This dissertation examines the potential of data generated from SAR sensors to differentiate between mangrove genera using polarimetric data, biomass estimates based on forest structure, and by producing thematic mangrove forest maps, using multi-frequency SAR images. Polarimetric SAR data (Radarsat2), multi-frequency (bands X, C and L) were used to achieve this goal. The study area is the Ajuruteua Peninsula, located in the eastern sector of the Amazon Coastal Zone. Mangroves in this region are considered preserved with the following species present: Rhizophora mangle, Avicennia germinans, Avicennia schaueriana and Laguncularia racemose, where R. mangle is the dominant species in this region. Through the polarimetric image Radarsat-2, band C, the polarimetric response of the parcels, dominated by these two principal genera (Rhizophora e Avicennia), were analyzed. Through the analysis of the polarimetric response, we were unable to define a pattern per genus to allow for the differentiation of the genera using this parameter. Analysis of the scattering mechanism was conducted through polarimetric decomposition of the coherence and covariance matrix. In the H-α plane, all vegetation parcels were classified as zone 5 of average entropy, attributed to vegetation scattering, and as zone 6, also of average entropy, associated to an increase in the surface-level roughness. Only the field class presented an H value lower that the other classes, standing out from the others. The images of scattering mechanisms: double-bounce, volumetric and surface-level did not permit the separation of dominant genera in the region. Using annual time series images Sentinel1-A, the behavior of the σ° polarizations VV and VH were similar and did not present differences in relation to total biomass values. Variations in radar backscatter over the year were related to environmental conditions (precipitation and tidal regime), canopy alterations (phenology) and incidence angle. The σ° values were greater during the month of May to the end of August, and in the same period, σ°VH/σ°VV rate values were lower, which is a reflex of mangrove canopy dynamics in the region given that during this period leaf litter production is greater, despite modest oscillation (1.5 dB). The use of multi-frequency SAR images for Random Forest classification of the environments on the peninsula resulted in the best Kappa index, 0.53, for the model that included the Sentinel1-A and ALOS-PALSAR images, with a Kappa mangrove class of 0.90. The dwarf mangrove class presented a global disagreement (up to 10%) higher than the others, principally in types exchange with mangrove classes, hypersaline plains and others, being that the last two had the lowest Kappa indices per class in all the classifications. The mangrove class showed a global disagreement maximum of 5% and a Kappa index greater than 0.90 in all classifications. Based on the above considerations, we conclude that the mangrove genera approach, even with a greater n sample, did not produced significant differences to distinguish genera through polarimetric techniques, be them polarimetric responses or scatter techniques. Two advances can be identified in the study of mangroves using time-series Sentinel-1A: σ°VH/σ°VV polarization rate is related to canopy dynamics, and the use of trimester images, representing different seasons for the classification of mangrove forests. In addition, the potential of SAR images for mapping mangrove forests using time-series images from C and L bands through machine learning techniques is also recognized.
|
