Yu et al. (2026) Vine copula regional dry-wet evaluation index incorporating baseflow and XGBoost-SHAP early warning system in Huzhou, China

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-09-19
• Authors: Tongtong Yu, Hao Chen, Yukun Wang, Jiaqi Tan, Zhishao Li, Yu Qiao, Jianuo Li, Saihua Huang, Yue‐Ping Xu, Yuxue Guo, Hui Nie
• DOI: 10.1016/j.ejrh.2026.104002

Research Groups

• School of Hydraulic Engineering, Zhejiang University of Water Resources and Electric Power, Hangzhou, China
• Zhejiang Key Laboratory of River-Lake Water Network Health Restoration, Zhejiang University of Water Resources and Electric Power, Hangzhou, China
• Estuarine and Coastal Disaster Prevention and Mitigation and Ecological Governance Institute, Zhejiang University of Water Resources and Electric Power, Hangzhou, China
• Institute of Hydrology and Water Resources, College of Civil Engineering and Architecture, Zhejiang University, Hangzhou, China

Short Summary

This study developed a Comprehensive Dry-Wet Index (CDI) for Huzhou City, China, by integrating baseflow, precipitation, and evaporation using vine copulas. It also established an XGBoost-SHAP early warning system, revealing significant drying trends and spatially heterogeneous driving factors for dry-wet conditions.

Objective

• To separate baseflow from surface runoff using the Lyne-Hollick digital filter and construct a comprehensive dry-wet index (CDI) by coupling precipitation, evapotranspiration, and the baseflow index (BFI) via a copula function, thereby improving the accuracy and comprehensiveness of dry-wet condition assessments.
• To develop a monthly-scale dry-wet early warning model based on the XGBoost algorithm, incorporating multiple predictors (precipitation, evapotranspiration, BFI, teleconnection indices, and relative soil moisture) to achieve high-precision prediction of dry-wet severity grades.
• To quantify the contribution of each driving factor to dry-wet evolution using the SHAP framework, analyze how climate change and human activities modulate watershed dry-wet conditions, and reveal the spatiotemporal evolution patterns of dry-wet states.

Study Configuration

• Spatial Scale: Huzhou City, northern Zhejiang Province, southeastern China, covering approximately 5818 square kilometers. The study focused on five hydrological stations: Duihekou (DHK), Fushi (FS), Hengtangcun (HTC), Gangkou (GK), and Laoshikan (LSK).
• Temporal Scale: Monthly resolution data covering the period from 2012 to 2024.

Methodology and Data

• Models used:
  • Lyne-Hollick digital filter (for baseflow separation)
  • Vine Copula (C-vine and D-vine, for CDI construction)
  • XGBoost (Extreme Gradient Boosting) algorithm (for monthly dry-wet early warning model)
  • SHapley Additive exPlanations (SHAP) framework (for interpreting driving mechanisms)
  • Modified Mann-Kendall trend test (for trend analysis)
  • Heuristic segmentation algorithm (for detecting abrupt change points)
  • Run theory (for quantifying dry/wet event characteristics)
• Data sources:
  • Hydrological stations (Duihekou, Fushi, Hengtangcun, Gangkou, Laoshikan): Monthly runoff, precipitation, and evapotranspiration data (2012–2024).
  • Hydrological Historical Database and China Meteorological Data Service Center (http://data.cma.cn): Daily precipitation, pan evaporation observations, and mean temperature data.
  • Sunspot Index and Long-term Solar Observations (SILSO) of the World Data Center (http://sidc.oma.be/silso/dayssnplot): Sunspot data.
  • National Oceanic and Atmospheric Administration (NOAA) Earth System Research Laboratory and NOAA National Centers for Environmental Information (http://www.ncdc.noaa.gov/teleconnections/ao.php): Teleconnection indices (El Niño-Southern Oscillation (ENSO), Pacific Decadal Oscillation (PDO), Arctic Oscillation (AO), Indian Ocean Dipole (IOD), Atlantic Multidecadal Oscillation (AMO), North Atlantic Oscillation (NAO), Southern Oscillation (SO)).
  • Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2): Root-zone soil moisture (0–1 meter).

Main Results

• Drying Trends: Statistically significant decreasing dry-wet trends were detected at four of the five hydrological stations (DHK, FS, GK, and LSK), while HTC showed no significant monotonic change.
• CDI Performance: The Comprehensive Dry-Wet Index (CDI) successfully integrated precipitation, evaporation, and baseflow signals, exhibiting strong correlations with standardized precipitation index (SPI), standardized evapotranspiration index (SEI), and standardized baseflow index (SBI) at multiple temporal scales. It achieved an overall hit rate of 0.88 (14 out of 16) for detecting documented historical dry and wet episodes, outperforming individual indices.
• Dry-Wet Evolution: The region's dry-wet conditions shifted from seasonal alternation towards more persistent states, with dry periods concentrated in winter (December to February) and wet spells largely coinciding with high baseflow during the rainy season.
• Early Warning Model Accuracy: The monthly XGBoost-based early warning model achieved an overall accuracy of 63.5%–76.0% across stations (mean 70.8%). While it reliably predicted the 'Moderate' state (Recall 89.5–95.8%), its performance for 'Dry' and 'Wet' states was more limited (Recall 32.8–46.0% and 33.2–40.8%, respectively).
• Driving Mechanisms: Local hydrometeorological inputs (precipitation, evaporation, baseflow, and baseflow index) accounted for 66.5%–71.5% of the normalized external-factor importance. Baseflow-related variables were the leading predictors at four stations, while evaporation was the primary driver at GK (24.0%). Teleconnection indices contributed 28.5%–33.5% and primarily modulated dry-wet evolution through interactions with local states.

Contributions

• Developed a novel Comprehensive Dry-Wet Index (CDI) by integrating baseflow with precipitation and evapotranspiration using vine copulas, providing a more holistic and accurate assessment of dry-wet conditions, particularly in human-influenced humid monsoon basins.
• Established an interpretable monthly dry-wet early warning system using the XGBoost-SHAP framework, which not only provides high-precision predictions but also elucidates the spatially heterogeneous driving mechanisms of dry-wet evolution, including the roles of local hydrometeorological factors and teleconnection indices.
• Provided a scientific basis for upgrading integrated drought and flood early warning systems, optimizing water resource allocation, and formulating targeted disaster prevention strategies in Huzhou City and other regions with analogous hydrological characteristics.

Funding

• National Natural Science Foundation of China (52509037, 52479028, and 52579023)
• Natural Science Foundation of Zhejiang Province in China (ZCLQ24E0901, LGEY25E090012)
• Students’ innovation and entrepreneurship training program of Zhejiang University of Water Resources and Electric Power (S202511481010 and 202511481008)

Citation

@article{Yu2026Vine,
  author = {Yu, Tongtong and Chen, Hao and Wang, Yukun and Tan, Jiaqi and Li, Zhishao and Qiao, Yu and Li, Jianuo and Huang, Saihua and Xu, Yue‐Ping and Guo, Yuxue and Nie, Hui},
  title = {Vine copula regional dry-wet evaluation index incorporating baseflow and XGBoost-SHAP early warning system in Huzhou, China},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.104002},
  url = {https://doi.org/10.1016/j.ejrh.2026.104002}
}