土壤有机碳空间预测与驱动解释研究进展

Advances in Spatial Prediction and Driver Interpretation of Soil Organic Carbon

摘要: 土壤有机碳空间预测是土壤资源管理和碳储量评估的基础，其应用价值取决于预测精度、区域适用性及解释依据。本文采用叙述性综述方式，梳理数字土壤制图的模型方法、环境协变量与尺度效应，重点讨论验证策略、过程融合和管理信息的利用。以改进RothC与随机森林融合的已发表研究为例，分析模拟样本传递、尺度匹配和校准验证的技术环节；结合分类、重复观测与层级数据，比较管理信息的建模路径。在此基础上，提出连接应用目标、数据条件、模型选择和验证方案的决策流程。现有证据尚不足以支持统一的算法优劣排序，空间分块验证与基于概率抽样的地图精度评价也不能相互替代。对于碳储量变化和水土保持应用，需要统一土层与质量核算口径，检验预测区间和适用域，并以重复观测或试验设计区分预测关联与管理效应。

Abstract: Spatial prediction of soil organic carbon (SOC) supports soil resource management and carbon stock assessment. Its value depends on predictive accuracy, regional applicability and the evidence underlying interpretation. This narrative review examines modelling approaches, environmental covariates and scale effects, with emphasis on validation, process-data integration and management information. A published study combining a modified RothC model with random forest illustrates the transfer of simulated training samples, scale matching, calibration and validation. Methods for categorical, repeated-measurement and hierarchical management data are compared. An application-oriented workflow then links study objectives and data availability to model selection and validation. Existing evidence does not support a universal ranking of algorithms. Spatially blocked validation and probability-based map accuracy assessment address different inferential targets and are not interchangeable. Applications to carbon stock change and soil and water conservation require consistent soil depths and mass accounting, evaluation of prediction intervals and areas of applicability, and repeated observations or experimental designs that distinguish predictive associations from management effects.