SEARCH

SEARCH BY CITATION

Keywords:

  • Cajamarca;
  • fragmentation;
  • land use change;
  • paramo;
  • Peru;
  • plant species diversity;
  • tropical Andes

Abstract

  1. Top of page
  2. AbstractResumen
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments
  8. Literature Cited
  9. Supporting Information

Habitat loss and fragmentation are considered major threats to biodiversity, especially in tropical mountain ecosystems. Most studies focus on the relationships between biodiversity and patch characteristics such as patch size, connectivity or degree of contrast with the surrounding matrix, but leave the rate of change within these variables little explored. We analyzed the importance of changes in patch characteristics over time on species diversity and species composition in the paramo of northern Peru, a tropical grassland ecosystem, locally known as jalca. We obtained land use/cover maps for 1987 and 2007 spanning an area of 6300 km2, and quantified land use change, jalca patch characteristics and their proportional changes over 20 yr. In 2009, 371 vascular plant species were recorded in 92 plots, each plot representative of single patches. Between 1987 and 2007, jalca cover decreased from 47 to 35 percent due to encroaching agriculture. This activity showed an upward shift probably favored by climate change. The number of jalca patches increased, mean patch size decreased, and the patches showed a higher contrast with the surrounding matrix. Multiple linear regression failed to show that species diversity relates to changes in patch characteristics. Canonical correspondence analysis indicated that species composition relates to the degree of contrast between the patch and its surrounding matrix and its changes through time. We concluded that changes in patch characteristics are important only for species composition. This study highlights the importance of considering matrix management with a long term perspective for conservation efforts.

Resumen

La pérdida de hábitat y la fragmentación son consideradas como unas de las mayores amenazas a la biodiversidad, especialmente en los Andes tropicales. La mayoría de estudios se enfoca en la relación entre biodiversidad y características de los parches como tamaño de parche, conectividad o grado de contraste con la matriz circundante, pero la tasa de cambio de estas variables ha sido aún poco explorada. En la presente investigación se analizó la importancia de los cambios a través del tiempo de las características de los parches en la diversidad y composición de especies de los páramos del norte de Perú, un ecosistema tropical de pasturas localmente conocido como jalca. En base a mapas de uso de suelo y cobertura de los años 1987 y 2007 y cubriendo un área de 6300 km2, cuantificamos el cambio de uso del suelo, las características de los parches de jalca y el cambio proporcional de estas en este periodo de 20 años. En el año 2009 se registraron 371 especies de plantas vasculares en 92 parcelas, cada parcela representando un parche diferente de jalca. Entre 1987 y el 2007, la cobertura de jalca disminuyó de 47 a 35 por ciento debido al avance de la agricultura, con énfasis en las zonas más altas, probablemente favorecido por el cambio climático. El número de parches de jalca se incrementó, el tamaño promedio de parche disminuyó y los parches mostraron un mayor contraste con su matriz circundante. Los resultados de la regresión lineal múltiple no mostraron relación alguna entre la diversidad de especies y los cambios en las características de los parches. El análisis de correspondencia canónica indicó que existe una relación no solo entre la composición de especies y el grado de contraste entre el parche y la matriz sino también con la tasa de cambio de este contraste con la matriz a través del tiempo. Concluimos que los cambios en las características de los parches influyen solo en la composición de especies. Este estudio resalta la importancia de considerar el manejo de la matriz con una perspectiva a largo plazo en los esfuerzos de conservación.

Habitatfragmentationduetotheexpansionofagriculture (Andrén 1994, Ewers & Didham 2006) and climate change (Peterson et al. 1997, Dirnböck et al. 2003) is one of the major threats to biodiversity (Tilman et al. 1994, Fahrig 2003), especially in high-mountain ecosystems. Fragmentation may produce a reduction of patch size, an increase of isolation between patches, the creation of boundaries with edge effects and an increase in the number of patches (Andrén 1994, McGarigal & Marks 1995, Gustafson 1998, Fahrig 2003, Turner 2005, Ewers & Didham 2006). Within single landscapes, patch size and connectivity have been associated with species richness (Bruun 2000, Helm et al. 2006), habitat occupancy (Hodgson et al. 2009) and species abundances (Eriksson et al. 1995). Also, the structure and quality of the matrix that surrounds patches have been correlated to biodiversity patterns (Ricketts 2001, Jules & Shahani 2003, Williams et al. 2006, Prugh et al. 2008).

