Brazil’s Iguaçu National Park threatened by illegal activities: predicting consequences of proposed downgrading and road construction

V M Prasniewski; N Szinwelski; A S Bertrand; F Martello; C R Brocardo; J Cuhna; C F Sperber; R Viana; B G dos Santos; P M Fearnside; T Sobral-Souza

doi:10.1088/1748-9326/ac4e39

1. Introduction

When Europeans arrived in what is now Brazil in 1500, the Atlantic Forest was 150 million ha in area (Ribeiro et al 2009), about the size of the US states of Texas and California combined. Over the succeeding centuries this rainforest was almost completely destroyed (Dean 1997), and what remains has been heavily fragmented into thousands of small (i.e. mostly $\lt$ 50 ha), highly degraded patches (Ribeiro et al 2009). Much biodiversity has been irrevocably lost, but what remains of the Atlantic Forest is considered to represent one of the world's five most important biodiversity 'hotspots' (Myers 1988, 2003, Myers et al 2000). The largest contiguous area of inland Atlantic Forest is protected by the Iguaçu National Park in Brazil and the adjacent Iguazú National Park and Urugua-í Provincial Park in Argentina (Ribeiro et al 2009, Rezende et al 2018). These parks were created around the iconic Iguassu waterfalls that are shared by the two countries. Biodiversity in the parks is under pressure from a series of illegal activities, and it now faces a major threat from the proposed 'Caminho-do-Colono' road that would bisect the Iguaçu National Park.

Biodiversity losses in many parts of the world are being accelerated by a lack of environmental legislation, enactment of laws that reduce restrictions, and policy reversals for areas that were already protected (downsizing, downgrading and degazetting, or 'PADDD'; Bernard et al 2014, Campos-Silva et al 2015, de Marques and Peres 2015, Prasniewski et al 2020). Biodiversity loss and the need for its protection are being increasingly discussed worldwide, mainly due to the role biodiversity plays in providing ecosystem services and the severe, and sometimes irreversible, social and economic impacts that biodiversity loss causes (Sala et al 2000, 2005, Bellard et al 2012).

Brazil's biodiversity encompasses 8.5%–11.5% of the world's species, yet Brazil is undergoing a severe environmental crisis with an increase in deforestation, fires and other environmental setbacks (Carvalho et al 2019, Escobar 2019, Nobre 2019, Pereira et al 2019, Sheikh et al 2019, Nicolav 2020, Prasniewski et al 2020). Illegal activities have increased due to impunity, lack of supervision and dismantlement of the country's environmental agencies (Fearnside 2019, Ferrante and Fearnside 2019). Deforestation, logging, hunting and burning are examples of human activities harming the environment and biodiversity, resulting in defaunation and forest loss that can affect ecosystem services (Terborgh et al 2008, Dirzo et al 2014, Peres et al 2016, Villar et al 2020). Regrettably, the reporting of human illegal activities is not systematized and there are not enough law-enforcement field agents or patrols to cover all protected areas (Bertrand 2016).

An inconspicuous but important threat to biodiversity in protected areas is the unsustainable (and in most cases illegal) harvesting of wildlife and plants for human consumption. These activities, here represented by poaching, fishing and palm-heart extraction, can directly affect large endangered vertebrate species such as tapir (Tapirus terrestris), white-lipped peccary (Tayassu pecari), jaguar (Panthera onca), endemic fish species (e.g. Steindachneridion melanodermatum) and plants (e.g. Euterpe edulis). Consequently, protected areas suffer from species and/or population losses due to illegal harvesting (Galetti et al 2017). The Brazilian National System of Protected Areas categorizes the protected areas and regulated their uses (Brazil 2000). For example, the National Forest category (IUCN category VI) allows the sustainable direct use of natural resources according to specific management plans, while the National Park category (IUCN category II) only allows indirect use, such as tourism, education and scientific research, with activities such as hunting, fishing and the harvest of timber and non-timber forest products being prohibited. Commercial and recreational hunting are either prohibited (Brazil 1967) or regulated (Brazil 1998), and all hunting, including that for subsistence, is prohibited in national parks. In Brazil, the National Council of the Environment (CONAMA 2001) regulates the management of threatened plant species in the Atlantic Forest, a biome that is also protected by federal law (Brazil 2006). Nevertheless, biodiversity continues to be threatened due to successive policy changes and the lack of enforcement of environmental regulations, as well as proposed laws that are advancing through the congressional committee processes (Brazil 2016), especially after the control of both houses of the National Congress by the 'Centrão' coalition, which increases the likelihood of bills like these being approved (Ferrante and Fearnside 2021).

