Wang et al. (2025) Estimation of seasonal ecological water demand in arid zone of Northwest China: An approach using the LSTM-random forest regression model

Identification

• Journal: Journal of Environmental Management
• Year: 2025
• Date: 2025-12-06
• Authors: Chao Wang, Qi Zhang, Min Tao, Hong Hu, Chenyang Xue, Fan Xue, Zengchuan Dong
• DOI: 10.1016/j.jenvman.2025.128240

Research Groups

• State Key Laboratory of Water Disaster Prevention, Yangtze Institute for Conservation and Development, Hohai University, Nanjing, China
• College of Hydrology and Water Resources, Hohai University, Nanjing, China
• College of Water Conservancy and Hydropower Engineering, Hohai University, Nanjing, China

Short Summary

This study developed a coupled LSTM-Random Forest Regression model with a probability modification coefficient to dynamically assess seasonal ecological water demand in arid zones, overcoming the limitations of deterministic models by characterizing uncertainties. Applied to the Shiyang River Basin, the model accurately predicted fractional vegetation cover and revealed significant seasonal variations in ecological water demand.

Objective

• To construct a coupled LSTM-Random Forest Regression model, incorporating a probability modification coefficient, to dynamically assess seasonal ecological water demand in arid zones, addressing the uncertainties inherent in conventional deterministic models.

Study Configuration

• Spatial Scale: Arid zone of Northwest China, specifically the Shiyang River Basin (SRB).
• Temporal Scale: Seasonal (e.g., May–September, March–April), with predictions for 2022 and implications for future water management.

Methodology and Data

• Models used:
  • Coupled Long Short-Term Memory (LSTM) and Random Forest Regression model (LSTM-Random Forest Regression model)
  • Bayesian optimization (for model construction)
  • Probability Modification Coefficient (PMC) (for interval-range prediction)
  • Sen-Mann-Kendall (Sen-MK) trend test (for spatiotemporal evolution pattern analysis)
• Data sources:
  • Fractional vegetation cover (FVC) data (implied as a key input/output for the model)
  • Ecological water demand data (estimated by the model)

Main Results

• The coupled LSTM-Random Forest Regression model achieved high precision in predicting fractional vegetation cover (FVC), with a test set R² of 0.951 and a Mean Squared Error (MSE) of 0.597.
• The estimation of ecological water demand for the lower Shiyang River Basin (SRB) in 2022 showed a significant seasonal divergence pattern.
• Summer months (May–September) required a substantial increase in upstream water supply to the lower basin.
• Late spring (March–April) was characterized by low water consumption and high ecological benefits.
• Priority for future water management in the lower SRB should be given to early spring water replenishment and summer buffer water supply.

Contributions

• Developed a novel coupled LSTM-Random Forest Regression model integrated with Bayesian optimization and a Probability Modification Coefficient (PMC) to provide interval-range predictions, thereby overcoming the limitations of single-value predictions from conventional deterministic models.
• Significantly mitigated the issue of future predicted values deviating from actual values due to uncertain factors like climate change and anthropogenic influences.
• Provided a technical framework capable of resolving uncertainties in the dynamic assessment of ecological water demand in arid zones.
• Offered a scientific basis for improved water resources management strategies in arid regions by analyzing the spatiotemporal evolution pattern of FVC and its indicative significance for ecological water demand.

Funding

Not specified in the provided text.

Citation

@article{Wang2025Estimation,
  author = {Wang, Chao and Zhang, Qi and Tao, Min and Hu, Hong and Xue, Chenyang and Xue, Fan and Dong, Zengchuan},
  title = {Estimation of seasonal ecological water demand in arid zone of Northwest China: An approach using the LSTM-random forest regression model},
  journal = {Journal of Environmental Management},
  year = {2025},
  doi = {10.1016/j.jenvman.2025.128240},
  url = {https://doi.org/10.1016/j.jenvman.2025.128240}
}