology, Iligan City, Philippines (e-mail: The simulation domains are shown in Fig. 1. The WRF model
[email protected]). is built over parent domain (D1) with 25 km spatial resolution
R. C. Crampatanta, Jr., was with Electrical Engineering and Department, covering all of Mindanao. The first nested domain (D2), with
MSU-Iligan Institute of Technology, Iligan City, Philippines (e-mail:
[email protected]). spatial resolution of 5 km, covers the area of Region 9, 10 and
W. F. dela Cerna, Jr., was with Electrical Engineering and Department, area of Lake Lanao. The innermost domain (D3), which is the
MSU-Iligan Institute of Technology, Iligan City, Philippines (e-mail: focused of the study, has covered Lanao del Norte and some
[email protected]).
parts of Misamis Oriental. The vertical structure of the model Parametrization Available Schemes
contains 35 vertical levels with 10 m, 5 m and 2 m geographic Surface Land (SL) MM5, ETA, Pleim-Xiu
resolution for the 3 domains. Planetary Boundary Layer (PBL) Yonsei University,
Mellor-Yamada-Janjic, ACM2,
RUC
Land Surface Model (LSM) Noah, RUC, 5-layer, Pleim-Xiu
Long-wave Radiation Rapid Radiative Transfer Model
Short-wave Radiation Dudhai
Cumulus Kain-Fritsch
Microphysics WSM6

Table 2. Parametrization of each schemes

First Case Second Case
Parametrization
Parameters Parameters
Surface Land (SL) MM5 Pleim-Xiu
Planetary Boundary Yonsei University ACM2
Layer (PBL)
Land Surface Model Noah Pleim-Xiu
(LSM)
Figure 1. Simulation domain projection. Long-wave Radiation Rapid Radiative Rapid Radiative
Transfer Model Transfer Model
The sensitivity test performed in this study is divided into Short-wave Radiation Dudhai Dudhai
different categories: seasonal representation, physical options, Cumulus Kain-Fritsch Kain-Fritsch
simulation domain resolution and validation of the model. Microphysics WSM6 WSM6
Two different months were considered that represent the two
seasons in the Philippines: the wet and dry seasons. The To validate the simulation output data, statistical
months of July and January were chosen for wet and dry parameters such as Mean Absolute Error, Root Mean
season, respectively. Squared Error, Standard Deviation Error, and BIAS were
used to evaluate the performance of the WRF model.
For physical options, adaptive time step was used to
facilitate faster runtime in the simulation. Since the focus of
C. Data Extraction and Visualization
this study is to simulate near surface winds, physical options
that deals with the interaction of the atmosphere and the land The simulation using WRF Model gives an output file in
surface will be utilized. The available parameter options for netCDF format. This format is a representation of scientific
the simulation is shown in Table 1. The physical options variables in an array-oriented data. WRF model output is a
meteorological data. In this study, two scientific variables
related to the boundary processes parameterization (SL, PBL,
were extracted from the output data which are the wind speed
AND LSM) are the one that will have a larger influence on an
and wind direction.
accurate near surface wind simulation. Although more
physical options are available in the model (for cumulus, For developing a wind resource map, open source GIS
radiation, microphysics etc.), it is not feasible to include them based software will be used for visualization. Using the
in the sensitivity analysis. Other options will be leaved as extracted wind speed, the data visualized with a projection of
default values in the configuration of the WRF Model WGS 1984. An interpolation method was used to have a
Version 3.6.1 ARW core. smooth raster data. Lastly, identification of the area where
high wind speeds are observed
In order to account the effect of geographical
representation of the area modeled, the spatial resolution is
set as 25 km, 5 km, and 1 km. Higher domain resolution III. RESULTS AND DISCUSSION
results to accurate domain representation. The duration of the simulation was approximately 1 month
using 8 logical processors and adaptive time step. Simulation
of each cases with corresponding month was approximately 5
days. The comparison of the simulated wind speed data from
the real data are shown in Fig. 2 and Fig. 3. Each case for the
month of January and July are shown differently.

Table 1. Parametrization Schemes

Figure 2. Wind speed plot for January.

Figure 4. Wind Rose Plot.

For the month of July, wind rose (c) and (d) shows its wind
rose plot. The 1st and 2nd case shows similar results as wind
blows particularly from the west. This result coincides with
the effect of southwest monsoon that the country experienced
during this month.

The different wind power maps are shown in Fig. 5 to Fig.

8. Projection of the simulated data using GIS environment
Figure 3. Wind speed plot for July. shows the specific places in the area of Mindanao which
exhibits larger wind speeds. The data is projected in WGS
Using the statistical methods, the value of MAE is 1984 and employed raster smoothing technique to acquire a
0.87096, RMSE is 13.716, Bias is -0.87096 and STDE is better visualization of the output data. From the map, the area
13.688 for 2nd case parameter for the month of January extracted which exhibits viable potential for wind project is
which exhibits lesser values in comparison to the 1st case the area of Panguil Bay. Funneling effect occurs in the bay
parameter for the same month. This shows that the model which shows large wind speed values. This simulation
parameters for 2nd case used outperformed the model replicated the phenomenon.
parameters used in the 1st case parameters. For the month of
July, 1st case parameter shows better results compared to the Wind Power calculation of the area is based in the output
2nd case parameters. The values for each statistical parameter wind speed at 80m and 100m. Wind power varies for a
for the 1st case parameter are as follows: MAE = 0.88699, specific wind turbine design. The specific design used in the
RMSE = 13.968, Bias = -0.88699 and STDE = 13.94. As calculation is discussed by [11]. This design was chosen
observed, the Bias value is negative which means the model because it is cost effective in compare to other designs.
has underestimated the predicted wind speed values.
Notable area in the municipality of Tubod, Lanao del Norte
The wind rose plot of each cases for each month is and in the bay area of Ozamiz city. The highest simulated
visualized in Fig. 4. Wind Rose (a) and (b) are the plots for wind speed in this area is 7.90791 m/s.
the month of January which show that most of the time the
wind blows from the western region. The length of each
“spoke” around the circle is related to the frequency that the
wind blows from a particular direction per unit time. As
depicted in the two wind roses, 6% of the total time the wind
blows from the western region at the speed between 2 to 3
m/s. This also shows that the wind blows rarely from the
north and south directions and also shows larger wind speeds
in the eastern and western regions.
Figure 5. Wind Power Map at 80m for January. Figure 7. Wind Power Map at 100m for January.

