Research On Cotton Planting Areas Extraction Based On MODIS-EVI Time-series Data Base

Remote sensing technology has been widely applied to cultivatedland identification, yield estimation and growth condition monitoring. Butwhen it comes to the different crops condition monitoring in cultivatedland, the research is not mature enough yet. Crop information extractionis the basis of crop monitoring. The acquisition of high resolution dataalways needs a long cycle and it also needs a large amount of data andmoney to cover large cultivated scale. Therefore, it is hardly to extractcrop planting information with high resolution data sets. It is vital to getlarge regional scale crop spatial distribution and monitoring crop growthcondition in time.In this paper，with Xinjiang Province as the research site，Ireconstructed the MODIS EVI time-series data using S-G filter method.After processing the datasets, the reconstructed EVI time-series curve canrepresent cotton key growth period preferably. And then, we choose thefield investment sample points and compare the Euclidean distance todivide Xinjiang province into three identification subareas. Throughextracting an ideal time-series EVI curve of cotton in each subarea, wegot the planting area extraction results. This study also used the fieldsurvey plots in the year of2010as regions of interest, checked out theaccuracy of the plant distribution of cotton in2010using confusionmatrix. Through comparing the classification result and tested data, the average precision increased to91%. Taking Xinjiang cotton as theresearch object, this article is aimed at exploring a new method based onmedium resolution data available to large regional scale crop plantingarea extraction. The conclusions are as follows:（1） In this paper, we evaluated the clustering accuracy of threedistance measure methods(i.e., Euclidean distance, spectral informationdivergence, correlation coefficient and spectral angle cosine-Euclideandistance) based on the shape and amplitude characteristics of the EVItime-series data. The results show that the correlation coefficient and SSSmethod which fully utilizes the curve shape and the amplitudecharacteristics of the EVI time-series data shows the highest clusteringaccuracy among the three methods.（2）In Xinjiang, we used to treat mount Tianshan as the boundary ofsouth and north Xinjiang. By analyzing the data, we find a distinctphenomenon: MODIS-EVI time-series curves are in transition changes inmiddle Xinjiang (Aksu, northern Kashi). Therefore, finding the newboundary of south and north Xinjiang to extract ideal cotton time-seriescurve is the basis of cotton extracting. Taking transition changes inmiddle Xinjiang into consideration, this paper come up with an idea thatdivide Xinjiang into three subareas and the middle one is transitionalzone.Based on this idea, we choose Euclidean distance as the unique index. After stacking the primary subarea data and EVI time-series datatogether, we extract20sample plots out of the field investigation plots.Among them, there are10plots in middle Xinjiang. The central densesample points are to identify the boundary of transitional zone. At last, wedivided Xinjiang into three classification subareas. These subareas notonly can be used to improve the accuracy of extraction, but also provide amethod to monitor crop phenology regulations.