Human population and socioeconomic modulators of conservation performance in 788 Amazonian and Atlantic Forest reserves

Protected areas form a quintessential component of the global strategy to perpetuate tropical biodiversity within relatively undisturbed wildlands, but they are becoming increasingly isolated by rapid agricultural encroachment. Here we consider a network of 788 forest protected areas (PAs) in the world’s largest tropical country to examine the degree to which they remain intact, and their responses to multiple biophysical and socioeconomic variables potentially affecting natural habitat loss under varying contexts of rural development. PAs within the complex Brazilian National System of Conservation Units (SNUC) are broken down into two main classes—strictly protected and sustainable use. Collectively, these account for 22.6% of the forest biomes within Brazil’s national territory, primarily within the Amazon and the Atlantic Forest, but are widely variable in size, ecoregional representation, management strategy, and the degree to which they are threatened by human activities both within and outside reserve boundaries. In particular, we examine the variation in habitat conversion rates in both strictly protected and sustainable use reserves as a function of the internal and external human population density, and levels of land-use revenue in adjacent human-dominated landscapes. Our results show that PAs surrounded by heavily settled agro-pastoral landscapes face much greater challenges in retaining their natural vegetation, and that strictly protected areas are considerably less degraded than sustainable use reserves, which can rival levels of habitat degradation within adjacent 10-km buffer areas outside.

Tropical forest regions, in particular, have undergone rapid changes in land-use intensification, 50 leading to increasing isolation and habitat degradation of existing PAs (DeFries et al., 2005;Wright, 2005). These mounting pressures combined with conflicts with powerful economic interests have also led to formal alterations in existing environmental legislation, ultimately resulting in the downsizing, downgrading, and even degazettement of many formally established PAs (Mascia & Pailler, 2011;Marques & Peres, 2015). 55 Brazil is the largest tropical country on Earth, and contains some 41% of world's remaining tropical forests and approximately 13% of all known species (Lewinsohn & Prado, 2005).
Between 2000 and 2005, however, Brazil lost an average of ~33,000 km² of forest each year, the fastest absolute tropical deforestation rate in human history (Hansen, Stehman & Potapov, 2010).
Credible projections suggest that primary habitat conversion will continue to increase as the 60 country becomes one of the largest emergent agricultural and industrial economies, amounting to a rapid escalation in demand for new arable cropland, energy, and raw materials. To boost economic growth, the Brazilian government has launched an ambitious macroeconomic development blueprint -the Growth Acceleration Plan (PAC) -which envisages to deliver many mega infrastructure projects, including major hydroelectric dams, power transmission lines, and highways and waterways, to hitherto poorly accessible 'hinterland' regions. Such concerted geopolitical strategies will clearly have a major impact on natural ecosystems, especially in remote parts of Amazonia.
Brazil is also the only country hosting two of the world's major tropical forest biomes, which are disjunct across a wide latitudinal gradient spanning the central-northern (Amazonia) and eastern-

Manuscript to be reviewed
Atlantic Forest domain is highly heterogeneous and includes coastal and montane evergreen 80 forests, semideciduous seasonally-dry forests, dunes, marshes along coastal plains, and native grasslands, all of which amount to a global biodiversity hotspot containing 19,355 plant species, ~40% of which endemic (Forzza et al., 2012). Of the original Atlantic Forest vegetation cover of 1.3 million km 2 (13% of Brazil's territory), only 22.2% remains, considering all ecoregions and centers of endemism (official estimates from IBAMA, 2012). Unlike the vast forest tracts of 85 Amazonia, forest remnants in the Atlantic Forest are now highly fragmented and largely restricted to existing reserves, which protect ~9% of the remaining vegetation cover (Ribeiro at al., 2009;Tabarelli et al. 2010). Growing demographic and economic pressures over five centuries have rendered the Atlantic Forest into one of the most threatened biodiversity hotspots worldwide (Fundação SOS Mata Atlântica & INPE, 2013). 90 In contrast, Brazilian Amazonia hosts the largest unbroken tract of tropical forest controlled by a single country (~4.2 million km 2 ), and one of the highest levels of known and unknown biological diversity anywhere (Peres 2005;Pimm et al. 2014). The region accounts for some 55% of Brazil's territory but contains only about 11.8% of the Brazilian population. In addition to the largest surface and underground freshwater reserves, this region contains one of the largest 95 untapped mineral reserves on Earth, as well as vast areas of cheap agricultural land and a flat relief that facilitates mechanized farming and cattle ranching. The historical context is also in marked contrast to the Atlantic Forest, as Brazilian Amazonia remained entirely roadless and unexploited by agropastoral interests until 1970, when only <1% of the region had been deforested and the first major paved highway linked eastern Amazonia to the rest of Brazil (Peres reserve types (IUCN categories I to VI). We therefore restricted our analysis to PAs under SNUC 110 regulations, comprising all formally established conservation units within Amazonia and the Atlantic Forest. Unlike protected forests in Europe and North America, where IUCN categories V and VI are most frequently adopted (Schmitt et al., 2009), these two major tropical forest biomes host a considerable number of PAs under all management categories.
