SPATIO-TEMPORAL ANALYSIS OF LANDSCAPE PATTERNS IN THE CATOLÉ WATERSHED, NORTHERN MINAS GERAIS1

ANÁLISE ESPAÇO-TEMPORAL DOS FRAGMENTOS FLORESTAIS NA BACIA DO RIO CATOLÉ, NORTE DE MINAS GERAIS

Rodrigo Nogueira Martins Selma Alves Abrahão Danilo Pereira Ribeiro Ana Paula Ferreira Colares Marco Antonio Zanella About the authors

ABSTRACT

The aim of this study was to quantify the spatio-temporal changes in land use/ cover (LULC), as well as analyze landscape patterns over a 20-year period (1995 - 2015) in the Catolé watershed, northern Minas Gerais State, using landscape metrics. The LULC maps were obtained using Landsat 5 and 8 data (Processing level 1) through supervised classification using the maximum likelihood classifier. Seven thematic classes were identified: dense vegetation, sparse vegetation, riparian vegetation, cropland, planted forest, bare soil, and water. From the LULC maps, classes related to the natural landscape (dense, sparse, and riparian vegetation) were grouped into forest patches, which was then ordered by size: very small (< 5 ha); small (5 - 10 ha); medium (10 - 100 ha); large (100 ha); and a general class (no distinction of patch size). Then, metrics of area, size and density, edge, shape, proximity and core area were calculated. The dense vegetation portion of the study area decreased considerably within a given time, while the portion of cropland and bare soil increased. Overall, in the Catolé river basin, the total area of natural vegetation decreased by 3,273 hectares (4.62%). Landscape metrics analysis exhibited a reduction in the number of very small patches, although the study area was still considered as fragmented. Moreover, a maximum edge distance of 50 m is suggested for conducting studies involving core area metrics in the Catolé watershed, as values above this distance would eliminate the very small patches.

Keywords:
Land use/ cover; Landscape metrics; Forest patches

RESUMO

O objetivo deste estudo foi quantificar as mudanças espaço-temporais no uso e cobertura da terra (land use/ cover - LULC), assim como, analisar o padrão da paisagem ao longo de 20 anos (1995 - 2015) na bacia do rio Catolé, utilizando métricas da paisagem. Os mapas foram obtidos utilizando imagens do Landsat 5 e 8 (nível de processamento 1) por meio de classificação supervisionada usando o classificador de máxima verossimilhança. Foram identificadas sete classes temáticas: vegetação densa, vegetação esparsa, mata ciliar, cultura agrícola, floresta plantada, solo exposto e água. Em seguida, as classes relacionadas à paisagem (vegetação densa, esparsa e mata ciliar) foram agrupadas em uma classe de fragmentos florestais, a qual foi ordenada por tamanho: muito pequeno (< 5 ha), pequeno (5 - 10 ha), médio (10 - 100 ha), grande (100 ha) e uma classe geral (sem distinção de tamanho). Em seguida, calcularam-se as métricas de área, tamanho e densidade, borda, forma, proximidade e área central. A vegetação densa na área de estudo diminuiu consideravelmente dentro do período avaliado, enquanto a área de culturas agrícolas e solo exposto aumentaram. No geral, na bacia do rio Catolé, a área total de vegetação natural reduziu 3.273 hectares (4,62%). As métricas da paisagem mostraram uma redução no número de fragmentos muito pequenos, embora a área ainda esteja fragmentada. Além disso, uma distância de borda máxima de 50 m é sugerida para a realização de estudos envolvendo métricas de área central na bacia do rio Catolé, pois valores acima desta distância eliminariam os fragmentos muito pequenos.

Palavras-Chave:
Uso e cobertura da terra; Métricas da paisagem; Fragmentos florestais

1.INTRODUCTION

Progressive changes in natural environments have led to habitat loss and landscape fragmentation worldwide (Cardille and Foley, 2003Cardille JA, Foley JA. Agricultural land-use change in Brazilian Amazonia between 1980 and 1995: Evidence from integrated satellite and census data. Remote Sensing of Environment. 2003;87(4):551-62.; Vieira et al., 2003Vieira ICG, Almeida AS, Davidson EA, Stone TA, Carvalho CJR, Guerrero JB. Classifying successional forests using Landsat spectral properties and ecological characteristics in eastern Amazonia. Remote Sensing of Environment. 2003;87(4):470-81.). Hence, conserving biological diversity has been recognized as a key issue in the scope of environmental protection. Human activities have modified the environment to the extent that most common landscape patterns exhibit mosaics of human settlements, agricultural land, and scattered fragments of natural ecosystems (Midha and Mathur, 2010Midha N, Mathur PK. Assessment of forest fragmentation in the conservation priority Dudhwa landscape, India using FRAGSTATS computed class level metrics. Journal of the Indian Society of Remote Sensing. 2010;38(3):487-500.). Owing to all these interferences, conservation areas are becoming increasingly surrounded by intensively modified environments, and, in the long-term, appear to be condemned to function as isolated natural ecosystems.

