Estimation method of straw burned area based on multi-source satellite remote sensing

Abstract: Straw burning often occurs in June every year in Huang-Huai region of China, which results in the wasting of agricultural resources, atmospheric pollution and traffic accident. Straw-burning area is one of important factors for estimating the impact of straw burning. In China, there are many researches for monitoring straw-burning hot spots and estimating burned areas. However, there are two problems in estimating straw-burning area efficiently in Huang-Huai region: one is the mixture of multiple types of underlying surfaces in the pixel coverage with middle or lower resolution, such as FY-3/MERSI, EOS/MODIS with 250 m resolution which may contain cropland, water body, residential land and other types in one pixel; and the other is fast speed of cropland ploughing that may be only one or two days after harvest or straw burning according to the field survey, which causes the difficulty in the estimation by using land resource satellites such as TM, SPOT and HJ/CCD. So, it is difficult to use single type of satellite data for estimating the straw-burning area in Huang-Huai region. To solve this issue of time sensitivity of estimating straw-burning area, a new method was presented in this paper, which used multiple satellites data including FY-3/MERSI and GF-1. FY-3/MERSI has three times of revisiting period per day and lower resolution (250 m), and GF-1 has high resolution (16 m) and long revisiting period (four days). The new method combined the advantages in each type of satellite. FY-3/MERSI data was used to acquire the straw-burning scar region timely, and GF-1 data was used to provide detailed distribution of multiple underlying types, especially the ratio of cropland in a pixel. The estimation was based on spectral analysis of satellite data and mixed pixel separation technology. There were two important parameters provided in this new method. One was cropland ratio, and the other was burned extent. Land cover information of study region including cropland, surface water and residential area was classified based on GF-1 data, using the decision-tree classification method. The accuracy of classification reached 90%. The cropland ratio was calculated by the above land cover classification data. The information derived from GF-1 could improve the accuracy of straw-burning area estimation by providing the accurate ratio of cropland in a pixel. Straw-burning scar region could be extracted utilizing the near infrared band of FY-3/MERSI data which is more sensitive than other MERSI bands to distinguish the straw-burning scar region according to the spectrum analysis. The burned extent of straw could be estimated based on the difference between the near infrared bands before and after the straw burning. During the process of mixed pixel separation, selecting the pure end-member pixels of burned and un-burned cropland was the critical step because the end-member value may be different to different observation images, so dynamically deciding the end-member value based on searching window should be more reasonable. Furthermore, in order to simplify the method of estimating burned extent using two images in different observations, the equation was given to estimate burned extent based on just one image observed after the straw burning, which could avoid the errors resulted from the different atmospheric conditions of two-time images. After getting the two parameters, the straw-burning area could be calculated. Using this method, this paper estimated the straw-burning area in Zhengyang and Pingyu County, Henan Province on June 6, 2014. The result of estimation was validated by GF-1 data, and the accuracy reached more than 94%, which indicated that the method was effective.