Object-Based Classification of the veredas wetland macrohabitat using multispectral imagery from a Remotely Piloted Aircraft System.

Abstract

Abstract Veredas are wetlands of relevant ecological and social value that may be closely related to the maintenance of the water regime of the springs. Remotely Piloted Aircraft Systems (RPAS) have proved to be great allies in the space-time monitoring of wetlands. This study evaluates the effectiveness of multispectral sensors attached to an RPAS to discriminate habitats from paths through the Object-Based Image Analysis (OBIA) approach. Multispectral camera overflights were performed on September 25, 2020 (dry) and January 28, 2021 (wet). Radiometrically corrected orthomosaics were generated with five spectral bands. Multiscale segmentations were applied, and later the classification by the OBIA approach through the classifier of the nearest neighbor, the results were post-processed by applying the algorithm of a class assignment. The classification separated the objects into 14 and 12 classes with an overall accuracy of 92.21% and 88.01% (kappa 0.92 and 0.87), for September and January, respectively. Among these, are the phytophysiognomies of Cerrado stricto sensu (surrounding) and Gallery forest (centralized), in addition to eight classes of habitats in the vereda. The multispectral sensor was sensitive to differentiate these habitats in the vereda and the occurrence of areas covered by the pteridophyte Dicranopteris flexuosa, its distribution, and physiological stages. The classification of two seasonal seasons made it possible to characterize the behavior of habitats according to water availability. The multispectral sensor on board the RPAS is a powerful tool to determine the diagnosis and management of wetlands, contributing to the establishment of public policies for the conservation of vereda environments.