A reduction in natural vegetation cover has resulted in imbalances of the physical and biotic aspects of environments, which is a serious issue for the survival of numerous species in many ecosystems. Moreover, it jeopardizes the maintenance of the dynamic balance of landscapes and can also lead to a decrease in gene flow, reproductive isolation of species, and, consequently, a loss in genetic diversity (Neves et al., 2014Neves LFS, Neves SMAS, Canale GR. Análise da fragmentação de cerrado na bacia hidrográfica do rio Aguapei, Porto Esperidião (MT): um estudo de caso a partir das geotecnologias e métricas da paisagem. Ateliê Geográfico. 2014;8(2):130-49.; Dantas et al., 2017Dantas MS, Almeida NV, Medeiros IS, Silva MD. Diagnóstico da vegetação remanescente de Mata Atlântica e ecossistemas associados em espaços urbanos. Journal of Environmental Analysis and Progress. 2017;2(1):87-97.). Thus, increasing human interference is changing the structural and functional patterns of landscapes worldwide and has significantly impacted biodiversity (Sinha and Charma, 2006Sinha RK, Sharma A. Landscape level disturbance gradient analysis in Daltonganj south forest division. Journal of the Indian Society of Remote Sensing. 2006;34(3):234-43.).

The São Francisco river basin is the main water source in the northern, semi-arid region of Minas Gerais State, and is important for the development of irrigated agriculture and human consumption. In this region, the Pandeiros river basin is the breeding ground for many fish species from the São Francisco river. Due to this, its diverse array of plant species, and as the basin is located in a transition zone from the Cerrado to the Caatinga biomes, the Pandeiros river basin was converted into an Environmental ProtectionArea (EPA) in accordance with State Law 11,901 on September 1st , 1995 (Minas Gerais, 1995Minas Gerais. Lei n.º 11.901, de 01 de setembro de 1995. Declara de proteção ambiental as áreas de interesse ecológico situadas na Bacia Hidrográfica do Rio Pandeiros. 1995. [acessado em: 17 jul 2017]. Disponível em http://www.sfrancisco.bio.br/ legislac/l11901.html.
http://www.sfrancisco.bio.br/ legislac/l...
). This EPA shelters several Cerrado phyto-physiognomies, such as the Cerrado sensu stricto, deciduous seasonal forest (Mata Seca), riparian forests, floodplain areas, and Veredas, a type of Cerrado vegetation (Bethonico, 2009Bethonico MBM. Rio Pandeiros: território e história de uma área de proteção ambiental no Norte de Minas Gerais. Revista Acta Geográfica. 2009;3(5)23-38.; Nunes et al., 2009Nunes YRF, Azevedo IFP, Neves WV, Veloso MDM, Souza RM, Fernandes GW. Pandeiros: o Pantanal Mineiro. MG. Biota. 2009;2(1):4-17.).

One of the main tributaries of the Pandeiros river is the Catolé river, the basin of which is suffering from erosive processes caused by deforestation for charcoal production, the expansion of agriculture in Veredas and riparian vegetation, and poor conservation of unpaved roads (Fonseca et al., 2011Fonseca DDSR, Nascimento CR, Miranda WDA, Figueiredo FP. Diagnóstico do uso do solo e degradação ambiental na bacia hidrográfica do Pandeiros-MG como subsídio para estudos de impacto ambiental. Revista Geoaraguaia. 2011;1(1):1-20.). To evaluate the fragmentation of forest remnants, and propose methods of managing their conservation, similar studies on landscape ecology have been conducted in recent years (Abdalla and Cruz, 2015Abdalla LS, Cruz CBM. Análise de fragmentação florestal no Município de Silva Jardim, Apa do Rio São João, RJ. Revista Brasileira de Cartografia. 2015;67(1):169-84.; Silva et al., 2015Silva KGD, Santos ARD, Silva AGD, Peluzio JBE, Fiedler NC, Zanetti SS. Análise da dinâmica espaço-temporal dos fragmentos florestais da subbacia hidrográfica do Rio Alegre, ES. Cerne. 2015;21(2):311-8.; Saito et al., 2016Saito NS, Moreira MA, Santos ARD, Eugenio FC, Figueiredo AC. Landscape Geotechnology and ecology in monitoring forest fragmentation. Revista Floresta e Ambiente. 2016;23(2):201-10.).

Landscape ecology involves the study of landscape patterns, interactions among patches within a landscape mosaic, and how these patterns and interactions change over time (Mcgarigal and Marks, 1995Mcgarigal K, Marks BJ. Fragstats: Spatial Pattern Analysis Program for Quantifying Landscape Structure. Reference manual. 1995. 53p.). Through landscape ecology, landscape metrics has allowed the characterization of the complexity of landscapes and the determination of the conditions of forest fragments so that conservation measures can be developed (Silva and Souza, 2014Silva MSF, Souza RM. Padrões espaciais de fragmentação florestal na Flona do Ibura - Sergipe. Mercator. 2014;13(3):121-37.). Landscape metrics are tools to characterize the geometric and spatial properties of a patch (a spatially homogeneous entity) or a mosaic of patches (Fortin, 1999Fortin MJ. Spatial statistics in landscape ecology. In: Klopatek J, Gardner R, editors. Landscape Ecological Analysis: Issues and Applications. New York: Springer; 1999. cap.12. p.253-79.). In addition, landscape ecology involves the application of these principles in the formulation and solution of real-world problems.

The use of geographic information systems (GIS), remote sensing (RS), and multitemporal products can provide important information on landscape fragmentation mechanisms (Coelho et al., 2014Coelho VHR, Montenegro SGL, Almeida CN, Lima VER; Neto, AR, Moura GSS. Dinâmica do uso e ocupação do solo em uma bacia hidrográfica do semi-árido brasileiro. Revista Brasileira de Engenharia Agrícola e Ambiental. 2014;18(1):64-72.). Therefore, the aim of this study was to quantify the spatio-temporal changes in land use and land cover (LULC), as well as analyze landscape patterns over a 20-year period (1995 - 2015) in the Catolé watershed, northern Minas Gerais State, using landscape metrics.

