Climate-driven optimal canal water allocation in the Changgang Irrigation District, Northeast China, using LSTM (Long Short-Term Memory)-iTransformer forecasting and bilevel fractional programming

Climate change is profoundly altering the water supply and demand pattern by altering precipitation, air temperature, and evapotranspiration processes, thereby posing a major challenge to agricultural water resource management. Existing studies have rarely integrated climate-driven forecasting of water supply and demand with hierarchical canal water allocation optimization, limiting their ability to support future requirements for refined scheduling. To enhance water allocation efficiency in rice irrigation districts under future climate change, this study establishes a comprehensive analytical framework that integrates an LSTM-iTransformer prediction model with a bilevel fractional programming (BFP)-based optimal canal scheduling model. The framework was applied to the Changgang Irrigation District in Heilongjiang Province and solved using a nested NSGA-II-Dinkelbach optimization approach. Multiple SSP-RCP climate scenarios derived from the BCC-CSM2-MR global climate model of CMIP6 were incorporated into the optimization framework to generate adaptive canal water allocation schemes for future climatic conditions. The results show that the LSTM-iTransformer model achieves the best performance in predicting both runoff and ET 0 , with R 2 values of 0.8946 and 0.8415, respectively, representing improvements of 5.96% and 7.66% compared to the LSTM model. Optimization results show that the BFP model effectively coordinates canal system operational efficiency with field irrigation benefits. Under the condition of meeting crop water requirements, it shortens the total water allocation time during the critical growth period from 648 h to 543.1 h, a reduction of 16.19%. Simultaneously, canal water conveyance loss is reduced from 115,000 m 3 calculated using traditional empirical coefficients to 87,672 m 3 , a reduction of 23.76%. Future climate scenario analysis indicates that with increasing emission intensity, the supply-demand imbalance and pressure to balance objectives in irrigation districts gradually intensify. Under the SSP5-RCP8.5 scenario, water conveyance loss in high-load canals 25 increases by 23.8% compared to the baseline scenario. The model, by optimizing water conveyance timing and the proportion of water allocation during the critical growth period, improves the system's adaptability to climate change.