Navarro et al. (2025) Seamless seasonal to multi-annual predictions of temperature and Standardized Precipitation Index by constraining transient climate model simulations

Identification

• Journal: Earth System Dynamics
• Year: 2025
• Date: 2025-10-15
• Authors: Juan C. Acosta Navarro, Alvise Aranyossy, Paolo De Luca, Markus G. Donat, Arthur Hrast Essenfelder, Rashed Mahmood, Andrea Toreti, Danila Volpi
• DOI: 10.5194/esd-16-1723-2025

Research Groups

• European Commission, Joint Research Centre, Ispra, Italy
• Barcelona Supercomputing Center (BSC), Barcelona, Spain
• Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain
• National Center for Climate Research (NCKF), Danish Meteorological Institute, Copenhagen, Denmark

Short Summary

This study develops a computationally inexpensive analog-based method to generate seamless seasonal to multi-annual predictions of surface air temperature (TAS) and Standardized Precipitation Index (SPI) by constraining CMIP6 simulations, demonstrating competitive skill compared to state-of-the-art initialized systems and offering continuous monthly initializations.

Objective

• To demonstrate that a computationally inexpensive climate model analog method can fill the gap in continuous operational seasonal to multi-annual climate predictions by constraining variability in non-initialized CMIP6 simulations, providing skillful forecasts of surface air temperature and Standardized Precipitation Index.

Study Configuration

• Spatial Scale: Global coverage, with data bilinearly interpolated to a common grid of 5° × 5° for TAS and approximately 2.8° × 2.8° for sea surface temperature (SST) analog search and SPI.
• Temporal Scale: Forecast ranges from 3 months (seasonal) to 1, 2, and 4 years (multi-annual). Evaluation period is 1962–2018, with 3-month predictions evaluated from 1982–2018. CMIP6 data used spans 1960–2030.

Methodology and Data

• Models used:
  • Analog method (developed in this study, based on CMIP6)
  • Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-model ensemble (19 climate models, 149 simulations)
  • European Centre for Medium-Range Weather Forecasting SEAS51 (operational seasonal forecast system, benchmark for 3-month predictions)
  • EC-Earth3 (initialized climate model, benchmark for 12-, 24-, and 48-month predictions)
• Data sources:
  • NOAA Extended Reconstructed Sea Surface Temperature (ERSSTv5) for SST analog search.
  • Berkeley Earth Surface Temperatures (BEST) for surface air temperature (TAS) evaluation.
  • Global Precipitation Climatology Center (GPCC) for Standardized Precipitation Index (SPI) evaluation.
  • CMIP6 historical emissions (before 2015) and Shared Socioeconomic Pathway 2-4.5 (SSP2-4.5) scenario emissions (after 2015) for model forcing.

Main Results

• The analog-based method provides skillful forecasts for TAS and SPI across seasonal to multi-annual timescales, showing similar spatial skill patterns to initialized numerical predictions.
• Seasonal (3-month) TAS forecasts exhibit positive statistically significant correlation in the tropics, subtropics, most oceans, and the Arctic, offering substantial added value over the uninitialized CMIP6 ensemble.
• Seasonal (3-month) SPI forecasts are less skillful than TAS but still outperform the uninitialized CMIP6 ensemble, with skill primarily driven by internal climate variability alignment.
• Annual (1-year) TAS predictions show high skill (anomaly correlation coefficient > 0.8, mean absolute error skill score > 0.3) across most tropical areas, the North Atlantic, and many extratropical land regions, generally outperforming CMIP6.
• Biennial (2-year) TAS predictions maintain high skill, with residual skill (from initialization) evident in tropical South America, South Asia, Australia, and sub-Saharan Africa.
• Quadrennial (4-year) TAS predictions show higher overall skill metrics but reduced benefit from initialization, indicating a greater dominance of external forcing.
• The analog-based system demonstrates competitive skill compared to operational systems like EC-Earth3 for annual and biennial forecasts, and comparable skill for quadrennial forecasts when initialized monthly.
• A key advantage is the ability to provide continuous monthly initializations of forecasts across various timescales at low computational cost, which is not available from traditional seasonal or decadal prediction systems.

Contributions

• Develops and validates a computationally inexpensive analog-based method for generating seamless seasonal to multi-annual climate predictions of TAS and SPI, addressing a critical gap in continuous operational forecasting.
• Demonstrates that this method achieves competitive predictive skill compared to state-of-the-art initialized dynamical prediction systems (ECMWF SEAS51, EC-Earth3), particularly for annual and biennial forecast ranges.
• Enables the provision of continuous monthly initializations for forecasts across various timescales, offering more timely and comprehensive climate information throughout the year than traditional systems.
• Leverages large ensembles of non-initialized CMIP6 simulations, providing a valuable and viable complement to existing operational prediction systems by avoiding initialization shocks and drift.

Funding

• EU HORIZON EUROPE Climate, Energy and Mobility (grant no. 101081460)

Citation

@article{Navarro2025Seamless,
  author = {Navarro, Juan C. Acosta and Aranyossy, Alvise and Luca, Paolo De and Donat, Markus G. and Essenfelder, Arthur Hrast and Mahmood, Rashed and Toreti, Andrea and Volpi, Danila},
  title = {Seamless seasonal to multi-annual predictions of temperature and Standardized Precipitation Index by constraining transient climate model simulations},
  journal = {Earth System Dynamics},
  year = {2025},
  doi = {10.5194/esd-16-1723-2025},
  url = {https://doi.org/10.5194/esd-16-1723-2025}
}