Daliri et al. (2026) Water resources projection using CMIP6 global climate models and water balance uncertainty

Identification

• Journal: ENVIRONMENTAL SYSTEMS RESEARCH
• Year: 2026
• Date: 2026-01-18
• Authors: Farhad Daliri, Vijay P. Singh
• DOI: 10.1186/s40068-025-00449-9

Research Groups

• Absam EPC Co, Civil and Environmental Engineering Research Department, Tehran, Iran
• Independent Researcher, Dallas, TX, USA

Short Summary

This study evaluates near-future (2024–2054) hydro-climatic projections and water balance uncertainty for Lashkenar Village, northern Iran, using CMIP6 models under the SSP2-4.5 scenario. It projects a transition towards drought conditions with a significant decline in precipitation and increase in temperature, while introducing novel skewness-based diagnostics to reveal hidden distributional shifts and quantify precipitation-driven uncertainty in water balance components.

Objective

• To evaluate the efficiency of selected downscaling (DS) and multi-model ensemble (MME) methods in adjusting CMIP6 outputs for predicting and interpreting near-future (2024–2054) climate change (CC) impacts on annual temperature (T) and precipitation (Pr) in Lashkenar Village under the SSP2-4.5 scenario.
• To assess the effect of precipitation uncertainty propagation on the water balance (WB) of Lashkenar Village over the next three decades.

Study Configuration

• Spatial Scale: Lashkenar Village, Nowshahr, Mazandaran Province, northern Iran, spanning 10 square kilometers, with elevations ranging from 1525 meters to 2388 meters.
• Temporal Scale: Historical period: 2007–2023. Near-future projection period: 2024–2054 (30 years).

Methodology and Data

• Models used:
  • CMIP6 Global Climate Models (GCMs): MPI-ESM1–2-HR (Model 1) and ACCESS-ESM1.5 (Model 2) were selected.
  • Downscaling (DS): Support Vector Regression (SVR) (most robust), Linear Regression (LR), Random Forests (RF).
  • Bias Correction (BC): Quantile Mapping (QM).
  • Lead-Lag Correction (LLC): Dynamic Time Warping (DTW) algorithm.
  • Multi-Model Ensembling (MME): Bayesian Model Averaging (BMA) (most robust), Weighted Averaging Ensemble (WA-MME1).
  • Water Balance (WB) components: SCS-CN method for surface runoff, Turc formula for actual evapotranspiration, Maillet's exponential recession formula for aquifer storage/spring discharge.
  • Uncertainty Quantification: Expanded Uncertainty (EU), Bootstrapping (BS) (static and dynamic, with Bias-Corrected and Accelerated (BCa) method).
  • Trend Analysis: Mann-Kendall (M-K) Test, Sen’s Slope Estimator, Polynomial Fitting (Least-Squares Regression or LSR), Loess Smoothing (LS), Skewness–Torque (ST) framework, Yearly Skewness Trend (YST).
  • Change Point Detection: Pettitt homogeneity test.
• Data sources:
  • Observational data: Kojour Synoptic Station (2007–2023, elevation 1550 m), Kandolus Rain Gauge (last six years, elevation 1715 m).
  • Climate model data: CMIP6 GCMs from Earth System Grid Federation (ESGF) under SSP2-4.5 scenario.
  • Local information: Villagers' data on spring flow, local evidence.

Main Results

• Support Vector Regression (SVR) for downscaling and Bayesian Model Averaging (BMA) for ensembling provided the most robust performance, achieving a precipitation correlation (R) of 0.97 and temperature R of 0.6.
• Near-future projections (2024–2054) indicate a statistically significant decline in annual precipitation of -1.03 mm per year and an increase in annual temperature of +0.018 °C per year.
• Change-point detection identified 2014 for temperature and 2021 for precipitation as critical shifts, with subsequent trends indicating warming and declining rainfall.
• The novel Skewness–Torque (ST) framework and Yearly Skewness Trend (YST) method revealed distributional asymmetry, time lags (e.g., a 7-year lag between temperature and precipitation shifts), and resilience thresholds, providing early-warning signals of regime shifts not captured by traditional Mann–Kendall tests.
• Expanded uncertainty and dynamic Bootstrapping analyses quantified precipitation-driven uncertainty, showing that approximately ±11% of annual precipitation propagated into water balance components, equivalent to ±0.38 meters of groundwater fluctuation.
• Results suggest that Lashkenar Village is transitioning toward drought conditions, with precipitation frequently clustering between 350 mm and 360 mm in the simulated period, compared to 350 mm and 400 mm historically.

Contributions

• Developed a structured framework for adjusting CMIP6 outputs (lead-lag correction, bias correction, statistical downscaling, and multi-model ensembling) for reliable local-scale hydro-climatic projections in data-scarce mountainous catchments.
• Introduced a novel diagnostic framework, the Yearly Skewness Trend (YST) curve and Skewness–Torque (ST) method, which enhances the detection of subtle distributional shifts, time lags, and resilience dynamics in hydro-climatic time series beyond conventional trend tests.
• Explicitly quantified the propagation of precipitation uncertainty (±11%) into water balance components, translating to ±0.38 meters of groundwater fluctuation, providing a transparent framework for evaluating the robustness of water resource projections.
• Provided high-resolution insights into micro-scale climate–hydrology interactions for an ungaged mountainous catchment, demonstrating how adjusted CMIP6 outputs and quantified water balance uncertainty can inform climate-resilient planning.

Funding

This research received no external funding.

Citation

@article{Daliri2026Water,
  author = {Daliri, Farhad and Singh, Vijay P.},
  title = {Water resources projection using CMIP6 global climate models and water balance uncertainty},
  journal = {ENVIRONMENTAL SYSTEMS RESEARCH},
  year = {2026},
  doi = {10.1186/s40068-025-00449-9},
  url = {https://doi.org/10.1186/s40068-025-00449-9}
}