Real-time flood forecasting using satellite precipitation product and machine learning approach in Bagmati river basin, India

Abstract

Real-time flood forecasting is crucial for early flood warnings. It relies on real-time hydrological and meteorological data. Satellite Precipitation Products offer real-time global precipitation estimates and have emerged as a suitable option for rainfall input in flood forecasting models. This study first compared the daily Satellite Precipitation Products of Integrated Multi-Satellite Retrievals for Global Precipitation Measurement (IMERG) with observed rainfall data of India Meteorological Department from the year 2001 to 2009 using contingency tests. Rainfall data of IMERG are used to build four Real-time flood forecasting models based on machine learning: feedforward neural network (FFNN), extreme learning machine (ELM), wavelet-based feedforward neural network, and wavelet-based extreme learning machine. The models consider the IMERG gridded data at 1 h resolution as input to predict water level at Hayaghat gauging station of Bagmati River with lead times from 1 h to 10 days. These models have been trained and tested with the observed water level data. The model performance was also evaluated using various statistical criteria. Results showed good correlation between IMERG and observed data with a probability of detection of 85.42%. Overall, wavelet-based models outperformed their singular counterparts. Among the singular models, the FFNN model performed better than ELM with satisfactory predictions up to 5 days of lead time. For a 7 days lead time, only wavelet-based-FFNN performs well, whereas none of the models produced satisfactory results for 10 days lead time.