Assessing the performance of the WRF model cumulus parameterization schemes for the simulation of five heavy rainfall events over the Pol-Dokhtar, Iran during 1999–2019

The increase in number and intensity of extreme precipitation events has been remarkable over the Pol-Dokhtar, a city in the west of Iran, during 1999–2019. Five of the most extreme and unprecedented rainfall events were chosen to assess the performance of Cumulus Parameterization Schemes (CPSs) of the Weather Research and Forecasting model in reproducing rainfall amount and temperature. The numerical experiments were carried out to investigate the skills of the Kain-Fritsch (KF), Betts-Miller-Janjic (BMJ), Grell–Devenyi, Grell 3D (G3D), Grell-Freitas, KIAPS Simplified Arakawa-Schubert (KSAS), and Tiedtke schemes. The simulated outputs are compared quantitatively with gridded data and the in-situ observations using statistical metrics. Therefore, the evaluation process included both pattern- and point-comparison. The in-situ precipitation indicated that the KSAS-CPS produced the best performance in winter/autumn rainfall events, while the KF-CPS reached the best in spring/summer rainfall events. All schemes performed similarly in predicting in-situ temperature. Comparing with precipitation pattern from the gridded data revealed that the CPSs performance was seasonal. Most of the CPSs overestimated the precipitation in spring/summer cases, while they led to dry bias in autumn/winter cases. Objective comparison between reanalysis and simulated surface temperature patterns illustrated that relatively all CPSs produced warm bias during the most simulations. The calculated statistical indices using gridded data showed that Tiedtke-CPS produced the highest correlation and less mean absolute error in predicting the space-averaged precipitation values, while the BMJ-CPS led to the worst results. Compared with the gridded data, the G3D-CPS (KSAS-CPS) led to the lowest (greatest) dry bias for precipitation distribution. In most simulations, all CPSs, except the GD- and G3D-CPSs, produced the maximum precipitation rate with a lag time (of about 1–4 h) compared with the gridded data.