Todaro et al. (2026) Skill of CMIP6 decadal climate predictions at the subregional scale

Identification

• Journal: Stochastic Environmental Research and Risk Assessment
• Year: 2026
• Date: 2026-04-01
• Authors: Valeria Todaro, Marco D’Oria, Maria Giovanna Tanda
• DOI: 10.1007/s00477-026-03218-x

Research Groups

• Department of Engineering and Architecture, University of Parma, Parma, Italy

Short Summary

This study assesses the skill of a CMIP6 decadal climate prediction model (HadGEM3-GC31-MM) in simulating subregional climate conditions for precipitation and temperature in the Emilia-Romagna region, Italy. It finds that while drift correction improves performance, particularly for temperature, substantial uncertainties and challenges remain, especially for precipitation in areas with complex topography.

Objective

• To assess the skill of a CMIP6 decadal climate model from the Decadal Climate Prediction Project (DCPP) in simulating subregional climate conditions (precipitation, minimum, and maximum temperatures) in the Emilia-Romagna region of northern Italy.
• To determine if these predictions are sufficient to support regional-scale climate analyses relevant to water-resource-dependent sectors and local decision-making.

Study Configuration

• Spatial Scale: Emilia-Romagna region, northern Italy (approximately 22,500 km²). Analysis conducted at a spatial resolution of 0.05° (approximately 5 km). Native GCM resolution is approximately 60 km.
• Temporal Scale:
  • Observational data period: 1961–2019.
  • Hindcast simulations: Initialized annually from 1960 to 2018, each extending for approximately 10 years (120 months).
  • Analysis period for skill evaluation: January 2000 – December 2010.
  • Lead times: 1 to 120 months.

Methodology and Data

• Models used: Met Office Hadley Centre (MOHC) HadGEM3-GC31-MM model, a contributor to the CMIP6-DCPP initiative.
• Data sources:
  • Model data: Hindcast simulations from HadGEM3-GC31-MM (50 hindcast start years, 10 ensemble members each).
  • Observational data: ERG5 Eraclito dataset for Emilia-Romagna, Italy, providing daily gridded climate information at approximately 5 km resolution (1961–present).
  • Variables analyzed: Monthly precipitation, monthly mean minimum temperature, and monthly mean maximum temperature.
  • Preprocessing: Second-order conservative interpolation to regrid GCM data to 0.05° resolution, followed by inverse distance weighted remapping onto the observational grid.
  • Correction: Lead-time dependent mean drift correction.
  • Skill evaluation metrics: Root Mean Square Error (RMSE), Mean Squared Skill Score (MSSS), and Anomaly Correlation Coefficient (ACC).

Main Results

• Model Drift: Significant spatial and seasonal variability in model errors was observed for all variables.
  • Precipitation: Predominantly negative drift (underestimation) in the Apennine region and positive drift (overestimation) in flatter areas. No clear trend over lead-times, but a strong seasonal modulation.
  • Minimum and Maximum Temperatures: Exhibit spatial variability in drift (overestimation in Apennine areas, underestimation elsewhere) with a weak downward tendency over lead-times (maximum difference of approximately 0.7 °C) and a dominant seasonal bias.
• Skill Evaluation (after drift correction):
  • Precipitation: Drift correction improved skill, but RMSE remained high, particularly in the Apennine area (e.g., peak RMSE of approximately 160 mm uncorrected). MSSS and ACC showed significant improvement, with highest skill in the Apennine region due to high observed variability and a clear seasonal signal. The model showed limited ability to reproduce extreme precipitation events.
  • Minimum Temperature: RMSE decreased substantially across the region (94% of the area below 2 °C after correction, from maximum uncorrected values of approximately 4.5 °C). MSSS improved everywhere (exceeding 0.9 over approximately 89% of the region). ACC values consistently exceeded 0.9 after correction (from above 0.7 uncorrected).
  • Maximum Temperature: RMSE decreased across the region (87% of the area below 2 °C after correction, from maximum uncorrected values of approximately 7 °C). MSSS improved everywhere (exceeding 0.9 in 99% of the region). ACC values consistently exceeded 0.9 after correction (from above 0.7 uncorrected).
• Uncertainty: Substantial uncertainty was found for all variables, being particularly high for precipitation. While the ensemble spread generally encompassed observed monthly values, the model showed limited ability to reproduce extreme monthly conditions for both precipitation and temperature.

Contributions

• Provides a high-resolution (0.05°) assessment of CMIP6-DCPP model skill for decadal climate predictions of precipitation and temperature at a subregional scale in a complex topographical area (Emilia-Romagna, Italy).
• Highlights the significant limitations of global climate models, even after statistical downscaling and drift correction, in accurately representing local-scale climate dynamics, particularly for precipitation in mountainous regions.
• Emphasizes the need for caution when using GCM outputs for local impact assessments and adaptation planning, especially for precipitation-dependent sectors.
• Demonstrates that while mean drift correction is effective for improving temperature predictions at monthly scales, substantial challenges persist for precipitation.

Funding

• University of Parma, Bando di Ateneo 2023 per la ricerca
• PNRR MUR project ECS00000033ECOSISTER
• OurMED PRIMA Program project, funded by the European Union’s Horizon 2020 research and innovation, grant agreement No. 2222

Citation

@article{Todaro2026Skill,
  author = {Todaro, Valeria and D’Oria, Marco and Tanda, Maria Giovanna},
  title = {Skill of CMIP6 decadal climate predictions at the subregional scale},
  journal = {Stochastic Environmental Research and Risk Assessment},
  year = {2026},
  doi = {10.1007/s00477-026-03218-x},
  url = {https://doi.org/10.1007/s00477-026-03218-x}
}