Multicriterial evaluation in the definition of protected areas at the Piedade municipality, SP.

ABSTRACT The preservation in situ, through the establishment of legally protected areas, is one the approaches to mitigate environmental impact and protect biodiversity in the long term. The Piedade municipality shows a favorable set of conditions for consolidating protected areas (i.e. relatively low urbanization rates, the presence of forest remnants, and portions of relevant conservation units). In this context, the present study proposed the definition of priority areas for protection using Multicriteria Evaluation (MCE). The specific objectives were to identify the criteria and weights that are important for defining protected areas and to evaluate which method of MCE, whether the Weighted Linear Combination (WLC) or the Ordered Weighted Averaging (OWA) would be more appropriate. Using the Participatory Technique (PT), we defined criteria and weights. WLC and OWA made possible the identification of areas likely to become protected, but the first method produced more general and less flexible solutions. On the other hand, OWA provided a greater number of possible answers to the question of interest. Thus, showing the low-risk level, the OWA map was considered most suitable for the study proposal.


INTRODUCTION
Currently, biodiversity is under threat due to human consumption needs and expansion. So, the preservation in situ, through the establishment of legally protected areas, has been one approach proposed to mitigate environmental impact and protect biodiversity long term.
In this context we have the Piedade municipality, showing a favorable conditions for establishing protected areas, such as relatively low urbanization rates; presence of forest remnants, that occupy approximately 25% of the municipality; and portions of relevant Conservation Units (CU), that are named Itupararanga Environmental Protection Area (EPA), in the northeast region, and the Jurupara State Park (PEJU), in the southeast region.
Multicriteria Evaluation (MCE) is a technique that can support the protected areas defi nition. It allows the transformation and combination of diff erent criteria, considering its relevance, and infl uence level, in order to provide solutions/alternatives for the decision-making process.
The Weighted Linear Combination (WLC) and Ordered Weighted Averaging (OWA) are among the MCE methods. The fi rst involves a criteria normalization to a common scale, the defi nition of criteria weights (i.e. factors weights), and their aggregation using a weighted average (Voogd, 1983). Such methodology provides medium risk solutions, and it is considered an OWA variation (Malczewski;Rinner, 2015).
Diff erently, OWA not considers only the factor weights, having also a second group of weights called order weights (Yager, 1988), that control the trade-off levels among the criteria and the decision process risk-taking .
Using OWA, the solution can be located between the extremes (risk-taking and risk-averse), considering that this position (or factor order) will infl uence the risk level (Malczewski, 2004).
WLC has been used for many purposes as the environmental quality analyzes and valuation of ecosystem services (Comino, et al., 2014), selection of possible locations for solar power plants (Zoghi et al., 2015;Doorga;Rughooputh;Boojhawon, 2019), for biomass plants (Jeong;Ramírez-Gómez, 2017) and for anaerobic digestion facilities for food and biodegradable waste (Babalola, 2018); also studies on ecological-urban vulnerability Lin, 2015), evaluation of accessibility to lignite deposits (Blachowski, 2015), defi nition of priority areas for forest restoration in urban environments (Valente et al., 2017), mapping of landslide susceptible areas (Lorentz et al., 2016), monitoring of natural areas considered World Heritage Sites by UNESCO Du, 2016), indication of suitable locations for electrical vehicle charging stations (Costa et al., 2018) and disposal of hazardous waste (Danesh et al., 2019).
In this context, this study aimed at the identifying of priority areas for biodiversity protection at the Piedade municipality, using Multicriteria Evaluation Methods (MCE). The specifi c objectives were the criteria defi nition, considering their weights for the prioritization of areas; and the WLC and OWA methods evaluation for the defi nition of biodiversity protection areas.

Study area
The Piedade municipality is located at the southeast portion of the São Paulo state, between coordinates UTM 253551-257159 W and 7385600-7345310 S, having an area of approximately 747 km² and population around 55.092 habitants (IBGE, 2017).
In the northeast portion of the Piedade is the Itupararanga EPA and in the southeastern portion is the PEJU, highlighting its considerable richness of fl ora and fauna species (Figure 1).
The PEJU management plan mentions that the buff er zone is composed by portions of the municipalities (i.e. Piedade, Ibiúna, Juquitiba, Miracatu, Juquiá, and Tapiraí) and, also by parts of the Serra do Mar EPA and the Itupararanga EPA, resulting in a mosaic with diff erent landscapes (Fundação Florestal, 2010).
The municipality shows altitude of approximately 800 m and tropical altitude climate (Cwa -Köppen classifi cation) with mean annual temperatures varying from 13.5°C to 25.7°C and 1354.7 mm of rain (Cepagri, 2013). Located at the crystalline Atlantic Plateau, the municipality characterizes by sharp mountainous relief (Prefeitura de Piedade, 2012).

