Large-Scale Crop Mapping From Multisource Remote Sensing Images in Google Earth Engine

Abstract

Large-scale crop mapping is vitally important to agriculrural monitoring and management. However, traditional methods cannot well meet the needs of large-scale applications. Therefore, this study proposed a method for large-scale crop mapping based on multisource remote sensing images. To be specific, 1) harmonic analysis was conducted on normalized difference vegetation index time-series derived from moderate resolution imaging spectroradiometer images and synthetic aperture radar backscattering coefficient time-series derived from Sentinel-1 data, respectively, extracting harmonic-derived phenological features and harmonic-derived backscattering features, and then combined with spectral features from Landsat-8 and Sentinel-2 images to construct the final multisource feature set for crop classification; 2) it employed prior constraints of crop dominance and cropland distribution to reduce misclassifications in large scale crop mapping; and 3) the whole process was conducted on the Google Earth Engine online platform, which can reduce the computational burdens caused by the spatiotemporal data. In the experimental study, we evaluated three crops, including wheat, rapeseed, and corn in Qinhai in 2018, based on the classification and regression tree classifier. The results show that the Jeffries-Matusita distances between crop samples are close to 2, and the overall accuracy is 84.25%. Furthermore, this study found that the distribution of the crops in Qinghai is associated with climate, topography, and cultivation habits.