Tropical forests, particularly the Atlantic Forest, are among the most biodiverse and threatened ecosystems on Earth. Intense fragmentation, historical deforestation, and climate change have compromised their integrity and endangered endemic and rare species. The situation is even more critical in the Doce River Basin (DRB), a biodiversity hotspot in southeastern Brazil, due to the expansion of pastureland and mining, as well as the 2015 Mariana disaster. Understanding how future climate may affect the persistence of threatened tree species in the region is essential. This study evaluated the effects of climate scenarios on the distribution of 26 rare and/or threatened tree species in the DRB. We used Species Distribution Models (SDMs), combining algorithms (GLM, RDF, SVM, MaxEnt) into a consensus ensemble model under SSP3-7.0 and SSP5-8.5 scenarios for 2050 and 2090. Climatic, topographic, and edaphic variables were predictors, with climate explaining most environmental variation. Among the 26 species, 19 are projected to lose over 75% of their suitable area, with local extinction likely for range-restricted and specialized species such as Handroanthus arianeae and Grazielodendron rio-docensis . Species associated with mature forests and humid microclimates were most severely affected, while some generalists showed resilience, maintaining or expanding suitable areas. Although all species retain some suitable habitats within Protected Areas (PAs), results indicate the current PA network will be insufficient for long-term conservation. Complementary strategies such as assisted translocation, habitat connectivity, and priority-area identification based on climate projections are needed. This study provides critical insights for policies and land-use planning aimed at conserving the DRB’s threatened flora.

Ferruginous rupestrian vegetation in the Carajás National Forest comprises environmentally heterogeneous ecosystems exposed to strong seasonal variation. We evaluated how geoenvironment and seasonality are associated with the structure and floristic composition of herbaceous–shrub vegetation in the N1 area, Pará, Brazil. We sampled 56 permanent 1-m² plots arranged in 14 clusters across four geoenvironments - Shrubby Rupestrian Vegetation (SRV), Open Rupestrian Vegetation (ORV), Rock Outcrop (RO), and Grassy Field (GF) - during the rainy and dry seasons. We recorded 7,942 individuals, 94 species, and 31 families. Generalized linear mixed-effects models showed a significant geoenvironment × season interaction for species richness (χ² = 22.19; P < 0.001), whereas individual density was significantly affected by geoenvironment (χ² = 8.71; P = 0.033) and season (χ² = 22.81; P < 0.001), with no significant interaction. Richness differed among geoenvironments during the rainy season but not during the dry season. Density was higher during the rainy season, with significant seasonal declines in GF, RO, and SRV but not in ORV, while differences among geoenvironments were localized to the rainy season. Plot-level NMDS showed extensive floristic overlap among geoenvironments, and PERMANOVA detected no significant differences in compositional centroids among geoenvironments after accounting for cluster-level variation. Seasonality produced a significant but small overall compositional shift (R² = 0.017; P = 0.0028), while multivariate dispersion differed among geoenvironments in both seasons. These results show that geoenvironment and seasonality affect different dimensions of community organization, with differentiation expressed mainly through richness, density, and within-geoenvironment compositional heterogeneity. Conservation and monitoring should therefore consider the complete geoenvironmental mosaic, fine-scale spatial heterogeneity, and seasonal variation.

Forest edge effects are commonly interpreted as radial gradients from the edge toward the interior, but this assumption may oversimplify the spatial organization of heterogeneous tropical forest fragments. Here, we integrated field-based phytosociological data with Sentinel-2 spectral indices to evaluate whether edge-related structural interpretation depends on analytical scale in Semideciduous Seasonal Forest fragments embedded in the Brazilian Cerrado. Five fragments were analyzed using transect-based plots and continuous pixel-level modeling. Basal area showed a strong positive correlation with the Normalized Difference Vegetation Index (NDVI) (r = 0.95), supporting its use as the main spectral proxy for basal-area-related structural variation. Plot-level segmented regression detected statistical transitions along the edge–interior gradient, with breakpoints ranging from approximately 13 to 39 m. However, pixel-level modeling revealed scale-dependent responses, including shorter estimated transition distances in the rural Ipameri fragment (IF) and the Campo Alegre de Goiás fragment (AC), an intermediate transition in the Caldas Novas fragment (CN), a longer estimated transition distance in the UEG–Ipameri fragment (IP), and high internal heterogeneity without a single dominant radial transition in Panga. The first two axes of the Principal Component Analysis (PCA) explained approximately 81% of the total variance, reinforcing the structural–spectral correspondence. These findings show that edge-related spectral–structural patterns are detectable but are not adequately represented by fixed radial zones alone. Pixel-level Sentinel-2 modeling provides a complementary spatial framework for interpreting edge-related heterogeneity in fragmented tropical forests when integrated with field-based structural data.

Forest edge effects are commonly interpreted as radial gradients from the edge toward the interior, but this assumption may oversimplify the spatial organization of heterogeneous tropical forest fragments. Here, we integrated field-based phytosociological data with Sentinel-2 spectral indices to evaluate whether edge-effect interpretation depends on analytical scale in Semideciduous Seasonal Forest fragments embedded in the Brazilian Cerrado. Five fragments were analyzed using transect-based plots and continuous pixel-level modeling. Basal area showed a strong positive correlation with NDVI (r = 0.95), supporting its use as the main spectral proxy for vegetation structure. Plot-level segmented regression detected edge-to-interior transitions, with breakpoints ranging from approximately 13 to 39 m. However, pixel-level modeling revealed scale-dependent responses, including shallow gradients in IF and AC, an intermediate transition in CN, a deeper gradient in IP, and high internal heterogeneity without a single dominant radial transition in Panga. The first two PCA axes explained approximately 81% of the total variance, reinforcing the structural–spectral correspondence. These findings show that edge effects are detectable but not adequately represented by fixed radial zones alone. Pixel-level Sentinel-2 modeling improves the spatial interpretation of fragmented tropical forests.