The most important task of carrying out the research on land use/land cover changes is to obtain accurate land cover data timely and efficiently. In recent years, many researches have been carried out to improve the accuracy of land cover classification, but they mainly focused on the aspects of land cover classification methods, classification features, classification systems and reducing noise pollution of remote sensing images. There were also some researches in which time series NDVI (normalized difference vegetation index) was directly applied to classify land cover or detect change of land cover. However, the researches that enhance classification accuracy for low classification precision areas on the basis of the existing land cover products using time series NDVI have not been found yet. In this paper, Henan Province in central China was taken as the study area, the regions possessing low classification precision with the existing MODIS land cover product (MOD12Q1) were selected, and the study on improving classification accuracy in these regions was carried out by means of the construction of time series NDVI database and the measurement of similarity of time series NDVI curves. The results are as follows: 1) Time series NDVI is an important feature for land cover classification, and there is a strong correlation between time series NDVI and land cover categories. The overall accuracy of the land cover data extracted based on time series MODIS-NDVI database in the study area was as high as 79.57% (sample evaluation based on field survey) and 73.45% (comparative evaluation with land use map at scale of 1:100 000) respectively. It is demonstrated that it is feasible to improve the accuracy of land cover classification by means of time series NDVI. 2) It was shown from a practical application that classification accuracy of existing land cover products could be significantly improved by using time series MODIS-NDVI database. The overall classification accuracy of the new land cover product, which was obtained by revising original MCD12Q1 based on time series MODIS-NDVI database, increased by 10.29% (from 72.76% to 83.05%, comparative evaluation) and 17.20% (from 64.52% to 81.72%, sample evaluation) respectively. 3) Degree of accuracy improvement for various land cover categories was different. The producer's accuracy for forest land, grassland, cultivated land, artificial land surface and water body increased by 35.36%, 29.51%, 2.98%, 6.96% and 6.11% respectively. 4) With regard to 2 specific methods that were used for judging similarity of NDVI curves in time series, the minimum distance (MD) method is superior to spectral angle matching (SAM) method in general. In conclusions, it is an effective method of improving precision of existing land cover products to retain parts of land cover information with high precision and reclassify the parts with low precision using time series NDVI. The classification accuracy of land cover products can be significantly improved by revising them based on time series NDVI database.%为提高MODIS土地覆盖产品的分类精度,该文以河南省为试验区,首先将MODIS土地覆盖产品(MCD12Q1)分为高精度区域和低精度区域,然后通过构建时序NDVI图谱库并利用图谱曲线相似性测定方法,改进MCD12Q1低精度区域的分类精度.结果表明:1)时序 NDVI 是土地覆盖的重要分类特征,二者之间具有较强的关联性.2)利用时序NDVI图谱库能够明显提高MODIS土地覆盖产品的分类精度,改进后的MCD12Q1的总体分类精度分别由72.76%(比较评价)、64.52%(样本评价)提高到83.05%和81.72%.3)不同土地覆盖类别精度提高的程度不同,林地、草地、耕地、人工地表以及水体的生产者精度分别提高35.36%、29.51%、2.98%、6.96%和6.11%.4)对于判定时序NDVI曲线相似度的2种具体方法而言,最小距离法(minimum distance, MD)总体上优于光谱角度匹配法(spectral angle mapper, SAM).综上,保留现有土地覆盖产品中分类精度较高的部分,基于时序NDVI图谱库改进分类精度较低的部分,是提高现有土地覆盖产品分类精度的有效方法.
展开▼
