Spatial Analysis of Forest Fragmentation in the Atlantic Forest Bioma Areas

The goal of this work was to calculate landscape ecology metrics using the R language, allowing the analysis of forest fragments under the Atlantic Forest domain located in the sub-basin of Arroio Jaquirana, Rio Grande do Sul, Brazil. For the mapping of the forest fragments, we used images from the REIS/RapidEye sensor dated 2016, and the classification was supervised through the Bhattacharya algorithm. The fragments were analyzed in seven size classes, to separate them and to calculate the landscape metrics it was used R language. The results attained demonstrated that the native forest occupied 34.01% of the study area, covering a total of 1,995 fragments, of which 93.43% were less than 5 ha. The highest values of edge and perimeter-area ratio were found in the small fragments indicating a greater edge effect, with the central areas of these remnants being exposed to the external matrix effects. Thus, it is concluded that the Atlantic Forest is highly fragmented and is extremely important to establish measures to minimize the effects and/or increase the connectivity between the fragments through ecological corridors using the smaller fragments, in addition, it makes necessary the development of public policies and research for the management of the region in order to preserve the remnants.


Introduction
For decades the deforestation resulting from the process of anthropization of the landscape, has caused the destruction of natural resources.Native forests gave way mainly to agriculture, livestock and urbanization, with the formation of fragments, often isolated from one another.The loss of areas, the increasing of the isolation and the greater exposure of the borders of the fragments to human actions cause changes in the structure and function of these fragments (Lindenmayer & Fischer, 2006).
The process of forest fragmentation and loss of areas are present in the different biomes.However, the Atlantic Forest is one of the most threatened biomes in the world, being one of the most deforested (Jenkins et al., 2013).Thus, this biome was conferred the title of hotspots due to it being a threatened area, yet rich in biodiversity (Laurance, 2009).Originally the forest area covered approximately 1,315,460 km 2 , and currently there are only 12.5% of this original area, when the fragments above 3 hectares were recorded (SOS Mata Atlântica, 2015).In addition, it is highly fragmented, with 83% of the fragments smaller than 50 ha (Ribeiro, Metzger, Martensen, Ponzoni, & Hirota, 2009), being the conservation of the remaining remnants primordial.Thus, it is necessary to carry out studies related to the spatial characterization of forest fragments in order to establish conservation strategies (Pirovani et al., 2014).In this context, the ecology of the landscape is inserted, since the search for ecological knowledge about the forest fragments of a given area allows applying a correct environmental management as to these forests management (Calegari, Martins, Gleriani, Silva, & Busato, 2010;Negrini et al., 2014).In landscape ecology, the main object to be measured is the landscape structure, which provides characteristics of the constituent elements of the landscape, being commonly expressed through landscape metrics (Skokanová & Eremiášová, 2013).The landscape metrics contribute to the maintenance of biodiversity, since the analysis of these contributes to the determination of management techniques with the purpose of conserving and recovering the remaining forest fragments (Juvanhol et al., 2011).Thus, to evaluate .The The sub-basin of Arroio Jaquirana is an affluent of the Jacuizinho River, which, in turn, flows into the Jacuí River, thus, is inserted in the catchment area of Alto Jacuí.The main phytogeographic type is the Decidual Seasonal Forest, belonging to the Atlantic Forest biome, limiting the pastures of the southern campaign and the northern plateau (Rambo, 1956).According to the classification of Köppen the climate is a Humid Subtropical (Cfa), with average temperatures that vary between 16 ºC and 20 ºC.The precipitations are well distributed throughout the year, varying between 1,600 and 1,900 mm (Alvares, Stape, Sentelhas, Gonçalves, & Sparovek, 2014).

