Estimation of Rape-Cultivated Area Based on Decision Tree and Mixed Pixel Decomposition
2021
Yang, Huan | Deng, Fan | Fu, Hancong | Zhang, Jiahua
The timely and accurate information of rape-cultivated area in Jianghan Plain is of great significance for crop yield estimation and food security in China. MODIS-EVI time series images were used to estimate rape-cultivated area of Jianghan Plain in 2015–2017 based on decision tree and mixed pixel decomposition methods. By analyzing the features of EVI time series curves of different objects, rape distribution was preliminarily extracted by establishing the decision tree model; however, due to the problem of mixed pixels, there are heterogeneous objects in a single pixel, which will affect the extraction accuracy of crop area by decision tree method. To solve this problem, according to EVI spectral curve of the endmembers, the linear spectral mixture model was adopted to further improve the precision of crop extraction, and rape-cultivated area was extracted accurately by using computed rape endmember abundance. The results showed that the average estimation accuracy of rape area was 96.6% with city-level R² (coefficient of determination) greater than 0.93. It demonstrated that the vegetation index time series data with medium resolution can accurately extract the rape-cultivated area, which can provide reference for rape monitoring over a large scale.
显示更多 [+] 显示较少 [-]