2.MATERIAL AND METHODS

2.1. Study Area

The Catolé watershed is located in northern Minas Gerais State, between 14º 58' 28" and 15º 25' 50" S and 44º 39' 33" and 44º 55' 24" W, and it covers an area of 70,692 hectares. It lies in the lower portion of the middle São Francisco river, within the boundaries of the Pandeiros EPA in the municipality of Bonito de Minas - MG (Figure 1). The climate of the region is semi-arid, with well-defined dry and rainy seasons. The mean annual temperature is approximately 21 to 24 ºC, and the region’s elevation ranges from 485 to 515 m. The average rainfall is approximately 1,050 mm per year,with precipitation mostly occurringfrom October to March (Fonseca et al., 2011Fonseca DDSR, Nascimento CR, Miranda WDA, Figueiredo FP. Diagnóstico do uso do solo e degradação ambiental na bacia hidrográfica do Pandeiros-MG como subsídio para estudos de impacto ambiental. Revista Geoaraguaia. 2011;1(1):1-20.).

2.2. Land Use/Cover Maps

Temporal remote sensing data were collected to extract the LULC information. The data were obtained using the Landsat 5 Thematic Mapper (TM; overpass: July 30, 1995) and Landsat 8 Operational Land Imager (OLI) sensors (overpass: June 19, 2015), which have a spatial resolution of 30 x 30 m (path: 219; rows: 70 and 71). The QGIS 2.18 software was used to preprocess the data, which included geometric and radiometric correction, mosaic, subset, and layer stacking.

Landsat 5 images were geometrically corrected using a second-degree polynomial model, which presented a mean square error (MSE) smaller than 1 pixel. Reference images for georeferencing the Landsat 5 data came from five Rapid Eye images at level 3A, which are orthorectified tile products with radiometric, geometric, and terrain corrections (Tile ID: 2331915, 2332015, 2332016, 2332115, and 2332116)(overpass from July 3, 2014 to August 31, 2015). All images were radiometrically corrected using the dark object subtraction (DOS) (Chavez, 1988Chavez, J.P.S. An improved dark-object subtraction technique for atmospheric scattering correction for multispectral data. Remote Sensing of Environment. 1988;24(3):459-79.). Supervised classification was conducted using the Maximum Likelihood classifier, where red, green, blue, and near-infrared bands were used for both Landsat 5 and 8 images. Seven thematic classes were identified: dense vegetation, sparse vegetation, riparian vegetation, cropland, planted forest, bare soil, and water.

The accuracy was assessed using the error matrix of the 1995 and 2015 supervised classification maps, which considered four accuracy measures: the overall accuracy, user’s accuracy, producer’s accuracy, and Kappa coefficient. Twenty-five random samples were taken from each class to determine the accuracy of each classification. The 1995 LULC map samples were assessed using information from the inventory map of the flora and reforestation of Minas Gerais (Data from 2003 to 2007) and local forest fragment information from the georeferenced database (year 1998) in the local office of the State Forest Institute (IEF-MG). To assess the accuracy of the 2015 LULC map, samples were obtained through visual analysis of RapidEye images from 2014 and 2015. Moreover, the accuracy was assessed based on their omission (producer’s accuracy) and commission (user’s accuracy) errors, and the Kappa coefficient. The overall map accuracy was calculated by dividing the total number of correctly classified pixels (major diagonal of the error matrix) by the total number of pixels in the error matrix.

2.3. Forest Patches Mapping

From the LULCmaps, classes related to the natural landscape, such as dense, sparse, and riparian vegetation, were grouped into a new class denoted as forest fragments and individualized for calculating landscape metrics. To better compare the degrees of conservation and patch sizes, all patches were separated in classes according to their size: very small (< 5 hectares); small (5 - 10 hectares); medium (10 - 100 hectares); large (100 hectare); and a general class (no distinction of patch size).

The spatial dynamics of the forest landscape refers to the temporal changes in the size, number, shape, adjacency, and proximity of patches in a landscape (Çakir et al., 2008Çakir G, Sivrikaya F, Keles S. Forest cover changes and fragmentation using Landsat data in Macka State Forest Enterprise in Turkey. Environmental Monitoring and Assessment. 2008;137(1):51-66.). These landscape pattern metrics involve qualitative and quantitative measurements that express the characteristics of the landscape as a whole (Abdullah and Nakagoshi, 2006Abdullah SA, Nakagoshi N. Changes in landscape spatial pattern in the highly developing state of Selangor, peninsular Malaysia. Landscape and Urban Planning. 2006;77(3):263-75.). Accordingly, some metrics or measurements were used as a proxy to quantify and spatially analyze changes in the structure, as demonstrated by McGarigal and Marks (1995)Mcgarigal K, Marks BJ. Fragstats: Spatial Pattern Analysis Program for Quantifying Landscape Structure. Reference manual. 1995. 53p.: area, density and size, edge, shape, and proximity. The landscape metrics were computed using FRAGSTATS version 3.3 (Macgarigal et al., 2002Mcgarigal K, Cushman SA, Ene E. FRAGSTATS: Spatial Pattern Analysis Program for Categorical Maps and Continuous Maps. Computer software program produced by the authors at the University of Massachusetts, Amherst [accessed: Feb. 2017. Available at: http://www.umass.edu/ landeco/research/fragstats/fragstats.html.
http://www.umass.edu/ landeco/research/f...
). In addition, to calculate the core area metrics, edge distances of 50, 100, and 150 m were simulated to generate scenarios for analysis.

