GeoPearl: SOM-content under arable land 1000m

Model predictions on SOM-content under arable land( Report: https://edepot.wur.nl/556312). Predictions with a model ensemble at fixed depths of 15, 45, 80 and 120 cm. Predictions using an ensemble of Random Forests and Gradient Boosting, with covariates from the SoilGrids 2017 model (Hengl et al. 2017) and a spatial cross-validation procedure. The model was applied on log-transformed observations of SOC. Models result were back- transformed and converted to SOM by multiplying SOC predictions with 1.743. Prediction

该数据集针对耕地土壤有机质（Soil Organic Matter, SOM）含量开展模型预测（报告链接：https://edepot.wur.nl/556312）。本次预测采用模型集成（model ensemble）方法，针对15 cm、45 cm、80 cm及120 cm四个固定土层深度展开；预测模型为随机森林（Random Forest）与梯度提升（Gradient Boosting）的集成模型，输入协变量取自土壤网格2017（SoilGrids 2017，Hengl等，2017）模型，并采用空间交叉验证（spatial cross-validation）流程。研究中将模型应用于经对数变换的土壤有机碳（Soil Organic Carbon, SOC）观测数据，最终模型预测结果经反对数变换后，通过将土壤有机碳预测值乘以1.743，转换为土壤有机质（SOM）含量。预测