Kim et al. (2026) Constraining climate model projections with observations amplifies future runoff declines

Identification

• Journal: Communications Earth & Environment
• Year: 2026
• Date: 2026-01-27
• Authors: Hanjun Kim, Flavio Lehner, Katherine Dagon, David M. Lawrence, Sean Swenson, Andrew W. Wood
• DOI: 10.1038/s43247-026-03213-8

Research Groups

• Department of Earth and Atmospheric Sciences, Cornell University, Ithaca, NY, USA
• Center for Climate and Carbon Cycle Research, Korea Institute of Science and Technology, Seoul, Republic of Korea
• Climate and Global Dynamics Laboratory, NSF National Center for Atmospheric Research, Boulder, CO, USA
• Polar Bears International, Bozeman, MT, USA
• Department of Civil and Environmental Engineering, Colorado School of Mines, Golden, CO, USA

Short Summary

This study comprehensively assesses biases in climate model runoff sensitivities and constrains future runoff projections across global river basins using multiple observational datasets and climate model generations. It finds that observationally constrained projections indicate significantly stronger future runoff declines than raw model outputs, particularly due to underestimated temperature sensitivity in models.

Objective

• To assess biases in climate model runoff sensitivities (change in runoff per unit change of precipitation or temperature) compared to observations.
• To develop a comprehensive observational constraint approach for future runoff projections across major global river basins, accounting for observational uncertainty, non-stationarity of sensitivity, and internal climate variability.

Study Configuration

• Spatial Scale: Global, focusing on 131 major river basins.
• Temporal Scale: Historical period (1947–2017) for sensitivity estimation; Future projections (2030–2070) under SSP2-4.5 (CMIP6) and RCP4.5 (CMIP5) scenarios; 5-year averaged variations for runoff sensitivity calculations.

Methodology and Data

• Models used:
  • Coupled Model Intercomparison Project Phase 6 (CMIP6): 28 Earth System Models (ESMs).
  • Coupled Model Intercomparison Project Phase 5 (CMIP5): 22 ESMs.
  • Community Earth System Model 2 Large Ensemble (LENS2): 50-member ensemble simulations.
• Data sources:
  • Global Runoff Reanalysis (GRUN) ensemble data: Machine learning-based, 0.5° × 0.5° horizontal resolution, 1902–2019 (extended to 2017 for analysis), 100 members from 4 atmospheric forcing datasets (CRU TS v4.04, PGFv2, GSWP3-EWEMBI, GSWP3-W5E5) and 25 subsampling sets.
  • Station-based naturalized streamflow: For Columbia River, Northern Sierras, Upper Colorado River (Western United States, 1947–2017), and Murray River basins (Australia, 1947–2008).
  • GPCC precipitation data (for GRUN extension).
  • HydroSHEDS database (for river basin masks).
  • GRADES-hydroDL (for cross-validation).

Main Results

• Climate models generally underestimate the magnitude of negative temperature sensitivity (runoff decline per warming) and overestimate positive precipitation sensitivity compared to observations. Specifically, CMIP6 models show significantly different P sensitivities in 106 out of 131 basins (overestimating in 102), and T sensitivities in 90 out of 131 basins (underestimating negative magnitude in 75).
• The historical runoff sensitivities effectively predict model-simulated runoff changes for 97 out of 131 global river basins. Accounting for non-stationarity and internal climate variability improves this to 107 out of 131 basins.
• The uncertainty in runoff projections from climate models is comparably influenced by meteorological forcing (dominated by precipitation) and runoff sensitivity (dominated by temperature sensitivity).
• Observationally constrained projections indicate stronger runoff declines than raw model projections in 41 out of 131 global river basins, including the Amazon, Danube, Murray, Nile, and Mekong. This constraint is robust across CMIP5 and CMIP6 model generations.
• The stronger runoff declines are primarily driven by the bias in temperature sensitivity. For example, in CMIP6, runoff decline is exacerbated by 14.9% in the Amazon, 12.2% in the Danube, and 31.9% in the Murray. Projected increases in the Nile and Mekong are reduced by 16.9% and reversed to a decline of -16.1%, respectively.
• The observational constraint adjusts the central values downward but does not significantly reduce the inter-model spread, as the reduction from bias correction is offset by increased uncertainty from non-stationarity and internal variability.

Contributions

• Developed a more comprehensive and robust observational constraint approach for future runoff projections, explicitly accounting for observational uncertainty, potential non-stationarity of runoff sensitivity, and the impacts of internal climate variability.
• Utilized a recent 70-year global runoff dataset (GRUN) to detect emerging climate change signals in runoff sensitivity, which were likely missed in previous studies with shorter or pre-warming observational periods.
• Quantified that climate models systematically underestimate runoff declines in response to warming, leading to a projected drier future for 41 major global river basins than previously estimated by raw model outputs.
• Highlighted that uncertainty in runoff generation processes related to warming (temperature sensitivity) contributes significantly to regional runoff projection uncertainty, sometimes as much as precipitation uncertainty.
• Developed a runoff sensitivity metric package for integration into NOAA Model Diagnostics Task Force and International Land Model Benchmarking packages to improve model development.

Funding

• NOAA MAPP award NA21OAR4310349
• Korea Institute of Science and Technology Research Program (2E33621)
• National Center for Atmospheric Research (NSF Cooperative Agreement 1852977)
• RUBISCO Scientific Focus Area (U.S. Department of Energy Office of Science, Regional and Global Climate Modeling Program in the Climate and Environmental Sciences Division of the Office of Biological and Environmental Research)

Citation

@article{Kim2026Constraining,
  author = {Kim, Hanjun and Lehner, Flavio and Dagon, Katherine and Lawrence, David M. and Swenson, Sean and Wood, Andrew W.},
  title = {Constraining climate model projections with observations amplifies future runoff declines},
  journal = {Communications Earth & Environment},
  year = {2026},
  doi = {10.1038/s43247-026-03213-8},
  url = {https://doi.org/10.1038/s43247-026-03213-8}
}