Assessment of physical parameterization schemes in WRF over national capital region of India

Abstract

Increase in the extreme weather events around the world has necessitated application of numerical weather prediction (NWP) models to forecast these events and minimize consequences. Application of NWP models requires appropriate selection of physics parameterization options for close representation of atmospheric processes. In this study, the WRF model performance was evaluated for varying physical parameterization of surface processes in simulating meteorology with respect to varying (i) shortwave and longwave radiation schemes, (ii) planetary boundary layer (PBL) and corresponding surface layer (SL) schemes over Delhi NCR. A total of 11 simulation sets were curated with 7 PBL schemes (ACM2, GBM, UW, MYJ, SH, TEMF and BouLac), 4 surface layer schemes (Pleim-Xiu, Revised MM5, Eta and TEMF), 3 shortwave radiation schemes (Dudhia, New Goddard and RRTMG), 3 longwave radiation schemes (RRTM, New Goddard and RRTMG) and 2 land surface models (LSM) (Pleim-Xiu and Noah). Sensitivity experiments are performed at a fine resolution (1 km) with updated LULC input. Based on the sensitivity analysis, it is inferred that the simulation set which works best for the region is TEMF PBL, TEMF SL, Dudhia shortwave radiation, RRTM longwave radiation and Noah LSM schemes. The TEMF PBL scheme is designed as hybrid (local–nonlocal) scheme and thereby, with consideration of both local and nonlocal viewpoints it is noted that the near-surface meteorological parameters are depicted with greater accuracy. To further address the model biases it is important to refine the physical parameterizations schemes in the WRF model or using different bias correction and data assimilation techniques.

Appendices

Appendices

1.1 Appendix 1

See Table 7.

Table 7 Salient features of the surface physical parameterization schemes are listed below

1.2 Appendix 2

Statistical measures used for Model Evaluation based on previous studies (Willmott et al. 2012; Sati and Mohan 2016; Gunwani and Mohan, 2017; Emery et al. 2001).

Following statistical parameters have been used in the present study.

• Index of agreement (IOA)

  $$ IOA = 1 - \frac{{\mathop \sum \nolimits_{i = 1}^{n} \left| {Pi - Oi} \right|}}{{2\mathop \sum \nolimits_{i = 1}^{n} \left| {Oi - \bar{O}} \right|}} $$

  when $\mathop \sum \nolimits_{i = 1}^{n} \left| {Pi - Oi} \right| \le 2\mathop \sum \nolimits_{i = 1}^{n} \left| {Oi - \bar{O}} \right|$ and

  $$ IOA = \frac{{2\mathop \sum \nolimits_{i = 1}^{n} \left| {Oi - \bar{O}} \right|}}{{\mathop \sum \nolimits_{i = 1}^{n} \left| {Pi - Oi} \right|}} - 1 $$

  when $\mathop \sum \nolimits_{i = 1}^{n} \left| {Pi - Oi} \right| > 2\mathop \sum \nolimits_{i = 1}^{n} \left| {Oi - \bar{O}} \right|$ IOA ranges between -1 and 1 (Willmott et al. 2012).

• Mean bias (MB)

  $$ {\text{MB}} = {\bar{\text{P}}} - {\bar{\text{O}}} $$

• Mean absolute gross error (MAGE)

  $$ MAGE = \frac{1}{N}\mathop \sum \limits_{i = 1}^{N} \left| {Pi - Oi} \right| $$

• Root mean square error (RMSE)

  $$ {\text{RMSE}} = \sqrt {\frac{{\mathop \sum \nolimits_{{{\text{i}} = 1}}^{\text{N}} \left( {{\text{Pi}} - {\text{Oi}}} \right)^{2} }}{\text{N}}} $$

  where Pi, Oi, $\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle-}$}}{P}$ and $\underset{\raise0.3em\hbox{$\smash{\scriptscriptstyle-}$}}{O}$ represent predicted data, observed data, predicted mean and observed mean respectively. Statistical benchmarks for the meteorological parameters (Emery et al. 2001; Sati and Mohan 2016; Gunwani and Mohan 2017)—Temperature Mean Bias ± 0.5 K, MAGE 2 K, RMSE 2 K; Wind Speed Mean Bias ± 0.5 m/s, MAGE 2 m/s, RMSE 2.0 m/s; Relative Humidity Mean Bias < 10%, RMSE < 20%.