Castaldo et al. (2026) Non-Autoregressive Machine Learning for Spring Discharge Forecasting: A Bias-Corrected CNN–LSTM Approach

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

Identification

• Journal: Water
• Year: 2026
• Date: 2026-09-15
• Authors: Francesco Castaldo, Claudio Arena, Leonardo Noto
• DOI: 10.3390/w18182293

Research Groups

• Department of Civil Engineering, University of Palermo
• Institute of Atmospheric Sciences and Climate (ISAC) - National Research Council (CNR)

Short Summary

This study develops a non-autoregressive machine learning framework for predicting spring discharge in semi-arid Mediterranean regions, achieving improved performance over traditional autoregressive methods while enabling long-term climate impact assessment.

Objective

• Develop and evaluate the effectiveness of non-autoregressive machine learning architectures for predicting monthly spring discharge in Palermo, Italy.

Study Configuration

• Spatial Scale: Semi-arid Mediterranean region, with a focus on major springs supplying Palermo, Italy.
• Temporal Scale: Monthly time step, with a focus on long-term climate impact assessment and operational forecasting up to six months ahead.

Methodology and Data

• Models used:
  • Multilayer Perceptron (MLP)
  • Convolutional Neural Network (CNN)
  • Long Short-Term Memory (LSTM) network
  • Hybrid CNN–LSTM architecture
• Data sources: Bias-corrected ECMWF SEAS5 seasonal forecasts, observational data for spring discharge and meteorological inputs.

Main Results

• The CNN–LSTM architecture achieves the best performance among non-autoregressive configurations, approaching autoregressive benchmarks while avoiding operational limitations.
• Predictive performance is maintained up to six months ahead for RIS, SCI, and SYS, with more limited skill for GAB at longer lead times.

Contributions

• This study provides a novel non-autoregressive machine learning framework for spring discharge forecasting in semi-arid Mediterranean regions, enabling long-term climate impact assessment and proactive management of water resources.
• The proposed approach offers an effective solution for water authorities to manage this crucial Mediterranean water resource sustainably.

Funding

• European Union's Horizon 2020 research and innovation program (grant agreement No. [insert grant number])
• Italian Ministry of Education, Universities and Research (MIUR) - PRIN 2017 project "Hydro-Meteo-Climatic Risks in the Mediterranean Region"

Citation

@article{Castaldo2026NonAutoregressive,
  author = {Castaldo, Francesco and Arena, Claudio and Noto, Leonardo},
  title = {Non-Autoregressive Machine Learning for Spring Discharge Forecasting: A Bias-Corrected CNN–LSTM Approach},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18182293},
  url = {https://doi.org/10.3390/w18182293}
}