Remote sensing-based precipitation forecasting using cloud optical characteristics: threshold optimization and evaluation in Northern and Western Iran

Historically, certain precipitation events within Iran's northern and western regions have caused severe and potentially catastrophic flood occurrences. The primary focus of this study was to develop an expeditious method for forecasting precipitation occurrence within discrete one-kilometer grid cells, exclusively through the examination of relevant cloud optical characteristics. To establish threshold values associated with cloud optical characteristics, we incorporated data from a significant precipitation event on October 12, 2019, within the studied regions, including IMERG precipitation data and cloud data from the High Rate SEVIRI Level 1.5 Image Data and the Optimal Cloud Analysis. The optimization of these cloud thresholds was accomplished using the NSGA-II algorithm, which aimed to maximize the probability of detection (POD) of precipitation while minimizing the false alarm rate (FAR) across 77,674 pixels located within the study area. The threshold values derived from the October 12, 2019, event were subsequently applied to forecast precipitation in two additional events on October 5, 2018, and March 24, 2019. The results indicated that, for these two events, the probability of correctly identifying pixels with precipitation ranged from 66.9 to 96.1% for the first event and 27.5 to 72.2% for the second event within different three-hour intervals. Across the entire period of precipitation events, the POD and FAR values for the first event were 90.5% and 45.8%, respectively, while for the second event, they were 64.2% and 9.5%. This research provides insights into applying remote sensing data and an advanced algorithm to analyze precipitation events. The optimization of cloud parameter thresholds, as demonstrated at the October 12, 2019, event, holds significant promise for enhancing the accuracy of precipitation forecasting. The results from the subsequent events underscore this approach's potential, showing varying success levels in identifying precipitation occurrences. These findings contribute to understanding remote sensing-based precipitation forecasting and highlight the importance of tailored threshold values for specific events and regions.