Rainfall temporal variability-oriented optimization of urban water resources allocation

Beijing City, China. The temporal variability of precipitation induces significant spatiotemporal fluctuations and uncertainties in water availability and demand, rendering conventional annual water allocation strategies inadequate . To address this issue, this study developed a rainfall temporal variability-oriented optimization method for allocating urban water resources. The framework comprised three key components: (i) analysis of rainfall temporal distribution characteristics using long short-term memory (LSTM) neural network; (ii) temporal disaggregation of annual water demand into monthly scales based on historical averages, with coordinated multi-source supply through adaptive priority adjustment; (iii) development of a multi-objective mathematical model solved using the Improved Non-dominated Sorting Genetic Algorithm (INSGA-II). The results demonstrated strategic monthly adjustments in water allocation priorities across the four optimized schemes, leading to shifts in the annual contributions from various water sources. For example, during high precipitation periods, groundwater withdrawals declined, while contributions from surface water and non-conventional water increased accordingly. Compared to the baseline year, optimized schemes improved economic efficiency, with water consumption reductions of 0.35 × 10 8 m 3 , 0.55 × 10 8 m 3 , 0.70 × 10 8 m 3 , and 0.68 × 10 8 m 3 , respectively. • Optimize water allocation with LSTM-based rainfall temporal analysis. • Assess rainfall-period water demand shifts for adaptive allocation strategies. • Solve multi-objective water allocation models using INSGA-II. • Enable adaptive water source allocation through dynamic source prioritization.