The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLIII-B3-2020
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B3-2020, 259–265, 2020
https://doi.org/10.5194/isprs-archives-XLIII-B3-2020-259-2020
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B3-2020, 259–265, 2020
https://doi.org/10.5194/isprs-archives-XLIII-B3-2020-259-2020

  21 Aug 2020

21 Aug 2020

EXPLOITING SENTINEL-1 SAR TIME SERIES TO DETECT GRASSLANDS IN THE NORTHERN BRAZILIAN AMAZON

A. F. Carneiro, W. V. Oliveira, S. J. S. Sant'Anna, J. Doblas, and D. V. Vaz A. F. Carneiro et al.
  • Dept. of Remote Sensing, Brazilian National Institute for Space Research, São José dos Campos, Brazil

Keywords: Remote sensing, Time series data, SAR, Image classification, Grassland detection, Sentinel-1

Abstract. Recent advances in cloud-computing technologies and remote sensing data availability foster the development of studies based on the analysis of optical and SAR imagery time series. In this paper, we assess the potential of Sentinel-1 imagery time series for grassland detection in the northern Brazilian Amazon. We used the Google Earth Engine cloud-computing platform as an alternative to obtain and analyse Sentinel-1 imagery, acquired from 2017 to 2018 over the region of Mojuí dos Campos/PA, Brazil. We extracted several temporal metrics from the imagery time series and used the Random Forest algorithm to perform the classification. In addition, we analysed the time series considering different channels, including the VV and VH polarizations, both separately and in combination, and the CR, RGI and NL indices. We could efficiently discriminate areas of grasslands from forest and agricultural crops using either VH time features or features extracted from the combination of both VV and VH polarizations. The classification map that resulted from the combination of VV and VH data presented the highest accuracy, with an overall accuracy of 95.33% and a 0.93 kappa index. Despite simple, the approach adopted in this paper showed potential to differ grasslands from areas of agriculture and forest in the northern Brazilian Amazon.