Author(s): Romeji Ngangbam; Khelchandra Thongam; Saroja Potsangbam; Dinesh Chanam; Tomba Khumukcham Singh
Linked Author(s): Ngangbam Romeji Singh
Keywords: Ensemble FLEWS-Manipur WRF HEC-HMS LSTM predicted observed peak discharge bias corrections
Abstract: Urban pluvial floods are generally characterized by surge in flow discharges, resulting in high flow peaks and flooding depths resulting in huge socio-economic losses. Flood events recurrently occur in the northeastern Indian Himalayan Region of Manipur especially in the valley region, affecting the urbanized areas of Imphal East and West districts. The present study ascribes the application of an integrated hydrometeorological (numerical weather prediction–hydrological) and deep learning model block to forecast floods in the convoluted landform of Imphal urban, Manipur. Two river catchments: Imphal-Kongba-Iril and Nambul, have been selected as focal river catchments for the study as they primarily flow through Imphal urban capital region. Forecasted Rainfall hourly dataset using Numerical Weather Prediction (NWP) model – Weather Research & Forecasting (WRF) form the major input for the quasi-distributed hydrological model in HEC-HMS platform for computing forecasted flood discharge in the two river catchments. Three major flood events were selected and trained in the deep learning LSTM block for bias corrections in peak flood discharge forecasts along with time of peaks. The model simulations of the maximum forecasted flood peaks showed overestimations and underestimations for the three flood events. The study demonstrated that the predictive quality of the hybrid model is superior in various indices, such as the flood peak and discharge error margins, and more importantly time of peak, which augments robust flood early warning systems in intricate river catchments.
Year: 2025