Xu et al. (2025) Hybrid ITSP-LSTM Approach for Stochastic Citrus Water Allocation Addressing Trade-Offs Between Hydrological-Economic Factors and Spatial Heterogeneity

Identification

• Journal: Water
• Year: 2025
• Date: 2025-09-09
• Authors: Wen Xu, Rui Hu, Yifei Zheng, Ying Yu, Yanpeng Cai, Shijiang Zhu
• DOI: 10.3390/w17182665

Research Groups

• Hubei Key Laboratory of Hydropower Engineering Construction and Management, China Three Gorges University
• College of Hydraulic & Environmental Engineering, China Three Gorges University
• Key Laboratory for City Cluster Environmental Safety and Green Development of the Ministry of Education, Institute of Environmental and Ecological Engineering, Guangdong University of Technology
• Guangdong Provincial Key Laboratory of Water Quality Improvement and Ecological Restoration for Watersheds, Institute of Environmental and Ecological Engineering, Guangdong University of Technology

Short Summary

This study developed a hybrid Interval Two-Stage Stochastic Programming (ITSP) and Long Short-Term Memory (LSTM) model to optimize stochastic water allocation for fragmented citrus cultivation, demonstrating an 8.67% increase in system-wide benefits by balancing hydrological-economic factors and spatial heterogeneity.

Objective

• To develop and apply a hybrid Interval Two-Stage Stochastic Programming (ITSP) and Long Short-Term Memory (LSTM) approach for optimizing stochastic citrus water allocation, considering hydrological-economic trade-offs and spatial heterogeneity, to enhance sustainable water management in fragmented agricultural systems.

Study Configuration

• Spatial Scale: Anfusi Town, Zhijiang City, Hubei Province, China (middle-lower Yangtze River region). The study area was divided into three citrus-growing zones (A, B, C) and corresponding water supply zones (a, b, c). Village-level allocations were also considered.
• Temporal Scale: Historical meteorological data from 1971 to 2023 (52 years); historical surface water supply data for the past 15 years; current year scenario (2023); planning year (2025); monthly time step for LSTM predictions; seasonal citrus growth stages.

Methodology and Data

• Models used:
  • Hybrid Interval Two-Stage Stochastic Programming (ITSP) and Long Short-Term Memory (LSTM) approach.
  • Long Short-Term Memory (LSTM) neural networks for forecasting irrigation water demand and supply dynamics under climate variability.
  • Interval Two-Stage Stochastic Programming (ITSP) for optimizing dynamic allocation strategies, quantifying uncertainties through interval analysis, and balancing economic returns with hydrological risks.
  • Penman–Monteith model for calculating daily potential evapotranspiration (ET₀).
  • United States Department of Agriculture (USDA) method for calculating effective precipitation.
• Data sources:
  • Meteorological data from Zhijiang City Meteorological Station (1971–2023): daily precipitation, relative humidity, maximum and minimum temperatures, wind speed, sunlight, and evaporation.
  • Historical surface water supply data for citrus irrigation (past 15 years).
  • Field survey data for citrus acquisition prices (2023).
  • Irrigation quotas for various crops, planting areas, and annual water supply data from reservoirs and weirs in Anfusi Town.
  • Citrus planting area data.
  • Standard of citrus irrigation water quota and temperature indices for citrus growth stages in the study area.

Main Results

• The proposed ITSP-LSTM model achieved an 8.67% increase in system-wide benefits compared to baseline practices in the current year scenario.
• For the planning year (2025), the LSTM model predicted a net irrigation water demand of 330.53 mm (3300.00 m³/hm²).
• Optimal water distribution thresholds were identified for 2025: an upper limit of 3.85 × 10⁶ m³ for high-availability citrus planting zone A and lower limits of 1.62 × 10⁶ m³ for moderate-to-low-availability citrus planting zones B and C.
• The model minimizes water scarcity penalties while maximizing net benefits, prioritizing local over external water sources (calibrated at a 2:1 ratio) to reduce costs.
• Under 2023 conditions, citrus planting zone A achieved full irrigation satisfaction, while zones B and C faced significant deficits of 5.50 × 10⁶ m³ and 5.49 × 10⁶ m³, respectively.
• For 2025, deficits persisted in citrus planting zones B and C, ranging from [2.25, 5.87] × 10⁶ m³ and [4.23, 5.62] × 10⁶ m³, respectively, without inter-zonal redistribution.
• The stochastic framework dynamically adjusts allocations based on low, medium, and high inflow scenarios, effectively reducing water deficits (e.g., for citrus planting zone A, deficits from water supply zone B were reduced from [1.35, 1.62] × 10⁶ m³ under low inflow to [0, 0.41] × 10⁶ m³ under medium inflow, and eliminated under high inflow).

Contributions

• Innovates by integrating stochastic-economic analysis with spatial prioritization of high-marginal-benefit zones and uncertainty robustness through interval analysis and two-stage decision making.
• Bridges a critical research gap in citrus irrigation optimization, particularly for perennial tree crops in fragmented topography with complex management systems.
• Offers a scalable and adaptive solution for regions facing fragmented landscapes and climate-driven water scarcity.
• Advances sustainable water management practices in complex agricultural systems by providing a robust framework for balancing economic efficiency and operational reliability compared to traditional deterministic models.

Funding

• Youth Fund from the National Natural Science Foundation of China (52000120)
• Key Scientific Research Projects of Water Conservancy in Hubei Province (HBSLKY201919 and HBSLKY202124)
• 111 project of Hubei Province

Citation

@article{Xu2025Hybrid,
  author = {Xu, Wen and Hu, Rui and Zheng, Yifei and Yu, Ying and Cai, Yanpeng and Zhu, Shijiang},
  title = {Hybrid ITSP-LSTM Approach for Stochastic Citrus Water Allocation Addressing Trade-Offs Between Hydrological-Economic Factors and Spatial Heterogeneity},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17182665},
  url = {https://doi.org/10.3390/w17182665}
}