基于随机森林的农耕区土壤有机质空间分布预测

Prediction for spatial distribution of soil organic matter based on random forest model in cultivated area

以陕西省周至县农耕区为研究区,采集192个土壤样品,通过随机森林模型(random forest,RF)对土壤有机质含量进行回归预测,通过29个(15%)独立验证点对预测结果进行精度验证,并与普通克里格(ordinary kriging,OK)和协同克里格(cokriging,COK)插值结果进行对比分析。结果表明,研究区土壤有机质含量在训练集和验证集中均属于中等变异性,含量处于中等偏低水平,大致表现为中、南部黑河东岸土壤有机质含量相对较高,东北部渭河沿岸含量较低。对变量重要性进行排序,影响研究区土壤有机质的主要因素为数字高程(DEM)和降水量。与OK、COK相比,RF对土壤有机质的预测值和实测值的相关系数(0.782)更高,而平均绝对误差(0.618 g·kg^(-1))和均方根误差(2.062 g·kg^(-1))更低,说明RF能够更精确地反映局部土壤有机质含量的空间变化信息。

In the present study, the cultivated area of Zhouzhi County in Shaanxi Province was selected as study area with 192 soil samples collected, and the soil organic matter （SOM） content and distribution were predicted based on random forest （RF） model. The prediction accuracy was verified by 29 （15%） independent verification points, and the results were compared with ordinary kriging （OK） and cokriging （COK）. It was shown that SOM contents in the training set and verification set were all moderately variable, and were classified into the medium low level. The SOM content was relatively high at east coast of Heihe River in middle and southern area, whereas was relatively low at Weihe River coast in the north-eastern area. By ranking the importance of variables, it was revealed that the main factors affecting the soil organic matter in the study area were elevation and rainfall. Compared with OK and COK, the correlation coefficient of prediction value and actual value of RF （0.782） was higher, yet the mean absolute error （0.618 g ·kg -1 ） and root mean squre error （2.062 g ·kg -1 ） were lower, which suggested that RF model yielded a more realistic spatial distribution of SOM.