Iguaçu National Park in Brazil and the adjacent Iguazú National Park in Argentina, protect ca. 2520 km² of inland Atlantic Forest (Ribeiro et al 2009). This fragment hosts rich biodiversity, including rare and endangered species of fauna and flora (Crespo 1982, Straube and Urben-Filho 2004, Bertrand 2016, Brocardo et al 2019). However, the Brazilian portion of this forest continuum is threatened by two proposed laws that change Law 9985/2000 to create a 'parkway' category, downgrading the status of this park from 'integral protection' to 'sustainable use'. Both projects aim to create the 'Caminho-do-Colono' parkway, which would divide the Iguaçu National Park in its currently 'untouchable' protected core (Prasniewski et al 2020). Both the Brazilian and Argentinean parks are known for the iconic Iguassu waterfalls, and both are also surrounded by an agricultural and urban matrix where illegal activities abound (Bertrand 2016).

Illegal activities, such as poaching, are not randomly distributed in space, but rather are associated with rivers, roads, gentle slopes and park edges (Haines et al 2012, Bertrand et al 2018, Ferreguetti et al 2018, Torres et al 2018, Brodie and Fragoso 2020). Spatially explicit models can be used to predict these types of activities, as has been done with occupation models (Ferreguetti et al 2018, Moore et al 2018, Dias et al 2020, Marescot et al 2020). Occupation models require presence-absence data obtained by systematic sampling over space and time to estimate detectability and derive occupation values (Mackenzie et al 2004). This requires substantial time, money and human resources for sampling. Models that use only presence data, such as niche-based models, can be very useful and easy to parameterize because inspection agencies usually already have georeferenced data on illegal activities. Niche-based models have been used to model phenomena beyond the distribution of species (Banks et al 2006, 2013, Sobral-Souza et al 2015, Romero et al 2018, Guimarães et al 2020), thus, these models can be a powerful tool to spatially predict illegal activities by inferring locations (cells on the map) that are environmentally similar to locations where such activities occur, that is, how susceptible a particular location is to these activities.

We used a niche-based modeling approach to predict geographical sites susceptible to illegal activities in the Iguaçu and Iguazú National Parks to predict the possible impact that opening the 'Caminho-do-Colono' [CC] road would have on these activities. We hypothesized that proximity to edge, roads and rivers, in addition to flatter slopes make areas more susceptible to the following illegal activities: poaching, fishing, palm-heart extraction and establishing base camps for these activities. We also expected that the CC Road, as proposed, would increase the areas susceptible to illegal activities, further increasing the threats to biodiversity.

2. Materials and methods

2.1. Study area

The study area encompasses Iguaçu National Park in Brazil (185 262 ha) and Iguazú National Park in Argentina (67 620 ha, figure 1). Iguaçu National Park is surrounded by 10 municipalities (counties) and its perimeter is immersed in a deforested matrix, while Iguazú National Park is surrounded by three municipalities and, unlike Brazil, the matrix in which it is inserted is not predominantly deforested. Both inside and outside of both parks there are paved and unpaved roads. The parks are contained in the Iguaçu watershed, which contains seven main rivers (Iguaçu River, Floriano River, Gonçalves Dias River, Silva Jardim River, Belo River, Índios River and São João River) and several small rivers, terrain with varying slopes (0%–171.52%; see Environmental variables), and different types of land use and occupation, as agriculture, pastures and cities (for more information, see supplementary material available online at stacks.iop.org/ERL/17/024024/mmedia: Characteristics of the study area).

2.2. Shapefile design of the study area

The layers of geospatial information in vector format (shapefiles) for the Iguaçu and Iguazú National Parks were obtained from the Instituto Chico Mendes de Conservação da Biodiversidade (ICMBio) (https://cutt.ly/cObfQjY) and Instituto Geográfico Nacional (IGN) (https://cutt.ly/MhDe5nP), respectively. The shapefiles from both parks were combined and a 500 m buffer outside the park boundaries was delimited to include illegal activities registered in forest areas linked to the continuum of both parks, known as the buffer zone (Brazil 2000, APN 2018, ICMBio 2018).

2.3. Records of illegal activities

We obtained the georeferenced records of illegal activities from the park administrations and law-enforcement agents (97.75% of records), researchers and academic research documents (2.25% of records). These records characterize the locations of illegal activities detected by law enforcement agents during inspections. Inspections were undertaken by accessing the area by car, boat or helicopter, and from the access points, travelling through the forest interior on foot for distances ranging from 3 to 15 km. For all activities, occurrence points located within the same pixel (with $100 \times 100$ m cell resolution) were considered as a single occurrence. Due to the diversity of infractions, we classified the records into four major groups (figure 2):

