Abstract:Abstract. Mapping plastic-mulched landcover (PML) is an important agricultural monitoring task. Because of its daily revisiting, imagery from moderate-resolution imaging spectroradiometer (MODIS) has been widely used to detect PML over a large area. However, the coarse spatial resolution and small field size make subpixel PML a problem for accurate PML mapping from MODIS. This study applies the improved spatial attraction model (ISAM), which estimates the spatial attraction of the central subpixel of a moving window by all subpixels in the window, to map large-scale subpixel PML from MODIS imagery. The linear spectral mixing model is used to obtain fractions of PML and three other landcover classes in each MODIS pixel as the inputs to ISAM for obtaining the hard subpixel PML maps at spatial resolution of 31.25 m. The accuracy evaluation, with validated landcover classification derived from Landsat-8 imagery as the ground truth, shows that overall accuracy, Kappa coefficients, producer accuracy, and user accuracy are 83.51%, 0.69, 90.97%, and 81.51%, respectively, indicating large-scale PML mapping at spatial resolution comparable to Landsat-8 can be derived from MODIS images with acceptable accuracy by ISAM. This study provides a practical and economic way for mapping PML for a large area at ∼30-m spatial resolution.

Large-scale Subpixel Mapping of Landcover from MODIS Imagery Using the Improved Spatial Attraction Model