Evaluating water conservation capacity in the Yellow River water conservation area integrating ecological model and machine learning

Abstract

Accurately assessing the Water Conservation Capacity (WCC) of the Water Conservation Area (WCA) in the Yellow River Basin (YRB) is imperative for regional ecological security, yet it remains challenging because of the intricate interplay among climatic and anthropogenic drivers. This study proposes a novel integrated framework that couples the process-based InVEST model with a data-driven Random Forest (RF) algorithm to evaluate the spatiotemporal dynamics of WCC from 2000 to 2022 using annual water yield and Water Conservation Quantity (WCQ). Results reveal that annual water yield and WCQ range from 80.17 to 218.68 mm and 3.56 to 10.99 mm, and optimal Grade I WCC is predominantly concentrated in the central-southern Yellow River Source Area (YRSA), the Southern Mountain Tributary Area of the Wei River (SMTAWR), and the western Yiluo River Basin (YLRB). RF-based factor importance analysis indicates that climatic factors (precipitation, potential evapotranspiration) and anthropogenic factors (NDVI, population, GDP, flow velocity coefficient) are the primary drivers of WCC, while natural structural factors (soil depth, slope, saturated hydraulic conductivity, plant available water content) exert relatively minor effects. By quantitatively disentangling the relative contributions of climatic, natural structural, and anthropogenic factors to WCC, the proposed InVEST-RF framework advances watershed WCC assessment. Moreover, it provides a transferable methodological tool for ecohydrological evaluations in global watersheds, particularly under the context of changing climate and evolving land use trajectories.