پیش بینی بارش‌های سنگین منطقه‌ای در جنوب غرب ایران: رویکردی مبتنی بر داده‌کاوی

Accurate and timely forecasting of regional heavy precipitation plays a crucial role in preventing floods and minimizing the resulting damage. This study presents a novel definition of regional heavy precipitation based on the probabilistic pattern of thunderstorms. For this purpose, daily precipitation data (1987 to 2018) from 12 synoptic stations in southwest Iran were used. In addition to precipitation data, six synoptic variables at pressure levels of 1000 to 200 hPa, corresponding to one to five days before the occurrence of heavy precipitation, were considered as predictors. These variables cover a wide area beyond the study region. This research was conducted using two variable selection methods and seven machine learning models of the binary classification type. In the defined scenario, precipitation data from the days before the occurrence of heavy precipitation was used continuously. The present study showed that for accurate discrimination between heavy and normal precipitation, the use of precipitation data up to four days before the occurrence of precipitation in the Random Forest model with the Chi-Square variable selection method is the best method. The relative humidity variables at three pressure levels of 1000, 700, and 500 hPa and the meridional wind component at the 1000 hPa pressure level were the most frequently used variables in this model.