Criteria
The criteria were defi ned through the literature review and Participatory Technique (PT), considering the opinion of nine experts, representing areas as geoprocessing, nature conservation, and landscape ecology. Considering grades from 0 to 10, experts associated weights to criteria, took into account their importance to defi ne protection areas. We selected the criteria set most mention in the TP, which were: proximity among forest patches with larger core, neighbor of forest remnants, distance from disturbance sources, and proximity to protected areas (i.e. PEJU and Itupararanga EPA). The experts mentioned other criteria, but, some of them did not in accordance to the study objective or, they were indirectly included in selected criteria.
We produced the maps in a GIS environment (Idrisi Selva and ArcGis 10.0), a standardizing the cartographic database to Corrego Alegre Datum and UTM 23S coordinate system, following a pre-existing database.

Proximity among forest patches with larger core
Core areas correspond to an internal area of space elements and in the case of the forest remnants, the edge-equivalent part is excluded. Peripheral areas, due to edge eff ect, especially in the fi rst 35 m (Rodrigues, 1998), are generally avoided by sensitive animals to potential disturbances. Small patches or patches with great edge eff ect present core areas that tend to zero (Lang;Blaschke, 2009). Edge eff ects contribute to increasing rainforest destruction (Primack;Rodrigues, 2001).
Large and interconnected patches have a trend to shelter a greater number of species and populations less vulnerable to extinction than smaller isolated patches (Geneletti, 2005). A single patch can possess considerable width to hold some species but may not have enough core area to support a viable population (Turner; Gardner, 1990). According to Valente and Vettorazzi (2008), proximity among forest patches with larger core was considered the most important factor when defi ning priority areas for forest conservation. Therefore, for the present study, the regions with the largest core patches were considered more relevant.
The proximity among forest patches with larger core map ( Figure 2A) was produced by Gasparoto et al. (2011), considering the layer core area (i.e. core areas of the forest patches).
Firstly, the core areas (in hectares -ha) were classifi ed into fi ve size classes (smaller than 50 ha, 50 to 100 ha, 100 ha to 150 ha, 150 ha to 200 ha, and larger/equal to 200 ha). After we attributed weights (values from 1 to 5) for the classes, according to their size (i.e. lower weights were attributed to smaller forest patches).
After this fi rst step, distances between the fi ve classes were calculated, core areas were recombined and standardized on a scale from 0 to 255 bytes. Values close to 255 bytes corresponded to patches with large cores areas close to other patches also with large cores areas and were considered the most important regions.
Proximity among forest patches with larger core map is important because when isolation among forest patches with large cores areas decreases, the chances of reestablishing landscape structural connectivity rise, allowing future functional connection and genetic fl ow between populations.

Proximity to protected areas
Areas close to protected areas can promote landscape connectivity, ensuring the long-term viability of fl ora and fauna (Ribeiro et al., 2012). An ideal scenario would be one with large forest remnants, few recesses and an approximately circular shape (Matsumoto;Kumler;Baumgarten, 2012).
Fragment size has a strong eff ect on edge and inland species but is irrelevant to generalist species Contreras;Fahrig, 1998). Because of this, large and close fragments were prioritized in a connectivity study and planning of possible Integral Protection Conservation Units (Metzger et al., 2008). Krishnadas et al. (2018) found that protected areas, when away from highways, reduced forest loss (reducing deforestation by 88%) in a global biodiversity hotspot in India.
Considering the limits layer of the protected areas and a distance algorithm in the GIS, we obtained the distance maps to protected areas (i.e. Itupararanga EPA and PEJU), that were normalized to the common scale (i.e. 0 to 255 bytes), through a decreasing linear function ( Figures 2C and D). Thus, regions near to the limits of the protected areas are more appropriate to be a protected area.