The spatial setting of current patches in a matrix only represents the present-day result of fragmentation. It does not reveal any information about the rate and intensity of the changes to which a particular patch or matrix and the associated biodiversity have been subjected (McGarigal & Cushman 2002). Fragmentation studies, which included manipulation of the landscape to compare pre- and post-treatment effects on biodiversity generally, require complicated logistics, especially in tropical ecosystems (McGarigal & Cushman 2002). Intermediate approaches which link current biodiversity patterns with metrics of patches at some time in the past (Helm et al. 2006) or with dynamically changing patch metrics in time, remain little explored.

In the equatorial Andes above 3000 m altitude, the paramo is a tropical alpine grassland bordered by the snowline at its upper limit and by cloud forests and shrublands at its lower limit (Luteyn 1999). Paramo systems are naturally fragmented because of the steep terrain. At century and millennium scales patches of paramo vegetation have experienced continuous changes in isolation and connectivity, because of the alternate expansion and contraction of the paramo zone driven by climate change (van der Hammen 1979, Hooghiemstra & van der Hammen 2004). Adaptations of paramo plants to high levels of natural fragmentation include short regeneration times (Janzen 1973), a prevalence of anemochory and zoochory dispersal (Melcher et al. 2000) and effective short-range migration (Weigend 2002). In the past decades paramo vegetation has become increasingly threatened by fragmentation due to encroaching agriculture.

The aim of our study was to assess how paramo patches (locally known as jalca) changed through time and how such changes affected patterns of biodiversity in northern Peru. Our two main research questions were: (1) which changes have occurred in the jalca patches over the last 20 yr? (2) How did changes in patch characteristics affect plant diversity and species composition? To answer these questions we analyzed land use change and patch metrics of jalca vegetation on the basis of land use/cover maps derived from Landsat TM5 imagery of 1987 and 2007. We studied the diversity (Shannon diversity index) and composition of vascular plant species in 92 small plots stratified over patches of natural jalca vegetation, as function of the current patch metrics and their changes over the period of 20 yr. We specifically tested if jalca patches, which were small, isolated and surrounded by a more contrasting matrix in 2007, or that changed in that direction between 1987 and 2007 showed a lower diversity and a different composition of vascular plants compared to patches which were less affected by fragmentation.

Methods

  1. Top of page
  2. AbstractResumen
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments
  8. Literature Cited
  9. Supporting Information

Study area

The study was conducted in an area spanning 6300 km2, situated in the Andean region of south Cajamarca, north Peru (6°30′ S–7°30′ S; Fig. S1). The natural paramo (locally known as jalca) vegetation in this area is found between 3000 and 4200 m asl and typically consists of bunch grasses (Weberbauer 1945, Luteyn 1992). The total annual precipitation in the area ranges from 650 mm in the west to 1370 mm in the east (Hijmans et al. 2005). Mean annual temperature ranges from 5.7°C in the upper areas to 16.3°C in the lower areas (Hijmans et al. 2005). Paramo vegetation is well known for its high endemism and diversity (Luteyn 1992), and its capacity to regulate the water supply to the lowlands nearby (Buytaert et al. 2006). Despite these important environmental services, agriculture has largely replaced the natural jalca vegetation during the last years (Sánchez-Vega et al. 2005). The presence of two important dairy factories in Cajamarca has intensified overgrazing (Sanchez 2003). Mining has developed intensively since the beginning of 1990 (Sanchez 2003), and tree plantations (Pinus sp.) have increasingly replaced jalca vegetation.

