Environmental Management DOI 10.1007/s00267-014-0387-7
Human Impacts Affect Tree Community Features of 20 Forest Fragments of a Vanishing Neotropical Hotspot Jose´ Aldo Alves Pereira • Ary Teixeira de Oliveira-Filho Pedro V. Eisenlohr • Pedro L. S. Miranda • Jose´ Pires de Lemos Filho
•
Received: 8 July 2013 / Accepted: 11 October 2014 Ó Springer Science+Business Media New York 2014
Abstract The loss in forest area due to human occupancy is not the only threat to the remaining biodiversity: forest fragments are susceptible to additional human impact. Our aim was to investigate the effect of human impact on tree community features (species composition and abundance, and structural descriptors) and check if there was a decrease in the number of slender trees, an increase in the amount of large trees, and also a reduction in the number of tree species that occur in 20 fragments of Atlantic montane semideciduous forest in southeastern Brazil. We produced digital maps of each forest fragment using Landsat 7 satellite images and processed the maps to obtain morphometric variables. We used investigative questionnaires and field observations to survey the history of human impact. We then converted the information into scores given to the extent, severity, and duration of each impact, including
proportional border area, fire, trails, coppicing, logging, and cattle, and converted these scores into categorical levels. We used linear models to assess the effect of impacts on tree species abundance distribution and stand structural descriptors. Part of the variation in floristic patterns was significantly correlated to the impacts of fire, logging, and proportional border area. Structural descriptors were influenced by cattle and outer roads. Our results provided, for the first time, strong evidence that tree species occurrence and abundance, and forest structure of Atlantic seasonal forest fragments respond differently to various modes of disturbance by humans. Keywords Cattle Disturbance effects Edge effect Fire Linear model Logging Introduction
Electronic supplementary material The online version of this article (doi:10.1007/s00267-014-0387-7) contains supplementary material, which is available to authorized users. J. A. A. Pereira Departamento de Cieˆncias Florestais, Universidade Federal de Lavras, Lavras, MG 37200-000, Brazil e-mail: [email protected] A. T. de Oliveira-Filho P. V. Eisenlohr (&) J. P. de Lemos Filho Departamento de Botaˆnica, ICB, Universidade Federal de Minas Gerais, Belo Horizonte, MG 31270-901, Brazil e-mail: [email protected] Present Address: P. V. Eisenlohr Faculdade de Cieˆncias Agra´rias e Biolo´gicas, Universidade do Estado de Mato Grosso, Alta Floresta, MT 78580-000, Brazil P. L. S. Miranda School of Geosciences, The University of Edinburgh, Edinburgh EH9 3JN, UK
Human impacts on forest fragments are of great concern worldwide (e.g., Peres et al. 2010; Goudie 2013). The growth of human activities has led to the destruction, degradation, and fragmentation of habitats and may be considered responsible for the current decline of biodiversity (Whitmore 1997), with changes in the distribution and abundance of organisms (Laurance and Bierregaard 1997; Pimm and Raven 2000; Wright 2010). The destruction of tropical forests and their fragmentation are critical for species’ extinction (loss of diversity) or for increasing the species’ vulnerability to the extinction processes. Even when a fraction of the individuals of a species survive, they will have suffered a significant reduction in genetic variation (Kageyama and Lepsch-Cunha 2001). Comparisons between natural forest ecosystems and forest remnants show that lower values of biodiversity are found in those that have suffered the impacts of anthropogenic interference (Bierregaard and Dale 1996; Bennett and Saunders 2010).
123
Environmental Management
South American Atlantic forests are among the world’s most endangered biomes and have been treated as one of the top hotspots for biodiversity conservation (Myers et al. 2000; Mittermeier et al. 2004; Zachos and Habel 2011). The wide latitudinal and altitudinal range of the Atlantic Forest (about 30° S and 2,700 m, respectively) explains its high diversity and endemism, with more than 20,000 species of vascular plants, of which about 8,000 are endemic (Mittermeier et al. 2004). In contrast, the original extent of 1,315,460 km2 has been reduced to barely 7.91 %, if taking into account only fragments over 100 ha, or to 11.41 %, if the minimum fragment size is reduced to 3 ha (Fundac¸a˜o SOS Mata Atlaˆntica and INPE 2009). The strong pressure on the Atlantic Forest is related to the long-time human occupancy in the region, which is now inhabited by more than 120 million people (Metzger 2009; Ribeiro et al. 2011). Due to this occupancy, almost no Atlantic Forest fragments can be considered untouched or free of impacts caused by human activities (Fonseca 1985; Ribeiro et al. 2009). The majority of these remnants are secondary, i.e., they have already been deforested in the past, and their diversity and structure were altered to some extent (Fonseca 1985; Ribeiro et al. 2011). Most studies suggested that the influence of human impacts on natural ecosystems is quite negative and causes damage to community attributes (Metzger et al. 2009). Some authors also demonstrated that the original diversity observed in these sites cannot be recovered, even with the best of conservational efforts (Silva and Tabarelli 2000; Tabarelli et al. 2008, 2010). The loss in forest area due to human occupancy is not the only threat to the remaining biodiversity. Forest fragments are susceptible to additional human impact, and, even if these impacts are halted, the long-term fragmentation itself will bring about escalating genetic erosion and species impoverishment (Lopes et al. 2009; Tabarelli et al. 2010). Forest fragmentation brings about physical and biological effects, including limitations to dispersal and genetic decay of isolated populations (Kattan and AlvarezLo´pez 1996), deeply modifying the composition and structure of forest communities (Nascimento et al. 1999; Botrel et al. 2002; Hill and Curran 2005; Pereira et al. 2007). The two main causes of these processes are isolation of populations and edge effects (Pimm and Raven 2000; Hill and Curran 2003; Laurance et al. 2006; Olifiers and Cerqueira 2006; Pu¨tz et al. 2011). Among the current human impacts on Atlantic Forest remnants, felling and coppicing are ranked as top threats (Fundac¸a˜o SOS Mata Atlaˆntica and INPE 2009) and both persist in spite of a legal ban (0.25 % or 350 km2 year-1). Other impacts include cattle trampling, trails, fire, hunting, urban expansion, and adjacent farming and forestry (e.g., Marini and Garcia 2005; Pereira et al. 2007; Metzger 2009; Freitas et al. 2010; Eisenlohr et al. 2013).