Distance from disturbance sources
The implementation and expansion of urban areas often perform interconnected environmental impacts such as deforestation, disturbances in the hydrological cycle, imperviousness of soil (thereby reducing the recharge of aquifers) and increased runoff , favoring problems such as fl oods, and several types of pollution generated by anthropic occupation, which lead to the spread of diseases, among other consequences (Castro, 2007).
According to Kamwi et al. (2018), an enhance in the distance of protected areas from settlements reduces the likelihood of conversion from forest to bare soil or crops and pastures. Neighboring urban centers, even when showing low population density, exert strong pressure on forest remnants (Zimbres;Machado;Peres, 2018). In a study conducted by Rodríguez-Rodríguez and Martínez-Vega (2019), the distance from cities was the variable that most infl uenced on the isolation of protected areas. Palomino and Carrascal (2007) observed biotic uniformization in birds due to the proximity to an urban area and thereby, urban explorer birds were the only ones that had a positive association with roads. They noted that most of the disturbances associated with urban areas do not diff er much from those related to highways.
Road construction is considered one of the causes of fragmentation (Schonewald-Cox;Buechner,1992) that involves two distinct processes, the breaking apart of habitat and, habitat loss (Farihg,2003). The last one is considered the biggest threat to biodiversity worldwide (Tabarelli;Gascon, 2005). Moreover, Maués and De Oliveira (2010) mention that fragmentation can reduce gene fl ow rates and tree recruitment and fructifi cation, and increase the possibility of inbreeding, also aff ecting pollinator populations which can have narrow coevolutionary relationships to tree species.
Roads drive changes in land use and infl uence deforestation (Freitas et al., 2010). Another relevant aspect is that marginal areas of highways present a high risk of forest fi res due to car accidents and bonfi res produced by passers-by (Silveira et al., 2008). De Jesus Silva et al. (2006) reported the importance of considering the eff ects of road network density on biodiversity when implanting a CU because highways can disrupt the interaction between seeds dispersal by rodents and plants, impacting in regeneration and composition of forests (Chen et al. al., 2019). Besides, they make it possible to raise the invasion of exotic plant species into the landscape and facilitate future occupations (Parendes;Jones, 2000), leading to a homogenization of species richness and plant community patterns in mountainous regions (Haider et al., 2018;Medvecka et al., 2018). Sousa et al. (2009) also highlighted the need for considering road distance as a determining factor in the distribution of native vegetation and CU.
Another factor to be considered is that these structures enlarge the eff ective distance, i.e., the "cost" to cross them increases (Metzger, 2006). Some animals tend to avoid roads (Mcgregor et al., 2008), such as gazelles that prefer to move away from habitats with high noise pollution, causing loss of functionality of these sites (Ghadirian et al., 2019). Road lights can disorient birds and noise can aff ect their reproductive behavior (Glista;Devault;Dewoody, 2009). The opposite eff ect may also occur, since the presence of carcasses of roadkill may induce the approach of carnivores and other animals, such as skunks that are attracted by the accumulation of garbage, increasing the risk of new accidents (Bueno;and Almeida, 2010).
In this context, we considered the urban areas and the roads as disturbance sources. These features were extracted by the land-use/land-cover map (Figure 1), composing the disturbance sources map. Firstly, we calculated the distance from the disturbance sources and, after, the distance as from them. The distance map was normalized (0 to 255 bytes scale) using a decreasing linear function, which high priority areas (close to 255 bytes) corresponded to regions more distant from urban areas and roads. Viana and Pinheiro (1998) found that strategies for biodiversity conservation in CU must extrapolate their boundaries and consider the characteristics and conservation potential of neighboring fragments, based on a study carried out in a forest patch surrounded by both pastures and Pinus sp. plots, in which pastures exerted higher edge eff ects than plots. Therefore, depending on the neighborhood, more intense edge eff ects may or may not occur in forest patches.

Neighbor of forest remnants
Mazarolle and Villard (1999) noticed that local habitat conditions are not always adequate for explaining species abundance and boundaries and that neighboring characteristics (landscape context) should also be considered. Hence, both the surrounding matrix and core areas must be considered in landscape planning to ensure the persistence, resilience of ecosystem functions, biodiversity and ecosystem services (Svensson et al., 2019).
Thus, diff erent weights were attributed to landuse/land-cover classes in the map of neighboring forest remnants ( Figure 2E). The most important classes received higher weights, with the soil exposed receiving the same weight of the agriculture. On the other hand, we did not consider the drainage network. This fi rst map was overlapped with the distance to forest patches map.
So, considering the forest patches (also extracted from land-use/land-cover map), we calculated the distances among patches, that was normalized to the common scale, by a decreasing linear function. After, the two maps were overlapped and, the product was normalized again resulting in the neighbor of forest remnants criterion map.