Land use/cover maps and fragmentation

Two orthorectified LANDSAT images were analyzed. To reduce problems related to seasonality, we selected the available images from the dry season, which is known to minimize differences in reflectance of vegetation (Martinuzzi et al. 2007) and applied object-based classification, which is less dependent on weather conditions (Walter 2004) compared to a pixel-based approach and which leads to better classification results (Dingle Robertson & King 2011). The images were taken on 23 June 1987 and 24 August 2007 and downloaded from the Global Land Cover Facility site (http://www.landcover.org/index.shtml) and re-projected to UTM zone 17S, datum WGS84. The images were first segmented into image objects and then classified using the Definiens Developer 7 (Definiens 2007) software. First we applied a multi-resolution segmentation algorithm (Baatz & Schäpe 2000) that clusters individual pixels into image objects based on three criteria: reflectance values (color), shape, and compactness (Benz et al. 2004). For land use applications the reflectance values are most influential when classifying satellite images. In this part of the Andes, land cover categories occur as units with highly irregular shapes. Therefore, a low weight was given to the parameters of shape and compactness. We used a weight of 0.5 for color, 0.3 for shape and 0.2 for compactness. The nearest neighbor classifier was subsequently used in an object-based supervised classification of the image objects. The classification step was repeated to correct initially misclassified image objects by visual inspection of the satellite images. In this procedure the following land cover/use classes were identified: jalca, shrublands, cloud forest, lakes, agriculture, mining, and tree plantations. The map was again visually inspected for small patches (< 0.5 ha) incorrectly classified as jalca. These were merged with the desired adjacent class. The classification accuracy assessment for the 1987 image was performed following the method described by Congalton (1991) by comparing the classification results with a reference dataset and presented as a confusion matrix with the User's, the Producer's and the Overall accuracy and KHAT statistics. We used 198 randomly placed points in the image in the accuracy assessment of the land use/cover classes. For the accuracy assessment of the 2007 image we used the same 198 points and 40 additional GPS points taken during the fieldwork campaign. The overall accuracy for both years was 78 percent (Table 1).

Table 1. Accuracy assessment of the classification for LANDSAT 5TM of 1987 (a) and 2007 (b).
 Classification 
Reference dataCloudsJalcaShrublandsCloud forestAgricultureMiningTree plantationsTotalUser's accuracy (%)Producer's accuracy (%)Overall accuracy (%)KHAT
(a) 1987
Clouds20000002100100780.65
Jalca068101000797686  
Shrublands0980800256732  
Cloud forest0021000310033  
Agriculture012107200858085  
Mining00000202100100  
Tree plantations00000022100100  
Total2891219022198    
(b) 2007
Clouds1000000150100780.62
Jalca058101521777775  
Shrublands0220900134015  
Cloud forest0011000210050  
Agriculture11510111001288087  
Mining0000030360100  
Tree plantations00004010149171  
Total27551139511238    

Patches of jalca were extracted from these land use/cover maps using FRAGSTATS 3.3 (McGarigal & Marks 1995). We used the eight pixels rule to mark out a patch. Both orthogonal and diagonal adjacencies defined continuous patches. Metrics at patch and at class level were calculated for the jalca from the 2007 land use/cover map (Table 2) following McGarigal and Marks (1995). Patch core areas were defined using an inside buffer of 100 m (McGarigal & McComb 1995, Laurance et al. 1998). The search radius for the PROX variable was set at 300 m (Gustafson & Parker 1992). The weighted values of the contrast variable ECON were assigned considering similarity with the jalca vegetation as follows: a weight value of one was used for lakes, bare soils, and mining areas, an intermediate contrast level was used for agricultural (0.7) and plantation areas (0.6), and a low contrast level was taken for natural vegetation types (0.1 for shrublands and 0.3 for cloud forest). Class level metrics were only calculated for the jalca. Finally, the proportional change of each patch metric in two decades was calculated according to equation (1):

  • display math(1)

where p2007 represents the value of the patch metric derived from the 2007 image, and p1987 the value derived from the 1987 image (Table 2).