3.RESULTS

The overall accuracy of classification was 95.61% for the 1995 LULC map from the Landsat 5 images and 98.46% for the 2015 map from the Landsat 8 data. The Kappa coefficientswere 91.8%(1995)and98.06% (2015), respectively. The user’s accuracy of most classes in 1995 exceeded 97%, except for bare soil and cropland, which presented user’s accuracies of 76.41% and 60.00%, respectively. The producer’s accuracy for all classes in the same year was high, with the lowest values of 89.41% and 87.49% being observed for bare soil and sparse vegetation classes. In 2015, both the user’s and producer’s accuracies reached 94.65% for all classes. In general, all forest cover classes were classified accurately, while non-forest classes were not.

The results from the LULC maps (Figure1) exhibited major changes in the dense vegetation class, which decreased by 18.52% (6,987 hectares) when comparing the 1995 and 2015 LULC maps (Table 1). Most of the area was converted into sparse vegetation (4,286 hectares, 14.78%), followed by cropland (an increase of 2,473%, 1,397 hectares), and bare soil,which increased by 125.64% (1,714 hectares). The coverage of riparian vegetation considerably reduced (23.00%; 572 hectares) as it was mostly converted into cropland alongside the Catolé river.

Table 1
Evolution of land use/ cover classes from 1995 to 2015 in the Catolé watershed.
Tabela 1
Evolução do uso e cobertura da terra de 1995 a 2015 na bacia do rio Catolé.

Overall, in the Catolé watershed, the total area of natural vegetation decreased by 3,273 hectares (4.62%) from 1995 to 2015. The detailed losses and gains among the seven land use/cover classes over the 20-year study period are presented in Table 1.

Forest patch analyses were conducted at the landscape level, with patches grouped into different size classes. In the Catolé watershed, considerable changes in the distribution of forest patches were observed over the study period (Table 2). In both years (1995 and 2015), forest area (CA) was mainly observed in larger patches (larger than 100 hectares), covering 73.08% and 79.64% of the area, respectively. The remaining forest area existed in smaller patches, with 7.67% and 6% as very small patches, 4.19% and 3.24% with areas ranging from 5 to 10 hectares (small patches), and 15.05% and 11.12% as medium patches in 1995 and 2015, respectively.

Table 2
Landscape metrics of forest patches in the Catolé watershed, MG.
Tabela 2
Métricas da paisagem de fragmentos florestais na bacia do rio Catolé, MG.

The number of patches (NP) can be used to quantify the spatial heterogeneity of the entire landscape. Even with a decrease in the number of patches in all classes, the landscape was still predominantly represented by very small patches, which covered 2,974 hectares(77.09%) and 2,056 hectares (76%) in 1995 and 2015, respectively. As fewer patches were larger than 100 hectares, the forest may become more fragile and fragmented. The mean patch size (MPS) increased in all classes, except the small patches class, in which the MPS decreased from 7.1 to 6.85 hectares. The NP and MPS should be used complementarily as high NP and low MPS values reinforce the interpretation of a fragmented landscape.

The patch size standard deviation (PSSD) and patch size coefficient of variance (PSCov) remained almost unaltered for all classes, excluding larger patches, where substantial increases from 1,716 to 2,541 hectares and 196.87% to 232.34% were observed, respectively. Similarly, there was an increase in the values for the general class.

Among the total edge metrics, all classes exhibited a reduction in the length, which was associated with the reduction in NP. In contrast, the edge density (ED), i.e., the amount of edge relative to the landscape area, which is also related to the degree of spatial heterogeneity, exhibited different trends depending between LULC classes. The ED increased in the small and medium patch size classes, while it decreased in the very small, large, and general classes (without size distinction).

Similar to TE and ED, the mean shape index (MSI) is also related to the degree of spatial heterogeneity in the landscape. However, this serves as a measure of the shapes of patches. The former denotes the average shape or average perimeter-area ratio for all patches in the landscape. The lowest index value (MSI = 1) is achieved when patches are circular (vector) or square (raster), and increases as they become more irregular in shape.

Comparing the MSI and TE indices reveals a decrease in the TE for all classes, while the MSI decreases for some classes and increases for others; in general, the patches became more irregular as the MSI increased from1.79 to 1.83. Thisindicates that MSI ismore sensitive to changes in the number of patches than changes in their perimeters. The results of the area-weighted mean shape index (AWMSI) were similar to those of the MSI, however, the patches became more regular in the general class as the AWMSI decreased from 8.99 to 8.23.

The mean nearest neighbor (MNN) defines the average edge-to-edge distance between a patch and its nearest neighbor in the landscape. The results exhibited an increase in the isolation degree of patches smaller than 100 hectares as all classes within this area increased the distance from edge-to-edge. However, the patches in the general class became more connected, as its distance decreased from 203.20 to 70.68 m.

The core area metrics obtained through simulations of 50, 100, and 150 m edge distances (Table 3) showed that the number of core areas (NCA) in all classes (excluding very small patches) exceeded the NP at an edge distance of 50 m. However, as the edge distance increases, the NCA value decreases, indicating how the edge effect would affect the entire patch area. The total core area (TCA) values were proportional to the patch size. The total core area index (TCAI) allows a better understanding of the edge effect over the different patch sizes. TCAI values decreased significantly with the increase in edge distance, reaching 0% at an edge distance of 150 m for patches smaller than 10 hectares. That is, at this point, all patches smaller than 10 hectares experienced the edge effect, which would directly affect its conservation degree.

Table 3
Core area metrics of forest patches in the Catolé watershed, MG.
Tabela 3
Métricas de área central para os fragmentos florestais na bacia do rio Catolé, MG.