123
In addition to diminishing the Atlantic Forest area and making forest patches even more fragmented, impacts such as urban expansion, isolation, farming, forestry, and the use of fire to prepare pastures and agricultural fields, are capable of reducing tree flora diversity (e.g., Hill and Curran 2003; Toniato and Oliveira-Filho 2004; Tabarelli et al. 2008) by changing the local environmental conditions and of the forest remnants nearby. By doing so, these impacts prevent some species from occupying such regions, reducing their diversity (Tabarelli et al. 2012). Felling, coppicing, and livestock are not only the main factors responsible for structural changes in forest patches, but are also related to tree diversity loss (Villela et al. 2006; Silva Costa et al. 2012; Clark and Covey 2012). When the whole area is not deforested, the first two activities are usually responsible for removing trees in reproductive state, just those considered as the main fruit and seed sources, besides being one of the main carbon accumulators (Vieira et al. 2008). Soil compression, and the removal of seedlings and small growing trees by livestock also alter the forest structure by slowing down or completely halting the regeneration process of these fragments (e.g., Trimble and Mendel 1995). The opening and use of trails and roads are also related to biodiversity loss and changes in forest structure (Goosem 2007; Freitas et al. 2010; Eisenlohr et al. 2013). Much has been done regarding the protection, conservation, and restoration of remaining Atlantic Forest fragments over the last decades (Rodrigues et al. 2009; Metzger 2009), and there was a widespread effort to determine how economical activities can influence vegetation (e.g., Cullen et al. 2000; Saatchi et al. 2001; Toniato and Oliveira-Filho 2004). However, we still know little about how much and in which ways the remaining Atlantic Forest has been modified due to human occupancy, especially in some regions, such as those located at semideciduous forests, which has only recently been considered part of the Atlantic Forest (Oliveira-Filho and Fontes 2000). In the present contribution, we investigated the central hypothesis that the impact history, which includes opening and usage of trails and roads, fire, coppicing, felling, fragmentation, border effect, and livestock, would decrease the number of slender trees and proportionally increase the amount of large trees found on 20 Atlantic Semideciduous Forest fragments located in a region with no severe climate variation. We also expected that the human impact would reduce the number of tree species that could inhabit these areas, thereby affecting negatively the tree species abundance and composition. To test these hypotheses, we produced a set of predictive variables related to the disturbance history of each fragment and sets of response variables describing tree species composition, abundance, and structure.
Environmental Management
Materials and Methods Forest Fragments The 20 forest fragments are situated in the Alto Rio Grande region, state of Minas Gerais, southeastern Brazil, between 21°000 and 21°430 S and 43°500 and 45°050 W (Fig. 1). This region lies on the continental side of the Mantiqueira range and is mostly composed of sloping highlands, with altitudes between 700 and 1200 m a.s.l. The region is part of the Alto Parana´ Interior Forests Ecoregion, which is characterized by seasonal climates, with rainy-warm summers and dry-cool winters (Olson et al. 2001). The vegetation of the fragments is classified as montane seasonal semideciduous forest, according to IBGE (2012). Descriptive information on the fragments is given in Table 1. Predictive Variables A flow chart representing the methodological framework is given in Fig. 2. Predictive variables (hereafter, predictors) obtained for each forest fragment are given in Table 2. These variables are classified into two sets: those related to edge effects and therefore to fragment shape (morphometric variables)—‘border area’ (proportional border) and ‘heterogeneity of the border/interior ratio’ (HBI)—and those related to the history of environmental impacts (impact variables)—‘inner roads’ (roads that pass through the fragment), ‘outer roads’ (roads that pass nearby, but never pass through the fragment), ‘fire,’ ‘trails,’ ‘coppicing,’ ‘logging,’ and ‘cattle.’ (i) Morphometric variables: We produced maps of the forest fragments in SPRING 3.4 and ENVI 3.1 using Landsat 7 ETM? satellite images and then processed the maps in FRAGSTATS 3.0 (McGarigal et al. 2002) which yields a number of morphometric indices. For the satellite images, we chose the bands 3, 4, and 5 to facilitate identification of soil reflectance. We also used band 8, which is panchromatic (resolution of 15 m), in order to identify and accurately separate the fragments studied from their surroundings, besides distinguishing which one of those surroundings are covered by managed forests, pastures, and croplands and mixed landscapes. As required by the software, we provided border-width values for each fragment. To this end, we searched the literature on edge effects on the tree communities of fragmented tropical forests (e.g., Kapos 1989; Laurance and Yensen 1991; Murcia 1995) including regional studies (Oliveira-Filho et al. 1997; Espı´rito-Santo et al. 2002; Higuchi et al. 2008; Machado et al. 2010). Based on these studies, we eventually assigned three border widths to the forest fragments, according to the surrounding landscape: 100 m, for open pastures and
croplands; 50 m, for managed forests; and 75 m, for mixed landscapes. By doing this, we were able to differentiate and include in our analyses the influence that the surrounding matrix has over our 20 Atlantic Forest fragments. Of all the many morphometric indices yielded by FRAGSTATS, we chose the ‘border area’ (proportional border) as the best descriptor of edge effects to be used as a predictive variable, since the impact influence on the tree flora is more severe on those areas, and this index is the best way of bringing that to the analyses (Murcia 1995; Kupfer 2012). An additional variable was produced from the FRAGSTATS indices; the ‘heterogeneity of the border/ interior ratio,’ HBI, aimed at expressing a balance between the interior (I) and border (B) areas of a fragment: HBI = (minimum (B,I)/(maximum (B,I). HBI therefore ranges from 1 (a fragment with equal areas of border and interior) to 0 (fragments composed only of border or interior areas, the latter obviously non-existent in the case of fragments). By doing so, we were able to quantify how intensely the fragments’ area is influenced by the edge effect and how intensely is not affected by it (core areas). Both predictors were arcsine square root transformed (Zar 2010). (ii) Impact variables: In order to obtain information on the history of human intervention in the forest fragments, we used the Rapid Participatory Rural Appraisal technique (Mueller et al. 2010), through investigative questionnaires that aimed at reporting not only the impacts but also the history of land use in surrounding areas. In the case of areas that are protected by law or other legal mechanisms (PB, IB, and TD), we also consulted management plans and other documents. We then organized the information into an Interaction Matrix (Leopold et al. 1971) to assess the impact of seven modes of human intervention: ‘inner roads’ (roads that pass through the fragment), ‘outer roads’ (roads that pass nearby, but never pass through the fragment), ‘fire,’ ‘trails,’ ‘coppicing,’ ‘logging,’ and ‘cattle.’ We assigned scores to the severity, spatial extent, and time span of each intervention mode and summed them up. We based our scoring of impacts on Dean (1996) and also on the 25-year experience of the research team of the Universidade Federal de Lavras responsible for the continuous surveys of permanent plots (see Pereira et al. 2007 for further information). As the scoring process could be biased by subjectivity, we only ascribed scores after careful discussion with the whole research team. The different levels of impact based on scoring were used as categorical levels of each variable (Table 2). Each variable was evaluated on a scale ranging from 0 to 7, being zero the absence of impact and seven, the gravest stage of impact found in the fragments.
