Radargrammetric approaches to the flat relief of the amazon coast using COSMO-SkyMed and TerraSAR-X datasets
MetadataShow full item record
The Amazonian coast consists of extensive flood plains and plateaus characterized by a high discharge of water and sediment from the Amazon River. This wide landscape occurs under a tropical climate with heavy rains and high cloud cover, making it unsuitable for conventional mapping based on optical images. Additionally, the flat relief and vegetation structure of the Brazilian Amazon coast define an incoherent to partially coherent behavior for the microwave signal, rendering radargrammetric models more suitable for the three-dimensional mapping of its surface. This study aimed to assess the digital surface models (DSMs) provided by Cosmo-SkyMed (CSK) and TerraSAR-X (TSX) Stripmap datasets throughout the radargrammetric models from SARscape and Toutin. The DSMs were generated from SAR (synthetic aperture radar) data with an acquisition geometry that addressed the need for a compromise between the intersection angles and low temporal decorrelation. The radargrammetric SARscape and Toutin's models were developed from different amounts of stereo ground control points (SGCP). The generated DSMs were evaluated considering a set of 40 independent checkpoints (ICP) measured by GNSS in the field, in their entirety and disaggregated by coastal environment. The vertical accuracy was based on the estimation of the discrepancies, bias and precision (standard deviation and root mean square error – RMSE), and the Taylor and Target diagrams were used for a more comprehensive comparison. In the vertical accuracy analysis using all ICPs measured in situ, the DSM obtained by the SARscape's model from the CSK SAR data resulted in the lowest RMSE (4.34 m) and mean discrepancy (0.05 m), but Toutin's model had the lowest standard deviation (2.58 m) of the discrepancies. The Taylor and Target diagrams showed fluctuations in accuracy that alternated the DSMs generated from the two types of SAR data, indicating that TSX produced more stable models and CSK produced better vertical accuracy. The Amazon Coastal Plateau and Fluvial Marine Terrace environments defined three-dimensional representations with lower RMSEs (better than 7.8 and 8.9 m, respectively), regardless of the type of SAR data or the radargrammetric model used. The worst performance, which was for the Fluvial Marine Plain, was influenced by the specific characteristics of this coastal environment, such as the structure of the mangrove vegetation and the shoreline. In general, the high resolution and good ability to revisit the SAR data used, together with the radargrammetric models, allowed for the accurate mapping of the flat relief of the Amazon coastal environments, providing detailed spatial information that can be acquired in severe rainfall conditions in a region of intense morphological dynamics.