黑土区土壤有机质和全氮含量遥感反演研究  

Remote Sensing Inversion of Soil Organic Matter and Total Nitrogen in Black Soil Region

作　　者：郑淼 王翔 李思佳[2] 张丽[1] 宋开山[2] Zheng Miao;Wang Xiang;Li Sijia;Zhang Li;Song Kaishan(College of Tourism and Geographical Science,Jilin Normal University,Siping 136000,Jilin,China;Northeast Institute of Geography and Agroecology,Chinese Academy of Sciences,Changchun 130102,Jilin,China)

机构地区：[1]吉林师范大学旅游与地理科学学院,吉林四平136000 [2]中国科学院东北地理与农业生态研究所,吉林长春130102

出　　处：《地理科学》2022年第8期1336-1347,共12页Scientia Geographica Sinica

基　　金：中国科学院战略性先导科技专项(XDA28050400)资助。

摘　　要：以东北典型黑土区耕地为研究区,以Sentinel-2A(全球环境与安全监测计划的第二颗卫星,于2015年6月23日发射)影像作为数据源,构建光谱指数,分别采用多元逐步线性回归(Multiple Stepwise Linear Regression,MSLR)和随机森林(Random Forest, RF)算法建立土壤有机质(SOM)和土壤全氮(STN)预测模型,并采用十折交叉验证方法评估模型的性能。研究对比分析了不同气候、土壤类型和地形下土壤有机质和全氮的空间分布差异。研究表明:①海伦示范区的SOM和STN含量最高,其年均温最低,高程最高,年降水量多,SOM含量升高,其年均温最低,年降水量多,STN含量升高;②与基于多元逐步线性回归算法建立的SOM和STN预测模型相比,随机森林算法建立的SOM和STN预测模型,有着更高的精度和稳定性;③运用RF算法建立的SOM反演模型的R^(2)为0.96,均方根误差为5.49 g/kg,STN反演模型的R^(2)为0.95,均方根误差为0.27 g/kg;④不同示范区统一建立SOM和STN预测模型,有助于提高预测精度,实现跨区域建模与制图。Soil Organic Matter(SOM) and Soil Total Nitrogen(STN) provide nutrients for plant growth,and they are important indexes for soil quality evaluation. In this study,the soil sample data were obtained from the cultivated land in the typical black soil area of Heilongjiang, Jilin and Liaoning provinces. The Sentinel-2A images were used as data sources to calculate spectral indices, and SOM and STN prediction models were established by Multiple Stepwise Linear Regression(MSLR) and Random Forest(RF) algorithm respectively, and the performance of the models was evaluated by 10-fold cross-validation. The spatial distribution differences of SOM and STN under different climate, soil types and terrains were compared and analyzed. Results showed that: 1) The contents of SOM and STN were the highest in Hailun demonstration area, anf low annual temperature, high elevation and annual precipitation lead to increasement of SOM, and low annual temperature and high annual precipitation lead to increasement of STN. 2) Compared with SOM and STN prediction models based on MSLR algorithm, SOM and STN prediction models based on RF have higher accuracy and stability. 3) The R^(2) of SOM inversion model established by RF algorithm is 0.96 and RMSE is 5.49 g/kg, while the R^(2) of STN inversion model is 0.95 and RMSE is 0.27 g/kg. 4) The unified establishment of SOM and STN prediction models in different demonstration areas is helpful to improve prediction accuracy and realize cross-regional modeling and mapping.

关 键 词：土壤有机质(SOM) 土壤全氮(STN) 多元逐步线性回归 随机森林 Sentinel-2A 

分 类 号：S127[农业科学—农业基础科学]