123
Environmental Management Fig. 1 Geographic situation of the a Alto Rio Grande region in southeastern Brazil and b the 20 forest fragments studied in the region () identified by two-letter codes (full names in Table 1)
Table 1 Identification (code and denomination), total area, central geographic coordinates, and altitude range of the 20 forest fragments
123
Code
Denomination
LV
Lavras
Area (ha) 5.72
Latitude (S)
Longitude (W)
21°140 4200 0
Altitude (m)
44°580 1000
918–937
00
44°580 5200
925–1,210
PB
Poc¸o Bonito
88.56
21°20 17
BS
Bom Sucesso
83.45
21°090 2700
44°540 1000
806–832
MD
Madre de Deus
20.66
21°290 0300
44°220 3200
915–980
0
00
0
00
913–945 913–960
IT
Itutinga
3.77
21°21 05
CM
Camargos
10.36
21°210 1800
44°360 4900
TD
Tiradentes
1,013.98
21°040 5200
44°080 0600
920–1,340
IB CP
Ibitipoca Capivari
95.05 9.78
21°420 4800 21°160 2300
43°530 0700 44°520 5300
1,150–1,510 825–875
SE
Subestac¸a˜o
8.73
21°130 1700
44°570 4700
910–940
0
00
44°36 29
0
00
ML
Mata da Lagoa
3.97
21°13 00
PI
Piedade
24.95
21°290 1600
44°060 0200
1,040–1,150
CR
35.98
21°360 2900
IN
Carrancas Ingaı´
LU
Lumina´rias
44°58 49
855–902
44°360 3800
1,440–1,513
16.14
21°24 26
00
44°530 3200
860–890
77.91
21°290 1100
44°440 2000
880–1,000
0
0
00
0
00
820–980
IU
Ibituruna
59.75
21°10 00
LA
Lafite
25.95
21°120 5800
44°480 0800
825–863
MI
Itumirim
54.97
21°130 3100
44°470 3100
815–910
PN
Pedra Negra
72.34
21°050 2500
44°560 4200
835–885
RM
Rio das Mortes
14.31
21°060 4900
44°450 0000
845–980
44°50 25
Environmental Management
Fig. 2 Flow chart summarizing the hypothesis, data collection, and analytical procedures used to investigate the interference of human impacts on 20 Atlantic forest fragments. LM linear models
Response Variables We produced response variables related to the tree community of the forest fragments from the dataset produced by the continuous surveys, including the species name, dbh (diameter at breast height) and total height of all live trees with dbh C 5.0 cm. We chose the dataset of the surveys carried out between 1998 and 2002 to minimize temporal bias. All response variables were generated using images from March 2001. Since total sample area varies among forest fragments, we used subsamples near 1 ha in all cases. As we wished to verify the influence of human impacts also on floristic composition and its abundance, we applied an ordination analysis to reduce the dimensionality of our species checklist. Thus, we obtained a set of response variables from the scores of each axis of two NMS ordinations (nonmetric multidimensional scaling; McCune and Grace 2002; Wildi 2010). The first NMS (hereafter, NMS1) aimed at representing the variations in species composition (binary presence matrix) and the second (NMS2), the variation in species abundance (relative density matrix). We assessed the significance of the axes by Monte Carlo tests (999 permutations), and the solution obtained was three-dimensional. We certified that the stress remained stable at the last iterations (McCune and Grace 2002). We
also obtained the proportion of the original n-dimensional ordination explained by each axis (R2). In order to assess the influence of impact history on regeneration guilds, we extracted the proportion of trees of pioneer species (%). Pioneers were defined here as those forming soil seed banks and requiring direct light radiation for germination (Swaine and Whitmore 1988). We produced two additional structural variables from regressing tree heights on log-transformed diameters: slope and intercept at the minimum dbh (5 cm), hereafter, h-slope and h-dbh5, respectively. These variables are intended to represent overall tree slenderness (higher slope shows that the trees are being able to grow taller without having to invest much in diameter acquisition and vice versa) and the slenderness of smaller trees, respectively (intercept values show the diameter size of the smaller trees). We also obtained the following structural variables for the tree community of each forest fragment: third quartiles of tree heights (hQ3) (to represent the height limits of trees, as those above the 75 % threshold along ranked height values); third quartiles of tree diameters (dbhQ3); basal area (BA, m2ha-1); density (D, trees.ha-1); and densities of trees with heights below or equal to the third quartile (D B hQ3) and above (D [ hQ3) (intended to represent the density of understory trees and canopy, respectively).
