Ptak et al. (2026) A novel approach for forecasting lake heatwaves using MSAPE-ML: A deep learning framework with WRF-derived physical features coupled with the air2water model

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-10-12
• Authors: Mariusz Ptak, Teerachai Amnuaylojaroen
• DOI: 10.1016/j.jhydrol.2026.136603

Research Groups

• Department of Hydrology and Water Management, Adam Mickiewicz University
• School of Energy and Environment, University of Phayao
• Atmospheric Pollution and Climate Research Unit, School of Energy and Environment, University of Phayao

Short Summary

This study introduces MSAPE-ML, a deep learning framework that uses WRF-derived physical features coupled with the air2water model to forecast lake heatwaves. The framework projects heatwave characteristics for 37 Polish lakes under three SSPs, achieving water temperature RMSE of 1.89 ± 0.15 ◦C (R2 = 0.94).

Objective

• Investigate the feasibility of using a deep learning framework with WRF-derived physical features to forecast lake heatwaves.

Study Configuration

• Spatial Scale: Regional scale, focusing on 37 Polish lakes.
• Temporal Scale: Long-term projections for end-of-century under three SSPs (SSP1-2.6, SSP2-4.5, and SSP5-8.5).

Methodology and Data

• Models used:
  • MSAPE-ML: A deep learning framework combining LSTM-Attention model with WRF-derived physical features.
  • air2water ODE lake model.
  • Bayesian model averaging of CMIP6 ensembles.
• Data sources: CMIP6 reanalysis data, Polish lake surface temperature observations.

Main Results

• The framework projects a significant increase in mean annual heatwave days for all lakes under SSP5-8.5 by end-of-century (334).
• Under SSP5-8.5, discrete events coalesce until 97% of lakes exceed 350 days yr⁻1 above the fixed 1985–2014 90th-percentile envelope by 2100.
• Severity shifts from predominantly Moderate (87%) to Strong-or-above (80%), and occurrence expands into all seasons.

Contributions

• This study establishes MSAPE-ML as a transferable framework for lake heatwave assessment based on physics-derived features.
• The projections provide an evidence base for freshwater-specific climate adaptation, including revised thermal discharge regulations and prioritised monitoring of shallow lakes most vulnerable to regime shifts.

Funding

• Not specified in the paper.

Citation

@article{Ptak2026novel,
  author = {Ptak, Mariusz and Amnuaylojaroen, Teerachai},
  title = {A novel approach for forecasting lake heatwaves using MSAPE-ML: A deep learning framework with WRF-derived physical features coupled with the air2water model},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136603},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136603}
}