Study On Mapping For Winter Wheat Distribution In China Based On Sentinel-1&2 Images

Winter wheat is one of China’s main food crops.Timely and effective access to winter wheat planting area is a key link to ensure national food security.The fragmentation degree of Chinese winter wheat is high,and high-resolution images such as Sentinel provide a new important data source for remote sensing identification of winter wheat in China.However,in the wide range of geospatial scales,winter wheat remote sensing identification still faces the following three problems:(1)Uncertainty of winter wheat image characteristics caused by significant differences in crop phenology on a wide range of geospatial scales,inconsistent image imaging time,and uncertainty of image trend periods caused by cloud rain;(2)China winter wheat plot fragmentation is high,there is still no quantitative study on the effect of different spatial resolution images on winter wheat recognition accuracy;(3)Winter wheat has similar spectral characteristics and timing with rapeseed and garlic.It is difficult to accurately solve the scientific problem that winter wheat is easy to be confused with rapeseed and garlic.Therefore,this study takes China’s main winter wheat production area(106°E ～ 122.8°E,28°N ～ 42°N)as the research area,based on Google Earth Engine cloud computing technology,synthesizing and coupling Sentinel-1 microwave image and Sentinel-2 Optical image,we accurately produced winter wheat distribution data of 2018 in China’s main producing areas with a spatial resolution of 10 m.The overall accuracy reached 96%,the kappa coefficient was 0.92,the user accuracy of winter wheat was 95.21%,and the producer accuracy was 97.61%.This paper studies and solves the main scientific problems or problems faced by winter wheat remote sensing identification on a wide range of geospatial scales,and has the following innovative results:1.Synthetic time series Sentinel-2 optical image to solve data redundancy and enhance winter wheat image information.Based on the phenological characteristics of winter wheat and MODIS-EVI data,the geospatial differentiation characteristics of winter wheat phenology in the study area were obtained.The study area was divided into four sub-areas,namely 28°N～32°N and 32°N～35°N.35°N～38°N,38°N～42°N.Obtain the effective time window of Sentinel-2 EVI image synthesis for each sub-area.For example,in the range of 38°N～42°N,the EVI high-value time window is 2017.11.1～2018.4.30,and the EVI low-value time window is 2017.10.1～2017.10.30 and 2018.6.10～2018.7.10.Defines the synthesis rules for Sentinel-2 EVI images,i.e.,the Sentinel-2 EVI maximum for the synthetic timing during the EVI high-value period,and the Sentinel-2 EVI minimum and median for the synthetic timing during the EVI low-value period.The method effectively reduces the redundancy problem of remote sensing data,significantly enhances the image information of winter wheat,and provides an important data basis for the selection of representative training samples and the improvement of classification accuracy.2.The argument indicates that Sentinel image has obvious advantages in the field of winter wheat remote sensing recognition in China.In this paper,the effects of spatial resolution images of 10 m,30 m,250 m and 500 m on the recognition results of winter wheat were quantitatively analyzed on the scale of winter wheat main producing areas in China.Based on the 10 m spatial resolution Sentinel imagery,the winter wheat identification results were compared with the statistical data.The errors of the five major producing provinces of Henan,Shandong,Anhui,Hebei,and Jiangsu were 2.06%,0.00%,1.21%,3.23%,and 4.10%,respectively.Accuracy evaluation was carried out through 72 1 km × 1 km verification samples.The overall accuracy of winter wheat identification of 10 m,30 m,250 m and 500 m spatial resolution images were 96.19%,89.87%,71.25%,and 64.48%,respectively.Winter wheat recognition accuracy decreases as spatial resolution decreases.Sentinel images can accurately identify the boundary information of winter wheat plots with the advantages of high spatial and temporal resolution.The misclassification rate or leak rate of winter wheat identification is less than 5%,which has obvious advantages for remote sensing identification of high-break winter wheat plots in China.3.Coupled active and passive remote sensing data to solve the problem of foreign matter homology and improve the accuracy of winter wheat remote sensing recognition.The Sentinel-1 microwave image was sensitive to the response of vegetation plants.By analyzing the Sentinel-1 spectroscopy curve of winter wheat,it was found that the valley value of winter wheat appeared in April,and the minimum values of VH and VV bands were lower than-20 d B and-15 d B,respectively.Garlic peaked,and the maximum values of VH and VV were greater than-15 d B and-10 d B,respectively.According to the definition of sequential Sentinel-1 image synthesis scheme,the image features of winter wheat in synthetic images are extracted: the VH and VV bands in the synthetic images in April are less than-15 d B and-9.5d B respectively;the difference between the April composite image in the VH band and the composite image in May is greater than-3 d B.Based on this,the winter wheat-rape mixed area Hubei province and winter wheat-garlic mixed area Shandong Jinxiang County were used as research area,and the threshold classification method was used to optimize the winter wheat recognition results based on Sentinel-2 optical image and improve the accuracy of winter wheat remote sensing recognition.