Wang et al. (2026) Exploring the effects of antecedent rainfall characteristic on streamflow variability in a karst catchment

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2026
• Date: 2026-04-12
• Authors: Fa Wang, Hongsong Chen, Jing Zhang, Zhiyong Fu, Jinjiao Lian, Yunpeng Nie
• DOI: 10.1016/j.ejrh.2026.103440

Research Groups

• Institute of Subtropical Agriculture, Chinese Academy of Sciences, Changsha, China
• Guangxi Key Laboratory of Karst Ecological Processes and Services, Huanjiang Observation and Research Station for Karst Ecosystems, Chinese Academy of Sciences, Huanjiang, Guangxi, China

Short Summary

This study investigates the influence of antecedent rainfall characteristics on streamflow dynamics in a karst catchment using machine learning models. It found that antecedent rainfall, particularly extreme events and consecutive drought days, critically influences streamflow, with climate change being the predominant driver (73.8%) of variability.

Objective

• To investigate the influence of antecedent rainfall characteristics on streamflow dynamics in karst catchments.
• To identify key antecedent rainfall indicators and enhance streamflow predictions using machine learning models.
• To explore ecohydrological mechanisms linking rainfall patterns to streamflow variability.
• To quantify the relative contributions of vegetation dynamics, climate variability, and rainfall pattern changes to streamflow regime shifts.

Study Configuration

• Spatial Scale: A typical peak-cluster depression karst catchment in Huanjiang County, Guangxi Province, Southwest China, with a total area of 1.14 km².
• Temporal Scale: Daily rainfall and streamflow observations from 2017 to 2019 were used for model training and validation. Daily streamflow was reconstructed for the period 2006–2023. The hydrological memory time was determined to be 29 days.

Methodology and Data

• Models used:
  • Machine Learning Models: Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN) for streamflow prediction.
  • Statistical Analyses: Autocorrelation and cross-correlation analyses for memory time estimation, Recursive Feature Elimination (RFE) and Variance Inflation Factor (VIF) for indicator selection, Mann-Kendall test for trend detection, Principal Component Analysis (PCA) for rainfall variability, and elasticity method for sensitivity analysis.
• Data sources:
  • Precipitation: Daily data recorded by a rain gauge (HOBO BHW-PRO-CD) and a pendant event logger (UA-003–64) with 0.2 mm resolution at a meteorological station within the catchment.
  • Stream Discharge: Daily data converted from 1-hour interval measurements by Manta 2 (Eureka, USA) at a weir near the catchment outlet.
  • Enhanced Vegetation Index (EVI): MOD13A1 product at 500 m resolution, with 16-day temporal resolution, linearly interpolated to daily scale.
  • Meteorological Variables: Daily average temperature (AT), maximum temperature (MT), and potential evapotranspiration (PET).

Main Results

• The average hydrological memory time for streamflow response to rainfall in the karst catchment was determined to be 29 days.
• Incorporating antecedent rainfall characteristics significantly improved the predictive accuracy of machine learning models, with the ANN model showing the best overall performance (mean testing R² of 0.823 with 12 parameters).
• Key antecedent rainfall indicators influencing streamflow included current day total rainfall (TR0), maximum consecutive 5-day precipitation (RX5), and consecutive drought days (CDD).
• Climate change was identified as the predominant driver of streamflow variability from 2006 to 2023, contributing 73.8% to observed changes.
• Vegetation restoration had a negative impact on streamflow, contributing -3.8%, suggesting increased evapotranspiration and water retention.
• Extreme rainfall events were the major drivers of interannual streamflow variability, while moderate rainfall events primarily affected streamflow persistence.
• Annual stream discharge showed a non-significant decreasing trend (Z = -0.912) from 2006 to 2023, while EVI exhibited a significant increasing trend (Z = 1.67).

Contributions

• Identified the hydrological memory time from long-term observations in a karst catchment.
• Established a reproducible framework for constructing and screening hydrologically meaningful antecedent rainfall indicators.
• Evaluated the added value of these indicators for machine learning prediction, process interpretation, and attribution analysis in karst streamflow modeling.

Funding

• National Natural Science Foundation of China (U2344201 and 42571054)

Citation

@article{Wang2026Exploring,
  author = {Wang, Fa and Chen, Hongsong and Zhang, Jing and Fu, Zhiyong and Lian, Jinjiao and Nie, Yunpeng},
  title = {Exploring the effects of antecedent rainfall characteristic on streamflow variability in a karst catchment},
  journal = {Journal of Hydrology Regional Studies},
  year = {2026},
  doi = {10.1016/j.ejrh.2026.103440},
  url = {https://doi.org/10.1016/j.ejrh.2026.103440}
}