Detailed georeferenced data on species richness and diversity are only available at selected sites, 115 either within or outside reserves, and most studies use deforestation (or the lack thereof) as the proxy of protected area performance in preserving forest biodiversity (e.g. Nepstad et al. 2006, Beresford et al. 2013, Paiva et al. 2015, Pfaff et al. 2015, Ren et al. 2015, Bowei et al. 2016, Miranda et al. 2016). Here, we provide a quantitative assessment of the degree to which the natural vegetation cover of 788 Brazilian forest reserves within the Amazonian and the Atlantic 120 Forest domains have been converted into different patterns of land use. In particular, we examine the role of reserve management category (under the jurisdiction of either federal or state-level management agencies) on this metric of conservation performance both within and around reserve boundaries. Unlike other studies addressing a small number of reserves or presenting aggregate deforestation data from areas within and outside PAs, we relate the absolute and 125 relative amounts of vegetation conversion within reserve polygons to areas immediately outside.
To test the assumption that human dwellings and economic activities have a negative impact on PAs, and that PA category determines the degree to which PAs succumb to such adverse effects, we explicitly consider the effects of reserve size, human population density, and per capita wealth within and around each reserve on measures of reserve performance. Finally, we consider how 130 human population density, socioeconomic context and conservation investment capacity as both part of the problem and the solution governing the fate of tropical forest reserves.

Protected Area Categories
We examined all forest protected areas encompassed by both the Brazilian Amazon and the 135 Brazilian Atlantic Forest biomes, which had been legally sanctioned under the SNUC legislation.
Official PAs within the SNUC comprise federal, state and municipal county conservation units and are divided into two main management classes: (1) strictly protected reserves, with five Manuscript to be reviewed categories; and (2) sustainable use reserves, with seven categories. The first group essentially aims to target conservation objectives, and are restricted to non-consumptive natural resource use 140 (equivalent to IUCN categories I-IV). The second group legally recognizes human occupation and sustainable use of natural resources, typically allowing human occupation of the areas, including agricultural activities (IUCN categories IV-VI). A reserve under IUCN category IV may be either strictly protected or sustainable use, depending on the nominal SNUC category it belongs to (but in this analysis we grouped all PAs according to their legal restrictions within 145 Brazil). For analytical purposes, we grouped all reserves into eight categories, in terms of their overall group, level of protection, land-use restrictions, and conservation objectives (Table 1). We excluded from the analysis other types of PAs that may also afford legal protection, such as indigenous lands, quilombolas, and private areas of restricted use, but are not governed under the SNUC legislation.

