Khalil et al. (2025) Inter-rater reliability adaptive weighting (IRRAWE) - a novel ensemble scheme for improved precipitation projections using CMIP6 climate models

Identification

• Journal: Theoretical and Applied Climatology
• Year: 2025
• Date: 2025-12-01
• Authors: Rashida Khalil, Zulfiqar Ali
• DOI: 10.1007/s00704-025-05947-5

Research Groups

• College of Statistical Sciences, University of the Punjab, Lahore, Pakistan
• Department of Statistics, Lahore College for Women University, Lahore, Pakistan

Short Summary

This study introduces IRRAWE, a novel spatio-temporal weighting scheme for CMIP6 multimodel ensembles, which significantly improves precipitation projections by achieving higher correlation and lower Normalized Root Mean Square Error compared to simple model averaging.

Objective

• To develop and evaluate a novel multimodel ensemble weighting scheme (IRRAWE) that uses inter-rater reliability (linearly weighted kappa statistic) and point-to-point divergence to determine optimal weights for CMIP6 General Circulation Models (GCMs, thereby improving precipitation projections.

Study Configuration

• Spatial Scale: Tibetan Plateau (TP) region in China, covering approximately 2.5 million square kilometers, with an average elevation exceeding 4,000 meters, analyzed at 32 distinct grid points.
• Temporal Scale: Monthly precipitation simulations covering the period from 1961 to 2014.

Methodology and Data

• Models used:
  • Inter-rater Reliability Adaptive Weighting Ensemble (IRRAWE) - proposed scheme.
  • Simple Model Averaged Ensemble (SMAE) - for comparative assessment.
  • 18 General Circulation Models (GCMs) from the Coupled Model Intercomparison Project Phase 6 (CMIP6).
  • K-Components Gaussian Mixture Models (K-CGMM) - for fitting precipitation probability distributions.
  • Standardized Precipitation Index (SPI) - for drought characterization.
  • Statistical methods: Linearly weighted kappa statistic, Expectation-Maximization (EM) algorithm, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC).
• Data sources:
  • Historical observed monthly precipitation data from 32 grid points within the Tibetan Plateau.
  • Simulated monthly precipitation data from 18 CMIP6 GCMs.

Main Results

• The proposed IRRAWE scheme achieved a 1.6% higher correlation (0.6625 vs. 0.6518) and a 3.8% lower Normalized Root Mean Square Error (NRMSE) (0.4798 vs. 0.4989) on average compared to the Simple Model Averaged Ensemble (SMAE).
• IRRAWE predictions were consistently closer to observations and exhibited narrower error bars (reduced ensemble spread) across most locations compared to SMAE.
• K-CGMM with three components provided a better fit to observed precipitation data than various univariate distributions, as indicated by lower BIC values.
• The inter-rater reliability (model skill) of individual GCMs varied significantly across the Tibetan Plateau, with no single model consistently outperforming others at all locations.
• IRRAWE effectively mitigates the inherent overestimation of the model ensemble and reduces overall uncertainty in precipitation forecasts.

Contributions

• Introduces a novel spatio-temporal weighting scheme (IRRAWE) for multimodel ensembles, which integrates inter-rater reliability (using the linearly weighted kappa statistic) and point-to-point divergence between simulated and observed data.
• Proposes a unique approach to transform numerical GCM simulations into a standardized, unit-free index for performance evaluation, considering both categorical and continuous aspects of model agreement.
• Demonstrates significant improvements in predictive accuracy and reduction in ensemble spread for precipitation projections compared to traditional simple model averaging.
• Provides a more robust and reliable framework for weighting ensemble predictions, particularly beneficial for applications like drought characterization using the Standardized Precipitation Index.

Funding

Authors received no funding for preparing this manuscript.

Citation

@article{Khalil2025Interrater,
  author = {Khalil, Rashida and Ali, Zulfiqar},
  title = {Inter-rater reliability adaptive weighting (IRRAWE) - a novel ensemble scheme for improved precipitation projections using CMIP6 climate models},
  journal = {Theoretical and Applied Climatology},
  year = {2025},
  doi = {10.1007/s00704-025-05947-5},
  url = {https://doi.org/10.1007/s00704-025-05947-5}
}