【摘要】 以我国典型的土壤侵蚀敏感区黄河流域为研究对象，通过构建融合InVEST-USLE、RF、XGBoost、Attention-LSTM的多模型框架，基于2000—2023年多源遥感与气候数据，模拟2024—2050年6种未来情景下土壤侵蚀动态演变，系统评估气候变化与生态修复措施的耦合影响，揭示土壤侵蚀的驱动机制与调控阈值。结果表明：1)降水量是土壤侵蚀的关键驱动因子，植被调控效应具有显著阈值特征；2)模型识别出土壤侵蚀具有明显的“近时主导”特征，且长时间序列预测效果更优；3)情景模拟显示，高排放路径下若无生态修复，土壤侵蚀模数将持续增大，而高强度生态修复可显著减轻土壤侵蚀，但减蚀效益存在滞后性。研究成果为黄河流域水土保持分区治理与动态调控机制提供了定量支撑，建议将关键阈值和时滞效应纳入未来生态治理政策框架，以提升区域生态管理的精准性和有效性。更多还原

【Abstract】 The Yellow River Basin, a typical soil erosion-sensitive region in China, was selected as the research target. By constructing a multi-model framework integrating InVEST-USLE, Random Forest, XGBoost, and Attention-LSTM, and using multi-source remote sensing and climate data from 2000 to 2023, the dynamic evolution of soil erosion under six future scenarios from 2024 to 2050 was simulated. The coupled impacts of climate change and ecological restoration measures were systematically evaluated, revealing the driving mechanisms and regulatory thresholds of soil erosion. The results show that: a) precipitation is the primary driver of soil erosion, and vegetation regulation exhibits significant threshold characteristics; b) the model identifies a distinct “recent dominance” of soil erosion, with better prediction performance for long-term series; c) scenario simulations reveal that under high-emission pathways, without ecological restoration, the soil erosion modulus will continue to increase. While a high-intensity ecological restoration can significantly mitigate erosion trends, the erosion reduction benefits are delayed. These findings provide quantitative support for the zoning and dynamic regulatory mechanisms of soil and water conservation in the Yellow River Basin. It is recommended that critical thresholds and time-lag effects be incorporated into future ecological governance policy frameworks to improve the precision and effectiveness of regional ecological management.更多还原