123
Environmental Management Table 2 Impact variables of the 20 Atlantic forest fragments used in LMs
Fragments are identified by their codes in Table 1. HBI (heterogeneity of the border/ interior ratio) and border area are proportions while the others are categorical variables
Code
Morphometric variables
Impact variables
Border area
HBI
Inner road
Outer road
Fire
Trails
Coppicing
Logging
Cattle
LV
1
0
1
7
1
3
1
6
0
PB
0.46
0.85
4
5
4
1
2
2
1
BS
0.59
0.69
1
1
1
1
1
2
2
MD
0.72
0.39
1
1
1
2
1
2
3
IT
0.98
0.02
1
1
2
0
1
1
1
CM
0.82
0.22
1
1
2
0
1
1
1
TD
0.12
0.14
3
1
2
2
4
4
1
IB
0.22
0.28
1
1
1
0
1
1
0
CP
0.96
0.04
1
3
2
2
5
5
1
SE
0.72
0.39
1
6
1
1
3
2
0
ML PI
1 0.42
0 0.72
3 1
6 2
2 1
3 0
3 1
3 2
3 1
CR
0.48
0.92
1
1
3
0
2
2
1
IN
0.63
0.59
1
1
1
1
1
2
2
LU
0.33
0.49
2
1
2
1
3
2
1
IU
0.39
0.64
1
7
5
1
4
2
1
LA
0.6
0.67
3
4
1
2
1
4
3
MI
0.65
0.54
4
5
2
1
6
3
2
PN
0.47
0.89
1
3
1
1
1
2
2
RM
0.59
0.69
2
5
2
1
1
2
1
Linear Models For each response variable, we prepared a general linear model (LM; Quinn and Keough 2002) based on least squares, with a significance level of 5 %. We first selected a set of potential relevant predictors in each model according to the existing literature and discarding variables with R2 (coefficient of determination) \ 0.3. Then we selected the best model based on the lowest AICc—corrected Akaike criterion (Burnham and Anderson 2002) using maximized log-likelihood and confirmed the lack of collinearities among predictor variables through the variance inflation factor (VIF), assuming a VIF cut-off of 10 (Quinn and Keough 2002). When we found VIF [ 10 in the best model, we discarded it and chose the best LM (i.e., that with the lowest AICc) among the remaining candidate models without collinearity. The spatial independence is a critical assumption of general statistical tests (Legendre et al. 2002). We therefore produced Moran’s I correlograms, following the SAM 4.0 (Rangel et al. 2010) defaults, to assess the significance of spatial structure of the residuals (Diniz-Filho et al. 2003) using Monte Carlo tests (999 permutations). We applied sequential Bonferroni correction to test the global spatial structure of the correlograms (Fortin and Dale 2005). When spatial structure was detected, we included spatial filters based on Moran’s Eigenvector Maps—MEM method by means of Delaunay triangulation (Dray et al. 2006). In this
123
step, we used only the MEMs required to remove spatial structure in the LM residuals. If spatial structure was present before the AICc selection procedure (see above), spatial filters were necessarily included in all candidate models (Diniz-Filho et al. 2008). Since the inflation of type I error may occur when both predictor and response variables are spatially structured (Legendre et al. 2002), we also obtained correlograms for these data. When both predictor and response variables were spatially structured, we followed Peres-Neto and Legendre (2010) in order to partition out the variance after a forward selection of MEMs, thereby isolating the fraction that causes the inflation of type I error and confirming the results of the aforementioned significance tests. We checked the assumptions of linearity and homogeneity of variances by plotting residuals on estimated values and observing the balance below and above the zero residual line, without any systematic structure such as curves or cones (Quinn and Keough 2002). When we suspected that the homogeneity of variances was violated, we performed Levene test (Zar 2010). We also tested the models for normality of residuals, using the Shapiro–Wilk test (Zar 2010). When any of the assumptions failed, we eliminated outliers (i.e., forest fragments with extreme studentized residuals). After this removal, we tested all the assumptions again, including the spatial independence, and checked the significance and coefficient of determination of the final model.
Environmental Management Table 3 Significant linear models used to explain the variation in NMS2 (relative density ordination) scores of 20 Atlantic forest fragments df
F
P
NMS2—Axis 1 Intercept
1
11.173
0.010
Fire Logging
4 5
14.291 6.265
0.001 0.012
MEM 1
1
8.710
0.018
Error
8
–
–
Intercept
1
13.874
0.002
Proportional border
1
13.249
0.002
–
–
NMS2—Axis 3
Error
17
Full results are shown in S1 (online supplementary material). Significant results (P B 0.05) in bold df degrees of freedom
We performed the analyses in PC-ORD 6.0 (McCune and Mefford 2011), PAST 2.0 (Hammer et al. 2001), SAM 4.0 (Rangel et al. 2010), and ‘vegan’ (Oksanen et al. 2011) and ‘spacemakeR’ (Dray et al. 2006) packages of R Statistical Environment (The R Foundation for Statistical Computing 2013).
Results The NMS2 ordination, which reflected the main trends of variation in species abundance, was highly influenced by
morphometric and, or, impact variables (Table 3). We found two linear models that were able to show the variables related to tree species abundance: for Axis 1 of NMS2 (F = 11.49, P \ 0.01, R2 = 93.49 %; R2adj = 85.35 %) and for Axis 3 of NMS2 (F = 13.25, P \ 0.01, R2 = 43.80 %, R2adj = 40.49 %). These axes explained 49.5 and 14.6 % (total of 64.1 %) of the floristic variance in the original n-dimensional space, respectively. The stress in real data was significantly lower than in randomized data (P = 0.02 in both axes). The trend of variation represented by Axis 1 of NMS2 was significantly related to fire, logging, and one spatial filter (MEM1) (Table 3). Intermediate levels of fire were related to higher scores (Fig. 3a). The greater the impact of logging, the higher the scores (Fig. 3b). The opposite was observed for the trends represented by Axis 3, where the higher the proportional border, the lower the scores (Fig. 3c). The structural descriptors of the vegetation were significant for third quartiles of tree diameters (dbhQ3), basal area (BA), and densities of trees with heights below or equal to the third quartile (D B hQ3) (Table 4; Fig. 4). The LM explaining the variations of dbhQ3 included only cattle as a significant predictor (Table 4). Only at the highest impact level of cattle, the dbhQ3 was significantly higher (Fig. 4a). In turn, two predictors were significant for BA: outer roads (impact variable) and one spatial filter (Table 4). The highest levels of outer roads were associated with lower values of BA (Fig. 4b). Finally, outer roads were also significantly related to D B hQ3 (Table 4). Here, in general, the larger the outer roads, the
Fig. 3 Ordination of the 20 Atlantic forest fragments by nonmetric multidimensional scaling of the relative species density (NMS2) per fragment. Diagrams show the proportional influence of each significant predictor in the linear model constructed to explain the variation in the NMS axis. Triangle size is proportional to the effect of fire in a, logging in b, and proportional border in c
123
Environmental Management Table 4 Significant linear models used to explain the variation in community structural variables of 20 Atlantic forest fragments df