Table 2. Patch and class metrics following McGarigal and Marks (1995). The metrics representing the percentual differences of patch characteristics between the 1987 and the 2007 Landsat imagery were labeled with a capital D at the start of the label (D_AREA, D_GYRATE, D_FRAC, D_CONTIG, D_CORE, D_NCORE, D_ENN, D_PROX, D_ECON).
LabelDescription
  1. a

    Variables were log-transformed in further analyses.

Patch metrics
AREAaArea of patch (ha)
GYRATEaMean distance between each cell in the patch and the patch centroid (m)
FRACShape complexity. FRAC approaches 1 for very simple perimeters (i.e., squares) and approaches 2 for highly convoluted perimeters
CONTIGConnectedness between the pixels of the same patch. CONTIG equals 0 for a one-pixel patch and increases to 1 as patch contiguity increases
COREaRemaining area of a patch after removing a depth-of-edge distance from the patch perimeter (ha). We used a distance of 100 m
NCOREaNumber of disjoint core areas in one patch after removing the depth of edge distance
ENNEuclidean nearest distance to the closest patch (edge to edge) (m)
PROXaProximity index measures the proportion between the total area of a patch and its neighboring patches (within a specified radius) and the sum of edge to edge distances from the focal patch to the other patches. We used a search radius of 300. PROX = 0 if a patch has no neighbors and it increases as the neighborhood is increasingly occupied by patches of the same type and as those patches become closer in distribution
ECONEdge contrast index (%). ECON equals the sum of the patch perimeter segment lengths (m) multiplied by their corresponding contrast weights, divided by total patch perimeter (m). ECON = 100 when the entire patch perimeter is maximum-contrast edge. Values of weights for each land use/land cover are described in methods
Class metrics
NPNumber of total patches
CPLANDTotal core area of jalca patches divided by total study area (%)
LPILargest patch index: largest patch area divided by total study area (%)

Fieldwork

In 2009, field sampling was done in 92 jalca patches (Fig. S1), which were selected from the 2007 land use/cover map. This selection was stratified according to the range of patch sizes and the range of distances to the nearest patch. Within the limits of terrain accessibility, the vegetation plots were located as close as possible to the centroid of each patch, using a handheld GPS device. In each selected patch one plot of 2 × 2 m was sampled. In all plots the abundance and cover percentage were recorded for each vascular plant species. Specimens from all species were collected to allow species identification. Position (UTM zone 17S coordinates), percentage of exposed rocks and altitude were also registered for each plot. We took one soil sample at 0–5 cm depth in the centre of each plot, after removing litter. Soil analyses regarding texture (Hydrometer), total N (micro Kjeldahl), available P (modified Olsen), available K (sulfuric acid), pH (water), C (modified Walkley-Black) and exchangeable Al (extracted with 1 N KCl) were conducted at the Instituto Nacional de Investigación y Extensión Agraria (INIA) in Cajamarca (Table 3).

Table 3. Vegetation measures, soil properties and terrain characteristics obtained in 92 plot (4 m2) samples of jalca patches.
NameMeanSDRange
  1. a

    Variables were log-transformed in further analyses.

Plant density (number per 4-m2 plot)1767442–433
Species richness (number per 4-m2 plot)1857–30
Shannon Index2.30.40.7–3
Vegetation cover (%)902525–161
N (%)0.60.30.09–2.04
Pa (ppm)9.710.01.91–53.9
K (ppm)221.363.4149–380
pH4.41.23–7
C_org (%)8.04.51.11–26.29
Ala (Meq/100 g)2.22.50–8.26
Clay (%)28.712.27–73
Sand (%)48.315.617–87
Rock cover (%)14170–60
Elevation asl (m)34802703030–4130

