Trajetória de recuperação de uma floresta manejada na amazônia central: integração entre inventário florestal contínuo e sensoriamento remoto

Abstract

The present study aimed to propose the integration of field data and remote sensing for monitoring the recovery of three Annual Production Units (UPAs B, C, and D) subjected to Sustainable Forest Management (SFM) and logged in 1996, 1997, and 1998, respectively. Logging intensity varied among the areas, being highest in UPA B, followed by D and C. The field analyses were based on continuous forest inventory data collected from 41 permanent plots. Using ANOVA and Tukey's tests, volume, biomass, and basal area were quantified and compared across pre- and post-logging periods to verify the effects of the intervention and the forest's resilience. In terms of remote sensing, six vegetation indices (NDFI, NDVI, AFRI, EVI2, MSAVI, GNDVI), derived from Landsat 5 and 8 imagery, were employed to monitor these processes. The punctual analysis on the permanent plots, validated by Friedman and Dunn-Bonferroni tests, showed that the indices captured the logging event through a sharp drop in their values. AFRI and NDFI were the indices that best translated the field observations. However, the spatial analysis, using 100 virtual plots in each UPA, revealed that the index behavior does not indicate a complete recovery of the area in terms of canopy surface spectral response, given the persistent statistical difference between current and pre-logging conditions, observed even after two decades. Moving average graphs generated for shorter time intervals aided in analyzing the indices' recovery trajectory, confirming the importance of following a strict remeasurement schedule, and indicating that AFRI and NDFI show fewer fluctuations over time. The predictive potential of the indices was also evaluated. AFRI and NDFI were the only ones to show moderate to slight negative correlations with logging variables and positive correlations with stock variables, demonstrating their capacity to detect the impact of logging and forest stock. This correlation validated the use of the GAMLSS (Generalized Additive Models for Location, Scale and Shape) regression model, with a Generalized Gamma (GG) family distribution, for volume estimation. The model showed high predictive efficiency, with AFRI proving to be a reliable predictor for the mean volume value (µ) and NDFI being able to model the volume variability (σ) indicating that an increase in NDFI corresponds to a decrease in variability.