Spatial distribution prediction of soil pH in arable land of Jiangxi Province based on multi-source environmental variables and the random forest model
-
摘要: 研究多源环境变量条件下随机森林(random forest,RF)模型和普通克里格(ordinary Kriging,OK)模型对大尺度耕地土壤pH值空间预测的性能差异,研究分析RF模型对提升土壤pH值预测精度的参考价值。以中国江西省为研究区,基于气候、地形和植被等环境协变量信息,结合土壤属性和耕地利用条件,利用RF模型对江西省耕地土壤pH值进行空间预测,识别土壤pH值空间变异的影响因素,并与OK模型计算精度进行对比。结果表明,增加土壤属性和耕地利用条件作为环境变量的RF-A模型预测耕地土壤pH值的精度优于以地形、气候、植被属性作为环境变量的RF-B模型和OK模型的预测结果,气候因素是决定土壤pH值的最重要因素,地形地貌因子和人为因素对pH值变异也有重要影响。研究结果表明该方法对提升大尺度耕地土壤pH值预测制图精度具有一定的理论和现实意义。Abstract: This study aims to compare the accuracy of random forest(RF) and ordinary kriging(OK) model for predicting spatial distribution of soil pH in arable land of Jiangxi Province using different covariates combination, and assess the feasibility and potential of RF method for improving the prediction accuracy of soil pH value. The RF algorithm is used to predict the pH value of cultivated soil in Jiangxi Province based on environmental covariate such as climate, topography and vegetation, combined with soil properties and cultivated land use conditions, identify the main influencing factors. The results produced by the RF was compared with the classical OK interpolation model. Our results showed that the accuracy of RF-A model with soil properties and cultivated land use conditions as environmental variables is better than that of RF-B model which only including terrain, climate and vegetation attributes as environmental variables. Climatic condition is the dominate factor which control the spatial variation of soil pH. the topographic factors and anthropogenic factors also have essential effect on spatial variability of soil pH. Thus, this study proved RF method has theoretical and practical significance for improving the accuracy of soil pH prediction at large-scale.
-
Key words:
- arable land /
- soil pH /
- random forest /
- Jiangxi Province /
- influencing factor
-
-
计量
- 文章访问数: 59
- PDF下载数: 5
- 施引文献: 0
下载: