Spatiotemporal dynamics of flood susceptibility under future precipitation variability, population growth, and land cover change

Overview

This study projects spatiotemporal flood susceptibility dynamics in the Kabul River Basin through 2100 by integrating climate change, population growth, and land cover alteration. The research addresses a gap in understanding how compounding environmental and demographic drivers will reshape flood risk in a transboundary, ecologically sensitive mountainous region. Flood susceptibility projections were generated across multiple future scenarios using machine learning classification.

Methods and approach

Flood susceptibility was modeled using an eXtreme Gradient Boosting (XGBoost) algorithm trained with nine static predictor variables and three dynamic predictors reflecting future environmental conditions. The model generated susceptibility classifications across the Kabul River Basin for decadal intervals from 2020 to 2100 under different socioeconomic and climatic scenarios. Model performance was evaluated using Area Under the Receiver Operating Characteristic Curve (AUC) metrics, cross-temporal consistency correlations, and bootstrap uncertainty analysis with confidence interval estimation.

Results

Very Highly susceptible areas expanded from 11.78% of the basin in 2020 to peak at 14.44% in 2060, declining slightly to 13.51% by 2100. Conversely, Very Low susceptibility areas contracted from 66.17% to 56.43% across the same period. The XGBoost model demonstrated high discriminatory accuracy with AUC values of 0.961-0.962 and strong cross-scenario consistency (correlation range 0.75-0.85). Bootstrap analysis confirmed robustness with mean AUCs of 0.9817-0.9834, minimal standard errors (0.0003), and narrow confidence intervals (0.9719-0.9887).

Implications

The sustained temporal increase in high-susceptibility areas necessitates revision of existing flood management protocols to account for dynamic environmental and demographic drivers. Integration of population growth trajectories and land cover projections into hazard assessment frameworks is essential for effective long-term flood risk reduction in the basin. The demonstrated predictive consistency of the XGBoost model supports its application to analogous climate-sensitive mountainous regions where flood risk assessment requires consideration of multiple interacting stressors.