Figure 6. Wind Power Map at 80m for July. Figure 8. Wind Power Map at 100m for July.
 IV. CONCLUSION [9] P. A. Jimenez and J. Dudhia, “On the ability of the WRF model to
reproduce the surface wind direction over complex terrain,” J.
A sensitivity analysis on WRF model is presented in this Appl.Meteorol. Climatol., vol. 52, no. 7, pp. 1610–1617, 2013.
chapter. The aim of this study is to evaluate the use of [10] E. Avolio, S. Federico, M. M. Miglietta, T. Lo Feudo, C. R. Calidonna,
different physical options in the simulation of the and A. M. Sempreviva, “Sensitivity analysis of WRF model PBL
schemes in simulating boundary-layer variables in southern Italy: An
near-surface wind speed and direction. The sensitivity experimental campaign,” Atmos. Res., vol. 192, no. December 2016,
analysis was applied to the area of Lanao del Norte and some pp. 58–71, 2017.
parts of Misamis Oriental as area of interest for wind project. [11] C. Levandowski, “Evaluating Tall Wind Turbine Towers in the Field.”
Two different sets of parameterization schemes regarding B.S. thesis, Iowa State University, 2015.
boundary layer (SL, PBL and LSM) were tested for different
seasonal conditions (wet and dry month). The SL-PBL-LSM
parameterization composed of Pleim-Xiu-ACM2-Pleim-Xiu
has better performance for the January and MM5, Yonsei
University and Noah for the month July. All simulation was
done with 1km innermost domain spatial resolution to take
into account the influence of high spatial resolution in the
WRF model. In testing simulated data for bias for both cases,
underestimation of the estimates was observed. Considerable
RMSE values and STDE was also observed for wind speed.

This study was able to identify the area of Tubod, Lanao

del Norte for a potential wind project site which agreed with
previous studies for its potential. Error minimization in the
wind simulation can be achieved by testing and choosing a
suitable model configuration for the region of interest.
Simulation grid characteristics should be a compromise
between the chosen resolution and available computational
resources.

ACKNOWLEDGMENT
All authors thank Philippine Atmospheric, Geophysical and
Astronomical Services Administration (PAGASA) EL
Salvador Station and MSU-IIT Phil-LiDAR 2 Program for
providing the datasets needed for this research.

REFERENCES

[1] J. T. Tolentino, M. V. Rejuso, J. K. Villanueva, L. C. Inocencio, M.

Rosario, and C. O. Ang, “Development of a Wind Resource
Assessment Framework Using Weather Research and Forecasting (
WRF ) Model, Python Scripting and Geographic Information Systems,
” vol. 9, no. 12, pp. 1338–1341, 2015.S. Sohmer and A. Davis, The
Genus Psychotria (Rubiaceae) in the Philippine Archipelago. Texas:
Botanical Research Institute of Texas, 2007.
[2] W. C. Skamarock et al., “A Description of the Advanced Research
WRF Version 3,” Tech. Rep., no. June, p. 113, 2008.
[3] D. Carvalho, A. Rocha, M. Gómez-Gesteira, and C. Santos, “A
sensitivity study of the WRF model in wind simulation for an area of
high wind energy,” Environ. Model. Softw., vol. 33, pp. 23–34, 2012.
[4] Wind Resource Assessment Handbook, New York State Energy
Research and Development Authority (NSYERDA), New York,
U.S.A., 2010.
[5] R. Borge, V. Alexandrov, J. José del Vas, J. Lumbreras, and E.
Rodríguez,“A comprehensive sensitivity analysis of the WRF model
for air quality applications over the Iberian Peninsula,” Atmos.
Environ., vol. 42, no. 37, 2008.
[6] F. J. Santos-Alamillos, D. P. V. Zquez, J. A. Ruiz-Arias, V.
Lara-Fanego, and J. Tovar-Pescador, “Analysis of WRF model wind
estimate sensitivity to physics parameterization choice and terrain
representation in Andalusia (Southern Spain),” J. Appl. Meteorol.
Climatol., vol. 52, no.7, pp. 1592–1609, 2013.H. Poor, An Introduction
to Signal Detection and Estimation; New York: Springer-Verlag, 1985,
ch. 4.
[7] M. Mohan and S. Bhati, “Analysis of WRF Model Performance over
Subtropical Region of Delhi, India,” Adv. Meteorol., vol. 2011, pp.
1–13, 2011.
[8] R. F. Banks et al., “Sensitivity of boundary-layer variables to PBL
schemes in the WRF model based on surface meteorological
observations, lidar, and radiosondes during the HygrA-CD campaign,”
Atmos. Res., vol. 176–177, pp. 185–201, 2016.