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

Area of Science:

• Environmental Science
• Climate Science
• Geospatial Analysis

Background:

• Climate change necessitates improved flood disaster prediction.
• Traditional flood risk assessment methods lack precision.
• Understanding spatio-temporal flood dynamics is vital for mitigation.

Purpose of the Study:

• To develop precise, machine learning-enhanced models for flood disaster risk assessment.
• To analyze temporal and spatial flood characteristics under various climate scenarios.
• To identify key factors influencing flood risk in Shanxi Province, China.

Main Methods:

• Integrated Particle Swarm Optimization-Machine Learning (PSO-ML) models with General Circulation Model (GCM) data.
• Applied PSO-XGBoost, PSO-RF, and PSO-KNN for enhanced prediction accuracy.
• Analyzed land use change, elevation, and slope as influential factors.

Main Results:

• Extreme precipitation days increased significantly from 1981 to 2021.
• A north-to-south gradient in extreme precipitation days was observed.
• PSO-ML models achieved high prediction accuracy (AUC up to 0.98).
• Land use, elevation, and slope were key flood risk drivers.
• Flood-prone areas are projected to shift southward, with significant increases under SSP370 and SSP585 scenarios.

Conclusions:

• PSO-ML models offer superior flood risk prediction compared to traditional methods.
• Climate change is exacerbating flood risks, particularly in southern regions.
• The study provides critical insights for regional flood risk management and disaster reduction.