基于随机森林的县域土壤有机碳密度及储量估算

Estimation of soil organic carbon density and reserves based on random forest model in county level

采用2009年云南省玉龙县土壤调查数据,基于土壤类型法将所测样点的土壤有机质含量转换为有机碳密度,经克里金插值进行空间化,再以2009年landsat7-Level2影像及SRTM 90 m数字高程模型数据为基础,提取归一化植被指数、亮度、绿度、湿度、坡度、坡向、曲率等与土壤有机碳形成密切相关的解释变量;通过随机森林模型模拟土壤有机碳密度及其空间分布,基于有机碳密度估算出0~20 cm表层土壤的有机碳总储量,并对两种模拟结果进行误差分析。结果显示:克里金和随机森林的估算结果分别为2.4×10^(8)和1.7×10^(8)t,均方根误差分别为20.77和14.11,普通克里金插值误差较大,且对采样点数量及空间分布有较强的依赖性;随机森林模型不仅能处理高维数据,还能给出多个变量的重要性,估算结果精度更高,也更接近区域实际情况,对小尺度的细节表现更佳,适于地形复杂且样点有限的县域土壤有机碳密度及其储量的估算。

Based on the soil survey data of Yulong county,Yunnan province in 2009,the soil organic matter content of the samples was converted to organic carbon density based on the soil type method,spatialized by Kriging interpolation,and then 2009 Landsat7-Level2 image and Shuttle Radar Topographic Mission 90 m DEM data were used to extract the explanatory variables which are closely related to soil organic carbon formation,such as NDVI,brightness,greenness,humidity,slope,aspect,and curvature.The density and spatial distribution of soil organic carbon were simulated through a random forest model,and the total organic carbon storage of 0~20 cm surface soil was estimated based on organic carbon density.The errors of the results by the two methods were analyzed.The results showed that the estimated results of Kriging and random forest were 2.4×10^(8)and 1.7×10^(8)t,respectively and the root-mean-square errors were 20.77 and 14.11,respectively.The ordinary Kriging interpolation error was larger and strongly depended on the number and spatial distribution of sampling points.The random forest model can not only handle high-dimensional data,but also calculate the importance of multiple variables.The accuracy of the estimation result by the random forest model is higher and closer to the actual situation of the region,and the performance of small-scale details is better and suitable.It is used to estimate the density and storage of soil organic carbon in counties with complex terrain and limited sample points.