F
P
dbhQ3 Intercept
1
1033.875
Outer road
6
2.763
0.075
Cattle
3
4.520
0.030
Error
10 298.968 3.447
Human Impacts Affect Tree Community Features of 20 Forest Fragments of a Vanishing Neotropical Hotspot Jose´ Aldo Alves Pereira • Ary Teixeira de Oliveira-Filho Pedro V. Eisenlohr • Pedro L. S. Miranda • Jose´ Pires de Lemos Filho
•
Received: 8 July 2013 / Accepted: 11 October 2014 Ó Springer Science+Business Media New York 2014
Abstract The loss in forest area due to human occupancy is not the only threat to the remaining biodiversity: forest fragments are susceptible to additional human impact. Our aim was to investigate the effect of human impact on tree community features (species composition and abundance, and structural descriptors) and check if there was a decrease in the number of slender trees, an increase in the amount of large trees, and also a reduction in the number of tree species that occur in 20 fragments of Atlantic montane semideciduous forest in southeastern Brazil. We produced digital maps of each forest fragment using Landsat 7 satellite images and processed the maps to obtain morphometric variables. We used investigative questionnaires and field observations to survey the history of human impact. We then converted the information into scores given to the extent, severity, and duration of each impact, including
proportional border area, fire, trails, coppicing, logging, and cattle, and converted these scores into categorical levels. We used linear models to assess the effect of impacts on tree species abundance distribution and stand structural descriptors. Part of the variation in floristic patterns was significantly correlated to the impacts of fire, logging, and proportional border area. Structural descriptors were influenced by cattle and outer roads. Our results provided, for the first time, strong evidence that tree species occurrence and abundance, and forest structure of Atlantic seasonal forest fragments respond differently to various modes of disturbance by humans. Keywords Cattle Disturbance effects Edge effect Fire Linear model Logging Introduction
Electronic supplementary material The online version of this article (doi:10.1007/s00267-014-0387-7) contains supplementary material, which is available to authorized users. J. A. A. Pereira Departamento de Cieˆncias Florestais, Universidade Federal de Lavras, Lavras, MG 37200-000, Brazil e-mail: [email protected] A. T. de Oliveira-Filho P. V. Eisenlohr (&) J. P. de Lemos Filho Departamento de Botaˆnica, ICB, Universidade Federal de Minas Gerais, Belo Horizonte, MG 31270-901, Brazil e-mail: [email protected] Present Address: P. V. Eisenlohr Faculdade de Cieˆncias Agra´rias e Biolo´gicas, Universidade do Estado de Mato Grosso, Alta Floresta, MT 78580-000, Brazil P. L. S. Miranda School of Geosciences, The University of Edinburgh, Edinburgh EH9 3JN, UK
Human impacts on forest fragments are of great concern worldwide (e.g., Peres et al. 2010; Goudie 2013). The growth of human activities has led to the destruction, degradation, and fragmentation of habitats and may be considered responsible for the current decline of biodiversity (Whitmore 1997), with changes in the distribution and abundance of organisms (Laurance and Bierregaard 1997; Pimm and Raven 2000; Wright 2010). The destruction of tropical forests and their fragmentation are critical for species’ extinction (loss of diversity) or for increasing the species’ vulnerability to the extinction processes. Even when a fraction of the individuals of a species survive, they will have suffered a significant reduction in genetic variation (Kageyama and Lepsch-Cunha 2001). Comparisons between natural forest ecosystems and forest remnants show that lower values of biodiversity are found in those that have suffered the impacts of anthropogenic interference (Bierregaard and Dale 1996; Bennett and Saunders 2010).
123
Environmental Management
South American Atlantic forests are among the world’s most endangered biomes and have been treated as one of the top hotspots for biodiversity conservation (Myers et al. 2000; Mittermeier et al. 2004; Zachos and Habel 2011). The wide latitudinal and altitudinal range of the Atlantic Forest (about 30° S and 2,700 m, respectively) explains its high diversity and endemism, with more than 20,000 species of vascular plants, of which about 8,000 are endemic (Mittermeier et al. 2004). In contrast, the original extent of 1,315,460 km2 has been reduced to barely 7.91 %, if taking into account only fragments over 100 ha, or to 11.41 %, if the minimum fragment size is reduced to 3 ha (Fundac¸a˜o SOS Mata Atlaˆntica and INPE 2009). The strong pressure on the Atlantic Forest is related to the long-time human occupancy in the region, which is now inhabited by more than 120 million people (Metzger 2009; Ribeiro et al. 2011). Due to this occupancy, almost no Atlantic Forest fragments can be considered untouched or free of impacts caused by human activities (Fonseca 1985; Ribeiro et al. 2009). The majority of these remnants are secondary, i.e., they have already been deforested in the past, and their diversity and structure were altered to some extent (Fonseca 1985; Ribeiro et al. 2011). Most studies suggested that the influence of human impacts on natural ecosystems is quite negative and causes damage to community attributes (Metzger et al. 2009). Some authors also demonstrated that the original diversity observed in these sites cannot be recovered, even with the best of conservational efforts (Silva and Tabarelli 2000; Tabarelli et al. 2008, 2010). The loss in forest area due to human occupancy is not the only threat to the remaining biodiversity. Forest fragments are susceptible to additional human impact, and, even if these impacts are halted, the long-term fragmentation itself will bring about escalating genetic erosion and species impoverishment (Lopes et al. 2009; Tabarelli et al. 2010). Forest fragmentation brings about physical and biological effects, including limitations to dispersal and genetic decay of isolated populations (Kattan and AlvarezLo´pez 1996), deeply modifying the composition and structure of forest communities (Nascimento et al. 1999; Botrel et al. 2002; Hill and Curran 2005; Pereira et al. 2007). The two main causes of these processes are isolation of populations and edge effects (Pimm and Raven 2000; Hill and Curran 2003; Laurance et al. 2006; Olifiers and Cerqueira 2006; Pu¨tz et al. 2011). Among the current human impacts on Atlantic Forest remnants, felling and coppicing are ranked as top threats (Fundac¸a˜o SOS Mata Atlaˆntica and INPE 2009) and both persist in spite of a legal ban (0.25 % or 350 km2 year-1). Other impacts include cattle trampling, trails, fire, hunting, urban expansion, and adjacent farming and forestry (e.g., Marini and Garcia 2005; Pereira et al. 2007; Metzger 2009; Freitas et al. 2010; Eisenlohr et al. 2013).