(a)
Base camps: these are exactly georeferenced sites where offenders camp and often process the products of their activities. Offenders usually spend at least two or three days consecutively at these sites, and the sites are likely to be used routinely over time. Some have considerable infrastructure, with showers, meat grinders, pans and smoke houses. Activities such as poaching and fishing do not occur directly at these sites, but rather in their proximity. The camps therefore facilitate illegal activities and increase the odds of success of these activities.
(b)
Poaching: sites where there was reliable and exact georeferenced evidence of poaching such as baits (known as ceva by local people). At these sites there is the addition of corn (maize), manioc (cassava) or fruits to lure animals. There are also salt licks (sites with the addition of mineral salt to attract animals; known as saleiros), standing hunting (hunting at ground level, without a structure), tree hunting (hunting using platforms in trees as support, known as jiraus), traps, gunshots, or sites where slaughtered animals were found.
(c)
Palm-heart extraction: exactly georeferenced sites where Euterpe edulis palm hearts were extracted. Euterpe edulis was originally a widely distributed palm tree and is a key species in this biome. It occurred throughout the Atlantic Forest region and also in part of the Cerrado biome. Although it still has a wide distribution, this palm tree is classified as 'vulnerable' by the National Center of Flora Conservation (CNCFlora 2021) because of illegal extraction.
(d)
Fishing: sites with reliable and exactly georeferenced evidence of fishing such as nets, fish pots, longlines with multiple hooks, fishing rods or fish captured at the fishing site.

**Figure 2.** Exact georeferenced records of illegal activities (base camps, poaching, palm-heart extraction and fishing) reported in both the Iguaçu and Iguazú National Parks.
Download figure:
Standard image High-resolution image

2.4. Environmental variables

We used four environmental variables (raster layers) in a 100 × 100 m resolution grid for all explanatory variables that were determined using QGIS software (QGIS Development Team 2018) to predict the occurrence of illegal activities in both parks. The variables were:

(a)
Edge distance: This represents park boundaries and reflects accessibility, and it is also close to the edge (up to 4 km) that poachers set traps or put out bait to attract animals (Wato et al 2006, Ferreguetti et al 2018). Edges can also offer escape points for offenders when the offenders see patrolling agents. Several edge lines now have cellphone signals, facilitating the communication between transgressors and informants about approaching law-enforcement forces. To estimate edge distance, we considered the limit of both parks, and we also manually drew the edges considering the fragments associated with both parks (supplementary material, figure S1). We then rasterized the edge, calculated the distance and cut the resulting raster map from the mask based on park limits and their buffers. The cells of the map assumed negative values outside the edge limits and positive values inside the limits of the delimited edge.
(b)
River distance: this reflects accessibility (Ferreguetti et al 2018), safety (because walking along rivers leaves no traces), water for drinking, cooking and bathing, harvesting, processing and places suitable for some species targeted by hunters. Three shapefiles were used to calculate river distance: we used the shapefiles from the hydrographic web and used the main rivers provided by the Iguazú and Iguaçu National Park administrations. The water mass shapefile was obtained from the Brazilian Institute of Geography and Statistics (IBGE) database (https://cutt.ly/wObg7ZS). We merged these shapefiles and deleted conflicting strokes to build the final shapefile, from which the Euclidean distance was calculated (supplementary material, figure S2).
(c)
Road distance: this reflects accessibility, especially by vehicles or motorcycles that facilitate access or escape by offenders (Watson et al 2013, Dias et al 2020). We used two layers, one with the distance calculated using all currently existing roads (current scenario) and the other including the proposed CC road as a hypothetical future scenario. For the current scenario, we obtained the shapefile of roads from the OpenStreetMap plugin in QGIS (https://cutt.ly/RhJPADx). We considered all roads (from highways to back roads) and we also manually drew the roads not included in the plugin. To calculate the road distance considering the opening of the CC road, we used the shapefile representing the current scenario and we considered the layout of the old road based on the route provided by Prasniewski et al (2020), saving it separately (supplementary material, figure S3). Euclidean distances were calculated from each shapefile.
(d)
Slope: this reflects accessibility and is related to the ease of movement of offenders especially on the gentle slopes. However, the proximity of steep slopes can also be important because this terrain can muffle the sound of gunshots (Haines et al 2012). We got the elevation layer from the NASA Shuttle Radar Topography Mission (SRTM) (https://cutt.ly/OhJJ5Bx; supplementary material, figure S4), from which the slope was calculated as a percentage, dividing the difference between the elevations of two points (rise) by the distance between them (run), then multiplied the quotient by 100. Values closed to 0 means flat slope while values higher than 60 represent very steep slopes.

2.5. Statistical analysis

2.5.1. Testing correlations and the effect of each explanatory variable on illegal activities

To test correlations between the explanatory variables, we calculated Spearman correlation with the Spearman method using the 'cor' function (stats package version 3.6.3) in the R environment (R Core Team 2020). Edges and roads showed high correlation (Pearson correlation = 0.70), and this can cause variance inflation in the models generating distorted predictions and making it impossible to separate the effects of edges and roads, so we restricted the modeling area to the location where there are roads inside the fragment, that is, the western portion of the parks (supplementary material, figure S5). This approach reduced the correlation between these variables (Pearson correlation = 0.40), enabling the dissociation of the effects of these variables and providing more reliable predictions in the scenario with the CC road.

We applied a Kolmogorov–Smirnov one-sample test using values of explanatory variables extracted from raster layers for each illegal activity to test if the observed frequency distribution was non-random. We implemented the Kolmogorov–Smirnov test using the function 'ks.test' (available in the R 'stats' package) to test the alternative hypothesis that the empirical frequency distribution is greater than random.