4.DISCUSSIONS

According to Landis and Koch (1977)Landis J, Koch GG. The measurements of agreement for categorical data. Biometrics. 1977;33(3):159-79., the Kappa index results obtained for the 1995 and 2015 classification maps were highly accurate when compared to the ground truth samples. This indicates that, even when using images with a mean spatial resolution of 30 m (such as those from Landsat 5 and 8), it was possible to accurately map the LULC in the study area. In general, forest cover classes are still predominant in the Catolé watershed. However, in 2015, there was a considerable increase in cropland and bare soil areas, and a decrease in riparian vegetation, indicating the occurrence of anthropization processes that are more intense in central areas close to the riverbed.

There are historical reports of agricultural activity in the rural communities located within the EPA’s boundaries. During a study on the Pandeiros and Catolé rivers, Bethonico (2009)Bethonico MBM. Rio Pandeiros: território e história de uma área de proteção ambiental no Norte de Minas Gerais. Revista Acta Geográfica. 2009;3(5)23-38. observed a steady growth in subsistence agriculture among those communities. Most of these areas were used to cultivate beans, rice, cassava, maize, and sugarcane. In addition, some farmers practice extensive livestock agriculture, which was included in the sparse vegetation class.

A large area of the Catolé watershed is covered by Neosol Quartzarenic soil, which is characterized by its sandy texture and is more susceptible to degradation when no soil conservation practices are established. There are several reports of illegal deforestation and charcoal production in the study area, and the use of fire for soil cleaning. These actions accelerate soil erosion by up to 20%, and are considered as the most degrading practices in natural forests (Cabacinha and Castro, 2010Cabacinha CD, Castro SS. Estrutura diamétrica e estado de conservação de fragmentos florestais no Cerrado Brasileiro. Revista Floresta e Ambiente. 2010;17(1):1-12.; Merten and Minella, 2013Merten GH, Minella JPG. The expansion of Brazilian agriculture: soil erosion scenarios. International Soil and Water Conservation Research. 2013;1(3):37-48.).

Owing to the increase in cropland and bare soil areas alongside the riverbed, there is a need to recover degraded areas and preserve riparian vegetation. Moreover, most of the changes were observed in the central and southern areas of the basin, where the terrain is mildly sloped, which favors agricultural exploration in the region.

During 1995-2015, the landscape was mostly represented by very small forest patches, which indicates high fragmentation and a low degree of conservation. As stated by Metzger and Sodhi (2009)Metzger JP, Sodhi N. Conservation issues in the Brazilian Atlantic forest. Biological Conservation. 2009;142(6):1137-52., such conditions can lead to a reduction in species richness as the patch becomes unsuitable for maintaining the survival of wild populations. There is an inverse relationship between the number of patches and its area, that is, there was a greater number of very small patches. However, when looking at the total area, it presents the smallest percentage of the landscape area. On the other hand, patches larger than 100 hectares present fewer units, though it has the highest amount of area.

Silva et al. (2015)Silva KGD, Santos ARD, Silva AGD, Peluzio JBE, Fiedler NC, Zanetti SS. Análise da dinâmica espaço-temporal dos fragmentos florestais da subbacia hidrográfica do Rio Alegre, ES. Cerne. 2015;21(2):311-8. observed similar conditions; 84.15% of forest patches in the River Plate basin of Ibiraçu andAracruz, ES, were smaller than five hectares, however, largeand mediumpatches accounted for 54.75% and 30.93% of the study area, respectively. Larger patches are important for maintaining ecological processes and ensuring biodiversity. However, small patches play different roles, such as increasing connectivity between different habitats (Liu et al., 2014Liu S, Dong Y, Deng L, Liu Q, Zhao H, Dong S. Forest fragmentation and landscape connectivity change associated with road network extension and city expansion: a case study in the Lancang River Valley. Ecological Indicators. 2014;36(1):160-8.), acting as ecological trampolines (Mortelliti et al., 2014Mortelliti A, Westgate MJ, Lindenmayer DB. Experimental evaluation shows limited influence of pine plantations on the connectivity of highly fragmented bird populations. Journal of Applied Ecology. 2014;51(5):1179-87.), and reducing the isolation degree between patches (Fahrig, 2003Fahrig L. Effects of habitat fragmentation on biodiversity. Annual Review of Ecology, Evolution and Systematics. 2003;34(1):487-515.).

High variability in patch size is common in landscape ecology studies, as described by several authors in recent years. After analyzing forest patches in the Itapemirin river basin(ES), Pirovaniet al. (2014)Pirovani DB, Silva AG, Santos AR, Cecílio RA, Gleriani JM, Martins SV. Análise espacial de fragmentos florestais na Bacia do Rio Itapemirim. Revista Árvore. 2014;38(2):271-81. observed high PSSD (164 hectares and 37 hectares) and PSCov metric (122.79% and 433.85%) values for large size patches (> 50 hectares) and the general class (all patches). Santos et al. (2016Santos CAP, Sano EE, Santos OS. Fronteira agrícola e a análise da estrutura da paisagem na Bacia do Rio Preto-Oeste da Bahia. Raega - O Espaço Geográfico em Análise. 2016;36(1):179-207.) found that the Black river basin, western Bahia State, was mainly occupied by small patches, and the MPS index ranged from 14.3 to 59.9 hectares in different regions within the basin. However, the PSSD values (148 hectares and 2,308 hectares) varied greatly among patch sizes, which indicated the existence of larger patches than average.