123
In addition to diminishing the Atlantic Forest area and making forest patches even more fragmented, impacts such as urban expansion, isolation, farming, forestry, and the use of fire to prepare pastures and agricultural fields, are capable of reducing tree flora diversity (e.g., Hill and Curran 2003; Toniato and Oliveira-Filho 2004; Tabarelli et al. 2008) by changing the local environmental conditions and of the forest remnants nearby. By doing so, these impacts prevent some species from occupying such regions, reducing their diversity (Tabarelli et al. 2012). Felling, coppicing, and livestock are not only the main factors responsible for structural changes in forest patches, but are also related to tree diversity loss (Villela et al. 2006; Silva Costa et al. 2012; Clark and Covey 2012). When the whole area is not deforested, the first two activities are usually responsible for removing trees in reproductive state, just those considered as the main fruit and seed sources, besides being one of the main carbon accumulators (Vieira et al. 2008). Soil compression, and the removal of seedlings and small growing trees by livestock also alter the forest structure by slowing down or completely halting the regeneration process of these fragments (e.g., Trimble and Mendel 1995). The opening and use of trails and roads are also related to biodiversity loss and changes in forest structure (Goosem 2007; Freitas et al. 2010; Eisenlohr et al. 2013). Much has been done regarding the protection, conservation, and restoration of remaining Atlantic Forest fragments over the last decades (Rodrigues et al. 2009; Metzger 2009), and there was a widespread effort to determine how economical activities can influence vegetation (e.g., Cullen et al. 2000; Saatchi et al. 2001; Toniato and Oliveira-Filho 2004). However, we still know little about how much and in which ways the remaining Atlantic Forest has been modified due to human occupancy, especially in some regions, such as those located at semideciduous forests, which has only recently been considered part of the Atlantic Forest (Oliveira-Filho and Fontes 2000). In the present contribution, we investigated the central hypothesis that the impact history, which includes opening and usage of trails and roads, fire, coppicing, felling, fragmentation, border effect, and livestock, would decrease the number of slender trees and proportionally increase the amount of large trees found on 20 Atlantic Semideciduous Forest fragments located in a region with no severe climate variation. We also expected that the human impact would reduce the number of tree species that could inhabit these areas, thereby affecting negatively the tree species abundance and composition. To test these hypotheses, we produced a set of predictive variables related to the disturbance history of each fragment and sets of response variables describing tree species composition, abundance, and structure.
Environmental Management
Materials and Methods Forest Fragments The 20 forest fragments are situated in the Alto Rio Grande region, state of Minas Gerais, southeastern Brazil, between 21°000 and 21°430 S and 43°500 and 45°050 W (Fig. 1). This region lies on the continental side of the Mantiqueira range and is mostly composed of sloping highlands, with altitudes between 700 and 1200 m a.s.l. The region is part of the Alto Parana´ Interior Forests Ecoregion, which is characterized by seasonal climates, with rainy-warm summers and dry-cool winters (Olson et al. 2001). The vegetation of the fragments is classified as montane seasonal semideciduous forest, according to IBGE (2012). Descriptive information on the fragments is given in Table 1. Predictive Variables A flow chart representing the methodological framework is given in Fig. 2. Predictive variables (hereafter, predictors) obtained for each forest fragment are given in Table 2. These variables are classified into two sets: those related to edge effects and therefore to fragment shape (morphometric variables)—‘border area’ (proportional border) and ‘heterogeneity of the border/interior ratio’ (HBI)—and those related to the history of environmental impacts (impact variables)—‘inner roads’ (roads that pass through the fragment), ‘outer roads’ (roads that pass nearby, but never pass through the fragment), ‘fire,’ ‘trails,’ ‘coppicing,’ ‘logging,’ and ‘cattle.’ (i) Morphometric variables: We produced maps of the forest fragments in SPRING 3.4 and ENVI 3.1 using Landsat 7 ETM? satellite images and then processed the maps in FRAGSTATS 3.0 (McGarigal et al. 2002) which yields a number of morphometric indices. For the satellite images, we chose the bands 3, 4, and 5 to facilitate identification of soil reflectance. We also used band 8, which is panchromatic (resolution of 15 m), in order to identify and accurately separate the fragments studied from their surroundings, besides distinguishing which one of those surroundings are covered by managed forests, pastures, and croplands and mixed landscapes. As required by the software, we provided border-width values for each fragment. To this end, we searched the literature on edge effects on the tree communities of fragmented tropical forests (e.g., Kapos 1989; Laurance and Yensen 1991; Murcia 1995) including regional studies (Oliveira-Filho et al. 1997; Espı´rito-Santo et al. 2002; Higuchi et al. 2008; Machado et al. 2010). Based on these studies, we eventually assigned three border widths to the forest fragments, according to the surrounding landscape: 100 m, for open pastures and
croplands; 50 m, for managed forests; and 75 m, for mixed landscapes. By doing this, we were able to differentiate and include in our analyses the influence that the surrounding matrix has over our 20 Atlantic Forest fragments. Of all the many morphometric indices yielded by FRAGSTATS, we chose the ‘border area’ (proportional border) as the best descriptor of edge effects to be used as a predictive variable, since the impact influence on the tree flora is more severe on those areas, and this index is the best way of bringing that to the analyses (Murcia 1995; Kupfer 2012). An additional variable was produced from the FRAGSTATS indices; the ‘heterogeneity of the border/ interior ratio,’ HBI, aimed at expressing a balance between the interior (I) and border (B) areas of a fragment: HBI = (minimum (B,I)/(maximum (B,I). HBI therefore ranges from 1 (a fragment with equal areas of border and interior) to 0 (fragments composed only of border or interior areas, the latter obviously non-existent in the case of fragments). By doing so, we were able to quantify how intensely the fragments’ area is influenced by the edge effect and how intensely is not affected by it (core areas). Both predictors were arcsine square root transformed (Zar 2010). (ii) Impact variables: In order to obtain information on the history of human intervention in the forest fragments, we used the Rapid Participatory Rural Appraisal technique (Mueller et al. 2010), through investigative questionnaires that aimed at reporting not only the impacts but also the history of land use in surrounding areas. In the case of areas that are protected by law or other legal mechanisms (PB, IB, and TD), we also consulted management plans and other documents. We then organized the information into an Interaction Matrix (Leopold et al. 1971) to assess the impact of seven modes of human intervention: ‘inner roads’ (roads that pass through the fragment), ‘outer roads’ (roads that pass nearby, but never pass through the fragment), ‘fire,’ ‘trails,’ ‘coppicing,’ ‘logging,’ and ‘cattle.’ We assigned scores to the severity, spatial extent, and time span of each intervention mode and summed them up. We based our scoring of impacts on Dean (1996) and also on the 25-year experience of the research team of the Universidade Federal de Lavras responsible for the continuous surveys of permanent plots (see Pereira et al. 2007 for further information). As the scoring process could be biased by subjectivity, we only ascribed scores after careful discussion with the whole research team. The different levels of impact based on scoring were used as categorical levels of each variable (Table 2). Each variable was evaluated on a scale ranging from 0 to 7, being zero the absence of impact and seven, the gravest stage of impact found in the fragments.
