Influence of the Spatio-Temporal Dynamics of Land Use and Land Cover on Flood Risk

Abstract

This study addresses the escalating threat of flash floods in Algeria, particularly in the Hodna basin, which is exacerbated by climate change and rapid urbanization. It proposes an innovative and integrated Geo-AI approach to flood mapping, combining machine learning (ML) techniques with geospatial data and Geographic Information Systems (GIS). The first part focuses on enhancing flash flood prediction, integrating diverse hydrological and topographical factors from multiple data sources. A stacking ensemble methodology was developed, combining CatBoost models with Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTMs), and Deep Belief Networks (DBNs). This approach demonstrated exceptional predictive performance, particularly CatBoost-CNN, which achieved an accuracy92% accuracy. The second part analyzes the complex interactions between flood risk and spatio- temporal dynamics of land use and land cover (LULC) changes over a 20-year period (2000- 2020) and projects future trends until 2040. Landsat data and a hybrid CA-Markov model were used for LULC classification and future predictions. Complementing these AI-driven predictive approaches, the thesis integrates detailed hydrodynamic simulations using HEC-RAS for critical sections of the Oued El Ksob in M'sila.