2.5.2. Building Niche-Based Models to predict susceptibility for illegal activities

To predict the geographical sites susceptible to illegal activities in the current scenario (without the CC road) and in the hypothetical future scenario (with the CC road), we used an ensemble forecasting approach based on different mathematical algorithms. We used two classes of algorithms. The first class was presence-only algorithms, represented by the Bioclim algorithm (Nix 1986) and the Domain/Gower distance algorithm (Carpenter et al 1993). The second class was presence/pseudo-absence algorithms, represented by the support-vector machines (SVM) algorithm (Tax and Duin 2004) and the maximum entropy (MaxEnt v. 3.3.3k) algorithm (Phillips and Dudík 2008). The combined use of different algorithms creates more-reliable predictions (Barry and Elith 2006, Diniz-Filho et al 2009). All niche-based models were built with both the dismo and kernlab R packages (Karatzoglou et al 2004, Hijmans et al 2020).

All models were generated separately for each illegal activity (i.e. base camps, fishing, palm-heart extraction and poaching). After fitting, we evaluated the models by randomizing the occurrence points of each illegal activity and dividing them into two subsets (training and testing) using k-fold sampling (k = 2) with 70% of the occurrence points used for training and 30% for testing. We repeated the randomization process 100 times for each algorithm to decrease the correlation between subsets, thus generating 400 models (100 replicates × 4 algorithms) for each illegal activity. The models were built for the current scenario for the western portion of parks and projected for the entire area for both the current and the hypothetical future scenario with the CC road. We then used ensemble forecasting (Araújo and New 2007) to determine a consensus map for each illegal activity. We estimated 'maximum specificity and sensitivity' threshold values for each model and then binarized the continuous suitability maps (presence or absence). Threshold values like the ones we derived are recommended when using presence-only algorithms for estimating commission and omission errors (Liu et al 2016).

Because we built the models for the western portion and predicted for the eastern portion, we needed to statistically test if the models were able to significantly predict occurrences both inside and outside of the modeling area. For this we used the formula $D = \sum{Xi(1 - Pi)}$ , where Xi represents the success or failure (1/0) of models to predict presence under a known occurrence and Pi is the proportion of the area predicted by the model as having presence (Pearson et al 2007). We associated the Monte Carlo simulations to calculate the empirical p-value for each generated model and for each algorithm in each repetition, both inside and outside of the modeling area. Therefore, inside the modeling area we calculate the D-observed at the test points, while outside of the modeling area D-observed was calculated using all existing occurrences in the area for each illegal activity. The formula for p-values was $P = 1-(r/n+1)$ , where r represents the number of times in which D-observed was greater than the expected D-random and n is the number of permutations. The expected D-random was calculated from 999 permutations considering a random distribution of occurrence points for each respective modeling area (west and east). To generate the suitability consensus map, we used only the significant binary maps both inside and outside the modeling area (west and east). The consensus maps (referred as the 'susceptibility map' for each illegal activity) had susceptibility values varying from 0 to the number of significant binary maps. After dividing by significant binary maps, the maps indicate the frequency of models that predicted the presence of illegal activities in each cell. Finally, we present a map for each illegal activity, with each cell showing values ranging from 0 (low susceptibility) to 1 (high susceptibility).

2.5.3. Predicting susceptibility differences with opening of the CC road

To test whether there were significant differences in susceptibility to illegal activities after the potential CC road opening, we performed a Wilcoxon paired signed-rank test with continuity correction among values of each illegal activity in the current scenario and with opening of the CC road. This analysis was performed only for the area under influence of the road, i.e. pixels that are now closer to roads in the hypothetical scenario (CC road opened).

3. Results

Considering just the exact coordinates where illegal activities occurred and were detected, we found 882 records of illegal activities (base camps = 121, fishing = 140, palm-heart extraction = 65, poaching = 556). After removing duplicate occurrences, leaving only one occurrence per cell for each illegal activity, and dividing the area into western and eastern portions, we successfully extracted 622 records of illegal activities to build susceptibility models with (base camps = 77/18 (west/east); fishing = 58/37; palm-heart extraction = 8/43, poaching = 260/121; figure 2). All illegal activities were most frequently closer to edges than would be expected if they occurred at random (figure 3), with these activities being more frequent than randomly expected in the first 2.5 km from an edge, but with the effect of edges extending up to 10 km. Except for palm-heart extraction, all illegal activities were also most frequently found closer to rivers than would be expected if their locations were random (figure 3), and these activities were more frequent than randomly expected in locations up to 500 m from rivers. Excluding base camps, all illegal activities were more associated with roads (especially in the case of distances up to 5 km) than would be expected if they occurred at random (figure 3). Finally, slope did not affected the occurrence of any illegal activity.

