John et al. (2026) Bottom-up assessment of climate change vulnerability of a large and complex river basin using emulator models

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-01-06
• Authors: A. Michael John, Avril Horne, Leah Traill, Keirnan Fowler, Rory Nathan
• DOI: 10.1016/j.ejrh.2025.103095

Research Groups

• Department of Infrastructure Engineering, The University of Melbourne, Melbourne, Australia
• Hydrology and Risk Consulting, Pty Ltd., Blackburn, Australia

Short Summary

This study conducts a bottom-up climate vulnerability assessment for the Murray-Darling Basin using computationally efficient machine learning-based emulator models. It reveals significant system sensitivities, non-linearities, and a critical threshold of 15% precipitation reduction, beyond which environmental targets are severely compromised.

Objective

• To assess climate change vulnerabilities to water resources and environmental management objectives in the large and complex Murray-Darling Basin (Australia) using computationally efficient machine learning-based emulator models within a bottom-up assessment framework.

Study Configuration

• Spatial Scale: The Murray-Darling Basin (approximately 1,000,000 km²), with data inputs generated at 77 sub-basin regions, aggregated to 21 major river valleys, and ecological outcomes reported at 24 key indicator sites.
• Temporal Scale: Baseline period 1976–2005, with stochastic data generated for 500 years. Climate change projections focused on 2060 (representing 2045–2074). Simulations were performed at a monthly timestep for river flows, disaggregated to daily for ecological outcomes.

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) emulator models, monthly WAPABA rainfall-runoff model (modified for snowmelt), and a monthly-to-daily disaggregation technique based on the method of fragments.
• Data sources: Outputs from existing complex jurisdictional water resource models (for LSTM training), stochastically generated hydroclimate data sequences, and bias-corrected climate change projections from a 37-member CMIP6 ensemble. Historic climate data (1890–2009) was used for validation.

Main Results

• The LSTM emulator models achieved high accuracy (median Kling-Gupta Efficiency > 0.95, percentage bias < ±10%) in simulating regulated river flows, enabling efficient bottom-up assessment.
• Runoff elasticity to precipitation ranged from approximately 2 to 6 (% change in runoff per % change in precipitation), and to temperature from 3.7% to 11.3% per °C, with northern basins generally more sensitive.
• A critical threshold was identified: a 15% reduction in precipitation significantly compromises the long-term performance of environmental targets across the basin, reducing the percentage of site-specific flow indicators (SFIs) meeting targets from 52% to approximately 26%.
• A 2 °C increase in temperature alone reduced the percentage of SFIs meeting targets to approximately 43%.
• High flows were found to be generally more sensitive to climate change than low flows.
• Significant non-linearities and spatial differences in hydrological response were observed between the northern and southern Murray-Darling Basin, particularly regarding sensitivity to precipitation seasonality. For example, Murray Barrage SFIs showed significant degradation after precipitation reductions of about 15%, with the target of 95% success for a 2 x 10^9 m³/year (2000 GL/year) 3-year rolling average flow rapidly degrading beyond 20% precipitation reduction.

Contributions

• Demonstrated a novel and computationally efficient bottom-up climate vulnerability assessment for large, complex river basins using machine learning-based emulator models.
• Enabled the simulation of 1089 climate scenarios over 500 years of stochastic data in less than one day on a standard laptop, a task previously intractable with complex water resource models.
• Revealed critical non-linear system sensitivities and thresholds (e.g., 15% precipitation reduction) in ecological management objectives that are often missed by traditional top-down assessments.
• Provided region-specific insights into hydrological responses and vulnerabilities across the northern and southern Murray-Darling Basin, informing targeted adaptation strategies.
• Offers a valuable framework for water managers to 'range-find' scenario assessments and identify high-impact planning scenarios, enhancing the robustness of adaptation measures globally.

Funding

• OneBasin CRC

Citation

@article{John2026Bottomup,
  author = {John, A. Michael and Horne, Avril and Traill, Leah and Fowler, Keirnan and Nathan, Rory},
  title = {Bottom-up assessment of climate change vulnerability of a large and complex river basin using emulator models},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2025.103095},
  url = {https://doi.org/10.1016/j.ejrh.2025.103095}
}