基于MODIS时间序列的松嫩平原盐渍地提取

Saline Land Extraction Using MODIS NDVI Time Series Data in Songnen Plain

松嫩平原是我国内陆盐渍土三大分布区之一,土壤盐渍化是该区最主要的环境问题。以多时相中分辨率成像光谱仪(MODIS)的归一化植被指数(NDVI)时间序列影像为主要数据源,通过Savizky-Golay滤波重构NDVI时序数据,依据研究区7种主要土地覆被类型的时间序列曲线差异性,应用分类回归树(Classification And Regression Tree,CART)方法确定像素归属类别,得到松嫩平原2013年盐渍土的分布数据;并基于不同盐渍化程度土壤的植被物候特征差异性建立CART决策树区分不同程度盐渍土。分类结果为:盐渍地掩膜提取精度达98.13%,Kappa系数为0.83;不同程度盐渍土识别的精度达到86.08%,Kappa系数为0.78。该研究表明多时相MODIS数据在大尺度盐渍土信息识别中具有可行性。

Songnen Plain is one of China′s inland saline soil distribution area and its salinization degree has aggravated,soil salinization is the main environmental problem in this area.It is necessary to monitor the saline soil information of Songnen Plain.Firstly,we applied Savitzky-Golay filter to multi-temporal MODIS NDVI data,the 16 day composite product with 250 meter resolution,to reconstruct NDVI time series curves.Based on the difference of NDVI curves of different land cover types,CART method was used to extract saline soil of Songnen Plain.Secondly,we also established a CART decision tree to identify different degrees of saline soil by using 11 phonological parameters.In the process of classification,the maximum value of NDVI plays the most important role to identify saline soil information which represents the strength of vegetation photosynthesis.It also explains that the significant difference of NDVI curves is valuable to distinguish saline soil information when NDVI at higher levels.Results showed that the accuracy for extract saline soil of Songnen Plain is 98.13%.At the same time,Kappa coefficient is0.83.The accuracy for identify different degrees of saline soil is 86.08%,and Kappa coefficient is 0.78.This study indicates that the applying of CART method to MODIS time series data is feasible to acquire salt information in a large scale.