Area Extraction Of Winter Wheat Using MODIS Imagery Data In North Of Anhui Province

Food security is one of the most important factors to maintain long-term stability and economy development of the country.Winter wheat is one of the most important food crops in northern China.Anhui Province is the main production area of winter wheat in China.It is vital to obtain winter wheat information including spatial distribution and area for strengthening management of agricultural apartment and furtherly formulating scientific and rational food production policies.At present,remote sensing technology is widely applied in agriculture with the advantages such as easy data acquisition,timeliness and wide coverage.In this paper,the six cities(Suzhou,Bozhou,Fuyang,Bengbu,Huainan and Huaibei)located in the north of Anhui Province were used as the research area.And the MODIS with the seven 500 meter(m)reflectance bands and the 8-day composite period was selected as the main data source.The time range for the remote sensing data was the growing season of 2016-2017.We employed three methods to extract the winter wheat acreage in the study area which included decision tree,MODIS-NDVI time series waveform matching and mixed pixel decomposition.Then,the quantity and spatial distribution precision of extraction results were validated based on the statistical data and the crop distribution calculated by the Landsat-8.The results are as follows:(1)The Savitzky-Golay filtering method was used to filter the time series and to eliminate the rough and abnormal jitter caused by adverse noises,such as weather conditions and observation angles.Therefore,the processed images could better reflect the growth trend for winter wheat.Considering the uneven greenness distribution caused by the concentration degree difference of different winter wheat planting regions,we divided planting area to 4 parts according to the average NDVI in winter from the county-level administrative units.This method took imbalanced planting density of winter wheat in different regions into full account and was convenient for extracting winter wheat.(2)According to the diversity characteristics of winter wheat’s greenness during growing season,screening rules was established by reasonable NDVI threshold and main phenological period trend based on the specificity of winter wheat and other vegetation in greenness and its trend.The planting area and spatial distribution of winter wheat were both extracted by the decision tree method.Comparing these results and the winter wheat planting area recorded in the statistical yearbook,the results indicated that the overall error was 25.7%and the error ranges of most counties were 10%-30%.In the study area,seven verification samples were randomly selected to test the extraction accuracy based on the distribution of winter wheat from Landsat8-OLI images.The omission error and commission error were from 5.91%to 16.63%and 12.27%to 31.13%,respectively.The Kappa coefficient range was from 0.43 to 0.62,and most samples were above 0.5.The results showed that this method was suitable for extracting planting area of winter wheat.(3)Referencing the hyperspectral data analysis method spectral Angle matching(SAM).The ENVI spectral hourglass tool was employed to process time series data.The typical time series profile of winter wheat endmembers were abstracted by minimum noise fraction rotation,pixel purity index calculation and n-dimensional visualization.The SAM classification algorithm was adopted to select the winter wheat pixels.Next,the quantity of winter wheat planting area at the county level was counted and compared with the sown area of that in the designated year.The results showed that the overall error was 20.44%and the error of most counties were less than 30%.The range of omission error and commission error were from 11.37%to 38.97%and 13.33%to 28.20%,respectively.The Kappa coefficient range was from 0.23 to 0.52,and most of the samples were above 0.43.This result was worse than that of the decision tree method.(4)Based on the typical time series profile characteristics of winter wheat,the fully-constrained least squares(FCLS)method in the mixed pixel decomposition model was used to extract the winter wheat planting area.Compared with the statistical data,the results showed that the overall error was 24.92%,and most of the counties were between 15%and 25%.The total precision of winter wheat was higher than those computed by the decision tree method.This method estimated the area abundance of winter wheat in any pixel from the sub-pixel scale and weakened the problem of a large number of mixed pixels caused by moderate spatial resolution,which is the character-istics of this paper.The research results of this paper were highly meaningful,which could provide reference to master the wheat planting information,guide agricultural production and formulate relevant policies in time for agriculture and govenment departments.