Are Protected Areas Effective In the Conservation of Priority Areas for Freshwater Insects In Brazil?

Biodiversity conservation has faced many challenges, especially the conversion of natural areas that compete with use for agriculture, energy production and mineral extraction. This problem is further aggravated by lack of knowledge of the biodiversity that exists and the geographical distribution of different groups. The objectives of our study were to examine the distributional pattern of Gerromorpha diversity in Brazil, create a map of conservation priority areas, measure the importance level of the current network of protected areas, and identify the size thresholds in geographical distributions that would allow species to be protected. We used species occurrences from the Water Bugs Distributional Database, and we used 19 bioclimatic variables to build models of the potential distributions of species using the MaxEnt program. Using the potential model results, we calculated diversity metrics and overlapped them with the current state and federal “conservation units” (protected areas for biodiversity) in Brazil. Total beta diversity and turnover portions were separated into two faunistic groups, one in northern and the other in southern Brazil. The Amazon has higher beta diversity than what was predicted by the null models. We detected a positive relationship between species distribution area and occurrence in conservation units. Conservation units with less than 250 km² do not protect Gerromorpha species. We reinforce the necessity of formulating new conservation strategies for this group, contemplating species with both restricted and ample distributions, because rare and specialist species are the most harmed by habitat reduction, given that they are more sensitive to environmental disturbance. extraction


Introduction
Biodiversity conservation has faced many challenges in the past few years, especially because of the speed with which natural areas have been converted to areas for agriculture, energy production, mineral extraction and real-estate development 1 . This challenge is even bigger when areas of economic interest overlap with areas that are important for biological conservation 2,3 , creating socio-economic con icts. In addition, the de nition of areas for conservation is intended for ag species, synergistic species or species with a social appeal 4,5 , leaving aquatic and terrestrial insects out of this planning 6 . One way of diminishing this problem is by creating and implementing conservation units to conciliate nature conservation with sustainable resource use, making a human presence in protected areas compatible with biodiversity conservation. In Brazil there are 12 types of "conservation units" (CUs) (Unidades de Conservação in Portuguese), or protected areas for biodiversity 7 . The different types of CUs differ in the level of access permitted; some are similar to ecological sanctuaries and have effective protection, being restricted to scienti c research. Some unit types are less restricted and allow use for recreation (e.g., National Parks) and even for resource exploitation (Extractive Reserves). Generally, each protected area has rules regarding its use. These areas shelter a high diversity of species, ecosystem services and traditions. CUs provide biodiversity maintenance, and (in some cases) biodiversity recovery 7 . However, not all protection measures are complete, sometimes being ine cient because the areas tend to be selected arbitrarily based either on the empirical knowledge of researchers or by being placed in areas with di cult access, which limits competing interests for agriculture or for real-estate development.
Selection may be done considering only the speci c characteristics of a restricted biological group, usually only accounting for terrestrial biodiversity 8 . Brazil's current CU network is insu cient and does not adequately protect biodiversity 6,9 . The selection of priority areas for conservation must consider an area's representativeness, irreplaceability, exibility, complementarity and persistence 10 . Area selection is based on biogeographical and ecologically georeferenced information, so as to permit analyses and interpretation at different scales 11 . However, it is often impossible to fully protect these components due to the lack of biogeographical information on species 3,12 . In the Neotropical region the lack of knowledge is acute regarding real species distribution (the "Wallacean shortfall") 12 . For instance, in Brazil, biogeographical information on species is scarce, even for biodiversity hotspots. Lack of information, together with the lack of nancial resources and the overlapping interests between agricultural development and conservation are the main challenges in de ning priority areas for conservation 6, 13 . In addition, we do not know whether the current reserve network is su cient to protect all the taxonomic groups. A study carried out with bats in the Cerrado showed that species with restricted distributions tend not to be covered by conservation units 14 . This result leads us to question the effectiveness of conservation units, especially for key species in ecosystems (such as aquatic insects, which are not charismatic).