Factor weights (Fw)
Factor weights indicated by the PT were readjusted according to a continuous scale of values and evaluated under the Analytical Hierarchical Process (AHP) (Saaty, 1977) using a Pairwise Comparison matrix. Consistency Rates (CR) were generated and the smallest value (0.06) diff erent from 0.00 was chosen. Factor weights obtained using the matrix were 0.3697 for proximity among forest patches with larger core; 0.2028 for distance from disturbance sources; 0.1571 for both proximity to PEJU and proximity to the Itupararanga EPA; and 0.1133 for neighbor of forest remnants.

Criteria aggregation by Weighted Linear Combination (WLC)
The criteria were combined using a weighted average (i.e. WLC algorithm), producing a fi nal map of priority areas for biodiversity protection. After, evaluation the map histogram, we defi ned the class range.
According to Jiang and Eastman (2000), WLC assumes 100% trade-off among criteria, considering that the factor weights cannot control the trade-off , because they only refl ect their importance into the decision-making process.

Criteria aggregation by Ordered Weighted Averaging (OWA)
OWA requires the defi nition of the order weights and trade-off level among criteria, which were defi ned according to Valente and Vettorazzi (2008). So, we obtained seven priority maps, using OWA to the criteria aggregation and we selected only two maps.
These maps showed 56% trade-off among criteria, having the fi rst medium-high risk-taking (R = 0.4650) and the second low risk-taking (R = 0.7187).
The same way, the histograms of the maps were evaluated, and these maps were reclassifi ed into priority classes, that support the comparison among their and the WLC map.

Weighted Linear Combination
According to the WLC priority map, southeast of Piedade was predominantly classifi ed as a high priority for protection. Furthermore, "longitudinal strips" are formed to the west of the municipality, that was associated with very high, medium, low and very low priorities. The latter priority class occupies a restrict area in the western extremity of the municipality, representing only 0.10% of the total area. Low to very low priority classes respectively represent 5.31%, 19.78%, 47.76% and 27.02% of the total area.

Medium-high risk-taking
The map produced by OWA, having mediumhigh risk-taking (R=0.4650) and 56% trade-off among criteria, presented the priority classes high and very high occupying similar areas of the Piedade (Figure 4).
The fi rst class occupied 38.16% of the municipality and the second, 39.36%, representing approximately 78% of Piedade. Therefore, the classes very low, low, and medium occupied respectively 0.07%, 3.88%, and 18.50%.

Low risk-taking
The map produced by OWA, having low risktaking (R=0.7187) and 56% trade-off among criteria, presented diff erences in relation to the fi rst OWA. According to this map, 56% of the Piedade was classifi ed as medium priority to protected ( Figure 5).
In this scenario, the priority classes very high occupied 5.86% of the municipality and high 26.01% and, both were concentrated further southeast of the municipality. Thus, the class very low represented only 0.75% of the study area, concentrating far west of the municipality.

Weighted Linear Combination
In the WLC map, the spatial distribution of priority classes refl ects the importance of the criteria, that were the most infl uential in the decisionmaking process (i.e. proximity to protected areas and Proximity among forest patches with larger core maps). Consequently, this map shows the priority classes very high and high totalizing 75% of the municipality.
The least infl uential criteria (i.e. distance from disturbance sources and neighbor of forest remnants) were practically disregarded in the fi nal product, even though the fi rst showed 20% importance (factor weight). However, as mentioned by Geneletti (2003), the presence of roads cannot be disregarded in a landscape, seeing that they are responsible for modifying habitat conditions and may aff ect the abundance and distribution pattern of the biota.
According to Chen et al. (2019), 90.7% of protected areas are disturbed by the presence of roads. De Jesus Silva et al. (2006) and Sousa et al. (2009) also stress the importance of considering roads when establishing a Conservation Unit. In this context, the WLC map was not adequate for selecting priority areas for protection in the municipality of Piedade.
As reported by Valente and Vettorazzi (2008), WLC maps are equivalent to those showing medium risk-taking and total trade-off between criteria obtained by OWA. Valente and Vettorazzi (2013) stated that WLC products are strongly infl uenced by their highest-ranking criteria, with one or two priority classes often occupying large regions of the study area.

