Forecasting rainfall has always been a challenging task due to its unpredictable nature. Nowcasting, or short-term forecasting, provides an effective solution by continuously updating with near-real-time input data to address the variability in rainfall systems. In this paper, we developed the Hydro-Estimator (HE) nowcast model using INSAT-3DR-derived HE precipitation intensity to provide precipitation nowcasts with a lead time of up to three hours. The model was developed within the PySTEPS framework, where motion field vectors from precipitation fields are generated using optical flow algorithms and then advected with an extrapolation method. We conducted a sensitivity study of the model using three different optical flow algorithms: Lucas-Kanade, Proesman, and DART. The analysis revealed that while Proesman achieved the highest skill scores, its processing limitations led us to select the Lucas-Kanade algorithm for further analysis. We validated the model by computing various skill scores, including a Fractional Skill Score (FSS) of 0.9, a Heidke Skill Score (HSS) of 0.58, a Gilbert Skill Score (GSS) of 0.47, and a Mean Absolute Error (MAE) of 0.6 mm/hr for precipitation forecasts over 3 months: April, May, and June 2023. We also analyzed diurnal variation trends in model accuracy, showing improved performance during the initial stages of the forecast. The paper presents case studies of heavy precipitation events in Northeast India on June 17, 2022, and Kerala on May 1, 2022, demonstrating the model’s capability to accurately predict high-intensity rainfall. The model achieved errors of less than 10
原文链接
分享
引用
