Unmixing Of Modis Data Mixed Wheat Identification And Area Estimation

The wheat production in the northern of China occupies an important position in the national grain annually production. Grasping sown area, growing situation and information of the wheat in time, is not only useful to strengthen the management of wheat and excavate productivity, but also significant to assist government department making the grain policy and making sure the food security. During the recent decades, the remote sensing has developed into the integrated technical disciplines. As the front technology of the earth observing system, it has the objective and good current situation features, which could survey the large area of the earth surface information in short time. On the base of the above superiority, the remote sensing is the directly and economy methods of crop watch surveying in the large area. The key of the wheat area extracting is the crop identified. Unfortunately, there are many constraints would impact the precision of classification, such as sensor resolution, surface features diversity, and mixed pixels existing.Linear spectral mixture model has the good features of mature, operable and good fitting. In this paper, we choose it to decompose and extract the wheat sown area by the MODIS mixed pixels. In the past decomposing model, researcher used the reflectance as the endmember. In contrast with the traditional model, we try to guide the Normalized Difference Vegetation Index (NDVI is for short), Enhance Vegetation Index (EVI is for short) and Difference Vegetation Index (DVI is for short) as the new factor. Through the precision evaluate system, EVI is the optimal mixed pixels decomposing factor.He Ze is the main wheat production base which is in the southwest of Shandong province. The plain is the main terrain and with the warm temperate continental monsoon climate. The suitable geographic location supplied the favorable situation for plant growing. So as the important role of the agricultures production, we choose He Ze for the research area. Considering of the hyperspectra, high time resolution and moderate spatial revolution, the MODIS data is the best choice. Within the step of the atmospheric correction,"Bow-tie"eliminates, geometric correction, mask and image fusion, we could build the time series vegetable index of the research area. The key role of the mixed pixels decomposing is the endmember and pixel purity choosing. Through the analysis of vegetable index curve changed in different step of the growing, we could discover that every feature have their own discipline. Then we try to guide the decision tree algorithm to find the threshold of the classification. This method would help us purifying the endmember for the vegetable index time series.At the same time, we use the method of the supervised classification to extract wheat growing area by Landsat5 TM image, resized and operated the result by pixel accumulation analysis. Then we could establish the evaluate index of mixed pixels decomposing precision. Within a large number of experiments, we found the following conclusions: (1) the kinds of surface features would deeply impact the precision. The decompose precision of single surface features is better than complex. (2) The result precision of decomposing which the endmember has been purified is better than never. The total and pixel decompose precision are separately increased by 6.32% and 2.60%. (3) In the aspect of the total precision, the endmember is purified by EVI time series could increase the precision by 5.12% than only use the traditional pixel reflectance model, which is also better than base on the NDVI 7.88% and DVI 1.45%. (4) In the aspect of the pixel precision, the mixed pixels decomposing by EVI time series could increase the precision by 2.83% than only use the traditional pixel reflectance model, which is also better than base on the NDVI 5.27%. But it is less than the result of DVI by 1.45%. Overall, using the EVI time series is the best method of the endmember purified and mixed pixels decomposing.By joint Chinese-German technical project (2007DFB70200) and Shandong provincial natural science foundation (Y2008E10),in view of crop watch needing, this paper studied on searching the efficient, operable and high precision business running model of the wheat identified and sown area extracted by hyperspectra. The research of this project has an important academic and practical role. It is not only could bring the more social and economic benefits, but also would promote the agriculture information survey modernization.