Ine ciency of reserve networks may be related to non-protection of rare species 6 or to choosing areas that are marginally adequate for species distributions 15 . One way of reducing ine ciency is by using other criteria for area selection, e.g., species richness, beta diversity, fauna complementarity and the number of endemic species 3,16 . These parameters, along with use of species-distribution models (SDMs) could reduce the problem of lack of knowledge about distributions that leads to protection of marginally adequate areas. This knowledge would allow selection of areas having high probability of occurrence of a species and that provide favorable environmental conditions for its ecological niche may be selected. Although this approach has been robust for conservation, the criteria used are generally based on a few groups of vertebrates and plants, which reduces the effectiveness of protected areas for conserving the biodiversity of all groups. Consequently, several biodiversity components, such as aquatic invertebrates, might not be included in the protected areas 6,9 .
In aquatic ecosystems, macroinvertebrates stand out from other groups because of their sensitivity to environmental impact 17,18 and their role in nutrient cycling and energy transfer in food chains. Insects in the suborder Heteroptera are predators, are at the top of food chains, can respond to changes happening in lower trophic levels, and are considered to be good models for evaluating environmental impact 19 . Heteroptera occupy an ample variety of habitats, including waterbodies that are either lotic or lentic and either perennial or temporary; these insects have an important role in biological control in waterbodies [20][21][22][23] . Heteroptera is composed of three infraorders: Nepomorpha (truly aquatic), Gerromorpha and Lepdomorpha (both semi-aquatic). The suborder Gerromorpha is divided into 20 families, 325 genera, and approximately 4700 species inhabiting freshwater ecosystems: 28% of these species are in the Neotropics 24 . Most Gerromorpha species live over the water column, either on oating plants or between plant roots on the edges of bodies of fresh water 25 . Thus, they are affected by forest removal and increased water ow in streams 22,26 . Approximately 2100 species have been described for Gerromorpha, which currently has eight families and about 160 genera 24 , of which 206 species and subspecies occur in Brazil 27 . This group was chosen because it occupies various types of aquatic ecosystems, has an ample geographical distribution, and perhaps acts as an umbrella or surrogate group encompassing additional aquatic species that inhabit the same sites 28,29 .
Due to the speed with which natural areas have been modi ed by anthropogenic activities and the lack of knowledge about the state of conservation of aquatic insects, our objectives are to: (i) describe distributional patterns of Gerromorpha diversity (Richness and Beta Diversity) among Brazilian biomes; (ii) create a map of conservation priority areas for this group; (iii) assess the importance of the current network of conservation units and (iv) identify the threshold in geographical distribution size needed to protect species under the current Brazilian system of conservation units.

Materials And Methods
Species occurrence data Species occurrences (Fig. 1) were obtained from the Water Bugs Distributional Database (available at https://sites.google.com/site/distributionaldatabase/). This database represents an extensive research effort to obtain information from the literature. The database contains occurrence information for all Gerromorpha species known in Brazil together with the sources from which the information was obtained. More information about the database is available Dias-Silva et al 30 ,

Environmental Variables
Initially, species-occurrence data were transformed into decimal degrees (in datum SAD 69), using the SpeciesLink website. In order to avoid any bias caused by geographically imprecise data, we eliminated occurrences where the distance between point coordinates and the seat of the municipality was less than 2 km 31,32 .
