Water-economic collaborative management in the Yellow River Basin by dual-channel adaptive spatio-temporal graph transformer

The Yellow River Basin faces severe water resource constraints, posing dual challenges of ecological protection and high-quality development.Traditional predictive models cannot simultaneously capture the topological constraints of natural hydrological processes and the interaction patterns of socio-economic trade networks.To address this, this study proposes a dual-channel adaptive spatio-temporal graph transformer (DA-STGT) model.The model integrates multi-source hydrological and economic data to construct a high-precision spatio-temporal dataset for the basin.It employs a parallel graph convolution structure for the 'physical water network' and the 'economic gravity network', effectively decoupling unidirectional water flows from bidirectional economic influences.Experimental results show that DA-STGT performs exceptionally in predicting basin-wide water supply and demand, reducing root mean square error (RMSE) to 3.31 (10 8 m 3 ).Statistical tests (p < 0.01) confirm its superiority over existing benchmarks.Scenario simulations further indicate that coordinated scheduling under extreme drought conditions can reduce total economic losses across the basin by 23.1%.This study establishes a robust, data-driven decision-making framework, offering quantitative support for sustainable basin management and providing methodological guidance for interdisciplinary fields such as socio-hydrology and water resource economics.