CAJ | 학술논문

　　为了解决MODIS 数据中普遍存在的混合像元问题，该文利用2008年和2009年多时相的MODIS13Q1影像，以经过优化的N-FINDR算法进行线性混合像元分解提取冬小麦种植面积，各省的误差均控制在正负4%左右。利用同期多时相的HJ-1星分类数据作为参考值，在试验区域选择14个均匀分布的样区验证混合像元分解结果。结果显示6个样区的相对误差在10%以内，其余8个样区的误差基本在15%左右。该研究可为冬小麦种植面积的监测提供参考。
　　위료해결MODIS 수거중보편존재적혼합상원문제，해문이용2008년화2009년다시상적MODIS13Q1영상，이경과우화적N-FINDR산법진행선성혼합상원분해제취동소맥충식면적，각성적오차균공제재정부4%좌우。이용동기다시상적HJ-1성분류수거작위삼고치，재시험구역선택14개균균분포적양구험증혼합상원분해결과。결과현시6개양구적상대오차재10%이내，기여8개양구적오차기본재15%좌우。해연구가위동소맥충식면적적감측제공삼고。
Winter wheat is one of the main food crops in the north of China. It is significant to monitor winter wheat planting areas for China’s grain policy and economic planning. The MODIS products are outstanding with the characteristics of large area coverage, frequent repeat, and free access to download. It offers a valuable application on long-term and large-area detection of winter wheat. Because of the coarse spatial resolution of MODIS products, the mixed pixels become the common problem existing in MODIS data. Therefore, it is necessary to solve the problem of mixed pixels in crop area extraction with MODIS data, In this study, we chose the Huanghuaihai Plain (including Hebei province, Shandong province, Henan province, Beijing, and Tianjin) as the study area, and used multi-temporal MODIS data in 2008 and 2009 to extract the winter wheat area with an optimized N-FINDR algorithm and linear unmixing method. In a traditional N-FINDR algorithm, all pixels in the image would be traversed to find the pixel group that can form a simplex with the maximum area. The optimized N-FINDR algorithm we used simplifies the procedure by finding the points set that can form a triangle with the maximum area in a two-dimensional plane composed by any two bands first, then the vertex of the triangle are taken as candidate endmembers, and final endmembers are obtained by traversing all the candidate endmembers. In order to find points set in a two-dimensional plane, we used the convex hull property of a polygon with rotating calipers. This optimized algorithm can improve time complexity from O(n3) to O(n2). Comparing this with national statistical data in 2009, the relative error of the extracted winter wheat planting area was less than 4%for each province. The results showed that the method we used was applicable for winter wheat area extraction on a large scale. In order to further validate the results, we selected 14 sample areas, and multi-temporal HJ-1 data at same period were taken to produce the winter wheat planting map as a reference for each sample area. The validation results showed that the spatial distribution of the unmixing results agreed with the classification maps of HJ-1. The relative error of winter wheat planting area was less than 5%for 5 sample areas, and larger than 15%for 4 sample areas. The error was relatively larger for the sample areas located in the urban area and the mountain area. The error was mainly caused by the error of endmember extraction, the internal difference of the winter wheat phenology and spectra for large area, the fragmentation of crop land, and the complexity of the land surface in the mountain area.