Land use/land cover characterization in Amazonia using COSMO-SkyMed multitemporal images

The use of radar imagery is an alternative source of information to support the monitoring of the Amazon region, since the optical images have imaging limitations in tropical areas due to the occurrence of clouds. Therefore, the goal of this study is to analyze the radar images in X-band multi-temporal polarized obtained by COSMO-SkyMed satellite (COnstellation of small Satellites for Mediterranean basin Observation), in the intensity mode, isolated and/or combined with textural information, to thematic characterization of land use/land cover in the Humaitá, Amazonas State region. The methodology used includes: analysis of the dual images obtained during two subsequent acquisitions, in order to explore the potential of the dataset as a quad-pol intensity; extraction of textural attributes from the co-occurrence matrix (Gray Level Co-occurrence Matrix) and subsequent contextual classification; statistical assessment of the thematic performance of the intensity and textural images, isolated and in polarized groups. Within the results achieved, the group formed only by the intensity images presented a better performance if compared to those containing the textural attributes. In this discrimination, the classes involved were forest, alluvial forest, reforestation, savannah, pasture and burned areas, yielding 66% overall accuracy and a Kappa value of 0.55. The results showed that X band images, from COSMO-SkyMed, StripMap mode (Ping-Pong), multi-polarized, presents a moderate potential to characterize and monitor the dynamics of land use/land land cover in the Brazilian Amazon.

radar; forest mapping; textural analysis; monitoring