Yu et al. (2026) Exploring xLSTM for Large-Sample Hourly Rainfall–Runoff Modelling

Identification

• Journal: Environmental Modelling & Software
• Year: 2026
• Date: 2026-10-01
• Authors: Qiutong Yu, Bryan Tolson, Jimmy Lin
• DOI: 10.1016/j.envsoft.2026.107193

Research Groups

• Department of Civil and Environmental Engineering, University of Waterloo, Canada
• David R. Cheriton School of Computer Science, University of Waterloo, Canada

Short Summary

This study explores the use of extended long short-term memory (xLSTM) for large-sample hourly rainfall–runoff modelling across 430 Canadian basins. The results show that xLSTM achieves comparable overall accuracy to standard LSTM-based benchmarks and improved performance in spatial extrapolation.

Objective

• To evaluate the effectiveness of xLSTM for large-sample hourly rainfall–runoff modelling
• To compare the performance of xLSTM with standard LSTM-based benchmarks

Study Configuration

• Spatial Scale: 430 Canadian basins, covering a broad range of hydroclimatic and physiographic conditions
• Temporal Scale: Hourly time step, with a one-year look-back window for input sequences

Methodology and Data

• Models used: xLSTM, sMTS-LSTM, MF-LSTM (benchmarks)
• Data sources: Canadian Surface Reanalysis version 3.1 (CaSR v3.1), Water Survey of Canada (WSC) hydrometric server, North American Land Change Monitoring System (NALCMS), Global Soil Dataset for Earth System Models (GSDE)

Main Results

• xLSTM achieved comparable overall accuracy to standard LSTM-based benchmarks in temporal extrapolation
• xLSTM improved performance in spatial extrapolation compared to standard LSTM-based benchmarks
• Kling–Gupta efficiency (KGE) and Nash–Sutcliffe efficiency (NSE) were evaluated across seven Canadian hydroclimatic regions

Contributions

• This study provides preliminary evidence on the potential of replacing the LSTM backbone with an xLSTM backbone in established hydrological models for future modelling applications
• The results suggest that xLSTM can be a viable alternative to standard LSTM-based benchmarks for large-sample hourly rainfall–runoff modelling

Funding

• Not specified

Citation

@article{Yu2026Exploring,
  author = {Yu, Qiutong and Tolson, Bryan and Lin, Jimmy},
  title = {Exploring xLSTM for Large-Sample Hourly Rainfall–Runoff Modelling},
  journal = {Environmental Modelling & Software},
  year = {2026},
  doi = {10.1016/j.envsoft.2026.107193},
  url = {https://doi.org/10.1016/j.envsoft.2026.107193}
}