Lagos-Castro et al. (2026) Towards a balancing performance, uncertainty coverage, and spatial consistency in climate model sub-selection

Identification

• Journal: Climatic Change
• Year: 2026
• Date: 2026-03-27
• Authors: Ivan Lagos-Castro, Héctor Macian-Sorribes, Manuel Pulido-Velazquez, María Pedro-Monzonís
• DOI: 10.1007/s10584-026-04161-0

Research Groups

• Research Institute of Water and Environmental Engineering (IIAMA), Universitat Politècnica de València, Valencia, Spain
• Global Omnium, Valencia, Spain

Short Summary

This study introduces a standardized multi-criteria framework to evaluate climate model sub-selection methods, integrating performance, uncertainty coverage, and spatial consistency across basins with contrasting hydrological regimes. It finds that methods based on future change diversity and adaptive consensus best balance accuracy and uncertainty representation, consistently outperforming mono-model strategies, and highlights the critical role of spatial variability in method evaluation.

Objective

• To develop and apply a standardized multi-criteria framework to cross-evaluate twelve climate model sub-selection methods, considering global performance and spatial equitability simultaneously, to identify methods that balance fidelity, uncertainty representation, and spatial consistency for robust climate impact and adaptation studies.

Study Configuration

• Spatial Scale: Turia River basin (approximately 6300 km²), divided into five sub-basins (Arquillo de San Blas, Benageber, Pueblos Castillo, Manises, l’Horta de València), and the aggregated basin scale. The study also references the Iberian Peninsula (for observed data) and European scales (for climate model ensembles).
• Temporal Scale:
  • Observed meteorological data: 1950-2015.
  • Observed hydrological data: From 1960 onwards.
  • Historical reference period for model calibration/bias adjustment: 1961-1990.
  • "Future" historical period for validation of projections: 1991-2015.
  • Climate change projections: EURO-CORDEX CMIP5–RCP8.5 scenario.

Methodology and Data

• Models used:
  • Hydrological model: Témez hydrological model (lumped, conceptual rainfall–runoff model).
  • Climate models: EURO-CORDEX ensemble (18 models, combinations of 10 General Circulation Models (GCMs) and 8 Regional Climate Models (RCMs)) for the CMIP5–RCP8.5 scenario.
  • Potential Evapotranspiration (PET) estimation: Thornthwaite equation and a modified Marcos-García (2019) Equation.
  • Bias correction: Empirical Quantile Mapping (EQM) method using the bias_correction Python library.
  • Sub-selection methods (12 tested): Best performing seasonal climate depiction (SCD), Best-performing seasonal hydrological depiction (SHD), Simple climate model weighing (CMW), Climate GCM diversity (CGCM), Climate RCM diversity (CRCM), Reliability Ensemble Average (REA), Reliability Ensemble on Historical Performance (REHP), Reliability Ensemble on Future Consensus (REFC), K-means Clustering on Reference Climatology (KCRC), Agglomerative Hierarchical Clustering on Reference Climatology (HCRC), K-means Clustering on Future Change (KCFC), Agglomerative Hierarchical Clustering on Future Change (HCFC).
• Data sources:
  • Observed meteorological data: SPAIN02 (monthly gridded precipitation and temperature dataset, 0.2° resolution, 1950-2015).
  • Observed hydrological data: PATRICAL model (monthly streamflow series, re-natured from historical gauged series in Spain, from 1960 onwards).
  • Climate change data: EURO-CORDEX initiative (monthly precipitation, temperature, in-month maximum and minimum temperatures for 18 climate models, RCP8.5, EUR-44 resolution).

Main Results

• No single sub-selection method optimizes all criteria (performance, uncertainty coverage, and spatial consistency), confirming the existence of inherent trade-offs.
• Methods based on future change diversity (K-means Clustering on Future Change (KCFC) and Agglomerative Hierarchical Clustering on Future Change (HCFC)) and adaptive consensus approaches (Reliability Ensemble on Historical Performance (REHP) and Reliability Ensemble on Future Consensus (REFC)) perform best, striking a balance between accuracy and uncertainty representation.
• These top-performing methods consistently outperform mono-model strategies, which show the poorest consistency and lowest average performances.
• Spatial variability is a decisive factor; methods performing adequately at aggregated scales often lose consistency across heterogeneous sub-basins, underscoring the value of spatial consistency as an explicit selection criterion.
• KCFC and HCFC demonstrated excellent future uncertainty coverage (70-90% of the full ensemble range) and effectiveness against random selection (50-85th percentiles).
• REFC excels in predominantly continental basins, while REHP performs best in markedly Mediterranean basins.
• An intermediate sub-ensemble size of 8-10 models provides a practical balance, reducing computational time by 45-55% while maintaining adequate uncertainty representation.
• The Témez hydrological model showed robust performance in the Turia River basin, with Kling-Gupta Efficiency (KGE) values ranging from 0.77 to 0.90 across sub-basins, 0.83 for the reference period, 0.68 for the future evaluation period, and 0.76 for capturing the observed shift between periods.
• The proposed low-computational-cost PET estimation method achieved a better-adjusted seasonal estimate (Root Mean Square Error (RMSE) = 9.98 mm) compared to other simplified methods (RMSE ranging from 12.84 to 22.49 mm).

Contributions

• Introduces a novel, standardized multi-criteria framework for evaluating climate model sub-selection methods that explicitly integrates average performance and spatial consistency, addressing a gap in existing literature.
• Demonstrates the critical importance of spatial robustness in climate model sub-selection, showing that good aggregated performance does not guarantee consistency across heterogeneous regions.
• Provides a reproducible and low-complexity procedure for identifying balanced sub-selection methods, offering transparent guidance for robust applications in climate impact and adaptation studies.
• Quantifies the trade-offs between accuracy, uncertainty coverage, and computational simplicity in climate model sub-selection, informing decision-making for specific application contexts.
• Identifies specific sub-selection methods (future diversity-based and adaptive consensus approaches) that offer superior balance and robustness for water resource management and adaptation planning.

Funding

• European Union’s Horizon 2020 research and innovation programme under the GoNEXUS project (reference code: 101003722).
• CRUE-Universitat Politècnica de València (funded the open access charge).

Citation

@article{LagosCastro2026Towards,
  author = {Lagos-Castro, Ivan and Macian-Sorribes, Héctor and Pulido-Velazquez, Manuel and Pedro-Monzonís, María},
  title = {Towards a balancing performance, uncertainty coverage, and spatial consistency in climate model sub-selection},
  journal = {Climatic Change},
  year = {2026},
  doi = {10.1007/s10584-026-04161-0},
  url = {https://doi.org/10.1007/s10584-026-04161-0}
}