Data acquisition and geoprocessing
Our study regions cover the two largest forest biomes in tropical South America, Amazonia and the Atlantic Forest, the phytogeographic boundaries of which are defined by the Brazilian Institute of Geography and Statistics (IBGE). Shapefiles describing the geographic boundaries of all conservation units were obtained from official sources (mapas.mma.gov.br and 155 www.icmbio.gov.br/portal/comunicacao/downloads.html). Complementary information about each protected area was extracted from the Brazilian National Registry of Conservation Units (CNUC -www.mma.gov.br/areas-protegidas/cadastro-nacional-de-ucs). In total, we consider reserve polygons of 788 federal, state and municipal county scale conservation units that met all of the following criteria: reserve boundaries included natural forest cover and overlapped one of 160 these two forest biomes; satisfactory correspondence in each reserve number code between its shapefile and that of the CNUC database; conservation units were terrestrial rather than marine reserves; and reserves were represented by a reliable polygon comprising an area of at least 2 ha (many small private forest reserves were represented by small circles in the shapefile, due to lack of accurate mapping, so were excluded from the analysis). Our overall sample corresponds to 165 82% and 62% of all Amazonian and Atlantic Forest conservation units, respectively (MMA, 2012). In the latter biome, most private reserves contained in shapefiles did not meet our size Manuscript to be reviewed criteria nor data quality control in terms of their spatial data and were therefore also excluded from the analysis.
There is considerable spatial overlap between conservation units in Brazil, partly because strictly 170 protected conservation units can be located within sustainable use conservation units [e.g. many Environmental Protection Areas (EPAs) may include parks, ecological stations, or adjacent private areas]. There are also mapping errors and competition between federal, state and municipal level environmental agencies, which may set aside overlaying PAs within existing reserves, thereby claiming jurisdiction over their respective territories. Therefore, to avoid 175 overestimating reserve areas and their respective classes of land cover, these overlaps were painstakingly manually removed from the vector files (Fig. S1). The following hierarchical structure was used to decide which conservation unit should prevail in cases of overlapping areas: (1) legal restrictions on land use (e.g. strictly protected reserves prevailed over sustainable use reserves, ecological stations over parks, and extractive reserve over EPAs); (2) official year of 180 decree (oldest reserves prevailed); (3) reserve boundaries completely enclosed within another conservation unit (if an EPA or Natural Monument overlapped a smaller private reserve, the latter was retained and the conservation unit larger than the overlap zone was subtracted of a corresponding area).
To compare each conservation unit with its surrounding landscapes, a 10-km external buffer was 185 created from the reserve perimeter, and any overlap between the 788 buffers and neighboring conservation units were also removed. A current land cover map was then generated overlaying the deforestation data (see below) to the vegetation map of Brazil (2002), at a scale of 1:250,000,

Manuscript to be reviewed
Human population density (HPD) was calculated considering data from the last national census of the Brazilian population (2010), from which a shapefile of 213,872 sub-municipal census district polygons (also known as 'census sectors') covering the two major forest biomes (Amazonia: 18,031; Atlantic Forest: 195,841) was generated (www.ibge.gov.br). Municipal county 200 boundaries were then linked to the county-level gross domestic product (GDP) dataset (www.ibge.gov. br) to estimate the county-scale per capita GDP as of 2009. The same boundaries were also linked to the Human Development Index (HDI) at the municipal scale (www.pnud.org.br). HPD and GDP, which were represented at a census district and municipal scale, respectively, were estimated for each conservation unit and its corresponding buffer zone, 205 on the basis of the area-weighted average by intersecting the PA polygon with either the census districts or municipal counties. All area calculations were carried out using the Albers Equal Area Conic projection, South American 1969 datum.

Data analysis
Generalized linear models (GLMs) were used to investigate predictors of natural forest loss 210 within all forest reserves of different denominations in each biome and both biomes combined.
Our response variable was the cumulative conversion rate of any natural vegetation within each of the 788 mapped protected areas across our bi-regional sample. Our predictors included several key variables describing the reserve size, reserve age (year of decree), reserve category, management class (strictly protected or sustainable use reserve), weighed mean human 215 population density (HPD) at the scale of census district both within each reserve polygon and the buffer zone neighboring this polygon (calculated on the basis of all terrestrial areas only), two socioeconomic variables [weighed mean Gross Domestic Product (GDP) and Human Development Index (HDI) at the scale of municipal counties], and reserve governance structure (at the level of federal, state-level or municipal administration). HPD (log 10 x + 1) within and 220 outside reserves was highly correlated (r = 0.789), so we used either one rather than both of these variables in any given model. In comparison, GDP and HDI within and outside reserves were less strongly correlated (r = 0.298 -0.331) and could be considered as independent from one another.