**Figure 3.** Change ratio of illegal activities (base camps, fishing, palm-heart extraction and poaching) in relation to edges, rivers, road distances and slope (%) in the western region (modeling area) compared to random.
Download figure:
Standard image High-resolution image

The ensemble frequency and the suitability consensus map for each illegal activity was built using the significant models, both inside and outside the modeling area (base camps: 227 from 400 binary maps; fishing: 108; palm-heart extraction: 23; poaching: 71). The niche-based models showed that sites closest to edges, roads and rivers were most susceptible to all illegal activities (figure 4), and if the CC road were opened, ca. 10 000 ha would be closer to roads (figure 5). Opening the CC road increased susceptibility in a fashion similar to what is currently seen in Iguazú National Park with its RN-101 road. The opening of the CC road, as suggested by the proposed laws, would increase the area susceptible to illegal activities (figure 6). Except for base camps, the areas under the influence of the road would all have significantly increased susceptibility, especially in sites that currently have low susceptibility (inside the fragment). For fishing, the median susceptibility value for illegal activities would increase from 0.009 to 0.101 with the opening of the CC road. The median for palm-heart extraction would increase from 0.087 to 0.260, and sites that currently have low susceptibility (0.05), would have a susceptibility of 0.5 with the opening of the CC road. For poaching, the median would increase from 0.324 to 0.334, and interior sites that current have low susceptibility (0.1) would increase their susceptibility up to 0.4 (figure 7).

**Figure 4.** Predicted current susceptibility to illegal activities in the Iguaçu and Iguazú National Parks. Values varies from 0 (low susceptibility) to 1 (high susceptibility).
Download figure:
Standard image High-resolution image

**Figure 5.** Area that will be affected by the opening of the CC road in Iguaçu National Park, in terms of distance to the nearby road.
Download figure:
Standard image High-resolution image

**Figure 6.** Predicted impact of the opening the CC road in Iguaçu National Park in terms of susceptibility to illegal activities and the difference between them. Predictions values varies from 0 (low susceptibility) to 1 (high susceptibility) while difference values varies from −1 (susceptibility decrease) to 1 (susceptibility increase).
Download figure:
Standard image High-resolution image

**Figure 7.** Changes in susceptibility to illegal activities in the area under the influence of the CC road if it were to be opened as proposed.
Download figure:
Standard image High-resolution image

4. Discussion

Our results corroborate other studies that show association of illegal activities (here represented by base camps, fishing, palm-heart extraction and poaching) with proximity to edges, roads, and rivers (Haines et al 2012, Bertrand et al 2018, Ferreguetti et al 2018, Torres et al 2018, Brodie and Fragoso 2020), but not with gentle slopes, probably because the slope variation in the region is low ( $7.07~\pm 4.45$ ; max = 65.06), except for the region of the Iguaçu Falls, which presents an accentuated and specific relief due to falling waters ( $24.02~\pm 20.79$ ; max = 171.52). In general, the region is formed by gentle and crossable slopes, which are characteristics that do not significantly affect the movement of offenders. The proposed opening of the CC road would increase the incidence of illegal activities in Brazil's Iguaçu National Park because it would make more sites accessible (ca. 10 000 ha). Both Argentina's Iguazú National Park and Brazil's Iguaçu National Park are affected by the adjacent deforested landscape that is formed by chemical-intensive agriculture and by pasture and urban areas, where the park boundaries are linked to a large network of edges and roads. The accessibility promoted by edges, rivers, official roads (BR-469, RN-12 and RN-101) and by unofficial roads are the main drivers of the increase in illegal activities susceptibility in these parks. Opening the CC road would make at least 10 000 ha of protected area be closer to roads, increasing this area's susceptibility to all of the illegal activities studied here except for base camps. Roads serve as pathways for threats to biodiversity worldwide and are an important factor for calculating the human footprint in ecosystems (Venter et al 2016).

The reliability of our prediction, of an increase in the susceptibility of illegal activities (except for base camps) after opening the CC road, is corroborated by the observed frequency of illegal activity along the roads inside and outside of the Iguaçu and Iguazú parks, especially the BR-469, RN-12 and RN-101 roads (see figure 4). The presence of these roads in Iguazú National Park makes almost its entire area susceptible to illegal activities (mainly fishing, palm-heart extraction and poaching). Since the models were good enough to represent the current context of the BR-469, RN-12 and RN-101, and the models took into account the existing roads and borders along the perimeter of the parks, their prediction of future change offers a robust projection of likely negative effects in Iguaçu National Park if the CC road were to be opened. The changes in susceptibility observed in the interior sites of the Iguaçu National Park when we added the trajectory of the proposed CC road is exclusively due to the effect of this internal road in the region, with the edges and external roads exerting their joint effects on the perimeter of the parks. So, the opening of the CC road represents a warning of the tremendous negative effects that downgrading of fully protected areas by economic pressures and political decisions can have on protected natural areas.