Medium-high risk-taking
Spatial distribution of priority classes in the medium to high risk-taking map is related to order weights. The lowest values of order weights associated with distance from disturbance sources and neighbor of forest remnants (both with order weights of 0.0080), which are the least infl uential criteria in the decision-making process. In this case, the tradeoff was low, and criteria continued to show reduced infl uence just as was seen using WLC.
Consequently, area percentage decreased in very low and low priority classes in comparison to WLC. Conversely, the highest order weight associated with proximity among forest patches with larger core (0.50) and proximity to PEJU (0.30), which already showed high infl uence, once forest continuity interconnects with core conservation, which in turn is considered a natural refuge (Svensson et al., 2019).
Proximity to the Itupararanga EPA, a very infl uential criteria, received the second lowest ranking weight (0.04), showing high trade-off between weights and causing a reduction of the factor's infl uence. Such fact did not signifi cantly alter the fi nal map confi guration due to order weights associated to the other two proximity maps.  The medium to high risk-taking map allowed prioritization in order to preserve forest patches that already showed established internal structure. According to Geneletti (2005), large interconnected patches tend to harbor more species and less vulnerable to extinction populations compared to smaller, isolated patches. However, as has occurred previously, the medium to high risk-taking map did not considered neighbor of forest remnants and disturbance sources and, thus, could not be considered adequate for selecting priority areas for protection in the municipality of Piedade.

Low risk-taking
Spatial distribution of low risk-taking maps portrays the interaction of order weights with factor weights. The lowest order weights associated with the most infl uential factors, in accordance with the study objectives and following the importance order.
For such purposes, very high and high priority classes locate in the southeast region of Piedade, that is, next to vegetation remnants and Conservation Units, as is shown in factor maps ( Figure 5). In those regions, predominant neighboring types are consequently from the native forest category. Ribeiro et al. (2012) mention that locations next to Conservation Units and areas located between forest remnants of signifi cant size and that harbor endemic and/or endangered species can promote landscape connectivity, ensuring long-term viability of fl ora and fauna.
The lowest concentration of disturbance sources (roads and urban centers) is observed in very high and high priority regions. Priority levels for such portions of the municipality became medium and, when disturbance sources enhance in concentration, especially roads, or when the region distanced from forested areas (remnants or conservation areas), priority levels fell from medium to low or very low. Highways drive change in land use, infl uence deforestation (Freitas et al., 2010;Bax et al. , 2016), biodiversity loss and raise extinction risk for some species (Ceia-Hasse et al., 2018).
In this context, low risk-taking maps show to be the best solution for selecting priority areas for protection among the three evaluated types and such solution is in accordance with the reality of the municipality. The area between PEJU and the Itupararanga EPA, associated to very high to medium priority levels, belongs to the PEJU buff er zone and therefore already classifi es as a protected area by law.
Such a level of priority reaffi rms the need to eff ectively transform the area into a protected one. The fi nal map, along with criteria maps, demonstrates this region has a potential to increase matrix permeability due to proximity between signifi cant remnants and protected areas. Such an endeavor would be possible by implementing one or more green corridors (Hamad;Kolo;Balzter, 2018) or ecological stepping stones (for example, using backyard agroforestry) (Cambui et al., 2017), which would provide structural connectivity (Metzger, 2006;Guzmán et al., 2016) for remnants, with future expectations of functional connection (Hernández et al., 2015).

CONCLUSIONS
Considering the specifi c conditions related to the study, we can conclude that: 1)The Participatory Technique, considering expert opinions and the Hierarchical Analytical Process, is satisfactory for the defi nition of the criteria (factors and restrictions) and their weights, that is considered a MCE diff erential because data on natural systems are hardly complete and fully understood. Another advantage of the Participatory Technique is allowing the solution of a confl ict.
2) Both WLC and OWA enabled the identifi cation of priority areas for protection. However, OWA off ers more versatile solutions, because it considers the risktaking of the process and trade-off among criteria. This way, OWA provides a greater number of possible answers to the interest question. WLC tends to provide more general and less fl exible solutions that, if not well evaluated, can conduct to uncertainties in landscape planning.
3)The correct use of these methodologies depends on a good understanding of the available techniques and the context of the landscape under study