利用MODIS-EVI时序数据对河南省土地覆盖进行分类

基于遥感的土地覆盖信息提取对农林业生产、环境监测具有直接的应用意义。该文选取河南省作为研究区域，利用2005年23个时相的MODIS EVI数据，结合农作物物候历、其他分类参考数据及河南省种植结构的相关文献，在对时序数据进行去云、平滑处理后，采用SVM（支持向量机）分类器，对河南省的土地覆盖进行分类。分类结果上，一方面参照2005年河南省农作物种植面积的统计数据得到面积精度，对大面积种植的农作物：小麦为81.47%、玉米94.87%、水稻82.43%；经济作物：油菜39.81%、大豆93.65%、棉花95.21%、花生74.27%；另一方面，参照2000年1：10万全国土地覆盖的分类数据，将2000年的对应数据和分类结果分别归并为：耕地、林地、草地、水体、建筑进行精度比较，结果表明总体识别精度为78.07%，Kappa系数为0.6556。从分类精度验证来看，表明MODIS植被指数时序数据及该文研究方法在农作物信息提取中的有效性。

Land-cover classification for Henan Province with time-series MODIS EVI data

The significance to agriculture, forestry production and environment monitoring is obvious that extraction of land cover information based on remote sensing data. So in this study, focusing on Henan Province, making use of MODIS 16-days composite EVI data at 2005, combining with crop phenology and other reference land cover data, the land-cover classification for Henan Province was performed. The raw EVI data was processed with cloud removing and smoothing, then the support vector machine (SVM) method was adopted for the classification. Refer to the classification result, compared with the statistics of crops acreage of Henan Province in 2005, the area accuracy of classification result was as following: for large-area planted crops, wheat got 81.47%, corn 94.87%, rice 82.43%; while for the economic crops, rape was 39.81%, soybean 93.65%, cotton 95.21%, peanut 74.27%. On the other hand, combining the classified land cover type into 5 types, farmland, woodland, grassland, water body, urban and built-up. The results were further compared with 1:100 000 land cover map which was produced by using the Landsat ETM+ and TM data in 2000. The overall accuracy and Kappa coefficient were 78.07% and 0.66, respectively. It turns out that the feasibility of MODIS time-series VI data and classification strategies adopted to extract crops information.