Mapping of Forest Fragments
Two SRTM (Shuttle Radar Topography Mission) scenes, captured by the Space Shuttle Endeavor, with a 30 m spatial resolution, were available for free from the USGS (United States Geological Survey).The area was automatically delimited using the GRASS modules version 7.0.4,coupled with the QGis software -Quantum GIS Geographic Information System (QGIS Development Team, 2016), version 2.14.7.
In the SPRING-Georeferenced Information Processing System, version 5.2.7 (Câmara, Souza, Freitas, Garrido, & Mitsuo, 1996), a spatial database with UTM (Universal Tranverse Mercator) and Datum SIRGAS 2000 cartography was created.For the mapping of the fragments of native forest four scenes REIS (RapidEye Earth Imaging System) sensors were needed, RapidEye satellite, with a resolution of 5 meters, composed of 5 multispectral bands, dated February 29, 2016.They were obtained by means of a project between the Federal University of Santa Maria and the Interstate Union of the Tobacco Industry (SINDITABACO), which aims to monitor forest cover in areas of Deciduous Seasonal Forest in the Central Sierra region of Rio Grande do Sul state.
Among the techniques of segmentation, this process precedes the classification, it was chosen the region's growth method, which consists of aggregating neighboring pixels to a given region that presents similar characteristics, thus occurring the increase of these regions (Happ, Feitosa, Bentes, & Farias, 2013).For this, it was necessary to define the parameters of similarity and area, being used the values of 10 and 100, since they were better suited to the study area.The classification was performed in a supervised manner using the Bhattacharya algorithm with a 99% acceptance threshold.With this algorithm the Bhattacharya distance measure is used to calculate the statistical separability between pairs of spectral classes, thus measuring the average distance between the probability distributions of the classes (Brites, Bias, & Rosa, 2012).
The efficiency of the mapping was verified using the Kappa index proposed by Congalton and Green (1999), according to Equation 1, and the quality of the generated data was evaluated according to the limits proposed by Landis and Koch (1977), according to Table 1.
Where, r = number of classes; X ij = number of elements sorted correctly; X i+ = total of elements classified for a category i; X +i = total of reference elements sampled for a category I; N = total number of samples.Source: Landis and Koch (1977).
In order to carry out the calculations, 200 points were distributed over the forest theme class, in a stratified random manner, according to Congalton's (1991) methodology, which suggested a minimum size of 50 samples for each category in areas that have a territorial coverage of up to approximately 4,050 Km².To verify the veracity of the classification we used high spatial resolution images of Google Earth Pro (Moreira, Barros, & Rudorff, 2008).

Analysis of Forest Fragmentation
The raster file containing the forest fragments was imported into the development environment integrated with the R version 3.3.0(R Development Core Team, 2016), for splitting the fragments into size classes, in which a sequence of commands in R language was employed.The fragments were grouped in adapted classes of Ribeiro et al. (2009), in order to compare the different sizes in the study area (Table 2).In order to avoid the use of redundant metrics, they encompassed different groups: area and density, edge, shape, central area and proximity.As shown in Table 3. Functions in R language were developed for the calculation of the metrics based on the equations described by McGarigal and Marks (1995), using as input parameter the raster file containing the forest fragments.In order to perform the calculations, it was needed the igraph package (Csardi & Nepusz, 2006)  The metric calculations were performed for each size class of the fragments as well as for the fragments as a whole, without stratification, in order to obtain general results of the study area and thus enable the comparison among those.The central area metrics were calculated using different edge distances, being 20, 60, 100 and 140 meters based on the method of Juvanhol et al. (2011), andPirovani et al. (2014), in order to analyze the different results obtained. jas.ccsenet.

Results
The thema measured categorize were divid Figure 2.

