Suitability of WRF model for simulating meteorological variables in rural, semi-urban and urban environments of Central India

Abstract

Accurate simulation of meteorological variables is a prerequisite for numerous downstream applications such as air quality modeling and weather forecasting. Weather Research and Forecasting (WRF) model is widely utilized to simulate various prognostic meteorological variables across multiple spatial scales. The suitability of the WRF-v3.9 model in simulation of surface meteorological variables and vertical thermodynamic profile is tested with available in situ surface and radiosonde observations collected from the locations representing rural, semi-urban, and urban environments of the central Indian region. Nested domains with 12 and 4 km grid spacing having 28 vertical layers are set up during the fair weather days of January and March 2018. The model sensitivity is tested by varying two non-local (Yonsei University, YSU and Asymmetric Convective Model, ACM2) and one local (Mellor-Yamada Eta, MY-E) closure Planetary Boundary Layer (PBL) schemes. Results indicate that no particular PBL scheme simulates best for all meteorological variables at different land uses. Overall, thermodynamic variables (temperature and relative humidity) are more accurately simulated than the dynamic variables (wind speed and direction). YSU and MY-E schemes have relatively better accuracy in simulating surface temperature in rural and semi-urban locations, while ACM2 performed better in the urban location. MY-E is relatively better in simulating relative humidity and wind speed in rural and semi-urban locations, while it poorly performed in the urban location. The vertical thermodynamic profile is perfectly correlated with radiosonde observations over the urban location during January and with a reasonably good fit during March. The study provides a comprehensive evaluation of boundary-layer meteorological variables simulated by the WRF model in Central India.

Appendix A

The metrics used for model evaluation in the study are defined as follows: where O is the observed value, P is the predicted value, n is the number of values, $\bar{O}$ and $\bar{P}$ represent the average over the data set, and σ is the standard deviation.

Mean Bias (MB)

$$\text{MB =}\;\frac{1}{n}{\sum }_{1}^{n}\left(P-O\right).$$

Normalized Mean Bias (NMB)

$$\text{NMB =}\;\frac{{\sum }_{1}^{n}\left(P-O\right)}{{\sum }_{1}^{n}O}.$$

Mean Gross Error (MGE)

$$\text{MGE =}\;\frac{1}{n}{\sum }_{1}^{n}\left|P-O\right|.$$

Normalized Mean Gross Error (NMGE)

$$\text{NMGE =}\;\frac{{\sum }_{1}^{n}\left|P-O\right|}{{\sum }_{1}^{n}O}.$$

Pearson’s correlation coefficient (r)

$$\text{r =}\;\frac{1}{\left(n-1\right)}{\sum }_{1}^{n}\left(\left(\frac{O-\bar{O}}{{\sigma }_{O}}\right)\times \left(\frac{P-\bar{P}}{{\sigma }_{P}}\right)\right).$$