Statistical analysis

Means of patch metrics of 1987 and 2007 were compared by means of Mann–Whitney U-tests using R v.2.11.1 (R Development Core Team 2008). We considered all patches identified on the 1987 and 2007 land use/cover maps. Species diversity was calculated for each 4-m2 plot using the Shannon Index (Shannon 1948). Applying multiple linear regression analyses in R v.2.11.1 this species diversity was related to four sets of explanatory variables: (1) current patch metrics (from the 2007 land use/cover map); (2) changes in metrics of patch characteristics; (3) soil and terrain properties; and (4) spatial variables (location of the plot). Following Borcard et al. (1992) and Legendre and Legendre (1998) we modeled our spatial information by adding all terms for a cubic trend surface regression, allowing to extract information in the response related to linear spatial patterns but also to non-linear patterns. Thus, the spatial variables were created by calculating the nine terms of the polynomial to the third degree (X, Y, X2, Y2, X3, Y3, X*Y, X*Y2, Y*X2) of the centered and standardized UTM (zone 17S) easting (X) and northing (Y). Some explanatory variables were log-transformed (Tables 2 and 3). Redundancy among the explanatory variables was examined by means of bivariate correlation coefficients (Table S1). Model simplification was done by backward stepwise regression analysis for each of the explanatory sets separately, using the step function in R. Models were evaluated using the corrected Akaike's information criterion (AICc), which is especially appropriate in situations of a low number of samples relative to the number of parameters (Burnham & Anderson 2002). In this process, the probability of each model was evaluated using so-called Akaike weights (wi) (Burnham & Anderson 2002, Whittingham et al. 2006). These values are based on the difference between the AICc value of each model and the AICc value of the best model (Δi) (Burnham & Anderson 2002). The importance of each variable was evaluated using the selection probability which is calculated by summing the wi values of all models in which the variable appears (Burnham & Anderson 2002, Whittingham et al. 2005). We included all variables with selection probabilities above 0.95 to build a final suite of regression models. The final parameter of each variable (β) was estimated by model averaging of that final suite (Burnham & Anderson 2002). This estimate was calculated as the sum, over all models of the final suite, of the product of a parameter found in a model and the Akaike's weight of that model. The model selection bias represented the difference between the averaged parameter and the parameter of the regression model including all variables, divided by the averaged parameter (Whittingham et al. 2005). Normality of residuals was checked for all the models. We calculated Moran's I of the residuals to check for spatial autocorrelation using the function Moran.I() in R.

With canonical correspondence analysis (CCA) in CANOCO for Windows 4.5 (ter Braak & Smilauer 2002), metrics of current patches (2007) and those of changes in patch characteristics were related to the species composition in the plots. CCA analyses were carried out using a biplot-scaling focused on inter-species distances. Preliminary CCA analyses were performed using each of the four above mentioned sets of explanatory variables separately. In each of these analyses, the most relevant variables were selected by means of a forward selection procedure (at P < 0.05) using Monte Carlo permutation test with 999 permutations. Subsequently, all relevant explanatory variables were used in the final CCA. Introduced species (Appendix S1) were excluded from all analyses.

Results

  1. Top of page
  2. AbstractResumen
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments
  8. Literature Cited
  9. Supporting Information

Habitat fragmentation

In 1987 jalca covered 47 percent of the study area, only slightly more than agriculture (43%). After two decades, agriculture had expanded to 55 percent and the jalca cover had diminished to 35 percent. A substantial part of this agricultural expansion took place at a relatively high elevation: 29 percent of the new agricultural patches were located at 3600–3800 m asl whereas only 12 percent were found at the lower limits of the jalca (3000–3200 m asl). Mining and tree plantations expanded as well (from 0.2% to 1.2% and 0.2% to 1.9%, respectively). Most important changes are shown in Figure 1. Most jalca patch metrics decreased significantly in this time lapse (Table 4). Also, patches showed a stronger contrast (ECON) with the surrounding matrix, implying that a higher proportion of the patch perimeters were exposed to more contrasting land use/cover classes. Shape complexity (FRAC) and number of core areas (NCORE) did not show a significant change. Between 1987 and 2007 the number of jalca patches (NP) increased from 744 to 1196. The total proportion of jalca core areas (CPLAND) decreased from 32.4 percent in 1987 to 19.3 percent in 2007, which is another indication that most jalca patches became smaller in this time lapse. In 1987 the largest jalca patch (LPI) still occupied almost a quarter of the study area (22.8%). In 2007 this class metric had reduced to 4.3 percent.