123
Environmental Management Fig. 1 Geographic situation of the a Alto Rio Grande region in southeastern Brazil and b the 20 forest fragments studied in the region () identified by two-letter codes (full names in Table 1)
Table 1 Identification (code and denomination), total area, central geographic coordinates, and altitude range of the 20 forest fragments
123
Code
Denomination
LV
Lavras
Area (ha) 5.72
Latitude (S)
Longitude (W)
21°140 4200 0
Altitude (m)
44°580 1000
918–937
00
44°580 5200
925–1,210
PB
Poc¸o Bonito
88.56
21°20 17
BS
Bom Sucesso
83.45
21°090 2700
44°540 1000
806–832
MD
Madre de Deus
20.66
21°290 0300
44°220 3200
915–980
0
00
0
00
913–945 913–960
IT
Itutinga
3.77
21°21 05
CM
Camargos
10.36
21°210 1800
44°360 4900
TD
Tiradentes
1,013.98
21°040 5200
44°080 0600
920–1,340
IB CP
Ibitipoca Capivari
95.05 9.78
21°420 4800 21°160 2300
43°530 0700 44°520 5300
1,150–1,510 825–875
SE
Subestac¸a˜o
8.73
21°130 1700
44°570 4700
910–940
0
00
44°36 29
0
00
ML
Mata da Lagoa
3.97
21°13 00
PI
Piedade
24.95
21°290 1600
44°060 0200
1,040–1,150
CR
35.98
21°360 2900
IN
Carrancas Ingaı´
LU
Lumina´rias
44°58 49
855–902
44°360 3800
1,440–1,513
16.14
21°24 26
00
44°530 3200
860–890
77.91
21°290 1100
44°440 2000
880–1,000
0
0
00
0
00
820–980
IU
Ibituruna
59.75
21°10 00
LA
Lafite
25.95
21°120 5800
44°480 0800
825–863
MI
Itumirim
54.97
21°130 3100
44°470 3100
815–910
PN
Pedra Negra
72.34
21°050 2500
44°560 4200
835–885
RM
Rio das Mortes
14.31
21°060 4900
44°450 0000
845–980
44°50 25
Environmental Management
Fig. 2 Flow chart summarizing the hypothesis, data collection, and analytical procedures used to investigate the interference of human impacts on 20 Atlantic forest fragments. LM linear models
Response Variables We produced response variables related to the tree community of the forest fragments from the dataset produced by the continuous surveys, including the species name, dbh (diameter at breast height) and total height of all live trees with dbh C 5.0 cm. We chose the dataset of the surveys carried out between 1998 and 2002 to minimize temporal bias. All response variables were generated using images from March 2001. Since total sample area varies among forest fragments, we used subsamples near 1 ha in all cases. As we wished to verify the influence of human impacts also on floristic composition and its abundance, we applied an ordination analysis to reduce the dimensionality of our species checklist. Thus, we obtained a set of response variables from the scores of each axis of two NMS ordinations (nonmetric multidimensional scaling; McCune and Grace 2002; Wildi 2010). The first NMS (hereafter, NMS1) aimed at representing the variations in species composition (binary presence matrix) and the second (NMS2), the variation in species abundance (relative density matrix). We assessed the significance of the axes by Monte Carlo tests (999 permutations), and the solution obtained was three-dimensional. We certified that the stress remained stable at the last iterations (McCune and Grace 2002). We
also obtained the proportion of the original n-dimensional ordination explained by each axis (R2). In order to assess the influence of impact history on regeneration guilds, we extracted the proportion of trees of pioneer species (%). Pioneers were defined here as those forming soil seed banks and requiring direct light radiation for germination (Swaine and Whitmore 1988). We produced two additional structural variables from regressing tree heights on log-transformed diameters: slope and intercept at the minimum dbh (5 cm), hereafter, h-slope and h-dbh5, respectively. These variables are intended to represent overall tree slenderness (higher slope shows that the trees are being able to grow taller without having to invest much in diameter acquisition and vice versa) and the slenderness of smaller trees, respectively (intercept values show the diameter size of the smaller trees). We also obtained the following structural variables for the tree community of each forest fragment: third quartiles of tree heights (hQ3) (to represent the height limits of trees, as those above the 75 % threshold along ranked height values); third quartiles of tree diameters (dbhQ3); basal area (BA, m2ha-1); density (D, trees.ha-1); and densities of trees with heights below or equal to the third quartile (D B hQ3) and above (D [ hQ3) (intended to represent the density of understory trees and canopy, respectively).