4.1. Spatial context of illegal activities, ecological consequences, and management actions

The illegal activities studied here were neither evenly nor randomly distributed in space. Base camps are located far from edges (1–10 km) and the proximity of roads (0–2 km) showed a negative effect on this activity; beyond 2 km the camps are randomly distributed. This could be a strategy to avoid being detected, since the camps and signs of camps such as campfires and conversation can be more easily detected if close to edges or to roads. Alternatively, there seems to be no advantage in establishing base camps close to edges and highways because offenders can easily enter and leave the area, spending less time there and, consequently, decreasing the chances of being apprehended. The concentration of base camps near rivers reflects the basic necessity of offenders who use water for drinking, cooking, bathing and to process the bushmeat and fish obtained illegally. Rivers are also used as 'roads' by offenders, leaving no mark of human passage and making it difficult to locate the camps and arrest the offenders. Field operations from the Iguaçu law-enforcement sector repeatedly observed this: the further from edges and roads, the more sophisticated were the base camps, which included smokehouses, many types of pans, meat grinders, tables and other items used in processing bushmeat and fish, as well as facilities for housing the offenders. The sophistication of these camps reveals that they are used for extended periods or visited periodically, enabling offenders to capture larger numbers of animals. The slope of the terrain does not affect the installation of base camps, probably because locations are chosen based on minimum requirements, such as proximity to rivers and isolation from roads, and because rough terrain does not affect the mobility of offenders over short distances.

Fishing activities necessarily occur in rivers. Rivers near edges and roads are preferred by offenders because edges and roads can reflect the aspects of accessibility, especially escape routes. Rivers close to an edge or a road enable these sites to be visited frequently, depending on the season or on conditions that are more favorable for fishing, such as immediately after a rain, which produces cloudy waters that, for example, increase the chance of capturing some Siluriformes (Baumgartner et al 2012, Matthews et al 2012). Fishing was also associated with lentic rivers, which are preferred by the most-targeted endemic fish species: Steindachneridion melanodermatum (de Assumpção et al 2017, ICMBio 2018). Slope does not affect fishing because it occurs in easily accessible areas (near edges and roads) or because offenders use boats, making walking and overland travel through rough terrain unnecessary.

Palm-heart extraction in both parks is directly associated with edges and roads, which provide accessibility and allow the quick flow of illegally extracted products. Offenders cut palms up to 5 km from edges and roads and transport the palm hearts to hidden locations close to an edge or a road, so neither accessibility nor movement are influenced by rough terrain. Using vehicles, usually at night, they quickly transport the palm hearts. Law-enforcement agents in both parks have reported up to 200 palm hearts apprehended at a single location and areas within the parks where more than 500 palms have been cut. Palm hearts are generally extracted in large quantities, and areas with large cuts are commonly found in both parks, increasing the pressure on this vulnerable species (CNCFlora 2021). The extensive perimeters of the parks make enforcement activities a major challenge (Bertrand et al 2018). River proximity did not influence palm-heart extraction, which was a surprise given that law enforcement agents report the use of boats for undertaking this activity. However, this may be due to the relatively low frequency of palm-heart extraction detections in the modelled area (8) compared to the predicted area (43), although palm hearts occur throughout the parks.

Poaching was often reported as associated with proximity to edges, roads and rivers, and this may indicate that offenders live close by and visit these locations whenever they feel safe. Because these locations are easily accessible, they both allow a quick escape and reduce the time spent in obtaining hunted animals. Poachers can also come from afar, using vehicles to transport their products. Poaching associated with water bodies reflects the fact that the strategies of offenders are based on the biology of target animals (Ferreguetti et al 2017, Dias et al 2020). Some species (such as Cuniculus paca, Hydrochoerus hydrochaeris, Tayassu pecari and Tapirus terrestris) use habitats more frequently near waterbodies (Mones and Ojati 1986, Reyna-Hurtado et al 2009, Ferreguetti et al 2017, 2018). Poachers build perches in trees that allow them to wait and shoot these animals as they approach the waterbodies. The non-significant influence of the slope of the terrain on the choice of poaching sites can be explained mainly by ease of movement in the region, mainly formed by flatter areas. This benefits poachers because they usually carry heavy items, such as water, weapons, and other supplies, and because they also must bring back the product of their activities (animals or their parts).

These illegal activities undoubtedly generate consequences for biodiversity and ecological processes. Poaching impacts fauna in both parks, reducing animal populations, and disturbing the ecological balance (Paviolo et al 2009, Bertrand 2016, da Silva et al 2018). For example, Tayassu pecari has its population reduced in Iguazú National Park (Paviolo et al 2008) and this species has been locally extinct for twenty years in Iguaçu National Park (Brocardo et al 2017). Loss of this species, which acts as an 'ecosystem engineer', impacts the plant community (Villar et al 2020), nutrient cycles (Villar et al 2021) and the size of top predator populations (Paviolo et al 2008). This also increases conflict between jaguars and cattle ranchers in the Iguaçu region; in the 1990s this conflict was attributed to the low availability of natural prey (de Azevedo 2008). Additionally, the decline of large animals, which are preferred by poachers (Fragoso et al 2011), may impact recruitment of zoochoric plants (Fadini et al 2009), plant gene flow (Giombini et al 2017) and the carbon-sink capacity of the forest (Bello et al 2015). Areas with palm-heart extraction have their Euterpe edulis densities decreased by up to 8 times, reducing seed and food availability and impacting animal species richness (Muler et al 2014). Overfishing impacts fish populations, skewing the size distributions of individuals and the composition of the aquatic food web (Allan et al 2005). In the Iguaçu River, fishing is concentrated on Steindachneridion melanodermatum, which may give this activity a particularly pernicious impact on the whole fish community because this fish species is endemic to the lower Iguaçu River and some of its tributaries (Garavello 2005) and because the main population is concentrated within the parks (de Assumpção et al 2017).

