基于随机森林模型的旺业甸实验林场土壤全氮数字制图  

Digital mapping of soil total nitrogen in Wangyedian experimental forest farm based on random forest model

作　　者：甄诚 王海燕[1] 雷相东[2] 赵晗 董齐琪 崔雪 仇皓雷 ZHEN Cheng;WANG Haiyan;LEI Xiangdong;ZHAO Han;DONG Qiqi;CUI Xue;QIU Haoei(College of Forestry/State Key Laboratory of Efficient Production of Forest Resources/Key Laboratory for Silviculture and Conservation of Ministry of Education,Beijing Forestry University,Beijing 100083,China;Institute of Forest Resource Information Techniques,Chinese Academy of Forestry,Beijing 100091,China)

机构地区：[1]北京林业大学林学院/林木资源高效生产全国重点实验室/森林培育与保护教育部重点实验室,北京100083 [2]中国林业科学研究院资源信息所,北京100091

出　　处：《华中农业大学学报》2024年第3期249-257,共9页Journal of Huazhong Agricultural University

基　　金：国家科技基础资源调查专项(2021FY100801);亚太森林组织项目。

摘　　要：为探究林场土壤全氮含量的空间分布特征及对环境因素的响应,以旺业甸实验林场为研究区,采用随机森林模型和Cubist模型建立了不同土层深度(0~10、10~30、30~50cm)土壤全氮含量与环境协变量(海拔、归一化植被指数、年平均降水量、年平均气温、y坐标和坡向)之间的定量关系模型,对该区土壤全氮含量进行预测并制图,并分析了影响土壤全氮空间变异的控制性因素。研究结果显示:0~10、10~30、30~50cm土层实测全氮含量的均值分别为3.20、2.02、1.47g/kg,土壤全氮的平均含量随土层深度的增加而降低;3个土层深度土壤全氮预测随机森林模型的决定系数R2分别为0.59、0.42和0.39,均优于决定系数R2分别为0.56、0.38和0.34的Cubist模型,2种模型预测精度都随土层深度的增加而降低,各环境因素对土壤全氮空间分布的影响作用随土层深度的增加而减小;从随机森林模型土壤全氮预测图来看,不同土层深度土壤全氮含量均呈现西部、北部和中部低,西南、东南和东部高的空间格局,不确定性图显示随机森林模型预测土壤全氮含量分布具有较低的标准差;海拔对土壤全氮含量的影响最大,其他依次为:归一化植被指数>年平均降水量>年平均气温>y坐标>坡向。结果表明,随机森林模型可以作为有效预测该林场不同土层深度土壤全氮含量的方法。To explore the spatial distribution characteristics of soil total nitrogen content in forest farm and its response to environmental factors,random forest model and Cubist model were used to establish a quantitative relationship between soil total nitrogen content and environmental covariates including elevation,normalized difference vegetation index,mean annual precipitation,mean annual temperature,y-coordinate and aspect at soil depths of 0-10,10-30 and 30-50 cm in Wangyedian experimental forest farm.Soil total nitrogen content of the area was predicted and mapped,and the controlling factors affecting the spatial variation of soil total nitrogen were analyzed.The results showed that the average content of soil total nitrogen at soil depth of 0-10,10-30 and 30-50 cm was 3.20,2.02 and 1.47 g/kg,respectively.It decreased with the increase of soil depth.The results of cross-validation showed that the R2 of the random forest model for predicting soil total nitrogen at the three soil depths was 0.59,0.42,and 0.39,respectively,better than the R2 of Cubist model with 0.56,0.38,and 0.34,respectively.The prediction accuracy of both models decreased with the increase of soil depth.The influence of various environmental factors on the spatial distribution of soil total nitrogen decreased with the increase of soil depth.From the prediction map of soil total nitrogen with the random forest model,the content of soil total nitrogen at different soil depth showed a spatial pattern of low in the western,northern,and central regions,and high in the southwestern,southeastern,and eastern regions.The uncertainty map showed that the random forest model had a low standard deviation in predicting the distribution of the content of soil total nitrogen.The elevation had the greatest impact on the content of soil total nitrogen,followed by the normalized difference vegetation index>mean annual precipitation>mean annual temperature>y-coordinate>aspect.It is indicated that the random forest model can serve as an effective method for predicti

关 键 词：土壤全氮 随机森林模型 空间分布 土层深度 数字土壤制图 

分 类 号：S714[农业科学—林学]