An Integrated Statistical-Physical-Machine Learning Framework: Quantifying Human-Induced Terrestrial Water Storage Loss

Abstract

Terrestrial water storage (TWS) is influenced by both climate changes and human activities, yet research on the impacts of human activities remains limited. Here, a data-driven methodology is presented, integrating statistical analysis, physical equations and machine learning models to assess and quantify the influence of human activities on catchment TWS arising from engineering projects and water usage. The results validate a diminished water storage capacity of the Han River Basin (HRB) due to urbanization and decreasing natural permeable areas, resulting in a decrease in the maximum lagged time range from 4 months to 1 month, with a higher forgetting rate of 0.5 per lagged time. The alternation in water storage capacity affected the precipitation–runoff process. While climate change contributes to over 60% of the total effects, the substantial influence of human activities on TWS remains pivotal. Prior to the construction of the Middle Route of the South-to-North Water Diversion Project (1980–2003), human activities led to a multi-annual average TWS reduction of 14.7 km³ within the HRB. Post-construction (2015–2019), this figure rose to 19 km³, with human water usage and the reduction of groundwater flux feedback contributing 14.8 and 4.2, respectively. The proposed method provides a novel perspective for exploring the human impacts on TWS, potentially applicable to various geographical regions.