Spatiotemporal evolution of droughts and floods in the Yellow River Basin: A novel approach combining CMADS-L evaluation, hydroclimatic zonation and CNN-LSTM prediction

Abstract

Study Region: Yellow River Basin, China, a region of critical ecological and economic significance. Study Focus: This research presents a novel integrated framework for regional drought-flood analysis and precipitation forecasting by evaluating CMADS-L and ERA5 reanalysis datasets against ground-based observations in the Yellow River Basin. We implement K-means clustering on Standardized Precipitation Evapotranspiration Index (SPEI) sequences to identify drought-flood regions and validate a CNN-LSTM hybrid model for regional precipitation forecasting, uniquely bridging traditional climate analysis with modern deep learning techniques for enhanced prediction accuracy. New Hydrological Insights for the Region: Our Analysis Reveals: (1) CMADS demonstrates superior performance over ERA5 in precipitation representation (r = 0.93 vs 0.92; MAE: 9.6 mm vs 17.5 mm; RMSE: 15.8 mm vs 25.9 mm), providing the first comprehensive evaluation of these datasets in complex terrain; (2) We identify three distinct hydroclimatic zones: Northeast (31.6 % area, ''wet-dry-wet'' sequences, quasi-4.5-year oscillations), Southeast (42.5 %, post-2000 drying trends, quasi-2.3-year cycles), and Western (25.9 %, post-2008 drying, quasi-12-year periodicities); (3) Our novel CNN-LSTM hybrid model achieves unprecedented prediction performance (R²: 0.70–0.85), with highest accuracy in the Western zone due to stable precipitation patterns. This integrated approach significantly advances regional hydroclimate understanding and provides a robust, transferable framework for water resource management under changing climate conditions, offering valuable methodological insights for similar river basins globally.

Keywords

• Climate Change
• Yellow River Basin
• Drought-Flood Evolution
• Deep Learning
• Medium to Long-Term Precipitation