Duas décadas de mudanças dos manguezais de meso e micromarés do litoral brasileiro a partir de imagens multisensores

Abstract

Mangroves are coastal environments that extend along the tropical and subtropical global coastal regions, whose constant monitoring is hampered by its large-scale distribution. With the advent of new computational technologies supported by remote sensing (Google Earth Engine - GEE), this problem has been partially solved. However, some limitations still persist, for example, the use of an image library using only optical sensors, making it difficult to map mangrove forests in areas frequently covered by clouds. Thus, this work aims to evaluate the classification and changes in the mangrove areas of the meso and micro-tidal regions of the Brazilian coastal zone in the last two decades through multi-remote sensor data (optical and microwaves) from geographic object-based image analysis (GEOBIA). Multitemporal images from the Landsat, Alos PalSar, JERS SAR and SRTM series were used. The remote sensing dataset were processed according to the GEOBIA approach, which determines the reduction of an image in homogeneous regions (objects) by grouping sets of pixels with similar characteristics. As a result, it was observed that in 1996 and 2016 the area under study contained 2625,38 km² and 2898,26 km² of mangrove areas, respectively. This demonstrates an increase of 273 km² in mangrove areas. From the analysis of the change detection, it was observed a total increase of 684.55 km², a reduction of 411.7 km² and an unchanged area of 2213.70 km² of mangrove. The classification was validated through statistical analysis of two error matrices (2008 and 2016). The 1996 error matrix presented overall accuracy = 0.92; Kappa index = 0.84; and Tau index = 0.84. For the year 2016, overall accuracy = 0.93; Kappa index = 0.85; and Tau index = 0.85. On the other hand the error matrix for change detection showed an overall accuracy of 78.43%, with a quantity disagreement of 11.86% and an allocation disagreement of 9.71%. Quantifications of mangrove loss was 414 ± 43 km², gains was 590 ± 48 km² and remained mangrove unchanged was 2305 ± 60.3 km². These results demonstrate the effectiveness of using object-oriented classification for mapping and analyzing mangrove dynamics on a large scale. The products obtained in this research can serve as a basis for future work on the dynamics of mangroves, contributing to the improvement of management and preservation of this important ecosystem.