Deep Learning Forecasting: An LSTM Neural Architecture based Approach to Rainfall and Flood Impact Predictions in Bihar

Abstract

The increasing prominence of natural disasters and the variability in climate change have underscored the significance of research in contemporary and prospective studies. This investigation zeroes in on floods, a ubiquitous natural calamity that plagues regions globally. Bihar stands out as a state exceptionally vulnerable to persistent and intense flooding, with a staggering 69.70 lakh hectares - equivalent to nearly 74 percent of its geographical expanse - enduring its detrimental consequences. The repercussions of such inundations are widespread, permeating multiple facets of life and erecting impediments to both the welfare and the developmental trajectory of local communities. Vital sectors, including infrastructure, agriculture, and livestock, bear the brunt of the damage, catalyzing the dislocation of innumerable households. In an agile response to these adversities, the state's administrative machinery has mobilized disaster response brigades, ensuring timely aid and succor to the distressed populace. To adeptly predict impending floods, this study harnesses data from 1991 to 2022, integrating pivotal metrics such as the 'Total Population affected (in lacs)', 'Mean Annual Rainfall', and 'Crop damage (in INR lakhs)'. By adopting an LSTM (Long Short-Term Memory) model, the research elucidates its findings via illustrative line plots, furnishing an accessible rendition of the flood trajectory predictions over the forthcoming five-year span. These predictive insights are instrumental for disaster management agencies, enhancing their strategic planning and readied responses to imminent flood events in Bihar. Capitalizing on these forecasts enables authorities to architect proactive interventions, ensuring reduced flood repercussions, bolstered community resilience, and an elevated state of disaster preparedness in the region. © 2024 Elsevier B.V.. All rights reserved.