Due to population growth, environmental pollution and climate change, the lack of water resources has become a critical factor which threatens sustainable agricultural development. Reasonable irrigation scheduling strategies can reduce the waste of water and enhance agricultural water-use efficiency. In the present study, the decomposition-coordination theory was adopted to analyze the hierarchical canal system. A novel nonlinear multi-level multi-objective optimization model for complex canal systems was established, taking account of the multiple demands from decision makers and realistic factors of canal operation. An interactive method of the technique for order preference using similarity algorithm and genetic algorithm was proposed to solve the developed model. The developed model was successfully applied for the operational strategy making of a canal system located in the arid area of northwest China. The results indicated that the optimization model could help shorten the operational duration by two days, achieve about 26% reduction of irrigation water consumption, and improve the efficiency of water delivery from 0.566 to 0.687. That will be very favorable for the promotion of the agricultural water productivity, the relief of water shortage crisis and the sustainable development of agriculture. The outcomes can provide a wide range of support for decision making and make irrigation decision-making more scientific and systematic.

Agricultural water productivity (AWP), which is associated with multiple factors, is an important index for measuring the effectiveness of agricultural water management. The purpose of this study is to promote AWP through optimally allocating limited agricultural water resources with the coordination of related elements. Firstly, the coordination effects of multiple factors related to AWP are quantified as relative optimum membership degrees based on the fuzzy optimum selecting theory. Secondly, based on the relative optimum membership degrees for various crops, a linear fractional programming model is established to maximize AWP in agricultural water resources allocation. Thirdly, the impacts of the allocation schemes on the development of social-economy and ecological environment are discussed using the multi-dimensional regulation theory. The developed integrated system has advantages in increasing agricultural water productivity through the coordination of multiple factors with aspects of economy, society and resources. Moreover, the system is capable of screening schemes considering harmonious development of resources, economy, society and ecology based on optimization results, providing decision makers with more sustainable schemes for irrigation water allocation. The integrated system including the aforementioned three parts is applied to a real-world case study in China to demonstrate its feasibility and applicability. Different water allocation schemes for various crops under different scenarios were obtained. The average value of AWP is 1.85 kg/m³, which is 0.31 kg/m³ higher than the current value of AWP. An optimum scheme with 1.1405 × 10⁸ m³ of water being allocated was also selected due to its highest level of coordination for resources, economy, society and ecology. The developed system can provide an effective method for AWP promotion. The obtained results can help local decision makers adjust water resources allocation schemes.

To optimize the water distribution of three industries based on the water demand prediction under multiple uncertainties, a fuzzy credibility-constrained interval two-stage stochastic programming (FCITSP) model base on the Cobb-Douglas production (CD) function was developed. The CD-FCITSP model integrated fuzzy credibility-constrained programming (FCP), an interval two-stage stochastic programming (ITSP) method and CD function. The developed model could deal with uncertainties with interval, random, and fuzzy features, reflect tradeoffs between different water use sectors, and provide water managers in arid regions with sustainable and reasonable water-allocation schemes under different credibility scenarios of local policies. Moreover, the relationships between economic benefits and water consumption were taken into consideration via the Cobb-Douglas production function. The developed model was applied to support the optimal allocation of limited water resources in Minqin County, northwest China. The obtained solution demonstrated that the developed method could help local water managers to effectively allocate limited water under multiple uncertainties and different credibility scenarios. In addition, water use efficiency could be promoted and the emissions of major pollutants could be reduced. The developed method could be extended to water management practices in other arid regions.

Accurate evaluation of evapotranspiration (ET) is crucial for efficient agricultural water management. Data‐driven models exhibit strong predictive ET capabilities, yet significant limitations like naive extrapolation hamper wider generalization. In this perspective, we explore a novel hybrid deep learning (DL) framework to integrate domain knowledge and demonstrate its potential for evaluating ET under the influence of soil salinity. Specifically, we integrated physical constraints from process models (Penman‐Monteith or Shuttleworth‐Wallace) and salinity‐induced stomatal stress mechanisms into the DL algorithm, and evaluated its performance by comparing four diverse scenarios. Results demonstrate that hybrid DL framework offers a promising alternative for ET estimation, achieving comparable accuracy to pure DL during training and validation. Nonetheless, due to the limited available measurements, data‐driven model may not adequately capture plant responses to salt stress, leading to significant prediction biases observed during independent testing. Encouragingly, the hybrid DL model (DL‐SS) integrating Shuttleworth‐Wallace and salinity‐induced stomatal stress mechanisms demonstrated enhanced interpretability, generalizability, and extrapolation capabilities. During testing, DL‐SS consistently showed optimal performance, yielding root mean square error (RMSE) values of 37.4 W m⁻² for sunflower and 39.2 W m⁻² for maize. Compared to traditional Jarvis‐type approaches (JPM and JSW) and pure DL model during testing, DL‐SS achieved substantial reductions in RMSE values: 51%, 33%, and 43% for sunflower, and 45%, 31%, and 35% for maize, respectively. These findings highlight the importance of integrating prior scientific knowledge into data‐driven models to enhance extrapolation capability of ET modeling, especially in salinized regions where conventional models may struggle.