Study On The Method Of Region Land-cover Classification Using Time Series NDVI Date

Time series NDVI data plays an important role in the remote sensing of land-cover classification. The time-series data record the changes in plant growth process. Analyzing the different time series NDVI data can improve land-cover classification accuracy, if the spatial resolution can get certain level. The research methods that time series NDVI data is used in the land-cover classification was discussed, based on the 1999 12 months'SPOT4-VEGETATION/NDVI dataFirstly, based on the NDVI data to classify the land-cover directly. Combined the 12 months'SPOT-VEGETATION/NDVI data and carried though the Principal component analysis. An unsuperised classification method, the ISODATA algorithm, was employed to identify the vegetation types based on the four most important components. According to the proposed land-cover classification system, the Asian Land Cover Database of 30 seconds which produced by the Japan's Chiba University and the China vegetation map , mergered the results of the unsuperised classification method. Got the result of the land-cover classification.Latterly, classified the land-cover baed on the time series arctgTS/NDVI data. NDVI and Ts are two parameters to descripton the land cover'characteristic. The change tendency is opposite in the process of vegetation landcape from densely to sparse. Thus is opposite to independent employment NDVI or the Ts, synthesizes the Ts and the NDVI infotmation can classify and appraisal the land cover accurately and can reflect an greater criterion land cover spaceand time change rules effectively. Constructed the Ts-NDVI space by the SPOT-VEGETATION'the time series NDVI images and the NOAA'the time series Ts images. In order to enable the data result all to maintain at the first quadrant, asks the arctg function to Ts/NDVI, obtains the arctgTS/NDVI remotesensing image. In same with the NDVI image processing method,carried on the unsupervised classification method to the forth principal components of the arctgTS/NDVI remote sensing image, also can get 8 kind of land cover types.Separately calculates two kind of classified results'error matrix by ENVI software. The classified precision respectively are 82.57% and 85.3%. And then calculate the Bhattacharya distance. The results showed that, the time series arctgTS/NDVI data has the higest precision, the time series NDVI data has the minimum accuracy. In the classification results, water has the higest precision. The Bhattacharya distance between other land covers could get 2.00. The classification accuracy is more than 90%. The results of separating distance among the woodland, cultivated crops and seasonal green vegetation are not exactly. The phenomenon of confusional classification is in evidence. The woodland, cultivated crops and brush need further classification.The time series analysis presented in this paper may have a good value on research about how to make the best of time series remote sensed data and land cover classification. Distill 16 metrics from the SPOT4-VEGETATION/NDVI data and NOAA/ Ts data. The metics reflect the seasonal variation. We futher classify the reasearch area based on the 16 metrics.