Endmember abundance calibration method for paddy rice area extraction from MODIS data based on independent component analysis

There is a large discrepancy between the actual abundance of land cover and the result derived from the endmember abundance calibration of Independent Component Analysis (ICA). In order to solve this problem, this paper proposed a new method for the endmember abundance calibration of ICA by combining regression analysis. The new method includes 3 steps: Firstly, decomposing the remote sensing time-series data to obtain the independent component of the object feature. Secondly, selecting a certain amount of samples with actual object feature abundance and then building the relationship between the actual abundance and derived independent component using regression analysis. Finally, using regression relationship to derive the abundance of object feature for each pixel. Based on the MODIS time-series data, the new method and the linear scaling method were applied in Xinghua county, Jiangsu province of China for the mapping of rice abundance. The results derived from these two methods were then compared with the actual rice abundance map of the study area. Results showed that the Root Mean Square Error (RMSE) and Bias of the rice map derived from the new method was all smaller than that by linear scaling method, where the determination coefficient (R2) was all higher than that by linear scaling method at different spatial scales. The new method can enhance the application of ICA model in crop acreage mapping and provide a basis for large-scale crop identification and area extraction.