Assessment of WRF model sensitivity for simulating super heavy Precipitation, "Case study: 17 to 22 March 2019"

The large-scale computational network of planetary models are not able to predict climatic variables on a regional scale. In other words, these models are affected by processes with a smaller scale than the model network in providing predictions of regional precipitation. Therefore, the model outputs should convert into a regional scale. The research purpose is to investigate the different configurations of the WRF model in the simulation of 5-days rainfall in March 17 to 22 March 2019 in Golestan province, which has caused devastating floods and heavy damage in the province.

The observation and quality control precipitation data was analyzed in 13 synoptic stations of Golestan province for a 5-days period from March 17 to 22 March 2019 in the form of 24 Hours (From 06 UTC the day before to 06 UTC the next day) and 6 hours (00, 06, 12 and 18 UTC are 3:30, 9:30, 15:30 and 21:30 local time, respectively). Also, two types of input data including initial condition data and boundary condition data were used in the WRF model. The boundary condition data was GFS data with 0.5-degree resolution. Furthermore, two domains were used in WRF model, 1) the large (mother) with a horizontal resolution of 18 km and 2) internal domain, which is the main domain and has 6 km horizontal resolution.

Two configuration was selected which showed better output results. The 5-days cumulative precipitation data which caused the flood show that the maximum 24-hour precipitation during the 5-days period is 06:00 UTC on March 18 to 06:00 UTC March 19 and the maximum cumulative rainfall of 6 hours is related to 06 to 12 UTC on March 18, 2019. Subsequently, by study similar research in Iran, different configurations for precipitation prediction were extracted and modeled. Then, in order to determine the accuracy of the model, the values obtained from the model in different configurations were compared with the values of synoptic stations. To ensure this comparison, MAE, d, R and ENS test statistics were used.

The results showed that the WRF model overestimate the precipitation data in most stations. In both configurations, results convey the precipitation cores well illustrated and the model accuracy was good enough in predicting precipitation. In maximum values of precipitation, the configuration of the first type show better results. Overall, the first type configuration performed more accurate than the second type configuration.