123
Environmental Management Table 2 Impact variables of the 20 Atlantic forest fragments used in LMs
Fragments are identified by their codes in Table 1. HBI (heterogeneity of the border/ interior ratio) and border area are proportions while the others are categorical variables
Code
Morphometric variables
Impact variables
Border area
HBI
Inner road
Outer road
Fire
Trails
Coppicing
Logging
Cattle
LV
1
0
1
7
1
3
1
6
0
PB
0.46
0.85
4
5
4
1
2
2
1
BS
0.59
0.69
1
1
1
1
1
2
2
MD
0.72
0.39
1
1
1
2
1
2
3
IT
0.98
0.02
1
1
2
0
1
1
1
CM
0.82
0.22
1
1
2
0
1
1
1
TD
0.12
0.14
3
1
2
2
4
4
1
IB
0.22
0.28
1
1
1
0
1
1
0
CP
0.96
0.04
1
3
2
2
5
5
1
SE
0.72
0.39
1
6
1
1
3
2
0
ML PI
1 0.42
0 0.72
3 1
6 2
2 1
3 0
3 1
3 2
3 1
CR
0.48
0.92
1
1
3
0
2
2
1
IN
0.63
0.59
1
1
1
1
1
2
2
LU
0.33
0.49
2
1
2
1
3
2
1
IU
0.39
0.64
1
7
5
1
4
2
1
LA
0.6
0.67
3
4
1
2
1
4
3
MI
0.65
0.54
4
5
2
1
6
3
2
PN
0.47
0.89
1
3
1
1
1
2
2
RM
0.59
0.69
2
5
2
1
1
2
1
Linear Models For each response variable, we prepared a general linear model (LM; Quinn and Keough 2002) based on least squares, with a significance level of 5 %. We first selected a set of potential relevant predictors in each model according to the existing literature and discarding variables with R2 (coefficient of determination) \ 0.3. Then we selected the best model based on the lowest AICc—corrected Akaike criterion (Burnham and Anderson 2002) using maximized log-likelihood and confirmed the lack of collinearities among predictor variables through the variance inflation factor (VIF), assuming a VIF cut-off of 10 (Quinn and Keough 2002). When we found VIF [ 10 in the best model, we discarded it and chose the best LM (i.e., that with the lowest AICc) among the remaining candidate models without collinearity. The spatial independence is a critical assumption of general statistical tests (Legendre et al. 2002). We therefore produced Moran’s I correlograms, following the SAM 4.0 (Rangel et al. 2010) defaults, to assess the significance of spatial structure of the residuals (Diniz-Filho et al. 2003) using Monte Carlo tests (999 permutations). We applied sequential Bonferroni correction to test the global spatial structure of the correlograms (Fortin and Dale 2005). When spatial structure was detected, we included spatial filters based on Moran’s Eigenvector Maps—MEM method by means of Delaunay triangulation (Dray et al. 2006). In this
123
step, we used only the MEMs required to remove spatial structure in the LM residuals. If spatial structure was present before the AICc selection procedure (see above), spatial filters were necessarily included in all candidate models (Diniz-Filho et al. 2008). Since the inflation of type I error may occur when both predictor and response variables are spatially structured (Legendre et al. 2002), we also obtained correlograms for these data. When both predictor and response variables were spatially structured, we followed Peres-Neto and Legendre (2010) in order to partition out the variance after a forward selection of MEMs, thereby isolating the fraction that causes the inflation of type I error and confirming the results of the aforementioned significance tests. We checked the assumptions of linearity and homogeneity of variances by plotting residuals on estimated values and observing the balance below and above the zero residual line, without any systematic structure such as curves or cones (Quinn and Keough 2002). When we suspected that the homogeneity of variances was violated, we performed Levene test (Zar 2010). We also tested the models for normality of residuals, using the Shapiro–Wilk test (Zar 2010). When any of the assumptions failed, we eliminated outliers (i.e., forest fragments with extreme studentized residuals). After this removal, we tested all the assumptions again, including the spatial independence, and checked the significance and coefficient of determination of the final model.
Environmental Management Table 3 Significant linear models used to explain the variation in NMS2 (relative density ordination) scores of 20 Atlantic forest fragments df
F
P
NMS2—Axis 1 Intercept
1
11.173
0.010
Fire Logging
4 5
14.291 6.265
0.001 0.012
MEM 1
1
8.710
0.018
Error
8
–
–
Intercept
1
13.874
0.002
Proportional border
1
13.249
0.002
–
–
NMS2—Axis 3
Error
17
Full results are shown in S1 (online supplementary material). Significant results (P B 0.05) in bold df degrees of freedom
We performed the analyses in PC-ORD 6.0 (McCune and Mefford 2011), PAST 2.0 (Hammer et al. 2001), SAM 4.0 (Rangel et al. 2010), and ‘vegan’ (Oksanen et al. 2011) and ‘spacemakeR’ (Dray et al. 2006) packages of R Statistical Environment (The R Foundation for Statistical Computing 2013).
Results The NMS2 ordination, which reflected the main trends of variation in species abundance, was highly influenced by
morphometric and, or, impact variables (Table 3). We found two linear models that were able to show the variables related to tree species abundance: for Axis 1 of NMS2 (F = 11.49, P \ 0.01, R2 = 93.49 %; R2adj = 85.35 %) and for Axis 3 of NMS2 (F = 13.25, P \ 0.01, R2 = 43.80 %, R2adj = 40.49 %). These axes explained 49.5 and 14.6 % (total of 64.1 %) of the floristic variance in the original n-dimensional space, respectively. The stress in real data was significantly lower than in randomized data (P = 0.02 in both axes). The trend of variation represented by Axis 1 of NMS2 was significantly related to fire, logging, and one spatial filter (MEM1) (Table 3). Intermediate levels of fire were related to higher scores (Fig. 3a). The greater the impact of logging, the higher the scores (Fig. 3b). The opposite was observed for the trends represented by Axis 3, where the higher the proportional border, the lower the scores (Fig. 3c). The structural descriptors of the vegetation were significant for third quartiles of tree diameters (dbhQ3), basal area (BA), and densities of trees with heights below or equal to the third quartile (D B hQ3) (Table 4; Fig. 4). The LM explaining the variations of dbhQ3 included only cattle as a significant predictor (Table 4). Only at the highest impact level of cattle, the dbhQ3 was significantly higher (Fig. 4a). In turn, two predictors were significant for BA: outer roads (impact variable) and one spatial filter (Table 4). The highest levels of outer roads were associated with lower values of BA (Fig. 4b). Finally, outer roads were also significantly related to D B hQ3 (Table 4). Here, in general, the larger the outer roads, the
Fig. 3 Ordination of the 20 Atlantic forest fragments by nonmetric multidimensional scaling of the relative species density (NMS2) per fragment. Diagrams show the proportional influence of each significant predictor in the linear model constructed to explain the variation in the NMS axis. Triangle size is proportional to the effect of fire in a, logging in b, and proportional border in c
123
Environmental Management Table 4 Significant linear models used to explain the variation in community structural variables of 20 Atlantic forest fragments df
F
P
dbhQ3 Intercept
1
1033.875
Outer road
6
2.763
0.075
Cattle
3
4.520
0.030
Error
10 298.968 3.447
