Methodology
This study focuses on a 33,707 km² region of the central Brazilian Amazon encompassing parts of the Lower Rio Negro basin near Manaus. The area includes a mosaic of land cover types ranging from intact terra firme rainforest and várzea floodplain forests to degraded pastures and secondary regrowth. It overlaps several protected areas within the Mosaico do Baixo Rio Negro (MBRN), including sustainable development reserves (RDSs), extractive reserves (RESEX), and national parks. The study area was delineated based on the availability of cloud-free satellite imagery and relevance to ongoing conservation and land use planning initiatives. Figure X illustrates the geographic extent of the study region, major rivers, conservation units, and selected field sites used for ground validation. This spatial framework supports analysis of land cover dynamics, forest regeneration, and anthropogenic disturbance across ecologically and socio-politically diverse landscapes.
To assess long-term land cover changes and vegetation dynamics in the Amazon River region of Brazil, multispectral satellite imagery from two time points -1990 and 2024 - was analyzed. Landsat 4–5 Thematic Mapper (TM) data from 1990 and Landsat 8 Operational Land Imager/Thermal Infrared Sensor (OLI/TIRS) data from 2024 were acquired through the USGS Earth Explorer platform. Imagery was selected from the dry season months to reduce atmospheric interference, minimize cloud cover, and ensure consistent phenological conditions for vegetation analysis.
Surface reflectance imagery from Landsat 4–5 TM (1990) and Landsat 8 OLI/TIRS (2024) was acquired from the USGS Earth Explorer as Level-2 products, which include pre-applied geometric and atmospheric corrections. Each Landsat image was clipped to the known study area extent (33,707 km²) using ArcGIS Pro 3.2.
Vegetation and surface moisture were quantified using the Normalized Difference Vegetation Index (NDVI) and Shortwave Infrared Reflectance (SWIR), calculated from the appropriate spectral bands. To detect temporal change, difference rasters were created by subtracting 1990 values from 2024 values for each index. These were then combined into a composite raster using the Composite Bands tool, allowing spatially aligned, pixel-level comparison across both vegetation and moisture gradients.
High-resolution PlanetScope imagery from October 2024 was used for visual validation and site-level verification. Because the imagery was delivered as multiple overlapping tiles, a preliminary step involved mosaicking the individual PlanetScope scenes into a single composite image, enabling more efficient comparison with Landsat-derived layers and better visualization of fine-scale land cover features.
To quantify vegetation change, NDVI values were reclassified into three categories: loss (ΔNDVI < -0.2), no significant change (ΔNDVI between -0.2 and +0.2), and gain (ΔNDVI > +0.2). The same classification thresholding was applied to SWIR to capture relative changes in surface moisture conditions. The resulting classified rasters were overlaid to produce a 3 × 3 transition matrix representing nine distinct land change classes. These classes included combinations such as NDVI loss with SWIR gain (interpreted as vegetation loss and surface drying) and NDVI gain with SWIR loss (potential forest recovery and surface re-wetting).
Area calculations for each class were performed using the Tabulate Area tool in ArcGIS Pro. To address potential overestimation from background pixels (e.g., cloud, water, or no-data areas), tabulated pixel counts were normalized relative to the actual study area extent. This ensured that final land change estimates reflected only valid terrestrial areas.
To strengthen the interpretation of remote sensing results, high-resolution PlanetScope 8-band imagery from October 2024 was used for visual verification of land cover and vegetation conditions. This imagery provided 3-meter spatial resolution, enabling more precise identification of land use patterns and fine-scale disturbances not easily resolved in Landsat data.
Further validation was conducted using geotagged field photographs collected during a site visit in June 2024. Fieldwork included visual assessments at multiple locations representing a range of land cover types, intact forest, degraded areas, agricultural fields, and secondary regrowth. These images were linked to GPS coordinates and examined alongside classified raster data to confirm vegetation condition, detect disturbances (e.g., fire scars, clearings), and support interpretations of regrowth or succession in ambiguous areas. These combined datasets offered critical contextual insight and served as qualitative validation of the remote sensing–based classifications.
This methodological framework closely reflects the official monitoring strategies outlined in both the Plano de Gestão Integrada do Mosaico do Baixo Rio Negro 2025 and the Plano de Monitoramento do Mosaico do Baixo Rio Negro (WCS Brasil, 2017). These plans call for the integration of remote sensing technologies such as Landsat-based NDVI and SWIR analysis, geospatial processing, and ground-based validation to monitor vegetation dynamics, land use change, and ecological recovery across conservation units. By aligning with these protocols, including the use of surface reflectance-corrected imagery, phenologically consistent timepoints, and validation through high-resolution data and field observations, this project generates restoration-relevant data that can support ongoing assessments of forest regeneration and land use trends in the central Amazon.