Forecasting Floods Using Deep Learning Models: A Longitudinal Case Study of Chenab River, Pakistan

This study presents an integrated methodological approach for forecasting future floods. Over the course of human history, floods have undoubtedly been the most common form of catastrophe. In this study, hydrological analysis was combined with time-series modeling, and this integrated approach was used to forecast daily river flow. The daily runoff of the catchment area was calculated using curve number modeling by incorporating static spatial statistics and temporal rainfall data. The applicability of time-series modeling is checked through two deep learning algorithms: Long Short-Term Memory (LSTM) and Multilayer Group Method of Data Handling (ML-GMDH). This integrated approach yielded an optimal set of attributes for river flow prediction. The proposed methodology was applied to the Chenab River Basin, Pakistan, where the absence of machine learning and deep learning models tailored for flood prediction emphasizes the urgent need for innovative and advanced tools that facilitate timely flood forecasts. The LSTM model obtained the best results, attaining the highest R2 of 0.91, and the highest coefficient of correlation of 0.96. Additionally, the ML-GMDH model signifies its capability with 0.88 R2. The findings of this study demonstrate that the proposed hybrid approach is a promising and reliable choice for river flow prediction, particularly for ungauged basins where upstream hydrological data is not available. The proposed model is advanced and effective, and can be employed to forecast future flood events in the Chenab Basin and similar locations worldwide.