Saleem et al. (2025) Projecting future climate extremes in the glacier-fed upper indus basin using machine learning based downscaling of CMIP6 GCMs

Identification

• Journal: Theoretical and Applied Climatology
• Year: 2025
• Date: 2025-10-13
• Authors: Muhammad Saleem, Muhammad Shoaib, Sarfraz Hashim, Hafiz Umar Farid, Muhammad Ismail, Mubashir Ali Ghaffar, Ahmad Mujtaba, Jinwook Lee, Azhar Inam, A. Ameen
• DOI: 10.1007/s00704-025-05793-5

Research Groups

• Department of Agricultural Engineering, Faculty of Agricultural Sciences & Technology, Bahauddin Zakariya University, Multan, Pakistan
• Department of Agricultural Engineering, MNS University of Agriculture, Multan, Pakistan
• Field Wing of Punjab Agricultural Department, Agriculture House, Lahore, Pakistan
• College of Water Resources and Architectural Engineering, Northwest A&F University, Yangling, Shaanxi, China
• MEENA Consultancy and General Contracting DMCC, Dubai, United Arab Emirates
• Department of Civil, Environmental and Construction Engineering, University of Hawaii at Manoa, Honolulu, Hawaii, United States

Short Summary

This study downscaled CMIP6 Global Circulation Model (GCM) data for the Upper Indus Basin (UIB) using Artificial Neural Networks (ANNs) and Convolutional Neural Networks (CNNs) to project future climate extremes. The research found that CNNs outperformed ANNs, revealing robust warming trends in temperature and uncertain precipitation trends across the UIB under future Shared Socioeconomic Pathways (SSP) scenarios.

Objective

• To apply Artificial Neural Networks (ANNs) and Convolutional Neural Networks (CNNs) to downscale CMIP6 Global Circulation Model (GCM) datasets for the Upper Indus Basin (UIB).
• To project future climate changes and assess extreme climate indices (ECI) in the UIB for near-term (2026–2055) and long-term (2056–2085) periods under Shared Socioeconomic Pathways SSP245 and SSP585.

Study Configuration

• Spatial Scale: Upper Indus Basin (UIB) of Pakistan (33°N to 37°N and 70°E to 77°E), focusing on 12 meteorological stations across high, medium, and low elevated regions.
• Temporal Scale:
  • Historical: 1985–2014 (for model training and testing).
  • Near-term (NT): 2026–2055.
  • Long-term (LT): 2056–2085.

Methodology and Data

• Models used:
  • Machine Learning (ML) algorithms: Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs).
  • Statistical tests: Modified Mann-Kendall (MMK) test for trend analysis.
  • Software: RClimdex for calculating 12 precipitation and temperature extreme climate indices (ECI) based on the ETCCDI approach.
  • Global Circulation Models (GCMs): Ten CMIP6 GCMs (MIROC6, MRI-CM6-1, CNRM-ESM-2, CanESM5, IPSL-CM6A-LR, among others), with the top five selected for future projections.
• Data sources:
  • Observed historical daily precipitation (P), maximum temperature (Tmax), and minimum temperature (Tmin) from 12 meteorological stations in UIB (1985–2014), collected from the Pakistan Meteorological Department (PMD) and the Water and Power Development Authority (WAPDA).
  • CMIP6 GCM daily precipitation (P), maximum temperature (Tmax), and minimum temperature (Tmin) data (1985–2014 and 2026–2085 under SSP245 and SSP585 scenarios), sourced from Australia’s National Computing Infrastructure (esgf-node.llnl.gov).

Main Results

• Convolutional Neural Networks (CNNs) outperformed Artificial Neural Networks (ANNs) in downscaling CMIP6 GCM data, with R² (0.93–0.71), Kling-Gupta Efficiency (KGE) (0.93–0.77), Root Mean Square Error (RMSE) (3.00–5.90), and Percent Bias (PBIAS) (–3.90–1.00%) values.
• The top five CMIP6 GCMs selected for future projections were MIROC6, MRI-CM6-1, CNRM-ESM-2, CanESM5, and IPSL-CM6A-LR.
• Trend analysis revealed robust and statistically significant warming signals in minimum and maximum temperatures across all downscaled CMIP6 GCMs by CNNs under both SSP245 and SSP585 scenarios.
• Precipitation trends remained uncertain and inconsistent among models and scenarios, showing both increasing and decreasing signals with lower robustness.
• Annual mean precipitation change ranged from +89.27% to −81.50% under SSP245 (near-term) and up to +86.00% to −77.63% under SSP585 (long-term).
• Projected mean maximum temperature (Tmax) exhibited increases up to +6.41 °C and decreases up to −5.89 °C.
• Projected mean minimum temperature (Tmin) exhibited increases up to +5.47 °C and decreases up to −6.56 °C.
• Extreme Climate Indices (ECI) for both precipitation (Sdii, rx1day, R10mm, R25mm, R95p, R99p, Prcptot) and temperature (Su25, Tr20, Txx, Txn, Tnx, Tnn) revealed pronounced warming trends, particularly under the higher-emission SSP585 scenario.
• Temperature intensity increased from high to low elevated regions, while precipitation intensity decreased along the same gradient.

Contributions

• First application of Machine Learning (ANNs and CNNs) algorithms to downscale CMIP6 GCM outputs for the Upper Indus Basin (UIB) of Pakistan.
• Provides high-resolution regional climate projections and extreme climate indices for the UIB, addressing a critical data gap for a region of significant economic and water resource importance.
• Demonstrates the superior performance of CNNs over ANNs for climate downscaling in this complex mountainous region.
• Offers critical insights for local environmental planning and risk reduction initiatives related to climatic and hydrological extremes in the UIB.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Citation

@article{Saleem2025Projecting,
  author = {Saleem, Muhammad and Shoaib, Muhammad and Hashim, Sarfraz and Shoaib, Muhammad and Farid, Hafiz Umar and Ismail, Muhammad and Ghaffar, Mubashir Ali and Mujtaba, Ahmad and Lee, Jinwook and Inam, Azhar and Ameen, A.},
  title = {Projecting future climate extremes in the glacier-fed upper indus basin using machine learning based downscaling of CMIP6 GCMs},
  journal = {Theoretical and Applied Climatology},
  year = {2025},
  doi = {10.1007/s00704-025-05793-5},
  url = {https://doi.org/10.1007/s00704-025-05793-5}
}