Large patches presented the highest TE values and lowest ED metric values in both years (1995 and 2015). These resultsindicate ahigh degree of conservation in this class, as the edge effect is one of the main causes of declining biodiversity in vegetation remnants, which corresponds to changes due to contact between the forest patch and the deforested area (Borges et al., 2004Borges LFR, Scolforo JR, Oliveira AD, Mello JM, Acerbi FW, Freitas GD. Inventário de fragmentos florestais nativos e propostas para o seu manejo e o da paisagem. Cerne. 2004;10(1):22-38.; Nascimento and Laurence, 2006Nascimento HEM, Laurance WF. Efeitos de área e de borda sobre a estrutura florestal em fragmentos de floresta de terra-firme após 13-17 anos de isolamento. Acta Amazônica. 2006;36(2):183-92.). Furthermore, the border region is mostly subject to pressure associated with anthropic activities that occur in the deforested area, which directly interfere with ecosystem dynamics.

Just as the TE and ED indices are important for landscape metrics analysis, patch shape metrics are necessary to evaluate landscape structure and disturbance in forest remnants. A standard format should be adopted when comparing the shapes of forest patches. Lang and Blaschke (2009)Lang S, Blaschke T. Análise da Paisagem com SIG. São Paulo: Oficina de Textos; 2009. suggested that the closer the value is to one, the better the shape. If the patch is more circular, the edge effect on the ecosystem is reduced. Thus, shape values closer to one indicated that the core area may be more protected, whereas values greater than one indicate a more elongated tendency, assuming that the fragment is more vulnerable to edge effects.

Analysis of the MSI results revealed that, in both years, the shape of very small patches was the most regular shape (1.57 and 1.59) among all classes. Higher values were obtained for the AWMSI index than those for the MSI, which indicates that the shape of the largest patches is more irregular than that of average patches. This is due to the calculation of the AWMSI index, which considersthe patch size.Thus, although smaller patches are more regular in shape, they are still more susceptible to edge effects.

Patch isolation was measured by the MNN index, which, according to Benderet al. (2003)Bender DJ, Tischendorf LE, Fahrig L. Using patch isolation metrics to predict animal movement in binary landscapes. Landscape Ecology. 2003;18(1):17-39., refersto the inaccessibility of living beings for migrating between patches. Small patches presented the highest degree of isolation, with distances of 315.39 and 318.61 m for 1995 and 2015, respectively. Small patches are more isolated than other size classes, however, their ecological importance should be considered as small patches are important forconnecting larger patches.

To reduce the degree of patch isolation in the Catolé watershed, degraded areas should be restored to create ecological corridors between the most isolated patches as isolation directly affects plants dispersion and wildlife movement. In addition,it is more difficult to connect habitats due to the reduction in species richness and composition (Collinge, 1996Collinge SK. Ecological consequences of habitat fragmentation: implications for landscape architecture and planning. Landscape and Urban Planning. 1996;36(1):59-77.; Costa et al., 2015Costa OB, Matricardi EAT, Pires JSR. Análise do processo de fragmentação da floresta nos Municípios de Corumbiara e Buritis-RO. Revista Floresta e Ambiente. 2015;22(1):334-44.).

Core area metrics were calculated by simulating different edge distances to evaluate which distance exerts the greatest influence on the core area results. The highest NCA values for smaller fragments were due to their irregular forms, which increases the difficulty of connecting core areas in the same forest patch. Silva et al. (2015)Silva KGD, Santos ARD, Silva AGD, Peluzio JBE, Fiedler NC, Zanetti SS. Análise da dinâmica espaço-temporal dos fragmentos florestais da subbacia hidrográfica do Rio Alegre, ES. Cerne. 2015;21(2):311-8. found that there is a decrease in NCA values with an increase in the edge distance, as edge effects act on the patch and its irregularities. That is, the edge effect transforms complex geometric figures into circular surfaces, and the results of this metric are obtained through several simulations.

In both years (1995 and 2015), the very small and small patches exhibited null values for core area indices; the very small patches exhibited null values for these metrics at 100 m, and small patches exhibited null values above an edge distance of 100 m. These results demonstrate that, under this condition, all patches within these classes are dominated by the edge effect would be susceptible to the matrix influence (Juvanhol et al. 2011Juvanhol RS, Fiedler NC, Santos AR, Pirovani DB, Louzada FLRO, Dias HM, et al. Análise espacial de fragmentos florestais: caso dos parques estaduais de Forno Grande e Pedra Azul, estado do Espírito Santo. Revista Floresta e Ambiente. 2011;18(1):353-64.). The TCA index, which represents the sum of core areas (excluding the edges), decreased as the edge distance increased in 1995 and 2015.

The results of TCAI were the most suitable to represent the edge effect on different patch sizes. In 1995, the TCAI result for very small patches was 1.78% of their area, therefore,98.22% ofthe classwas susceptible to edge effects in the first 50-m distance. Vidolin et al. (2011)Vidolin GP, Biondi D, Wandembruck A. Análise da estrutura da paisagem de um remanescente de floresta com Araucária, Paraná, Brasil. Revista Árvore. 2011;35(3):515-25. stated that core area metrics are measures of habitat quality as they indicate the remaining effective area of a patch after discounting the edge effect.

Several studies have discussed different edge distances, however, there is no consensus on the size of the edge to be considered. Abdalla and Cruz (2015)Abdalla LS, Cruz CBM. Análise de fragmentação florestal no Município de Silva Jardim, Apa do Rio São João, RJ. Revista Brasileira de Cartografia. 2015;67(1):169-84. analyzed fragmentation in the São João river EPA, and used an edge distance of 100 m. Fernandes et al. (2017)Fernandes M, Fernandes M, Almeida A, Gonzaga MIS, Gonçalves F. Ecologia da paisagem de uma Bacia Hidrográfica dos Tabuleiros Costeiros do Brasil. Revista Floresta e Ambiente. 2017;24(1):1-9. simulated distances of 30, 60, and 90 m in the Piauitinga river basin, southern Sergipe State, and found that the increase in distance directly affects the core area, especially for small patches. The same authors recommended the creation of ecological corridors for better conservation and a maximum distance of 30 m for edge effects analysis.