The result Calculatio
The numb equivalent well distri number of (LSI = 73.5),followed by the second class, with a value corresponding to 48.2, the lowest value found in class 3 (LSI = 20, 3).The LSI measures the perimeter-area ratio for the landscape as a whole, as a function of a standard shape being equal to one when the landscape contains only a regularly shaped fragment (in the case of a raster the standard shape is a square).This, increases according to the greater disaggregation of the class indicating greater heterogeneity of the landscape (McGarigal & Marks, 1995).Thus, it was noticed that class 1 obtained the highest value, due to the high number of fragments indicating a high degree of disaggregation.
The mean shape index indicates the degree of irregularity of the fragments.The lowest value was found in class 1 (MSI = 1.8), indicating that the fragments with up to 5 ha have a more regular shape when compared to the other size classes, being the last class with the highest value, equivalent to 22.7.Pirovani et al. (2014), analyzed areas of Atlantic Forest, obtained values close to the one found in this study, with MSI equal to 1.85 for fragments smaller than 5 ha, presenting a more regular format of the other size classes adopted by the authors.
The area-weighted mean shape index (AWMSI) were higher than those observed for the MSI, however, the values did not show any significant differences, since the AWMSI calculated the shape index and weighted it in relation to the area of the fragments.Thus, larger values indicate that fragments of larger area present more irregular forms than the average fragments.
For the mean perimeter-area ratio (MPAR), the value of 976.7 m ha -1 was obtained in the lowest class of fragments.The results obtained for this metric were directly proportional to the perimeter of the fragments; thus, indicate that high perimeter values lead to high MPAR values.The higher the perimeter, the greater the area exposed to the effects of the anthropic actions and the environmental conditions of the environment (Martins, Soares, Silva, & Brites, 2002), leading to a greater edge effect and consequently an intensification of deleterious actions on the remaining ones (Oliveira & Mattos, 2014).Thus, Colli et al. (2003) pointed out that the reduction of edge effects can be achieved by reducing the perimeter/area ratio of the fragments.
By means of the total edge (TE) and edge density (ED) metrics, it is possible to infer about the border effect in which the fragments are subject and the irregularity of them.Of the total border, 1,204,409.0m represented class 1, this value equivalent to border density of 29.0 m/ha.This class encompasses the smaller fragments, thus covering most of the fragments present in the study area, which resulted in a higher edge value, when compared to the other classes.Juvanhol et al. (2011) emphasize that in small fragments the transition between the matrix and the forest is quite abrupt, thus, an edge is developed exposing the forest to the effects of the external matrix, thus, causes microclimatic changes, which in turn will change the structure of vegetation and the composition of species.
Among the metrics that determine the isolation of fragments is that of the nearest neighbor, which is considered important, since it indicates the need for implementation or existence of connection elements (Vidolin, Biondi, & Wandembruck, 2011).According to Table 4, the average distance of the mean nearest neighbor (MNN), obtained for each size class, presented the lowest value among the fragments of the largest size class (Class 7) totaling 13.4 m.These fragments are located in sloping areas of the sub-basin, resulting in regions that are not suitable for agricultural cultivation.In this way, larger fragments located near each other were found, when compared to the other fragments of the study area.
Analyzing the fragments as a whole (Class 8), without stratifying by size classes, an average distance of 43.1 m was obtained, considered low compared to most size classes.This is due to the small fragments that soften the distance between the larger fragments, since they have high density and are well distributed in the study area.In this context, these fragments have an important ecological function, since they reduce the isolation of the larger fragments.Cemin, Perico, and Rampel (2009), when analyzing the distance of the nearest neighbor in deciduous forest areas in the Taquari Valley, Rio Grande do Sul, found a mean distance of 59.62 m, close to that obtained in the present study, which totaled 43.1 m.Pirovani et al. (2014) also found the shortest distance when analyzed the fragments as a whole.Thus, the authors emphasize the importance of considering smaller fragments as ecological trampolines, in addition reducing the distance between fragments of larger size.Silva and Souza (2014) point out that isolation reduces biological diversity as well as immigration rates.In this context, it is possible to perceive the importance that the smaller fragments present for the biodiversity maintenance, being able to be used as ecological corridors.
The central area is a better indication of the quality of forest fragments (McGarigaL & Marks, 1995).The results for the central area metrics are shown in Table 5.
jas.ccsenet.62.3 20.2 6.9 1.9 t for the 20 m A) of 264.8 h e total amount est number of uivalent to a ce f fragments an ted higher valu Figure 4).

Table 1 .
Quality of the classification according to Kappa coefficient intervals

Table 2 .
Size classes of forest fragments adopted for analysis of landscape ecology metrics

Table 3 .
Landscape ecology metrics used in the study