Tedesco et al. (2026) The Modèle Atmosphérique Régional – Intelligence Artificielle (MAR-IA): surface meltwater over Greenland

Identification

• Journal: ˜The œcryosphere
• Year: 2026
• Date: 2026-07-30
• Authors: Marco Tedesco, Racheet Matai, Xavier Fettweis
• DOI: 10.5194/tc-20-4235-2026

Research Groups

• Lamont-Doherty Earth Observatory, Columbia Climate School, Columbia University, New York, USA
• University of Liege, Liege, Belgium

Short Summary

This study introduces a machine-learning emulator, MAR-IA, trained on daily surface meltwater production over Greenland and enables attribution of melt drivers. The emulator achieves high test-set skill with R2 = 0.987, low root mean squared error (< 10 mm w.e. d−1), and negligible bias relative to MAR meltwater outputs.

Objective

• Develop a machine-learning emulator that can accurately predict daily surface meltwater production over Greenland and attribute the dominant melt drivers in space and time.

Study Configuration

• Spatial Scale: Regional scale, focusing on the Greenland Ice Sheet.
• Temporal Scale: Daily temporal resolution for surface meltwater production predictions.

Methodology and Data

• Models used: eXtreme Gradient Boosting (XGBoost) algorithm with gradient-boosted trees optimized via Bayesian hyperparameter search.
• Data sources: MAR regional climate model outputs, ERA5 reanalysis dataset, and predictor variables such as surface energy balance components and near-surface meteorological variables.

Main Results

• The MAR-IA emulator achieves high test-set skill (R2 = 0.987) with low root mean squared error (< 10 mm w.e. d−1) and negligible bias relative to MAR meltwater outputs.
• SHAP-based model interpretation reveals that 2m air temperature, downward shortwave radiation, albedo, and sensible heat flux are the most important predictors of surface meltwater production.

Contributions

• This study demonstrates the potential of machine-learning emulators in complementing traditional modeling frameworks for predicting daily surface meltwater production over Greenland.
• The MAR-IA emulator enables computationally efficient ensemble simulations and supports model-interpretation analyses, facilitating attribution of melt drivers under past and future climate conditions.

Funding

• This research was supported by [insert funding projects, programs, and reference codes].

Citation

@article{Tedesco2026Modèle,
  author = {Tedesco, Marco and Matai, Racheet and Fettweis, Xavier},
  title = {The Modèle Atmosphérique Régional – Intelligence Artificielle (MAR-IA): surface meltwater over Greenland},
  journal = {˜The œcryosphere},
  year = {2026},
  doi = {10.5194/tc-20-4235-2026},
  url = {https://doi.org/10.5194/tc-20-4235-2026}
}