The high number of illegal activities and their consequences reported here highlight the urgent need to mitigate and reduce illegal activities in these protected areas. To achieve this goal, understanding the spatial distribution of illegal activities in protected areas can provide valuable information both for understanding how offenders behave and for providing input for law-enforcement strategies. When decision makers and law-enforcement agents know the areas that are most susceptible to infractions, they can develop more-effective strategies for patrols. Long-term actions can also be planned, such as the construction of surveillance posts, campaigns, and socioenvironmental projects to raise awareness among the population adjacent to the parks. It is fundamental to rely on scientific assessments rather than political or economic interests when suggesting the opening of roads and other so-called 'development' projects inside or close to protected areas, since the impacts on fauna and flora may be both irreversible and economically disastrous. Such considerations must substantiate the environmental legislation of both countries and the management plans for their protected areas.

4.2. Road effects in susceptibility to illegal activities

Roads are an important factor in structuring natural populations and communities. Because sites close to the roads are more accessible, they have greater depletion of target species. Consequently, roads can have additive or worsening effects such as fragmentation (Laurance et al 2018), population isolation (Laurance et al 2009) and changes in microclimate and soil (Delgado et al 2007). Opening the CC road will not accentuate the already established edge effect because no edge will be created with the CC opening, but it will accentuate the deleterious effect of roads in protected areas, especially in the western part of the park, which is narrower and, if the opening were to be approved, this region would be separated from the most-preserved portion of the park. Area close to roads maintain less vertebrate richness and harbor lower numbers of threatened species than do large areas without roads (Pfeifer et al 2017), such as is Iguaçu National Park in its current state. Additional negative effects of roads on these natural environment include alteration of nutrient cycling and release of greenhouse gases (Truscott et al 2005, Lee et al 2012), as well as limiting the movement of many species (Andrews 1990, Andrews and Gibbons 2005), isolating populations (Laurance et al 2000) and reducing genetic diversity (Holderegger and Di Giulio 2010). The CC road can increase the roadkills of both vertebrates and invertebrates (Andrews 1990), besides acting as a facilitator of biological invasion, which is a growing problem in the biodiversity hotspot where the parks are located (i.e. the Atlantic Forest; Bellard et al 2014). For instance, researchers and law-enforcement agents found 15 exotic plant species in and around a road used for tourism in the Iguaçu National Park (Rodolfo et al 2008). The CC Road can also facilitate logging, fires and all types of smuggling, invasions for agriculture and pasture, and extraction of minerals and forest products (Laurance et al 2002, 2009, Adeney et al 2009). A major road that cuts through the core of a protected area would bring the worst of these effects, impairing biodiversity and its associated ecosystem services and contradicting the protected area's main objective of biodiversity conservation. This is even more worrying because the Iguaçu National Park is the last and largest remnant of inland Atlantic Forest under protection (Ribeiro et al 2009). Opening the CC road would destroy part of this irreplaceable ecosystem.

There are many social, cultural and economic pressures behind the proposal to open the CC road: people claim that their right to come and go has been undermined, that their previous commercial and personal relations with both sides of the park (Capanema and Serranópolis do Iguaçu, in particular), have been forgotten and that their freedom has been taken away. Arguments for the road emphasize the costs of transportation between the western and southwestern portions of the state of Paraná, which would be easier because of the shorter distance to be traveled if passing through the park. Alternative suggestions that integrate the socio-cultural-economic context for implementation of the CC road are presented in Prasniewski et al (2020).

Those arguing for opening the CC road often refer to the presence of Highway BR-469 at the western end of Iguaçu National Park, which gives access to the Iguassu Falls. However, BR-469 meets the criterion of a 'parkway', which is for roads that provide access to areas with extraordinary beauty, taking into account their cultural, economic, touristic, entertainment and educational importance (Carr 1998), as contrasted with the CC road's function of linking cities or diminishing the costs of transportation.