These results are consistent with the values found here, because, for the 100 and 150 m edge distances, null values were observed for core area metrics. This demonstrates the dominance of the edge effect on small-sized patches. Thus, the maximum distance for evaluating edge effects on the Catolé watershed landscape should be 50 m.

5.CONCLUSIONS

Landsat TM and OLI images were appropriate for generating LULC maps. The accuracy assessment results showed that both maps were accurately produced by applying the Maximum Likelihood classifier. The change detection results exhibited substantial changes in the LULC from 1995 to 2015. The dense vegetation portion of the study area decreased considerably within a given time, while the portion of cropland and bare soil increased. Overall, in the Catolé river basin, the total area of natural vegetation decreased by 3,273 hectares (4.62%).

Landscape metrics analysis exhibited a reduction in the number of very small patches, although the study area was still considered as fragmented. The more irregular the patch shape, the larger the total edge of the patch. Although larger patches have the most irregular shapes, their core area metrics were still better than those of smaller patches, even under the effect of larger edge distances. A maximum edge distance of 50 m is suggested for conducting studies involving core area metrics in the Catolé river basin, as values above this distance would eliminate the very small patches. In summary, these results can aid in defining strategies for land planning and design, and decision making for conservation priorities.

6.ACKNOWLEDGMENTS

The authors would like to thank FAPEMIG (The Minas Gerais State Foundation for Research Support) for their financial grant to perform this work.