We used 19 bioclimatic variables to build the models; the values were obtained from monthly data on temperature and precipitation available on the WorldClim platform (Hijmans et al., 2005). The climatic variables have a resolution of approximately 9 km (≈ 0.083 decimal degrees) and are highly colinear. In order to solve or reduce the multicollinearity problem, we performed a principal component analysis (PCA) 33 , from which we extracted seven axes that explained, in total, 95% of the variation in the original dataset. These axes were used as predictive environmental variables in the potential species distribution model (SDM). We tried therefore to reduce the multicollinearity problem of the original variables, making it possible to model species distributions with a small dataset on occurrence 34 . Distribution models and model evaluation Potential species distributions were estimated using the MaxEnt (Maximum Entropy Modeling) program version 3.3.3 35 . To analyze the performance of the potential species distribution model, we used two techniques: AUC (area under the curve) and TSS (true skilled statistics). The AUC measure, which is the area under the ROC (receiver operating characteristic) curve, has scores varying between 0 and 1, where values ≤ 0.5 represent similar predictive models obtained randomly and values ≥ 0.7 are considered acceptable 36,37 . True skilled statistics (TSS) was also used to evaluate model quality, this measure varies from − 1 to 1, where scores close to zero and negative scores are no better than random scores, and scores close to 1 have a perfect adjustment between the observed distribution and the predicted distribution 38 . Values ≥ 0.5 indicate acceptable models. To convert predicted species distributions into presence/absence maps, we used threshold values 39,40 that were derived from the ROC curve. Use of the maximum threshold value maximizes the speci city and sensitivity of models and tends to produce more-restricted distributions. Additionally, we performed a Monte Carlo simulation test to evaluate the dependency of AUC on the number of occurrences 6 ( Table 1). Thus, we divided the data into ve classes (with an interval of 15 locations between the classes) and considered the rst class (4-15 locations) to be the control, using the mean AUC as a critical value. For AUC value simulations we maintained the species number inside the classes equal and calculated the mean AUC of random species. This procedure was repeated 10,000 times and the simulated AUC values greater than or equal to the values for the control class were considered to be scores obtained at random. This procedure was repeated for all classes, and the result was not signi cant. This means that there is no relationship between the number of sites and AUC values, indicating that the models are reliable 41 . To identify the spatial distribution of Gerromorpha diversity in Brazil, we used a grid with a resolution of 0.083° (the same resolution as the SDM) and extracted information on species presence and absence from potential distribution maps. Therefore, each row represents a grid cell and each column represents a species, and the spreadsheet cells represent species presence or absence. Using this table, we calculated the total species richness of each cell based on the sum of lines and beta diversity, using the procedure described by Baselga et al. 42 , in which diversity is partitioned into portions for nestedness and turnover, and the sum of these portions represents total beta diversity. For the spatialization of beta diversity (the total of the nested and turnover portions), we performed a principal coordinate analysis (PCoA) 33 with each diversity matrix, and the rst PCoA axis was used as a synthesis of diversity per cell. Thus, we obtained values for richness, total beta diversity, nested beta diversity and turnover beta diversity for each grid cell. This allowed us to spatialize diversity values by categorizing the cells into color gradients. This procedure was performed for each of the diversity measures.
We used a Monte Carlo randomization procedure with 10,000 randomizations to evaluate diversity distribution among Brazilian biomes. For this procedure we developed a routine in the R environment that classi es grid cells according to the biome to which cells belong and calculates the mean and standard deviation of the diversity metrics for each biome. We chose cells randomly for the signi cance test (always controlling the number of observed cells in the biome) and calculated the mean random diversity.
To calculate the signi cance value, we identi ed the number of random values higher than the observed values and divided them by 10,001. This procedure was performed for richness and total beta diversity.
To determine priority areas for conservation of aquatic species in Brazil, we used the Zonation algorithm 43 , along with data on the potential distribution of an indicator organism (Gerromorpha). Zonation is a quantitative method that prioritizes conservation areas, aiming for long-term biodiversity persistence 44 .
Evaluation is done by randomly removing cells. This analysis can be performed using three algorithms: Core-area Zonation, Additive bene t function and Target-based planning. For this study, we used the additive-bene t function because it is best suited for a high number of species. This analysis initiates with the complete landscape, where sites are classi ed based on biological value (complementarity and irreplaceability), and the less-valuable cells are removed one (or more) at a time, producing a sequence of landscape structures with more resources for biodiversity 45 .