We then tested for multicollinearity among variables by examining the least moderately redundant or collinear Variation Inflation Factors, but no variables were sufficiently collinear at a 225 VIF ≥ 5 threshold (Dormann et al., 2013). The relative strength of these predictors was then Manuscript to be reviewed examined using multiple GLMs to understand their role as drivers of forest conversion rates.
Rather than treating rates of forest loss as proportional data which has a number of drawbacks (Warton & Hui, 2011), we explicitly considered the total extent of primary habitat loss by modelling the total area (ha) of forest conversion within each reserve, but used the total terrestrial 230 reserve size (ha) as an offset variable, and a quasipoisson error structure to avoid overdispersion.
We then repeated this modelling approach using a failure:success binomial error by considering the total number of hectares that were either converted to other land-uses or retained in apparently intact form within each reserve. Models were examined on the basis of minimum BIC and AIC c values, and there was good convergence in identifying the most parsimonious "best" model.

235
These analyses were performed with data from both biomes combined, as well as separately. All models were fitted using the 'lme4' package (Bates et al., 2007) within the R platform. To assess the relative strength of spatial effects on the characteristics of forest reserves across the two biomes, we used spatial multiple linear regression implemented with simultaneous autoregression using the spautolm function in the R package 'spdep'. The degree of within-reserve forest loss 240 was clearly structured in space across all Amazonian and the Atlantic Forest protected areas, and the autoregressive parameter λ indicated significant spatial autocorrelation across all reserves (β = 0.72, p < 1e-15). However, this successfully eliminated spatial autocorrelation of the residuals (Moran's I, p = 0.64). Finally, we used paired t-tests to examine differences in deforestation rates and HPD within and outside protected areas.

Results
Of the 788 forest reserves considered here, which encompassed a total area of 120,289,994 ha, 251 are distributed across the Amazon (111,334,941 ha) and 537 across the Atlantic Forest (8,955,053 ha) ( Fig. 1 and S2). In general Atlantic Forest reserves share an older history since they were first established (see Fig. S3), but account for only ~8.0% of the total area of 250 Amazonian reserves. Sustainable use reserves (hereafter, SURs) accounted for 64.5% (508) of all forest reserves and 63.9% of the total area of the reserves we considered, whereas strictly protected reserves (hereafter, SPRs) accounted for 35.5% (280) of the reserves and 36.2% of the total area ( Table 2). The overall proportion of the original natural vegetation (in almost all cases primary forest) converted within those reserves (Amazonia: 12.1%; Atlantic Forest: 44.5%) was exceptions for which vegetation conversion rates within reserves (33.9 ± 30.2%, N = 537) were actually greater than those outside (62.1 ± 24.7%), particularly for SURs in the Atlantic Forest.
These trends in land use change reflect marked regional differences in human population 260 densities, which were much higher both within and outside Atlantic Forest reserves than those within and outside Amazonian reserves (Fig. 2).
Human-modified areas within SURs were usually proportionately much larger than within SPRs (Table 2). In fact, although vegetation conversion rates inside reserve boundaries scaled to conversion rates in the surrounding buffer areas of the same reserves in both biomes (p<0.001),

265
SPRs comprised a more effective deterrence against deforestation than SURs, particularly for Atlantic Forest reserves for which major class of reserve management was a significant predictor of conversion rates (p<0.001; Fig. S4). Reserve age (year of decree) had an overall positive effect on conversion rates for Atlantic Forest reserves (p = 0.028) but not for Amazonian reserves (p = 0.717), once reserve management class and conversion rates in external buffers were controlled 270 for. Overall, 45.2% of the aggregate Atlantic Forest reserve area across all categories had been converted to other land-uses (mean = 33.9 ± 30.2%), and therefore fared far worse than Amazonian forest reserves, which had lost only 12.1% (mean = 10.1 ± 18.0%) of their total forest area.
