首页 >  2015, Vol. 19, Issue (6) : 998-1006

摘要

全文摘要次数: 2615 全文下载次数: 63
引用本文:

DOI:

10.11834/jrs.20154257

收稿日期:

2014-11-17

修改日期:

2015-05-16

PDF Free   HTML   EndNote   BibTeX
星地多源数据的区域土壤有机质数字制图
1.浙江大学环境与资源学院农业遥感与信息技术应用研究所, 浙江 杭州 310058;2.中国科学院南京土壤研究所土壤与农业可持续发展国家重点实验室, 江苏 南京 210008;3.中国农业科学院农业资源与农业区划研究所, 北京 100081
摘要:

土壤有机质(SOM)是全球碳循环、土壤养分的重要组成部分,精确估算土壤有机质含量具有重要意义。本文以中国东北-华北平原为研究区,收集了1078个土壤样本,以遥感数据(MODIS,TRMM和STRM数据)与土壤地面光谱数据为预测因子,运用基于树形结构的数据挖掘技术构建土壤有机质-环境预测因子模型进行数字土壤制图。通过不同建模样本数建模精度比较,选择300个样本数时的模型为最优模型。建模结果表明土壤光谱和气候因子是研究区SOM变异的主控因子,生物因子次之,而地形因子影响最小。预测结果经检验,RMSE为7.25,R2为0.69,RPD为1.53制图结果与基于第二次全国土壤普查数据的土壤有机质地图具有相似的分布规律,呈现SOM自东北向西南递减的趋势。通过比较分析发现,经过20年左右的土地开发与利用,研究区低SOM和高SOM含量土壤面积减少,而中等SOM含量土壤面积增加。

Regional scale mapping of soil organic matter using remote sensing and visible-near infrared spectroscopy
Abstract:

Soil Organic Matter(SOM) is one of the key variables in agronomy and environmental management. It controls soil fertility and has a significant impact on atmospheric CO2 concentration. Carbon sequestration in soil can not only reduce the emissions of the greenhouse gases but also improve the quality and productivity of soils. Therefore, accurate estimation of SOM distribution at large scale is needed for policy making, sustainable soil utilization and management. The aims of this study were to predict the SOM across the Northeast and North Plain in China using a model trees method with a large number of satellite-derived data and soil vis-NIR spectroscopy data. A total of 1078 soil samples were collected to estimate spatial variation of SOM in Northeast and North Plains, China. Remote sensing data, including MODIS, TRMM and STRM, and soil spectroscopy data were used as environmental predictors. 306 soil samples were used as external validation dataset and the others were used to build models. Decision-tree-based M5 algorithm was introduced to construct the prediction models between SOM and the environmental predictors through the modelling tool Cubist. The method converted to a series of rules, each with an associated linear model, that partition the data into regions with similar conditions defined by the characteristics of the predictor variables. Prediction models between SOM and predictors with different number of samples were tested and it was found that the optimal number of training samples is 300. For the evaluation on the validation dataset, the model showed an R2 of 0.69, RMSE of 7.25 g·kg-1 and RPD of 1.53. According to the S=f(s,c,o,r,p,t) function, it was noted in the predicted model that soil spectroscopy and climate predictors were the dominant factors in controlling the spatial variation of SOM, while organism predictors showed less importance and terrain factors had least impact. The predicted map showed a significant increasing trend of SOM content from southwest to northeast. Compared with the map produced by National Soil Survey Office, the predicted map presents similar pattern of the spatial variation of SOM in Northeast and North Plain in China. Nevertheless, the area of high SOM and low SOM decreases in about two decades due to the human activities and tillage in the area.The methodology in this study combines remote sensing with proximal soil spectroscopy using a rule-based soil mapping framework. The result shows that predicting SOM at large area is acceptable through Cubist. The climate factors and soil spectroscopy were the most dominant factors among the environmental factors while terrain factors contributed least. It is found that the spatial pattern of SOM generated by Cubist is consistent with that of the second national soil survey of China produced in early 1980s, meanwhile the area of high SOM and low SOM decreases.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