Uso de RPAS no monitoramento de restauração florestal: acurácia posicional de produtos RGB e LiDAR

Abstract

Due to global changes driven by greenhouse gas emissions and land use alterations, the conservation and restoration of forest ecosystems are essential to mitigate these impacts through carbon sequestration, biodiversity protection, and the provision of ecosystem services. In this context, accessible and cost-effective monitoring technologies—such as the use of drones and sensors—emerge as promising alternatives to complement field data collection and reduce associated costs. However, several challenges remain, particularly regarding the positional quality of geospatial data, which necessitates strategies such as optimizing data acquisition timing, selecting appropriate techniques, sensors, and flight parameters, as well as implementing control points to correct geometric distortions and enhance spatial accuracy. Additionally, limitations of the GNSS (Global Navigation Satellite System) and the lack of accuracy information underscore the need for complementary measures to improve the reliability of geospatial surveys. This thesis aimed to evaluate the positional accuracy of RGB (Red-Green-Blue) and LiDAR (Light Detection and Ranging) products, considering the integration of ground control points (GCPs) during data processing. To meet this objective, the first stage involved the establishment of the Cepegeo Network, comprising fifteen geodetic control points across diverse environments, surveyed using conventional topographic and geodetic methods. The collected data were adjusted and compiled into descriptive reports. In the second stage, GNSS methodologies were evaluated based on tracking duration and environmental scenarios, including areas adjacent to forests and more challenging environments, such as near water bodies or buildings. The analysis revealed that the relative static and Precise Point Positioning (PPP) techniques (with a minimum tracking time of 30 minutes) achieved accuracies ranging from 10.1 to 30 cm in open areas and low-density forests. In dense forests, accuracy decreased, ranging from 50.1 to 80 cm, highlighting the canopy's impact on signal quality. Regarding vertical accuracy, none of the GNSS methodologies achieved results compatible with geodetic adjustments. Among them, the relative static technique delivered the best performance in terms of positional precision and reliability, establishing itself as the most effective for surveys under varied conditions. The third stage focused on assessing the accuracy of products generated by RPAS (Remotely Piloted Aircraft Systems) equipped with RGB and LiDAR sensors over a forest restoration site. A total of 90 control points, both horizontal and vertical, were established using dual-frequency GNSS receivers, total stations, and differential leveling. Subsequent drone flights were conducted using RGB and LiDAR-equipped systems, and the data were processed in commercial software platforms utilizing their open-access tools. For the RGB-derived data, orthomosaics and digital elevation models were generated with varying numbers of GCPs. Results indicated that RGB product accuracy was rated as “optimal” when four GCPs were used, while the inclusion of more than 50 GCPs led to overfitting, resulting in reduced accuracy in certain areas. Regarding the LiDAR products, point cloud classification and canopy height model generation were performed. Planimetric analysis revealed an average horizontal displacement of 1.2 meters when data were processed without RTK base station corrections, along with minor deformations in the central area of the model. The evaluation of vertical accuracy suggested that the positional quality of LiDAR data cannot be considered absolute, given the variations introduced by sensor-software interactions, particularly in geotag correction processes. This research presents an innovative approach by evaluating the accuracy of products generated by RGB and LiDAR sensors, as well as analyzing the accuracy of the control points themselves used to enhance product precision. The integration of RPAS with these sensors, in conjunction with geodetic GCPs and advanced processing techniques, proved to be an effective strategy for mapping and analyzing forest restoration areas. The careful selection and balanced distribution of GCPs, along with RTK-based corrections, are critical to ensuring high positional quality. Although challenges such as GCP oversaturation and LiDAR data adjustments require meticulous planning, the combined use of these technologies enhances the efficiency and robustness of environmental surveys. Further research is recommended, including varying flight altitudes and the deployment of more advanced sensors—particularly for LiDAR—as well as the use of premium software versions to further improve the outcomes.