Uso de drones e geotecnologias abertas para o monitoramento da cobertura do solo em uma área de restauração da Mata Atlântica

Abstract

Large-scale ecological restoration in the Atlantic Forest is strategic for biodiversity conservation and environmental resilience and requires standardized, replicable, and auditable monitoring to support adaptive management and accountability. This study implemented and evaluated a digital monitoring workflow based on data obtained by unmanned aerial vehicles (UAVs) with RGB (red, green, and blue) and multispectral sensors, photogrammetry-derived structural metrics, including a canopy height model (CHM), spectral index (NDVI), supervised classification (Random Forest), and open-access tools (Google Earth Engine and QGIS). A pilot area of approximately 13.0 ha was assessed across two periods, T0 (April 2024) and T1 (January 2026; approximately 20 months after planting), generating land cover maps, net change estimates, class-transition analyses, and an RGB–multispectral comparison at T1. Results showed an increase in native vegetation and a reduction in exotic grass cover over time, with similar net changes across sensors (Δnative vegetation: +2.91 ha in RGB and +3.00 ha in multispectral; Δgrass: -3.17 ha in RGB and -3.28 ha in multispectral). Point-based validation indicated high performance (overall accuracy ≈ 0.90–0.91; Macro-F1 ≈ 0.89–0.91), and class areas were also reported as error-adjusted estimates with 95% confidence intervals, strengthening the inferential robustness of the results. Spatial agreement between RGB and multispectral classifications at T1 was high (86.7%); however, deviations (13.3%) were concentrated in the low-height stratum (CHM < 1 m) and were dominated by swaps between young native vegetation and grass, highlighting a critical uncertainty zone in early-stage surveys. Independent field-transect validation confirmed the map-based estimates for native vegetation cover. As an applied contribution, the study proposes an operational strategy based on the recurring use of RGB as a time-series baseline and the targeted use of multispectral data and CHM as layers for quality control and management prioritization in critical areas.