Remote sensing estimation of planting area for winter wheat by integrating seasonal rhythms and spectral characteristics

Abstract: Research on winter wheat has an important significance for timely and accurately obtaining the crop acreage and their spatial distribution at regional and national scales. In traditional methods combining medium-resolution and low-resolution remote sensing data, only the area percentage of crops in a low-resolution pixel is extracted, thus the crop area is obtained. For this limitation, this paper proposes a new crop identification method. The land cover of the study area is summarized in six categories (farmland, forestland, shrub land, grassland, waters, and other). Each type of land cover's purity is calculated in the corresponding MODIS pixel. First, NDVI time series curves are extracted for various types of land cover based on MODIS time advantage, analyzed for identifying characteristics of winter wheat on the seasonal rhythm, and used to build the identification model. Then, MODIS pixels are classified based on the purity of farmland, including farmland pure pixel, other crop pure pixel, mixed pixel from farmland and other land cover, mixed pixel from winter wheat and other crops, and other pixel. The MODIS pixels involving winter wheat include three types, i.e. the farmland pure pixel, mixed pixel from farmland and other land cover, mixed pixel from winter wheat, and/or other crops. For the farmland pure pixels, the winter wheat is identified according to seasonal characteristics of winter wheat. For the mixed pixel from farmland and other land cover, their sub-pixel NDVI time series are extracted based on the pixel un-mixing method, in order to identify whether the sub-pixel belongs to winter wheat. Further, the identification results are repositioned to the medium-resolution scales according to the spatial relationship. The mixed pixels area from winter wheat and other crops are identified based on spectral differences of Landsat TM remote sensing images. Finally, these three types of identified results can be integrated into the medium-resolution scales. In this paper, the winter wheat identified method is applied to the dominating agricultural area of the Yiluo basin. A total of 11 016 MODIS farmland pure pixels with 250 m spatial resolution, corresponding 1 101 600 farmland pixel with 25 m spatial resolution, were identified as winter wheat; 18 630 MODIS mixed pixels integrating farmland and other land cover, corresponding 882 192 farmland pixels, were identified as winter wheat; 10 275 MODIS mixed pixels integrating winter wheat and other crops, corresponding 595 296 farmland pixels, were identified as winter wheat. Winter wheat acreage of our study area is 161 193.00 hm2. By random sampling, the identified results of winter wheat show an accuracy of 96.3%. The error rate is 2.79% compared with statistical data of Yearbook. The superiority of this identified method, compared with the other methods combining medium-resolution and low-resolution remote sensing data, is that not only was the acreage of crops accurately extracted, but also its spatial distribution was determined at the medium-resolution scales. This paper provides a new way to solve problems for extraction of crop cultivation area and spatial distribution information. It can be applied not only to the identification of winter wheat, but also has important reference value for the identification of other types of crops.