image

Figure 1. Areas of major land use/land cover change between 1987 and 2007.

Download figure to PowerPoint

Table 4. Patch metrics of jalca vegetation in 1987 (744 patches) and 2007 (1196 patches). For the patch metrics the results of the Mann–Whitney U-test of the differences between 1987 and 2007 are given.
Metrics19872007Mann–Whitney test
MeanSDMeanSD
  1. ns = P ≥ 0.05.

  2. a

    0.01 ≤ P < 0.05.

  3. b

    0.001 ≤ P < 0.01.

  4. c

    P < 0.001.

AREA40055501801340a
GYRATE4001070340660a
FRAC1.10.061.10.06ns
CONTIG0.720.270.680.29c
CORE2804250100850a
NCORE214210ns
ENN260380200300c
PROX15,30060,700300010,600a
ECON46234922b

Floristic diversity

The 92 plots were located between 3000 and 4200 m (Table 3). In total 371 vascular species were observed (Appendix S1). Seven of these species were introduced and 55 were endemic to Peru. Asteraceae was the most species-rich family (97 species of which 19 endemic), followed by Poaceae (54; 6 endemic), Scrophulariaceae (27; 15 endemic) and Fabaceae (16; 3 endemic). On average, a plot contained about 18 species and 176 plants, which represented a Shannon Index of 2.3 (Table 3). Frequent species were Calamagrostis sp. (present in 62% of the plots), Bidens triplinervia (44%), Rumex acetosella (42%), Stevia puberula (37%), Paspalum tuberosum (32%), Hieracium peruanum (29%), Werneria nubigena (28%) and Paranephelius uniflorus (27%).

Regression models of diversity and composition

The preliminary regression analyses of species diversity against each of the four sets of explanatory variables, considering a variable selection probability of 95 percent or more, led to the selection of D_CONTIG, D_PROX, FRAG, CONTIG, AREA, ENN (distance to closest patch), N, Al, elevation, Y3 and X*Y (Table S2). In the final regression (Table 5) the lower polynomial terms (Y2, X, and Y) were included as well (Venables 1998) and steps 7–11 represent the so-called confidence set of models (Burnham & Anderson 2002, Whittingham et al. 2006), which best approximates the true model (summed weights > 95%). None of these models showed spatial autocorrelation in their residuals (= 0.61, = 0.72, = 0.78, = 0.85, = 0.77 for step 7 to 11, respectively). In these final models ENN, elevation, Y and X*Y showed selection probabilities at or above 95%. Only elevation showed a negative partial regression coefficient while the others showed a positive partial regression coefficient. This implied that a higher diversity was found in patches, which occurred at longer distances from other patches, at locations represented by a higher product of UTM northing and UTM easting, further north areas (Y) and in low elevation areas. Bias was higher for those variables with low selection probabilities.

