Xia et al. (2025) A multi-temporal scale framework for comprehensive quantification and attribution of anthropogenic impacts on runoff

Identification

• Journal: Scientific Reports
• Year: 2025
• Date: 2025-12-11
• Authors: Chunchen Xia, Limin Zhang, Ziyue Zhu, Hongwei Xia, Haoyong Tian
• DOI: 10.1038/s41598-025-32088-6

Research Groups

• College of Civil Engineering, Zhejiang University of Technology, Hangzhou, China
• Zhejiang Key Laboratory of Green Construction and Intelligent Operation & Maintenance for Coastal Infrastructure, Zhejiang University of Technology, Hangzhou, China
• POWERCHINA Huadong Engineering Corporation Limited, Hangzhou, China

Short Summary

This study developed a multi-temporal scale framework integrating machine learning and hydrological models with multi-source data to quantify and attribute anthropogenic impacts on runoff. Applied to the Lan River Basin, the framework identified optimal temporal scales for modeling and elucidated the dual role of human activities on runoff, providing a transferable approach for attribution analysis.

Objective

• To establish a multi-temporal scale methodological framework for runoff reconstruction, quantification, and attribution of anthropogenic impacts.
• To identify watershed-specific optimal temporal scales for runoff modeling, overcoming limitations of single-scale approaches.
• To optimize explanatory variable selection by integrating Spearman correlation and variance inflation factor (VIF) analysis, addressing multicollinearity.
• To reconstruct runoff using a hybrid approach combining Random Forest Regression Model (RFRM) and Soil and Water Assessment Tool (SWAT) with multi-metric validation.
• To integrate remote sensing and statistical data across multiple temporal scales to identify and quantify anthropogenic drivers of runoff variation.

Study Configuration

• Spatial Scale: Lan River Basin, west-central Zhejiang Province, China, encompassing a total area of 19,350 square kilometers.
• Temporal Scale: 1970–2019, with daily hydro-meteorological records aggregated to weekly, two-week, monthly, two-month, seasonal (three-month), and annual resolutions. The study period was divided into a base period (BP: 1970–1987), variation period 1 (VP1: 1988–2008), and variation period 2 (VP2: 2009–2019).

Methodology and Data

• Models used:
  • Random Forest Regression Model (RFRM)
  • Soil and Water Assessment Tool (SWAT)
• Data sources:
  • Hydrological data: Daily runoff records (1970–2019) from three hydrological stations (Lanxi, Quzhou, Jinhua) provided by the Zhejiang Provincial Hydrology Bureau.
  • Meteorological data: Daily data (1970–2019) from five in-basin stations (National Meteorological Data Centre of the China Meteorological Administration), including precipitation, evaporation, temperature, air pressure, relative humidity, and wind speed.
  • Topographic data: 90-meter resolution Digital Elevation Model (DEM) from the Geospatial Data Cloud platform.
  • Soil data: 1:1,000,000 scale soil classification data from the Nanjing Institute of Soil Sciences.
  • Land use data: 1-kilometer resolution datasets from the Resource and Environment Science and Data Center, Chinese Academy of Sciences.
  • Vegetation data: 30-meter annual maximum Normalized Difference Vegetation Index (NDVI) data (2000–2018) from the National Ecosystem Science Data Center.
  • Urbanization/Human activity data:
    • Annual 30-meter resolution Impervious Surface Area (ISA) data (2000, 2005, 2010, 2015, 2018).
    • DMSP-OLS-like Nighttime Light (NTL) composite dataset (1992–2019).
    • Socioeconomic and agroforestry statistics (1970–2019) for Jinhua and Quzhou Cities from municipal statistical yearbooks.

Main Results

• Precipitation was identified as the dominant meteorological driver for runoff, showing the strongest correlation (maximum Spearman's ρ = 0.90 at the seasonal scale).
• Monthly and bimonthly temporal scales were found to be optimal for runoff modeling, with models at these scales generally outperforming those at finer (daily) or coarser (annual) resolutions.
• The Random Forest Regression Model (RFRM) and Soil and Water Assessment Tool (SWAT) demonstrated complementary strengths: RFRM models, particularly the P-T-Month model, exhibited optimal and balanced performance across low, medium, and high runoff depth intervals (RBIAS values of 3.26%, -3.65%, 3.30%, and 0.10% for low, medium, high, and total runoff, respectively). The SWAT-Month model, while having slightly lower overall accuracy, showed a distinct advantage in simulating extreme high-flow events.
• Both RFRM and SWAT models consistently indicated that human activities were the dominant driver of runoff variation in VP1 (1988–2008), with contribution rates ranging from 77.67% to 93.96% (RFRM) and 78.10% (SWAT).
• In VP2 (2009–2019), the relative contribution of human activities decreased (46.91%–66.10% for RFRM, 67.30% for SWAT), but the absolute runoff change attributed to human activities significantly increased (e.g., monthly scale runoff change rose from 5.39–6.15 mm in VP1 to 8.58–11.79 mm in VP2, an increase of approximately 90%).
• Multi-source data (ISA, NTL, and socioeconomic statistics) confirmed rapid urbanization, economic development, and water management policy implementation in VP2, empirically validating the enhanced influence of human activities on runoff increase.

Contributions

• Developed a novel and transferable multi-temporal scale framework for comprehensive quantification and attribution of anthropogenic impacts on runoff, addressing limitations of single-scale and single-factor analyses.
• Introduced an innovative variable screening module that integrates Bootstrap resampling with Spearman correlation and Variance Inflation Factor (VIF) analysis to effectively handle data scarcity and multicollinearity, ensuring robust explanatory variable selection.
• Achieved methodological complementarity by comparatively analyzing data-driven (RFRM) and physics-based (SWAT) models across multiple temporal scales, providing insights into their respective strengths for different hydrological regimes (e.g., RFRM for overall accuracy, SWAT for extreme flows).
• Advanced the analysis of anthropogenic impacts by integrating long-term hydro-meteorological observations with multi-source remote sensing products (ISA, NTL, NDVI) and socioeconomic statistics (GDP, population, water use, policies), moving beyond traditional land-use-change-centric approaches.
• Provided specific methodological guidance for optimal temporal scale and model selection in runoff attribution studies, particularly relevant for humid and semi-humid urban watersheds.

Funding

• Joint Funds of the Zhejiang Provincial Natural Science Foundation of China (Grant No. LZJWZ23E090010)
• National Natural Science Foundation of China (Grant No. 12002310)

Citation

@article{Xia2025multitemporal,
  author = {Xia, Chunchen and Zhang, Limin and Zhu, Ziyue and Xia, Hongwei and Tian, Haoyong},
  title = {A multi-temporal scale framework for comprehensive quantification and attribution of anthropogenic impacts on runoff},
  journal = {Scientific Reports},
  year = {2025},
  doi = {10.1038/s41598-025-32088-6},
  url = {https://doi.org/10.1038/s41598-025-32088-6}
}