Cosmo-SkyMed and TerraSAR-X datasets for geomorphological mapping in the eastern of Marajó Island, Amazon coast

• Autor(a) principal: Silva Guimarães, Ulisses [UNESP]
• Data de Publicação: 2020
• Outros Autores: de Lourdes Bueno Trindade Galo, Maria [UNESP], da Silva Narvaes, Igor, de Queiroz da Silva, Arnaldo
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1016/j.geomorph.2019.106934; http://hdl.handle.net/11449/199625
• topic: Amazon coastal environments; Random Forest; Synthetic aperture radar

The Amazon coast is marked by the high discharge of sediments and freshwater, macrotidal influence, a wide continental shelf, extensive flood plains and lowered plateaus which make it unique as a delta and estuary landscape. Further, the tropical climate imposes heavy rains and incessant cloudiness that render microwave systems more suitable for cartography. This study proposed to recognize and map the Amazon coastal environments through the X-band Synthetic Aperture Radar, provided by Cosmo-SkyMed (CSK) and TerraSAR-X (TSX) systems. The SAR datasets consisted of interferometric and stereo pairs, restricted to single-revisit and obtained with small interval (1–11 days), under steeper (θ < 35°) and shallow (θ ≥ 35°) incidence angles, and during the rainy and dry seasons. From the 4 acquisitions of X-band SAR data, attributes such as the backscattering coefficient, coefficient of variation, texture, coherence, and Digital Surface Model (DSM) were derived, adding each variable in 5 scenarios. These combinations resulted in 20 models, which were submitted individually to the machine learning (ML) classification approach by Random Forest (RF). The backscattering and altimetry described the coastal environments which shared ambiguity and high dispersion, with the lowest separability for vegetated environments such as Mangrove, Vegetated Coastal Plateau and Vegetated Fluvial Marine Terrace. The coherence provided by interferometry was weak (<0.44), even during the dry season, in the other hand, the cross-correlation was strong (>0.60), during the rainy and dry season showing more suitability for radargrammetry on the Amazon coast. The RF models resulted in Kappa coefficient between 0.39 to 0.70, indicating that the use of X-band SAR images at an incidence angle greater than 44° and obtained in the dry season is more appropriated for the morphological mapping. The RF models given by TSX achieved the higher mapping accuracies per scenario of SAR products, in order of 0.48 to 0.63. Despite this, the best classification was carried out by 19 RF model with 0.70, provided by CSK in shallow incidence composed by intensity, texture, coherence and stereo DSM. The CSK and TSX data allowed to map the Amazon coast precisely, based on X-band at single polarization, high spatial resolution and revisit, which has demonstrated the support for detailed cartography scale (1:50,000) and frequent updating (monthly up to yearly).