Estudo espacial multi-resolução e espectro-temporal subpixel no mapeamento do uso e cobertura do solo Aline Warnke Hipólito Pelotas,

Abstract

The aim of this study was to evaluate the influence of the spatial resolution of remotely sensed images in land use land cover (LULC) mapping as well as the percentage of rice and soybean crops within the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor pixels in their Enhanced Vegetation Index (EVI) spectro-temporal profile. To do so, we map 64 LULC targets in the municipality of Tapes-RS during the 2015/16 crop year over a high spatial resolution image (0.5m) from Google Earth which was used as reference map. Subsequently, we map those targets over a set of images at different spatial resolution, which were: Landsat/Enhanced Thematic Mapper plus (ETM+) (30m), Sentinel-2A/MultiSpectal Scanner (MSI) (20 and 10m), and CBERS-4/Pan (5m). Besides, we map rice and soybean fields over six Landsat images through a careful visual multitemporal interpretation which were assessed with MODIS vegetation index product (MOD13Q1) images in order to evaluate the EVI spectro-temporal profile for different percentages of these crops within the MODIS pixel in comparison with pure pixels. The spatial analyses showed an overall area error of -1.9, -0.94, -0.13 and -0.18% for ETM+, MSI 20, MSI 10, and Pan images, respectively. In general, mapping and area errors decreased as the image spatial resolution improved and the target size increased. The percentage of rice and soybean within each MODIS pixel highly impacts their spectro-temporal profiles, mainly below 60%. Moreover, 72.7% of rice pixels were pure or showed spectro-temporal profile similar to pure ones. For soybean, only 31.1% of pixels were pure and showed unique spectro-temporal profile, mainly below 60% of the crop within MODIS pixel.