A Soil Bulk Density Pedotransfer Function Based on Machine Learning: A Case Study With The Kellogg Soil Survey Laboratory Database

Soils data from 41,878 horizons were extracted from the Kellogg Soil Survey Laboratory (KSSL) database and used to calibrate and validate the PTF. Environmental datasets included terrain attributes (elevation, slope, aspect, landform), national land cover classification, hierarchical ecosystem land classifications, and 19 bioclimatic indicators. The results of a 5‒fold cross-validation scheme showed that average root mean squared prediction error (RMSPE) was 0.13 g cm-3, and mean prediction error (MPE) was -0.001 g cm-3.

本研究从凯洛格土壤调查实验室（Kellogg Soil Survey Laboratory, KSSL）数据库中提取了41878个土壤发生层的相关数据，用于校准与验证土壤传递函数（Pedotransfer Function, PTF）。所用环境数据集涵盖地形属性（海拔、坡度、坡向、地貌类型）、全国土地覆盖分类数据、分级生态系统土地分类数据以及19项生物气候指标。通过5折交叉验证方案得到的结果显示，平均均方根预测误差（root mean squared prediction error, RMSPE）为0.13 g·cm⁻³，平均预测误差（mean prediction error, MPE）为-0.001 g·cm⁻³。