We have shown that all highways or roads that are built in conservation units result in high rates of susceptibility to illegal activities. In addition, vehicle traffic causes soil compaction and water pollution that can extend into the food web through bioaccumulation (Dawson 2008). Conservation units cut by roads, such as the Taim Ecological Station (Rio Grande do Sul, Brazil), which is crossed by BR-471 and Sooretama Biological Reserve (Espírito Santo, Brazil), which is crossed by BR-101, show alarming roadkill rates (da Rosa and Bager 2012, Bager and Fontoura 2013, Klippel et al 2015, Srbek-Araujo et al 2015). This also happens on the currently existing roads in the Iguaçu and Iguazú parks. Although roadkill monitoring is not yet fully systematized, there were numerous records of threatened species being run over, such as Panthera onca, Leopardus pardalis, Sylvilagus brasiliensis and Mazama americana, among other vertebrates (Brocardo et al 2019). Since 2016, a research project has been underway on the Brazilian side regarding roadkill mitigation, but the results have not yet been published. Opening the CC road for transportation purposes would add roadkill to the other environmental impacts of the road.

4.3. Effect of precipitated political decisions and implications for other parks

The effect of rash or negligent policy decisions on environmental issues can have severe and irreversible effects on ecosystems. Discussion and evaluation of the risks of certain environmental actions therefore need to be based on prior scientific studies and extensive discussion with local communities. This is lacking in Brazil. Proposed laws usually stem from promises made in political campaigns and aim to favor lobbies and financial groups, neglecting the universal principle described in Article 255 of the Brazilian constitution: 'Everyone has the right to an ecologically balanced environment and a shared use of the people that is essential for a healthy quality of life, imposing on the public authorities and the community the duty to defend and preserve it for present and future generations' (Brazil 1988; free translation). Throughout the world there are many proposals for protected-area downgrading, downsizing and degazettement (PADDD). More than 3700 PADDD events have been reported globally, of which 90 are in Brazil, affecting 110 000 km² (CI and WWF 2019; see www.padddtracker.org for more information). The downgrading proposal that is underway for Iguaçu National Park means that the government intends to legally decrease restrictions on human activities in this conservation area despite its being a reference for conservation in the country. It is evident that the reasons for PADDD in general are strictly political and economic (e.g. mining, agriculture, transportation and extraction of forest products) rather than serving cultural, social or conservation purposes.

4.4. Potential of ecological niche models for monitoring, prediction and political decisions on illegal activities

Our results demonstrate that niche-based models are useful tools for studying phenomena that go far beyond the ecological niches of species, as reported in other studies (Banks et al 2006, 2013, Sobral-Souza et al 2015, Romero et al 2018, Guimarães et al 2020). Our models presented spatial patterns similar to those found by authors who used presence-absence data to predict illegal activities (Ferreguetti et al 2018, Moore et al 2018, Dias et al 2020, Marescot et al 2020). For example, predictive hunting models based on detectability generated the same spatial pattern as historical presence-only hunting data, i.e. when using presence-only data to model poaching, the authors found spatial patterns like those from models using detectability data (Ferreguetti et al 2018). Additionally, to get true absence for illegal activities is arduous due to the offenders' behavior, which includes deliberately covering up their tracks and stealing monitoring equipment.

We have demonstrated that the susceptibility model can be a powerful and agile tool for predicting the spatial distribution of sites susceptible to illegal activities in conservation areas and for predicting the impacts of political decisions (such as PADDD and opening roads). Georeferenced information on the occurrences of these activities is required for correctly fitting the models. Data of this type are available in databases around the world, so this tool can be validated based on new studies and can be used to guide law-enforcement, conservation, and management measures in the short, medium, and long terms.

5. Conclusions

Brazil's Iguaçu National Park is currently subject to substantial impacts from illegal hunting, fishing and palm-heart extraction, and these impacts would increase if the park were to be bisected by a proposed road. These activities impact biodiversity in the largest remaining remnant of inland Atlantic Forest which is a global biodiversity 'hotspot'. A niche-based model shows the magnitude of the increase in illegal activities, indicating the high environmental cost that the road would have. The model also identifies the locations of probable increases in illegal activity, providing valuable information for law-enforcement. However, the benefits of the study for law-enforcement would not compensate for the high impact of the proposed road, which the results suggest should not be built. The modeling tools developed in this study can be applied to many similar situations elsewhere, both for improving environmental control and for projecting impacts of infrastructure proposals before decisions are made.

Acknowledgments

We thank all scientists, law-enforcement field agents and conservationists who contributed information for our manuscript. Due to the risk of political and social retaliations we cannot name some of them. We also thank Ana T B Guimarães for statistical and computational support and Ariel (patrol Argentina) and Adaildo Policena for information on field activities. VMP thanks CNPq for a doctoral fellowship (Process 141645/2020-2). PMF thanks CNPq (Process 31103/2015-4). CFS thanks CNPq (Process 310353/2019-0). NS thanks Fundação Araucária (Edital 09/2021).

Data availability statement

The data generated and/or analyzed during the current study are not publicly available for legal/ethical reasons but are available from the corresponding author on reasonable request.

Brazil's Iguaçu National Park threatened by illegal activities: predicting consequences of proposed downgrading and road construction

Article metrics

Submit

Author e-mails

Author affiliations

Author notes

ORCID iDs

Dates

Peer review information

Abstract

1. Introduction