Deforestation rates declined in increasingly larger reserves, which protect larger and more 275 ecologically viable forest areas from structural alterations in surrounding landscapes ( Fig. 3; Table 3). Among all reserve categories, Environmental Protection Areas (EPAs) and Wildlife Refuges experienced the highest levels of forest loss, whereas Parks and Reserves performed much better. This is expected since these strictly protected reserve categories aim to conserve natural ecosystems under the public domain, while EPAs are predominantly comprised of private 280 lands under special regulations, often encompassing agricultural and even urban areas.
Accordingly, generalized linear models showed that strictly protected reserves were far more effective than sustainable use reserves in deterring vegetation conversion once other variables were taken into account (Fig. S4).
When we combined all reserve categories across both biomes, human population density (HPD) 285 and human development index (HDI) were the only significant socio-economic variables explaining the degree to which reserves had been degraded ( Fig. 4-5, S4). As such, reserves were Manuscript to be reviewed more degraded in more densely settled areas, but particularly in more developed counties. HPD also consistently declined in increasingly larger reserves, except for sustainable use Atlantic Forest reserves where this relationship was not significant (Fig. 6).  Table 2).
Human-induced land cover change was lower inside forest reserves of any denomination 300 compared to their external buffers. We noted an outlier reserve for which forest conversion has been particularly elevated compared to even sustainable use reserves: some 22.2% of the total area of the Metrópole Wildlife Refuge, located within the metropolitan area of Belém, the state capital of Pará, had been deforested. This relatively small reserve (6,369 ha), however, was decreed in 2010 as a conservation unit from former private farmland, so most of this deforestation had experienced the highest levels of forest conversion ( Table 2). Levels of legally permitted human activities and high HPD within EPAs has resulted in a high degree of within-reserve degradation (58.7%). However, large strictly protected Atlantic Forest reserves have been much more effective at inhibiting forest conversion (Fig. 3). As such, reserve size failed to explain the 320 degree to which SURs had been degraded, with many large SURs also exhibiting high proportions of deforestation. The most intact biogeographic subregion of the Atlantic Forest lies within the Serra do Mar montane domain, which still retains some 36.5% of its original vegetation cover. This high-elevation subregion contains the largest strictly protected Atlantic Forest reserves within the densely populated states of São Paulo, Rio de Janeiro and Paraná, 325 which account for 35.5% of the Brazilian population.
In contrast, strictly protected reserves such as biological reserves, ecological stations and parks (IUCN categories Ia, Ib and II, respectively) exhibited much lower deforestation rates, typically well below 10%. Reserve management category was therefore more important than reserve size per se in deterring deforestation across the Atlantic Forest. High levels of degradation within 330 external buffers (mean = 62.0 ± 24.7%) indicate that the heavily settled landscapes surrounding Atlantic Forest reserves have become highly fragmented for both strictly protected and sustainable use reserves (Fig. 8).

Discussion
Several studies have considered the effectiveness of protected areas in terms of biodiversity 335 conservation (e.g. Brunner et al. 2001, Nepstad et al. 2006, Coetzee, Gaston & Chown 2014, Bradshaw, Craigie & Laurance 2015. These studies have typically examined a small number of PAs at global or continental scales. However, global analyses using small sample sizes per country can mask national or regional trends in the de facto protection and true effectiveness of protected areas (Schmitt et al. 2009). In contrast, we considered anthropogenic conversion of 340 natural vegetation into different forms of land use within 788 strictly protected or sustainable use reserves within the two largest neotropical forest domains. There were clear contrasts between the Amazon and the Atlantic Forest biomes, particularly in terms of the size structure of nature reserves, overall levels of natural habitat degradation, which reflect major differences in regional scale socio-economics and human population density across those two biomes. This in turn 345 results from clear differences in post-colonial trajectories in human occupation, frontier expansion and land use, which paved the way to arguably the greatest polarity in economic development for two major regions within a single tropical country.
The Atlantic Rainforest region was the first in Brazil to be settled by Europeans, following through several economic cycles based on resource extraction, including Pau-Brasil (Caesalpinia  disregards the overall poor performance and high forest conversion rate of many forest reserves (~44.5% in SPRs and ~58.3% in SURs, see Table 2).