To evaluate the difference in importance between protected and unprotected cells, we used a second routine in the R environment 46 based on the Monte Carlo randomization test with 10,000 randomizations. Cells were classi ed into those that were "protected" (if they were totally or at last 50% within a conservation unit) and those that were "unprotected," and the mean importance of protected cells was estimated. The same number of unprotected cells was selected randomly, and we estimated the mean importance. The estimate of the random importance value was calculated 10,000 times, thus creating the expected distribution of the importance values. This procedure was performed for the reserve network of Brazil as a whole and for the reserve network inside each biome. The signi cance value was obtained by dividing the number of values that were greater than or equal to the observed values by 10.001 47 . The reserve network considered for the analyses was the o cial map of conservation units in the "full protection" category, which is available on the Brazilian Ministry of Environment (MMA) website (http://mapas.mma.gov.br/i3geo/datadownload.htm). The o cial biome borders (also available on the MMA website) were considered to represent the historical distribution of the biological biomes.
To evaluate if there is a minimum size for a species to be protected by a conservation unit, we performed a logistic regression between species distributions and their presence/absence inside the units. Species distribution area was estimated using the sum of cell numbers in which the species occurs, and the cells are classi ed as being inside (1) or outside (0) of conservation units. A non-linear logistic regression model was performed 41,48 , where presence/absence was considered to be the dependent variable.

Results
Of 208 Gerromorpha species recorded in Brazil, we found 3541 occurrences and modeled 111 species (only species with three or more occurrences are suitable for this procedure) (Table S1). Signi cant portions of the species-occurrence sites were in the Southeastern Region of Brazil, in the Atlantic Forest biome, and in the Northern Region (Amazon) (Fig. 1). The areas with the highest richness are the northern portion of the Amazon, eastern and western Cerrado, and a small coastal portion in the northern Atlantic Forest (Fig. 2a). The Atlantic Forest and the Amazon have higher richness than predicted (Table 2); the northern portions of the Cerrado biome, the Caatinga biome and the central portion of the Amazon biome had the lowest expected richness (Fig. 2a), and these values did not differ from what was predicted. When evaluating total beta diversity, the highest values were found for the Northern Region of Brazil (Fig. 2b), and the Amazon biome had higher beta diversity than predicted (Table 2). However, when evaluating the components of beta diversity, the turnover portion (Fig. 2d) was greatest in the Southern Region of Brazil (Fig. 2c). On the other hand, in the nested portion we found the areas with the lowest species richness values (the northern Cerrado and Caatinga, as well as the central Amazon); these are the areas with the highest nestedness values for Gerromorpha (Fig. 2d). The map of cell importance for conservation (Fig. 3) shows that cells in the northern Amazon, the eastern and western Cerrado and in the coastal portion of the Atlantic Forest are important to Gerromorpha conservation, and many locations that are considered priority areas are outside of conservation units. In general, the biomes did not differ in terms of importance for the conservation of Gerromorpha (Table 3), and we observed for both the general model (evaluating conservation units in Brazil as a whole) and the regional models (evaluating the CUs in each biome) that conservation units do not protect either more or less areas that are considered to be important for Gerromorpha than what was randomly expected (Table 3). Lastly, we observed a positive relationship between species distribution areas and species occurrences in conservation units (x² (1) = 36.144 p < 0.001). The minimum occurrence for a species to be present within a conservation unit is approximately ve cells, i.e., 250 km² (Fig. 4).

