Predicting spatiotemporal changes in flood prone regions using PSO-ML coupling under climate change scenarios

In the context of climate change, exploring and predicting the spatio-temporal distribution of flood disasters is crucial for developing effective flood risk management and disaster reduction strategies. This study tackles the shortcomings of traditional methods used to measure the risk of regional flood disasters, which often lack precision. A series of machine learning models enhanced by Particle Swarm Optimization - Machine learning (PSO-ML) models were developed and integrated with General Circulation Model (GCM) data to analyze the temporal and spatial characteristics of flood disasters under different scenarios in Shanxi Province, China. Results indicate that the frequency of days with precipitation exceeding 50 mm in the study area increased from 23 in 1981 to 71 in 2021. The northernmost city, DT, experienced 18 extreme precipitation days, while the southernmost city, YC, recorded 71 such days. A gradual increasing trend in extreme precipitation days was observed from north to south and from distant to near areas. The PSO-ML models demonstrated notably improved performance compared to traditional models across all indices. PSO-XGBoost, PSO-RF, and PSO-KNN exhibited higher prediction accuracy than conventional single models, with AUC values of 0.98, 0.95, and 0.94, respectively. Land use change, elevation, and slope emerged as the most influential factors, with weights of 10.37%, 10.01%, and 8.76%, respectively. Across four scenarios (SSP126, SSP245, SSP370, and SSP585), flood-prone areas were projected to shift southward, with varying degrees of increase in risk areas. The SSP370 scenario showed gradual growth, projecting 7660.116 km² of at-risk area by 2100. The SSP585 scenario exhibited the most rapid growth, with a projected peak of 13,933.69 km² by 2070. This research proposes a novel approach to flood disaster risk assessment and offers insights for effectively mitigating regional flood risks.