Synoptic weather variables and data mining methods for predicting regional heavy precipitation over the southwest of Iran

Abstract

Abstract Due to the socioeconomic impacts, hazards, and losses associated with floods, it is crucial to adopt advanced and more accurate methods for predicting regional heavy precipitation events, especially in flood-prone areas like southwest Iran. This study is aimed to predict regional heavy precipitation events over the southwest Iran using synoptic weather variables and data mining methods. Regional heavy precipitation events are identified by utilizing an innovative multi-frequency-based approach over the study area. Daily total precipitation data were collected from 12 meteorological stations located in the southwest Iran spanning 1987–2018. Furthermore, NCEP/NCAR reanalysis gridded data of six synoptic variables (covering a broad geographical range, including the study area) are used as predictors one to five days before heavy precipitation. Four feature selection methods and ten binary classifier machine-learning models are utilized in this study according to two time-delay scenarios. The top models identified in each scenario were tested to determine their ability to predict regional heavy precipitation events. As a result of this study, the Random Forest classification model with the selected synoptic variables of 1–4 days before the event had the highest efficiency in distinguishing heavy precipitations from non-heavy ones. The proposed model exhibited successful predictions for four out of five recent heavy precipitation events in southwest Iran. Among the synoptic variables, relative humidity and wind speed are most frequent.