Table 5. Stepwise regression results evaluated using the corrected AIC (AICc), the difference between the AICc of each model and the best model (ΔAIC) and the Akaike weights (wi). For each variable the selection probability (s. prob.) was calculated by summing the weights of the models where they are present. Finally, the parameter for each variable was calculated by averaging all models (β) by using the weights as well and the bias of these averaged parameters from those of a model considering all variables.
StepD_CONTIGD_PROXFRAGCONTIGAREAENNNAlElevationY3Y2YXX*YAICcΔAICwi
1xxxxxxxxxxxxxx−120.4815.740.00
2xxxxxxxxxxxx x−122.4713.750.00
3xxxxxxxxxx x x−124.4511.770.00
4xxxxxx xxx x x−126.339.890.00
5xxx xx xxx x x−127.918.310.01
6xxx  x xxx x x−129.626.600.01
7xx   x xxx x x−131.564.660.03
8xx   x xx  x x−132.903.320.07
9x    x xx  x x−134.451.770.13
10     x xx  x x−135.840.380.25
11     x  x  x x−136.220.000.50
s. prob.0.250.130.020.000.011.000.000.501.000.060.001.000.001.00   
β0.00170.0000.01770.0075−0.00060.00020.0003−0.0358−0.00030.0000.0000.00390.00000.0002   
Bias−3.4−8.0−125.0−232.5−159.5−0.8−196.0−0.7−0.0−12.5−2119.5−0.4−7350.1−0.0   

In the CCA analyses of species composition, the stepwise procedures for each set of descriptor variables led to the selection of two metrics of changes in patch characteristics (D_GYRATE and D_ECON), five metrics of current patch characteristics (ENN, ECON, PROX, FRAG, NCORE), three soil and terrain variables (pH, altitude and percent of sand) and three spatial variables (X, X3, X2*Y). The final CCA considered all these variables and also the lower polynomial terms (X2, Y, and X*Y). The stepwise selection procedure in the overall CCA led to the elimination of PROX, FRAG and percent of sand. The first axis of this CCA accounted for 3.2 percent of the species composition, and the second axis 2.4 percent. Soil pH and altitude explained most of the species composition along these two axes (Table 6). The patch metrics ECON (edge contrast index) and NCORE (number of core areas), and the spatial variable X2*Y were strongly related to the second axis. When the effects of the soil, terrain and spatial variables were filtered out (by entering them as co-variables in the CCA), four patch metrics (ECON, ENN, D_ECON, and D_GYRATE) continued to show a significant effect on species composition (Fig. S2). Edge contrasts with the surrounding matrix, both in the current patch metric (ECON) and in the way these contrasts changed between 1987 and 2007 (D_ECON) explained most the species patterns. Species mostly related to strong edge contrasts and strong changes in this metric over the past two decades (showing optima along the negative side of the first CCA axis; Fig. 1) were Hesperomeles lanuginose, Pteridium aquilinum, Brachyotum naudinii and Stipa rosea. On the other hand, Dalea exilis, Tagetes filifolia, Setaria parviflora, Coreopsis townsendii and Luzula racemosa showed optima when patches had developed low edge contrast values.

Table 6. Canonical coefficients for standardized variables and intra-set correlation coefficients of the overall CCA considering only variables selected from the sub models.
VariablesCanonical coefficientsIntra-set correlation coefficients
AXIS1AXIS2AXIS1AXIS2
D_GYRATE0.11−0.010.47−0.01
D_ECON−0.04−0.110.000.16
ENN0.120.080.390.11
ECON−0.21−0.36−0.33−0.45
NCORE−0.10−0.69−0.03−0.51
pH0.57−0.710.79−0.43
Elevation−0.42−0.60−0.72−0.31
X0.330.110.130.15
X2−0.17−0.27−0.100.45
X3−0.120.19−0.020.37
Y−0.080.41−0.09−0.25
X*Y0.00−0.470.11−0.30
X2*Y−0.10−0.69−0.03−0.51

Discussion

  1. Top of page
  2. AbstractResumen
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments
  8. Literature Cited
  9. Supporting Information

