Shakeel et al. (2026) Optimizing GCM ensemble selection and weighted MME development for improved drought projection under global climate models simulations

Identification

• Journal: Natural Hazards
• Year: 2026
• Date: 2026-03-26
• Authors: Muhammad Shakeel, Z ALI, Hussnain Abbas, Muhammad Mohsin, Mrim M. Alnfiai, Naim Ahmad, Rizwan Niaz
• DOI: 10.1007/s11069-026-08082-0

Research Groups

• College of Statistical Sciences, University of the Punjab, Lahore, Pakistan
• Department of Information Technology, College of Computers and Information Technology, Taif University, Taif, Saudi Arabia
• College of Computer Science, King Khalid University, Abha, Saudi Arabia
• School of Energy and Environment Science, Yunnan Normal University, Kunming, China

Short Summary

This study proposes a novel framework for selecting optimal Global Climate Model (GCM) subsets and developing weighted Multi-Model Ensembles (MMEs) to improve drought projection accuracy. It introduces the Multi-Location Multimodel Standardized Drought Index (MLMSDI), demonstrating its effectiveness in assessing future drought across various Shared Socioeconomic Pathways (SSPs) and timescales in Punjab Province, Pakistan.

Objective

• To propose a novel framework for selecting top-performing GCMs regionally using the Mutual Information (MI)-based Conditional Mutual Information Maximization (CMIM) feature selection method.
• To evaluate five Multi-Model Ensemble (MME) methods and identify the best performer for drought projection.
• To develop a new drought index, the Multi-Location Multimodel Standardized Drought Index (MLMSDI), effective in identifying extreme droughts in various future scenarios.
• To examine Shared Socioeconomic Pathways (SSPs) under steady-state probabilities to understand the likelihood of extreme wet or dry conditions.

Study Configuration

• Spatial Scale: Punjab Province, Pakistan (28 locations); Gridded precipitation data at 0.5° × 0.5° resolution (approximately 50 km × 50 km).
• Temporal Scale:
  • Historical period: 1950–2014 (65 years)
  • Future projection period: 2015–2100 (86 years)
  • Drought assessment timescales: Seven different timescales (e.g., 9, 48 months).

Methodology and Data

• Models used:
  • Global Climate Models (GCMs): 22 GCMs from CMIP6 (a subset of 9 top-performing GCMs were selected).
  • Feature Selection: Conditional Mutual Information Maximization (CMIM).
  • GCM Subset Selection: Group Decision Making Method (GDMM).
  • Multi-Model Ensembles (MMEs): Simple Average Ensemble (SAE), Constrained Least Squares based Ensemble (CLSE), Bayesian Model Averaging (BMA), Trimmed Eigen Vector Ensemble (TEVE), Trimmed Bias Corrected Eigen Vector Ensemble (TBCEVE).
  • Drought Index: Multi-Location Multimodel Standardized Drought Index (MLMSDI), based on the Standardized Precipitation Index (SPI) framework, utilizing K-Component Gaussian Mixture Models (KCGMM) for standardization.
  • Evaluation Metrics: Extended Distance between Indices of Simulation and Observation (EDISO), Kling-Gupta Efficiency (KGE), Diebold-Mariano (DM) test.
  • Statistical Software: R (packages: 'praznik', 'GeomComb', 'BMS', 'mixtools', 'propagate').
• Data sources:
  • Observed Precipitation: Monthly gridded precipitation data from TS v4.04 (0.5° × 0.5°) for Punjab Province, Pakistan.
  • Simulated Precipitation: Monthly precipitation data from 22 GCMs under CMIP6.
  • Future Scenarios: Shared Socioeconomic Pathways (SSPs) — SSP1-2.6, SSP2-4.5, and SSP5-8.5.
  • Data Source Link: https://climate.copernicus.eu

Main Results

• A rigorous two-stage selection process using CMIM and GDMM successfully identified a subset of nine top-performing GCMs for the entire Punjab Province, ensuring regional suitability and reduced uncertainty.
• The Constrained Least Squares based Ensemble (CLSE) MME demonstrated superior performance compared to other ensemble approaches, exhibiting a minimum EDISO value of 1.098 and a Kling-Gupta Efficiency (KGE) of 0.372. The Diebold-Mariano test confirmed the statistically significant superiority of CLSE.
• The proposed Multi-Location Multimodel Standardized Drought Index (MLMSDI), which incorporates K-Component Gaussian Mixture Models (KCGMM) for precipitation standardization, showed a better fit to complex, multi-peak precipitation datasets compared to 32 univariate probability distributions, as indicated by lower Bayesian Information Criterion (BIC) values (e.g., -9097.49 for KCGMM vs. -680.28 for univariate inverse gamma at specific timescales).
• MLMSDI effectively characterized future drought behavior across various SSPs (SSP1-2.6, SSP2-4.5, SSP5-8.5) and seven timescales, illustrating temporal variations of wet and dry periods.
• Steady-state probabilities for future scenarios indicated a high likelihood of "no drought" conditions, but also revealed increasing drought risk with longer timescales and higher emission scenarios (SSPs), highlighting the need for long-term adaptation policies.

Contributions

• Proposed a novel, unified framework for regional GCM subset selection using Conditional Mutual Information Maximization (CMIM) and Group Decision Making Method (GDMM), which effectively addresses interdependencies, redundancy, and ensures regional applicability of selected models.
• Developed and validated a superior weighted Multi-Model Ensemble (MME) approach (CLSE) for drought projection, demonstrating its enhanced accuracy over other methods through comprehensive evaluation metrics (EDISO, KGE, and DM test).
• Introduced the Multi-Location Multimodel Standardized Drought Index (MLMSDI), which improves drought characterization by employing K-Component Gaussian Mixture Models (KCGMM) for standardization, better capturing the complex, multimodal nature of precipitation data.
• Provided a robust assessment of future drought trends and steady-state probabilities across multiple Shared Socioeconomic Pathways (SSPs) and timescales for a climate-vulnerable region (Punjab Province, Pakistan), offering valuable insights for resource management and policy formulation.

Funding

• Large Research Project under grant number RGP2/417/46 (funded by the Deanship of Research and Graduate Studies at King Khalid University).

Citation

@article{Shakeel2026Optimizing,
  author = {Shakeel, Muhammad and ALI, Z and Abbas, Hussnain and Mohsin, Muhammad and Alnfiai, Mrim M. and Ahmad, Naim and Niaz, Rizwan},
  title = {Optimizing GCM ensemble selection and weighted MME development for improved drought projection under global climate models simulations},
  journal = {Natural Hazards},
  year = {2026},
  doi = {10.1007/s11069-026-08082-0},
  url = {https://doi.org/10.1007/s11069-026-08082-0}
}