不同放牧策略对草原土壤性质的影响研究——基于机器学习
Study on the Effects of Different Grazing Strategies on Soil Properties of Grassland—Based on Machine Learning

草原作为分布最广的重要陆地植被类型之一，其放牧优化问题的研究可以为政府制定放牧政策提供科学依据。本文以内蒙古锡林郭勒草原为例，基于不同放牧策略下的监测点数据，首先运用岭回归模型、BP神经网络和XGBoost等机器学习算法研究不同放牧策略对该区域土壤性质的影响，其次建立不同深度土壤湿度值的预测模型，然后构建反映放牧强度对土壤化学性质影响的数学模型，最后根据历史数据对2022年土壤性质的相关指标展开预测。结果表明随着放牧强度的加大，植被生物量呈现先增后降的趋势，同时无牧和重牧也不利于植被生长；模型预测了锡林郭勒草原监测样地在不同放牧强度下2022年土壤同期有机碳、无机碳、全N、土壤C/N比的值，经验证拟合效果都较好。为提高模型的实用性，后续研究还应考虑到部分特征间的相关性。
As one of the most widely distributed major terrestrial vegetation types, the optimization of grazing in grasslands could provide scientific evidence for governments to formulate grazing policies. Taking the Xilin Gol Grassland in Inner Mongolia as an example and based on the monitoring data across sites under various grazing strategies, this study first leveraged machine learning algorithms including ridge regression, BP neural networks and XGBoost to investigate the impacts of different grazing strategies on soil properties in the region. Prediction models of soil moisture content across depths were then established. Afterwards, a mathematical model reflecting grazing intensity’s in-fluence on soil chemical properties was constructed. Finally, predictions were made with historical data on relevant indicators of soil properties in 2022. Results demonstrated that with the increase of grazing intensity, vegetation biomass first increased and then decreased, while no grazing and overgrazing were also detrimental to vegetation growth. The model predicted the 2022 values of soil organic carbon, inorganic carbon, total nitrogen and C/N ratio across monitoring sites in Xilin Gol Grassland under varied grazing intensities. Verification suggested relatively decent goodness of fit. To improve model applicability, future studies should consider correlations between certain features.