Li et al. (2026) Deriving phase-contingent dynamic drought-limited water levels: An adaptive framework for managing megadrought evolution

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-03-07
• Authors: Yanbin Li, Haoyu Li, Kai Feng
• DOI: 10.1016/j.ejrh.2026.103309

Research Groups

• College of Water Conservancy, North China University of Water Resources and Electric Power, Zhengzhou, China

Short Summary

This study develops an adaptive framework to derive dynamic, phase-contingent Drought-Limited Water Levels (DLWLs) for managing megadroughts in reservoirs, addressing the limitations of static thresholds. It demonstrates that a supervised Random Forest model, anchored in physically constrained hydrological benchmarks, reliably classifies drought severity across four evolutionary phases, enabling improved, resilient reservoir operation.

Objective

• To develop a framework for dynamic, staged, and graded Drought-Limited Water Levels (DLWLs) that adapt to the evolutionary phases of megadroughts, linking physical drought science with applied reservoir operation.
• To delineate the evolutionary phases of a typical megadrought event.
• To develop a supervised classification methodology for objective drought severity grading.
• To synthesize temporal staging and severity grading into a structured set of dynamic DLWLs.

Study Configuration

• Spatial Scale: Xiaolangdi Reservoir in the Yellow River Basin, China, with specific focus on the Sanmenxia-Xiaolangdi inter-basin area for meteorological and soil moisture data aggregation. Gridded data resolutions include 10 km for SPEI and precipitation, and 1 km for soil moisture.
• Temporal Scale: Primary analysis period from 2006 to 2018, with SPEI-1 data extending from 1980 to 2018. Daily hydro-meteorological and operational data were aggregated into ten-day periods. The 2010–2011 megadrought event was selected as the reference case.

Methodology and Data

• Models used:
  • Supervised Random Forest (RF) classifier for drought grade classification.
  • Unsupervised K-means clustering as a baseline for comparative evaluation.
  • Entropy weighting method for objective weighting of indicators in the Comprehensive Hydrological Drought Index (CDI).
• Data sources:
  • Standardized Precipitation Evapotranspiration Index (SPEI-1): CHMDrought dataset (10 km resolution, monthly).
  • Precipitation: CHMPRE V2 dataset (10 km resolution, daily).
  • Soil Moisture Content: "Daily all weather surface soil moisture data set with 1 km resolution in China" (1 km resolution, daily).
  • Reservoir Inflow: Xiaolangdi Reservoir Authority (daily).
  • Reservoir Water Level: Xiaolangdi Reservoir Authority (daily).
  • Drought Characterization Indices: Precipitation Anomaly Percentage (Pa) and Standardized Precipitation Evapotranspiration Index (SPEI-1) for megadrought phasing.
  • Comprehensive Hydrological Drought Index (CDI) as the objective hydrological benchmark.

Main Results

• The supervised Random Forest (RF) model demonstrated significantly superior diagnostic skill compared to the unsupervised K-means baseline (RF: AUC = 0.98, RMSE = 0.62, PBIAS = 0.00%, NSE = 0.66; K-means: RMSE = 1.59, PBIAS = -44.55%, NSE = -1.18).
• The RF model achieved high classification fidelity, particularly for "Extreme Drought" (Grade 1), with a class accuracy of 97.2%.
• Soil moisture was identified as the dominant indicator for predicting drought severity in the RF model (relative importance = 0.36), followed by precipitation (0.35), emphasizing the critical role of catchment storage memory.
• The framework successfully delineated four distinct megadrought evolutionary phases: Gradual Emergence, Escalation, Persistence, and Recovery.
• Dynamic, stage-contingent Drought-Limited Water Levels (DLWLs) were derived, showing distinct operational logics: preemptive hedging (steep negative gradient) during Emergence and Escalation, and a stabilized survival baseline (approaching 235–240 m) during Persistence for extreme drought.
• The adaptive framework prevents unnecessary deep drawdown during moderate dry spells while authorizing access to deep storage during true megadroughts, balancing supply security and resource utilization efficiency.

Contributions

• Development of a novel adaptive framework for deriving dynamic, stage-contingent Drought-Limited Water Levels (DLWLs) that explicitly account for the evolutionary phases of megadroughts, moving beyond static operational thresholds.
• Empirical demonstration of the critical necessity for physically constrained benchmarks (e.g., soil moisture memory) in data-driven drought management tools, highlighting the limitations of purely statistical clustering for operational decision-making.
• Introduction of a phase-dependent operational logic for reservoir management, where the response to a given drought severity is contingent upon its evolutionary stage (e.g., preemptive hedging during emergence vs. survival-based conservation during persistence).
• Formulation of a Comprehensive Hydrological Drought Index (CDI) as a robust, physically consistent multivariate benchmark for supervised machine learning models, integrating precipitation, soil moisture, and inflow.
• Articulation of a transferable, data-driven methodology that bridges physical drought science and applied reservoir regulation, enhancing long-term water security under intensifying hydro-climatic non-stationarity.

Funding

• National Key R&D Program of China (grant number 2023YFC3006603)
• National Natural Science Fund of China (grant number 42301024, and 52179015)
• Science and Technology Projects in Henan Province (grant number 242102321114)
• Henan Province University Science and Technology Innovation Team and Talent Support Plan (grant number 26IRTSTHN022 and 26HASTIT031)
• Key Science Foundation Project of Henan Provincial Natural Science Foundation (grant number 252300421259)
• Special Project for Key Research and Development Tasks of Xinjiang Autonomous Region (grant number 2022293120)
• Hebei Provincial Department of Water Resources Projects (2025-19 and 2025-47)

Citation

@article{Li2026Deriving,
  author = {Li, Yanbin and Li, Haoyu and Feng, Kai},
  title = {Deriving phase-contingent dynamic drought-limited water levels: An adaptive framework for managing megadrought evolution},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103309},
  url = {https://doi.org/10.1016/j.ejrh.2026.103309}
}