Search

Plastic Mulching (PM) is criticized for causing environmental problems in spite of the benefits to increase crop yield. Accurate map of Plastic-Mulched Farmland (PMF) is demanded to monitor the usage of plastic mulches and to mitigate its environmental damages. However, mapping PMF with remote sensing data over large scale is still technically challenging due to its variable spectral characteristic. This paper proposed a Random Forest (RF) workflow to map PMF by coupling multi-resource satellite observations, including high- and medium- resolution optical imageries and Synthetic Aperture Radar (SAR) imageries. Under the context of two typical PMF regions, Jizhou and Guyuan, China, we test the performance of the proposed method. Results indicated that coupling optical and SAR remote sensing data improved the mapping accuracy significantly. There were different strategies for optimal data coupling. Coupling of three data, which are the GaoFen-1 (GF-1) data, Landsat-8 data and Radarsat-2 data, generated the highest mapping accuracy. Besides, coupling the GF-1 data with SAR data produced better accuracy than coupling the Landsat-8 data with SAR data. Compared with the accuracies derived from the coupling of two data (such as coupling GF-1 data with Radarsat-2 data and coupling Landsat-8 data with Radarsat-2 data), coupling the three data improved the overall accuracies by 0.39 to 4.93%. Compared with the mapping accuracy produced from Radarsat-2 data alone, the overall accuracies generated from the optimum coupling increased by 20.96 and 26.06%, respectively, in Jizhou and Guyuan. When compared with the mapping accuracy of GF-1 data alone, the overall accuracies generated from the optimum coupling improved by 4.77% both in Jizhou and Guyuan. In general, contribution of optical remote sensing data was greater than that of SAR data for mapping PMF. Whereas, the SAR data also made great contribution to improve the mapping accuracy of PMF. [ABSTRACT FROM AUTHOR]