3 RESULTS
The models demonstrated high levels of accuracy according to the TSS (0.86 ~ 0.90) and ROC (all values 0.99), indicating models with predictive quality (Tab. S1 in Appendix S1). The sum of the pixels for current projections between SpG and SpM species showed high variation (SpG: 3.5 ~ 7.0 million km² and SpM: 2.0 ~ 7.2 million km²). Estimates made by AOGCM institutions, in general, maintained consistency, and the highest rates of change with more hostile effects were recorded by CCSM4 (CC) (Tab. S1 in Appendix S1, Figure 1 and S3 in Appendix S2).
The spatial pattern of suitability predicted in the ENMs, in general, indicated a potential expansion of the area considering the binary maps, and maintains a gradual increase in the predicted metrics in future scenarios, representing an increase in areas potentially suitable for the occurrence of these species in 2050 (Fig. 2, Tab. S1 in Appendix S1). Despite the expected increased range and regardless of their status (SpG or SpM), it is estimated that areas with high adequacy values will be reduced, fragmented and/or displaced as the current scenario moves towards the future ones (hotter colors in Fig. 1). In different scenarios, the species S. convallariodora will be the one that will lose the most highly suitable area. The projected occurrence range continue to coincide with the Brazilian savanna, but greater interpenetrations will potentially occur in the adjunct biomes, especially in the Amazon according to our predictions (Fig. 1 and S3 in Appendix S2). On the other hand, it was predicted that the ecotonal region of central-north Brazil will gain environmental adequacy, which was something observed in the scenarios for most of the species studied.
Still regarding future projections, it is worth noting a southwestward shift, with increased extensions in the states of Mato Grosso, Rondônia and Acre, and Bolivia. The same was not seen in the northeast of Pará and Amapá, or in the other disjoint savannas, contrary to expectations. The analyses also indicated that the scenario of extension of generalist species (refuge SpG, RCP 8.5 in Fig. 3) represents the most inclusive area of suitability for the species, and that the adequate habitats of the specialist species (SpM, RCP 8.5 in Fig. 3) coincide with the relictual refuge of the five scenarios including all species.
In general, the results showed that most projected species are actually being favored with increased potential area and indicated a gain of range. The average gain was of 18.5% in the most optimistic scenario (2050; RCP 4.5) and 26.9% in the worst (2050; RCP 8.5, Tab. S1 in Appendix S1). Comparing the species, H. articulatus had the largest environmentally suitable area predicted in the current scenario (Tab. S1 in Appendix S1, Figure 1), followed by B. crassifolia ,C. americana and B. virgilioides . The average gain of these species persists throughout the scenarios until 2050, only changing their relative positions. However, three species showed no variation or a loss of spatial range: P. reticulata (SpM) andS. convallariodora (SpG) showed reduction in the less pessimistic (8 and 11%, respectively) and in the most pessimistic (3 ~ 16%, respectively) scenarios. There was a similar prediction for H. articulatus (SpM), with negative rates (6 ~ 14%) in three of the five scenarios (Figure 2 and S4 in Appendix S2).
In future projections, the 278 protected areas of different categories that overlapped the focal areas will house only 4.6% (5,634 km²) of the refuges; only 4.8% (7,822 km²) and 8.1% (142,196 km²) of the marginal and generalist species, respectively, will be protected inside CUs (Figure 3). Less than 10 current CUs are found specifically in relictual areas, the largest being the Ilha do Bananal/Cantão State Environmental Protection Area in Tocantins and the Chapada das Mesas National Park in southern Maranhão. When associated with vegetation loss data, a drastic and alarming reduction in the proportion of residual habitats per grid cell was observed, showing the preferred areas to concentrate on conservationist measures (Fig. S5 in Appendix S2).