环境科学
水文学(农业)
土壤流失
随机森林
土壤科学
地质学
地貌学
腐蚀
岩土工程
计算机科学
机器学习
作者
Yulan Chen,Nan Wang,Juying Jiao,Jianjun Li,Leichao Bai,Yue Liang,Yanhong Wei,Ziqi Zhang,Xu Qian,Zhixin Zhang,Jiaxi Wang
摘要
Abstract Soil loss is a common land degradation process worldwide, which is impacted by land use and climate change. In this study, random forests (RF) were first used to establish a soil loss model at the scale of a small watershed in the hilly‐gully region of the Loess Plateau based on the field observation data. Subsequently, the model was used to predict soil loss in the Chabagou watershed under the historical (1990–2020) and future emission scenarios, namely SSP1–2.6 (low‐emission), SSP2–4.5 (medium‐emission) and SSP5–8.5 (high‐emission) (2030–2,100) from the Coupled Model Intercomparison Project Phases 6 (CMIP6). In the RF model, the coefficient of determination (R 2 ) and Nash‐Sutcliffe coefficient of efficiency (NS) were both greater than 0.86, and the RMSE‐observations standard deviation ratio (RSR) was less than 0.36. Additionally, the RF‐based model had higher simulation accuracy and robustness than those of the previous soil loss models, indicating its potential for wider applications in simulating soil loss. Compared with soil loss between 1990 and 1999, climate change led to a 35.36% increase in soil loss, while land use change resulted in an 11.13% reduction from 2000 to 2020 in the Chabagou watershed. This reveals that the current land use management could not effectively counterbalance the soil loss caused by rainstorms. Furthermore, compared with the historical period (1990–2020), under SSP1–2.6, SSP2–4.5 and SSP5–8.5 (2030–2,100), the soil loss rates without land use change would be increased by 6.01%, 19.11% and 35.35%, while the soil loss rates with land use change would be changed by −5.88%, +4.41% and +19.12%, respectively. These results help to provide a scientific basis for enhancing the capacity to respond to climate change and mitigation of soil and water loss on the Loess Plateau.
科研通智能强力驱动
Strongly Powered by AbleSci AI