Estimação de biomassa e caracterização estrutural de áreas restauradas da Mata Atlântica com dados 3D obtidos por fotogrametria aérea digital

Abstract

Deforestation for agriculture and livestock in the Atlantic Forest biome is the main generator of disturbed and/or degraded areas, and it is necessary to monitor them continuously. The problem is to reconcile a method of data collection, in areas of restoration, that is efficient, fast and with proven cost-benefit. The monitoring, whether of degraded areas or under restoration, usually used in different phases of degraded areas recovery plans (PRADs), allows to visualize the indicators/results obtained in comparison with the established goals, either by global agreements, protocols, or PRADs. However, in terms of space-time and costs, field campaigns become very costly. As an alternative to traditional inventories, 3D remote sensing data, such as Light Detection and Ranging (LiDAR), have been used to assist in monitoring these areas. Because LiDAR has a high cost when compared to inventory methods performed in the field, a low-cost technology, such as 3D point clouds of digital aerial photogrammetry (FAD) obtained via remotely piloted aircraft (RPA) has been improved. It is intended, therefore, with this research, to perform the (i) characterization of the vertical structure and cover and (ii) estimation of above-ground biomass (AGB) with 3D point clouds obtained by FAD-RPA in restored Atlantic Forest areas. These areas were grouped according to the restoration method: RAM (active restoration), RAR (active restoration intercropped with eucalyptus) and RPA (passive restoration). A total of 1,797 tree individuals were inventoried, distributed in 59 species and 26 families. Among the three restoration models, RAR presented the highest values of NA, H, AB and AGB. The areas under restoration that presented the highest floristic similarity (55%) were RAR and RPA. The numerical model of the terrain overestimated the elevation values of the terrain, with RMSE 1.5 m (3%) and a bias of - 0.7 m, (1.3%). On the other hand, the dominant height values of the trees of the plots were underestimated by 1 m (bias of 11.7%), with an error of 2.3 m (RMSE of 25.8%). It was possible to characterize and differentiate (group) the restoration systems analyzed with the height and coverage metrics of the FAD-RPA point cloud. The adjusted regression model showed good performance, with R2 values of 0.84 and RMSE of 11.1 Mg.ha-1 (39.2%). In the validation, R2 was 0.75 and the RMSE at the plot level was 13.0 Mg.ha-1 (46%). We can conclude that, despite overestimating the elevation values of the terrain and, consequently, underestimating the heights of the trees, the FAD-RPA data can be used to characterize the different restoration systems analyzed and to estimate the AGB values with satisfactory accuracy.