Investigation of satellite precipitation product driven rainfall-runoff model using deep learning approaches in two different catchments of India

Abstract Rainfall–runoff models are valuable tools for flood forecasting, management of water resources, and drought warning. With the advancement in space technology, a plethora satellite precipitation products (SPPs) available publicly. However, application data data-driven rainfall–runoff model is emerging requires careful investigation. In this work, two rainfall sets, namely Global Precipitation Measurement-Integrated Multi-Satellite Retrieval Product V6 (GPM-IMERG) Climate Hazards Group Infrared with Station (CHIRPS), evaluated development prediction 1-day ahead streamflow. The accuracy from SPPs compared to India Meteorological Department (IMD)-gridded set. Detection metrics showed that light (1–10 mm), probability detection (POD) value ranges between 0.67 0.75 an increasing range, i.e., medium heavy (10–50 mm >50 POD values ranged 0.24 0.45. These results indicate performs satisfactorily reference IMD-gridded Using daily nearly decades (2000–2018) over river basins India's eastern part, artificial neural network, extreme learning machine (ELM), long short-time memory (LSTM) developed modelling. One-day runoff using modelling confirmed both sufficient drive reasonable estimated Nash–Sutcliffe Efficiency coefficient, correlation root-mean-squared error. particular, streamflow forecasts Vamsadhara basin (VRB) LSTM GPM-IMERG inputs resulted Nash-Sutcliffe Coefficient (NSC) 0.68 0.67, while ELM Mahanadhi (MRB) same input NSC 0.86 0.87, respectively, during training validation stages. At time, CHIRPS VRB 0.65, MRB 0.89 0.88, indicated could reliably be used prediction. This paper highlights deep models, such as LSTM, can lead new horizon provide forecasting flood-prone catchments.