In terms of the size structure of existing reserves, at one extreme a large number of forest reserves have been set aside within private landholdings (NHPRs), mostly within the Atlantic Forest (IUCN Category IV). These typically small reserves tend to be embedded within highly Manuscript to be reviewed between large fragments, can boost population sizes, and operate as a refuge in case any major disturbances (e.g. wildfires) take place in larger protected areas (DeFries et al., 2005). NHPRs comprise the most ubiquitous and numerically dominant reserve category in the Atlantic Forest, and they continue to proliferate each year, mainly due to land tax incentives. These reserves may 410 not be self-sufficient but greatly complement landscape-scale conservation planning, and their start-up costs are virtually zero for the public treasury because they are privately owned and managed. Moreover, the risk of any intentional deforestation after a private reserve is created is negligible, as even a change in property ownership does not entitle any new landowner to any changes in land use. NHPRs already exceed forest reserves under public jurisdiction in both total 415 area and numbers, and will increasingly play an important role in balancing biodiversity conservation, provision of ecosystem services and human welfare particularly in heavily humanmodified landscapes (Melo et al., 2013) The Brazilian Amazon, on the other hand, has experienced unprecedented deforestation rates, but unlike the Atlantic Forest, this recent process of frontier expansion largely took place in the last of public lands, timber extraction and wildfires both within and outside poorly implemented and rarely enforced protected areas (Souza, Roberts & Cochrane, 2005). With rapid economic 430 changes, the region has seen an economic transition from resource extractivism to industrialization, with mineral exploitation and commodity production from agribusiness such as cattle and soybean gaining ground (Soares-Filho et al., 2005) and driving deforestation (Barona et al., 2010). Public policies for credit, subsidies, land occupation and resettlement of southern Brazilian farmers have also encouraged deforestation (Fearnside, 2005;Schneider & Peres, 435 2015). In line with these changes, Brazilian Amazonia exhibits the highest urban growth rate (Oliveira & Oliveira, 2011). This means that each economic activity contributes individually or Manuscript to be reviewed collectively to current or future deforestation rates (Fearnside, 2005). Under the most pessimistic deforestation forecasts, forest losses by 2050 may exceed 45% of the Brazilian Amazon (Soares-Filho et al., 2005). However, data from the Brazilian Space Agency (INPE, 2015) indicate an 440 average annual reduction of 13.5% in deforestation rates over the last decade lending room for some optimism.
The Amazon is seen worldwide as one of the last remaining natural capital frontiers on Earth (Becker, 2005), but faces high expectations within Brazil in terms of valuing standing forests.
Despite international instruments proposed to slow down deforestation, such as the Reduced Although the sheer size of many Amazonian forest reserves is decisive in slowing down deforestation, there are major differences between reserve categories in terms of forest conservation performance. For example, deforestation rates within EPAs was nine-fold greater 455 than those recorded in SPRs (Table 2), and this is facilitated by no legal restrictions on agricultural land use, including slash-and-burn subsistence agriculture, which reduces secondaryforest resilience and crop productivity (Jakovac et al., 2015), therefore demanding ever more forest conversion to support the livelihoods of a growing population. All other categories of sustainable use reserves, however, exhibited forest conversion rates lower than 5% at least for 460 now, even when their surrounding areas had already been deforested. Deterring these encroachment pressures, however, will require sustained government action, including effective vigilance, law enforcement and a good working relationship with local communities, particularly in legally occupied forest reserves.