7.REFERENCES

  • Abdalla LS, Cruz CBM. Análise de fragmentação florestal no Município de Silva Jardim, Apa do Rio São João, RJ. Revista Brasileira de Cartografia. 2015;67(1):169-84.
  • Abdullah SA, Nakagoshi N. Changes in landscape spatial pattern in the highly developing state of Selangor, peninsular Malaysia. Landscape and Urban Planning. 2006;77(3):263-75.
  • Bender DJ, Tischendorf LE, Fahrig L. Using patch isolation metrics to predict animal movement in binary landscapes. Landscape Ecology. 2003;18(1):17-39.
  • Bethonico MBM. Rio Pandeiros: território e história de uma área de proteção ambiental no Norte de Minas Gerais. Revista Acta Geográfica. 2009;3(5)23-38.
  • Borges LFR, Scolforo JR, Oliveira AD, Mello JM, Acerbi FW, Freitas GD. Inventário de fragmentos florestais nativos e propostas para o seu manejo e o da paisagem. Cerne. 2004;10(1):22-38.
  • Cabacinha CD, Castro SS. Estrutura diamétrica e estado de conservação de fragmentos florestais no Cerrado Brasileiro. Revista Floresta e Ambiente. 2010;17(1):1-12.
  • Cardille JA, Foley JA. Agricultural land-use change in Brazilian Amazonia between 1980 and 1995: Evidence from integrated satellite and census data. Remote Sensing of Environment. 2003;87(4):551-62.
  • Çakir G, Sivrikaya F, Keles S. Forest cover changes and fragmentation using Landsat data in Macka State Forest Enterprise in Turkey. Environmental Monitoring and Assessment. 2008;137(1):51-66.
  • Chavez, J.P.S. An improved dark-object subtraction technique for atmospheric scattering correction for multispectral data. Remote Sensing of Environment. 1988;24(3):459-79.
  • Coelho VHR, Montenegro SGL, Almeida CN, Lima VER; Neto, AR, Moura GSS. Dinâmica do uso e ocupação do solo em uma bacia hidrográfica do semi-árido brasileiro. Revista Brasileira de Engenharia Agrícola e Ambiental. 2014;18(1):64-72.
  • Collinge SK. Ecological consequences of habitat fragmentation: implications for landscape architecture and planning. Landscape and Urban Planning. 1996;36(1):59-77.
  • Costa OB, Matricardi EAT, Pires JSR. Análise do processo de fragmentação da floresta nos Municípios de Corumbiara e Buritis-RO. Revista Floresta e Ambiente. 2015;22(1):334-44.
  • Dantas MS, Almeida NV, Medeiros IS, Silva MD. Diagnóstico da vegetação remanescente de Mata Atlântica e ecossistemas associados em espaços urbanos. Journal of Environmental Analysis and Progress. 2017;2(1):87-97.
  • Fahrig L. Effects of habitat fragmentation on biodiversity. Annual Review of Ecology, Evolution and Systematics. 2003;34(1):487-515.
  • Fernandes M, Fernandes M, Almeida A, Gonzaga MIS, Gonçalves F. Ecologia da paisagem de uma Bacia Hidrográfica dos Tabuleiros Costeiros do Brasil. Revista Floresta e Ambiente. 2017;24(1):1-9.
  • Fonseca DDSR, Nascimento CR, Miranda WDA, Figueiredo FP. Diagnóstico do uso do solo e degradação ambiental na bacia hidrográfica do Pandeiros-MG como subsídio para estudos de impacto ambiental. Revista Geoaraguaia. 2011;1(1):1-20.
  • Fortin MJ. Spatial statistics in landscape ecology. In: Klopatek J, Gardner R, editors. Landscape Ecological Analysis: Issues and Applications. New York: Springer; 1999. cap.12. p.253-79.
  • Juvanhol RS, Fiedler NC, Santos AR, Pirovani DB, Louzada FLRO, Dias HM, et al. Análise espacial de fragmentos florestais: caso dos parques estaduais de Forno Grande e Pedra Azul, estado do Espírito Santo. Revista Floresta e Ambiente. 2011;18(1):353-64.
  • Landis J, Koch GG. The measurements of agreement for categorical data. Biometrics. 1977;33(3):159-79.
  • Lang S, Blaschke T. Análise da Paisagem com SIG. São Paulo: Oficina de Textos; 2009.
  • Liu S, Dong Y, Deng L, Liu Q, Zhao H, Dong S. Forest fragmentation and landscape connectivity change associated with road network extension and city expansion: a case study in the Lancang River Valley. Ecological Indicators. 2014;36(1):160-8.
  • Mcgarigal K, Marks BJ. Fragstats: Spatial Pattern Analysis Program for Quantifying Landscape Structure. Reference manual. 1995. 53p.
  • Mcgarigal K, Cushman SA, Ene E. FRAGSTATS: Spatial Pattern Analysis Program for Categorical Maps and Continuous Maps. Computer software program produced by the authors at the University of Massachusetts, Amherst [accessed: Feb. 2017. Available at: http://www.umass.edu/ landeco/research/fragstats/fragstats.html
    » http://www.umass.edu/ landeco/research/fragstats/fragstats.html
  • Merten GH, Minella JPG. The expansion of Brazilian agriculture: soil erosion scenarios. International Soil and Water Conservation Research. 2013;1(3):37-48.
  • Metzger JP, Sodhi N. Conservation issues in the Brazilian Atlantic forest. Biological Conservation. 2009;142(6):1137-52.
  • Midha N, Mathur PK. Assessment of forest fragmentation in the conservation priority Dudhwa landscape, India using FRAGSTATS computed class level metrics. Journal of the Indian Society of Remote Sensing. 2010;38(3):487-500.
  • Minas Gerais. Lei n.º 11.901, de 01 de setembro de 1995. Declara de proteção ambiental as áreas de interesse ecológico situadas na Bacia Hidrográfica do Rio Pandeiros. 1995. [acessado em: 17 jul 2017]. Disponível em http://www.sfrancisco.bio.br/ legislac/l11901.html
    » http://www.sfrancisco.bio.br/ legislac/l11901.html
  • Mortelliti A, Westgate MJ, Lindenmayer DB. Experimental evaluation shows limited influence of pine plantations on the connectivity of highly fragmented bird populations. Journal of Applied Ecology. 2014;51(5):1179-87.
  • Nascimento HEM, Laurance WF. Efeitos de área e de borda sobre a estrutura florestal em fragmentos de floresta de terra-firme após 13-17 anos de isolamento. Acta Amazônica. 2006;36(2):183-92.
  • Neves LFS, Neves SMAS, Canale GR. Análise da fragmentação de cerrado na bacia hidrográfica do rio Aguapei, Porto Esperidião (MT): um estudo de caso a partir das geotecnologias e métricas da paisagem. Ateliê Geográfico. 2014;8(2):130-49.
  • Nunes YRF, Azevedo IFP, Neves WV, Veloso MDM, Souza RM, Fernandes GW. Pandeiros: o Pantanal Mineiro. MG. Biota. 2009;2(1):4-17.
  • Pirovani DB, Silva AG, Santos AR, Cecílio RA, Gleriani JM, Martins SV. Análise espacial de fragmentos florestais na Bacia do Rio Itapemirim. Revista Árvore. 2014;38(2):271-81.
  • Saito NS, Moreira MA, Santos ARD, Eugenio FC, Figueiredo AC. Landscape Geotechnology and ecology in monitoring forest fragmentation. Revista Floresta e Ambiente. 2016;23(2):201-10.
  • Santos CAP, Sano EE, Santos OS. Fronteira agrícola e a análise da estrutura da paisagem na Bacia do Rio Preto-Oeste da Bahia. Raega - O Espaço Geográfico em Análise. 2016;36(1):179-207.
  • Silva KGD, Santos ARD, Silva AGD, Peluzio JBE, Fiedler NC, Zanetti SS. Análise da dinâmica espaço-temporal dos fragmentos florestais da subbacia hidrográfica do Rio Alegre, ES. Cerne. 2015;21(2):311-8.
  • Silva MSF, Souza RM. Padrões espaciais de fragmentação florestal na Flona do Ibura - Sergipe. Mercator. 2014;13(3):121-37.
  • Sinha RK, Sharma A. Landscape level disturbance gradient analysis in Daltonganj south forest division. Journal of the Indian Society of Remote Sensing. 2006;34(3):234-43.
  • Vidolin GP, Biondi D, Wandembruck A. Análise da estrutura da paisagem de um remanescente de floresta com Araucária, Paraná, Brasil. Revista Árvore. 2011;35(3):515-25.
  • Vieira ICG, Almeida AS, Davidson EA, Stone TA, Carvalho CJR, Guerrero JB. Classifying successional forests using Landsat spectral properties and ecological characteristics in eastern Amazonia. Remote Sensing of Environment. 2003;87(4):470-81.

Publication Dates

  • Publication in this collection
    08 Nov 2018
  • Date of issue
    2018

History

  • Received
    20 Jan 2018
  • Accepted
    06 Aug 2018
Sociedade de Investigações Florestais Universidade Federal de Viçosa, CEP: 36570-900 - Viçosa - Minas Gerais - Brazil, Tel: (55 31) 3612-3959 - Viçosa - MG - Brazil
E-mail: rarvore@sif.org.br