Zhong et al. (2026) Integrating seasonal autoregressive integrated moving average and long short-term memory for multi-scale standardized precipitation evapotranspiration index forecasting

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Journal of Hydroinformatics
• Year: 2026
• Date: 2026-10-17
• Authors: Wanying Zhong, Yaping Zhang, Xu Chen
• DOI: 10.2166/hydro.2026.017

Research Groups

Not specified in the abstract.

Short Summary

This study proposes a hybrid SARIMA-LSTM model with spatial attention and dynamic residual weighting to improve multi-scale Standardized Precipitation Evapotranspiration Index (SPEI) forecasting accuracy over the Tibetan Plateau. The model effectively integrates linear and nonlinear components, outperforming benchmark models across various time scales and enhancing Nash–Sutcliffe Efficiency (NSE) from 0.88 to 0.91.

Objective

• To develop a hybrid forecasting framework integrating SARIMA and LSTM to improve the accuracy of multi-scale Standardized Precipitation Evapotranspiration Index (SPEI) forecasting, particularly in regions with complex terrain, by addressing multi-temporal-scale variability, coupled linear–nonlinear dynamics, and insufficient exploitation of spatial correlation.

Study Configuration

• Spatial Scale: Qinghai–Tibet Plateau, using grid-based data.
• Temporal Scale: 1981–2022, evaluating SPEI at 1-month, 3-month, 6-month, and 12-month time scales (SPEI-1, SPEI-3, SPEI-6, SPEI-12).

Methodology and Data

• Models used: Seasonal Autoregressive Integrated Moving Average (SARIMA), Long Short-Term Memory (LSTM), and a proposed hybrid SARIMA-LSTM model incorporating a lightweight spatial attention module and a dynamic residual-weighting mechanism.
• Data sources: Grid-based multi-scale SPEI data collected over the Qinghai–Tibet Plateau.

Main Results

• The proposed SARIMA-LSTM hybrid framework consistently outperforms benchmark models across all considered time scales (SPEI-1, SPEI-3, SPEI-6, SPEI-12).
• The integration of the spatial attention module and dynamic residual-weighting mechanism improved the Nash–Sutcliffe Efficiency (NSE) from 0.88 to 0.91.
• The framework provides an effective approach for multi-scale drought forecasting in regions with complex terrain.

Contributions

• Introduction of a novel hybrid SARIMA-LSTM forecasting framework that adaptively integrates linear and nonlinear prediction components through a dynamic residual-weighting mechanism.
• Incorporation of a lightweight spatial attention module to enhance the representation and utilization of spatially correlated features in drought forecasting.
• Demonstrated improved multi-scale SPEI forecasting accuracy in regions with complex terrain, specifically the Qinghai–Tibet Plateau, addressing limitations of existing statistical or deep learning models alone.

Funding

Not specified in the abstract.

Citation

@article{Zhong2026Integrating,
  author = {Zhong, Wanying and Zhang, Yaping and Chen, Xu},
  title = {Integrating seasonal autoregressive integrated moving average and long short-term memory for multi-scale standardized precipitation evapotranspiration index forecasting},
  journal = {Journal of Hydroinformatics},
  year = {2026},
  doi = {10.2166/hydro.2026.017},
  url = {https://doi.org/10.2166/hydro.2026.017}
}