The official count of 1,602 continental conservation units in Brazil, which currently represent 465 17.2% of the Brazilian territory, indicates that current international targets in safeguarding native biodiversity have already been reached (MMA, 2012). However, when we assess the aggregate existing conservation acreage by management category, official assertions on the degree to which Brazilian ecosystems are effectively protected become greatly overestimated. Since 2003, we have witnessed an increase of 47.3% in the number of protected areas established, especially 470 sustainable use reserves in the Amazon. Sustainable use reserves, of often questionable longterm future, now far exceed strictly protected reserves both in Brazil (Peres 2011) and worldwide (Jenkins and Joppa, 2009;Schmitt et al., 2009), and have proved to be less effective than strictly protected areas in the Brazilian Cerrado (Paiva, Brites & Machado, 2015). In addition, the most permissive reserve category (EPAs), whose land-use restrictions are effectively negligible, 475 account for some 19% of the entire extent of protected areas across the Amazon and the Atlantic Forest. As a result, 58.9% and 21.4% of the total area of EPAs in the Atlantic Forest and Amazonia, respectively, has already been deforested.
One of the factors contributing to the expansion of multiple-use tropical forest reserves is the high financial cost of establishing strictly protected conservation units, especially in the Brazilian 480 Atlantic Forest, as this involves expropriating private land and removing the resident population as required by law. In Brazilian Amazonia, only 24% of all lands are privately owned and much of the region remains sparsely populated. This facilitates the creation of large forest reserves and indigenous territories, which cover an additional 21.7% of the entire Legal Amazon territory (Ricardo, 2011). Furthermore, sustainable use reserves are politically more viable and more 485 socially acceptable than strictly protected reserves, particularly in densely populated areas (Nelson & Chomitz, 2011). In many cases, when a protected area is set aside, government mandated implementation actions that ensure protection may not necessarily follow, such as developing the appropriate infrastructure, removing squatters and effective monitoring. Therefore, this often disconcerting lack of reserve implementation may justify official 490 downgrading from strictly protected to sustainable use reserves or downsizing reserve boundaries to exclude heavily degraded areas (Bernard, Penna & Araújo, 2014, Marques & Peres, 2015. There are also additional factors that exert social and economic pressure at a regional scale, which may provoke changes in federal environmental legislation affecting protected areas in both biomes. One risk factor is the political pressure for formal alterations to the legislative acts that affecting the status of conservation units and other protected areas in Brazil. Legal initiatives to reduce, cancel or otherwise alter the protection status of 27 federal reserves nationwide are currently under appreciation by Congress (Marques & Peres, 2015). These alterations translate 500 into further protected area losses, compromising national conservation targets to meet binding agreements under the Convention for Biological Diversity.
This study therefore shows that the mere presence of conservation units established on paper inhibits deforestation even under scenarios of dismal implementation investments after 5 years or more of reserve creation. However, in many regions, reserve size is less important to future 505 reserve performance than the management category. For instance, with the exception of private reserves, the forest conservation performance of sustainable use conservation units is very poor in heavily settled post-frontier regions, such as the Atlantic Forest and in increasingly degraded subregions of Amazonia where protected areas now contain all remaining forest cover (Pedlowski et al., 2005). In the Atlantic Forest, strictly protected reserves, preferably under the public 510 domain, continue to be essential in retaining relatively intact natural ecosystems. In contrast, human population pressure is much lower in most of the Amazon, so that physically demarcated reserves, be they sustainable use or strictly protected, are very efficient for now in maintaining relatively intact forest cover, with increasingly larger reserves performing well under different landscape contexts of external encroachment. This picture may change, however, as large 515 infrastructure projects pave the way to agricultural expansion and burgeoning human populations inflated by economic migrants. Investments in protected area defense and enforcement of reserve management plans will therefore need to scale to growing external pressure, or else we risk undoing much of the huge gains in conservation acreage over the last four decades that has earned Brazil a unique contribution in global scale environmental targets and protected area 520 expansion.        (1:1) ratios. Forest conversion rates are typically above these lines (i.e. lower within than outside the vast majority of reserves), except for Atlantic Forest SURs for which many conversion rates inside reserves were actually higher than those outside.      *AIC -Akaike Information Criterion; lag_yrs -time since conservation unit established (years); loghpd_uclogarithm of human population density estimated within the conservation unit; loghpd_buf -logarithm of human population density estimated in area surrounding the conservation unit; log_uc_dryarea -logarithm of the dry area of the conservation unit; logpib -logarithm of the weighted mean GDP of the area covered by the conservation unit; Categ2[EPA, etc.] -analysis categories for the conservation units.