Düzenli et al. (2026) Assessing the utility of statistical downscaling for subseasonal temperature forecasts

Identification

• Journal: Scientific Reports
• Year: 2026
• Date: 2026-03-26
• Authors: Eren Düzenli, Jaume Ramón, Verónica Torralba, Sam Pickard, Ángel G. Muñoz, Dragana Bojovic
• DOI: 10.1038/s41598-026-45067-2

Research Groups

Earth Sciences Department, Barcelona Supercomputing Center (BSC), Barcelona, Spain

Short Summary

This study benchmarks 27 statistical downscaling methods for subseasonal temperature forecasts, demonstrating that while most methods successfully transfer skill from coarse (~100 km) to local (~5 km) resolution, method choice is critical, with some enhancing and others degrading skill, and incorporating atmospheric patterns or using weekly predictors showing benefits.

Objective

• To assess the utility of statistical downscaling for subseasonal temperature forecasts by benchmarking 27 statistical methods and evaluating how they transfer forecast skill from coarse to local resolution at the weekly scale. The study also investigates the value of incorporating atmospheric patterns and the role of temporal resolution (daily vs. weekly data).

Study Configuration

• Spatial Scale: Retrospective forecasts from approximately 100 km resolution downscaled to approximately 5 km resolution.
• Temporal Scale: Weekly scale, with lead times of one to four weeks before each target week (subseasonal scale).

Methodology and Data

• Models used: Climate Forecast System version 2 (CFSv2) for retrospective forecasts. Statistical downscaling methods included configurations of bias correction, linear regression, logistic regression, and analogs.
• Data sources: CFSv2 sub-seasonal forecasts (NCEP NOMADS server), CERRA, E-OBS, and ERA5 data (Copernicus Climate Data Store).

Main Results

• Most statistical downscaling methods successfully transfer CFSv2 skill from coarse to higher resolution.
• Method choice is critical, as some downscaling methods can degrade forecast skill while others enhance it.
• Methods incorporating atmospheric patterns show promise at longer lead times, particularly when the relevant climatological pattern, a key driver of heat in the target week, is well predicted.
• At the subseasonal scale, downscaling performed with weekly predictors generally outperforms that based on daily predictors.
• Skill was assessed using the Brier Skill Score at the 10th and 90th percentiles for temperature extremes.

Contributions

• Provides a comprehensive benchmark of 27 statistical downscaling methods for subseasonal temperature forecasts, offering insights into their performance and suitability.
• Quantifies the impact of incorporating atmospheric patterns on downscaling skill, particularly at longer lead times.
• Demonstrates the importance of temporal resolution in predictor data, showing that weekly predictors are more effective than daily ones for subseasonal downscaling.
• Offers guidance for selecting appropriate statistical downscaling techniques to improve local-scale subseasonal temperature predictions, especially for extreme events.

Funding

• Horizon Europe project Impetus4Change (I4C, grant id. 101081555)
• European Union funded project ASPECT (grant id. 101081460)
• National project BOREAS (PID2022-140673OA-I00) funded by MICIU/AEI/10.13039/501100011033 and by ERDF, EU
• EU Horizon 2020 Marie Skłodowska-Curie grant 101152499 (SINFONIA)
• Grant RYC2021-034691-I, funded by MCIN/AEI/10.13039/501100011033 and the European Union NextGenerationEU/PRTR

Citation

@article{Düzenli2026Assessing,
  author = {Düzenli, Eren and Ramón, Jaume and Torralba, Verónica and Pickard, Sam and Muñoz, Ángel G. and Bojovic, Dragana},
  title = {Assessing the utility of statistical downscaling for subseasonal temperature forecasts},
  journal = {Scientific Reports},
  year = {2026},
  doi = {10.1038/s41598-026-45067-2},
  url = {https://doi.org/10.1038/s41598-026-45067-2}
}