Discussion
Cells protected by conservation units (CUs) have a lower degree of importance than expected. This low importance, determined according to Gerromorpha biodiversity, may be related to the fact CUs have been created arbitrarily, probably based on economic interest or to protect speci c taxonomic groups 3 that are not aquatic 9 . These results indicate that the current network of CUs does little to protect Gerromorpha species, especially those with restricted distributions, and especially species with distributions smaller than 250 km². CUs may be protecting sites with marginal environmental suitability for the occurrence of these species (e.g., Nóbrega  Creating new conservation units is a slow process because creating them depends on nancial resources to buy land, provide area maintenance and pay teams working in the unit, besides the various impediments imposed by the con icts-of-interest mentioned earlier. Thus, an economically viable possibility would be to maintain existing forest fragments and encourage farmers to maintain legal reserves and riparian vegetation. Forest remnants and legal-reserve areas would act as steppingstones for biodiversity, helping maintain connectivity among CUs. In addition, riparian vegetation maintenance keeps catchments fully preserved, further enhancing connectivity among remnants, legal reserves and CUs. Finally, we emphasize that, although the maintenance of riparian forests is considered to be su cient for maintaining the Gerromorpha stream community, we must include other areas besides the riparian strip. This maintenance should be done in strategic areas, such as the ones presented in this paper, in order to avoid the risk of preserving only species with wide occurrence ranges and leaving rare and restricted species outside the network of CUs. In terms of species richness, we found that the Atlantic Forest and the Amazon biomes shelter the highest number of Gerromorpha species per unit area, and the richness in these biomes is higher than expected, suggesting that the habitats of these biomes are more heterogeneous than in the other Brazilian biomes. This idea is supported by the results on total beta diversity, suggesting that the Gerromorpha composition in these biomes results from conspicuous environmental gradients. On the other hand, we observed the Pampas biome to have fewer species than expected, suggesting that these areas have low heterogeneity for the group. If we consider Brazil's vast river network we can conclude that the number of records we have is small, providing an example of the "Wallacean shortfall" 54 , where information about species distributions is still insu cient. Additional sampling is needed so that information on species distributions can be expanded and more occurrence sites can be identi ed, enabling the discovery of new species and the generation of information on species distributions to assist in choosing conservation areas 51 . We also emphasize the importance of evaluating the extreme northern portion of the Amazon and the southeastern Cerrado, which are very relevant for conservation and have high species richness but are not well studied. Furthermore, the southeastern portion of the Cerrado stands out for its large turnover component, suggesting a strong environmental lter, possibly due to the existence of small fragments of vegetation. The northern portion of the Cerrado and the Caatinga stand out for their nesting components, suggesting that dispersion is easily observed in these areas. This result has already been observed for bats in the Cerrado, and the nesting component was justi ed by the existence of large preserved areas in the northern portion of the biome, which is facilitated the mobility of bats 14 . The turnover component was justi ed by the local extinction of some species because the southern and southwestern portions of the Cerrado are highly fragmented, which hampers dispersal and may cause local extinctions of species. The high diversity of Gerromorpha in these regions is also congruent with other biological groups, e.g., birds, reptiles, amphibians and non-ying mammals 55,56 . Hence, by preserving the diversity of Gerromorpha, we are also preserving the diversity of other groups, and ensuring that ecosystems can sustain themselves and recover from stochastic events or even climate-change events. We therefore stress that CUs de ned with little or no planning have low importance for Gerromorpha conservation and do not contemplate species with restricted distributions (less than 250 km² for Gerromorpha). New strategies for the conservation of this group are necessary, especially strategies that conserve both species with restricted ranges and those with ample distributions 6 because rare and specialist species are the most affected by habitat reduction and are the most sensitive to environmental disturbance. There is no con ict of interest.

Data availability statements
All occurrences data are available in Water Bugs Distributional Database (https://sites.google.com/site/distributionaldatabase/) Figure 1 Gerromorpha species occurrences used in the species distribution models (SDM) Figure 2 Predicted spatial distributions of a) richness, b) total beta diversity, c) turnover and d) nestedness of Gerromorpha in Brazil Page 18/19

Figure 3
Priority areas for Gerromorpha conservation in Brazil, according to the Zonation algorithm. Values indicate the importance of each cell; the higher the value, the higher the importance