基于辅助变量的紫色土耕地土壤有机质空间预测  

Soil Organic Matter Prediction of Purple Soil Based on Auxiliary Variables

作　　者：刘雅璇 于慧 罗勇[3] LIU Yaxuan;YU Hui;LUO Yong(Institute of Mountain Hazards and Environment(IMHE),Chinese Academy of Sciences,Chengdu 610299,China;Faculty of Land Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China;Faculty of Geography and Planning,Chengdu University of Technology,Chengdu 610059,China)

机构地区：[1]中国科学院·水利部成都山地灾害与环境研究所,成都610299 [2]昆明理工大学国土资源工程学院,昆明650093 [3]成都理工大学地理与规划学院,成都610059

出　　处：《土壤》2024年第4期857-865,共9页Soils

基　　金：国家自然科学基金项目(41971273);四川省地质调查研究院财政资金项目(SCIGS-CZDXM-2024014)资助。

摘　　要：研究收集了川中丘陵区域紫色土耕地共135个土壤样本,基于GEE(GoogleEarthEngine)云平台调用高分辨率Sentinel-2A数据、SRTMGL1v3.0高程数据、Soil Grids土壤属性数据,并创新性地加入了纹理特征作为辅助变量,分别通过梯度提升决策树(GBDT)和随机森林(RF)构建两种预测模型反演研究区土壤有机质。结果表明:研究区内紫色土耕地土壤有机质含量偏低,养分级别为二~六级;GBDT算法构建的模型相比于RF算法预测精度更高,R2、r、RMSE分别为0.687、0.829、5.668 g/kg和0.514、0.717、6.765 g/kg;加入纹理特征的模型R2分别增加了6.80%和1.70%,为土壤有机质预测研究提供了新的思路。This study collected a total of 135 samples from purple soil farmlands in the hilly region of central Sichuan.Based on the GEE cloud platform,high-resolution Sentinel-2A data,SRTMGL1v3.0 elevation data,and SoilGrids soil attribute data were invoked,and texture feature data was innovatively added.Two prediction models were constructed by using gradient enhancement decision tree(GBDT)and random forest(RF)to invert SOM.The results showed that SOM content of purple soil farmlands in the study area was relatively low,with the level ranging from 2 to 6 levels.The models constructed by GBDT algorithm had higher prediction accuracy(R2=0.687,r=0.829,RMSE=5.668 g/kg)compared to RF algorithm(R2=0.514,r=0.717,RMSE=6.765 g/kg).The R2 with texture features increased by 6.80%and 1.70%,respectively.TGIS study can provide a new scientific approach for SOM prediction.

关 键 词：土壤有机质 机器学习 紫色土 GEE 

分 类 号：S127[农业科学—农业基础科学]