Changes in habitat configuration: fragmentation

The land use/cover change analysis between 1987 and 2007 clearly showed that the natural vegetation of the jalca system is fragmenting mainly by encroaching agriculture. As a consequence the number of jalca patches increased by the breaking up of larger patches. These patches became smaller and more different in physiognomy with the surrounding matrix, implying that the natural ecotones of forest-jalca or shrubland-jalca degraded. More patches became surrounded by arable fields and other human activities. The decrease of the edge-to-edge Euclidean distance between 1987 and 2007 is probably due to the new patches formed after a large one was broken up. Since such patches remain close to each other, fragmentation easily contributes to a decreasing mean edge-to-edge distance between patches. The patch proximity index was distinctly higher in 1987 than in 2007 indicating less proximity between patches in 2007. This index is weighted, however, by the size of the neighboring patches (larger patches nearby yield a higher index than smaller patches). Therefore, given the reduction in the mean distance between patches (shown by the decreasing edge-to-edge Euclidean distance between patches), the decreasing proximity index likely showed that neighboring patches have become smaller in the past 20 yr. The only patch metrics, which did not change in the fragmenting jalca landscape, were the shape complexity, representing by fractal value of the patches, and the number of core areas. The constancy through time in the fractal values of the jalca patches probably reflects the low degree of mechanization of the agricultural activities. Most arable field created in the past 20 yr were still coerced to follow the natural boundaries between the jalca patches (as these are imposed by slope, altitude, soil rock cover and exposed bedrock).

The conclusions regarding the differences in time of the CORE (and NCORE) and PROX metrics are potentially biased by our arbitrary choice of the depth-to-depth distance (100 m) and search radius (300 m) for these metrics, respectively. Also our conclusions about the effects of these metrics on species diversity and composition as discussed below suffer from these restrictions. To our knowledge, there is no information how to define the most useful or informative values of these distances in a region with a complex topography such as the Andes. This issue certainly deserves more attention in the future, but is beyond the scope of the present study.

No link between changes in patch characteristics and plant diversity

With the exception of the edge-to-edge distance between patches, our regression analyses did not detect evidence that current patch properties or the rate of change within patch characteristics affected the vascular plant species diversity of the jalca vegetation. This is in accordance with the pattern found for periurban forests (Guirado et al. 2007), where changes in patch characteristics were assessed in a quantitative way. Possibly the degradation of the jalca patches was not strong enough to effectively influence species diversity. Some authors suggested that only strong habitat disturbances or severe loss of habitats lead to detectable relations between patch area and species diversity (Andrén 1994, Arroyo-Rodríguez et al. 2009). Recently fragmented patches may still function as a refuge for animals, which contributes to increasing seed dispersal into these patches (Jules & Shahani 2003). Such patches may also experience an increased influx of plant species because of the larger edge boundary with the matrix (Laurance et al. 2001). Higher species diversity was found in patches, which occurred at longer distances from other patches. As argued above, recent fragmentation presumably creates numerous patches, which are at close distance to each other. In turn, this would imply that more distant patches tend to be less affected by recent fragmentation, which might explain their relatively high species diversity. Finally the spatial effect on species diversity showed that the species diversity in the jalca increases toward the northeast and the north in general. This suggests that jalca areas with higher humidity tend to contain more species on an area basis.

Effects of changes in patch characteristics on plant species composition

Soil acidity and altitude mostly explained patterns in jalca species composition. The strong pH effect, which was consistent with Cooper et al. (2010), reflects the impact of the outcropping calcareous formations. The shallow soils in those areas show a high pH (potentially invoking P limitation) and low water retention capacity but a high warmth capacity. Deeper soils are more acidic and show a high level of soil organic matter, which induces a favorable environment regarding water storage and rooting. The impact of altitude reflects the general tendency for temperature in tropical alpine environments to strongly limit growth and nutrient availability (Janzen 1973, Ramsay & Oxley 2001).

Even after cancelling the effects of soil, terrain, and spatial variables, the current patch metrics and those of changes in patch characteristics significantly explained the variation in species composition, especially the contrast of patch edges with the surrounding matrix and the way this contrast changed through time. The jalca patches, which were surrounded by agriculture, mining and bare soils in 2007 or which became more surrounded by these land use/cover types over the past two decades, showed different species assemblages than jalca patches surrounded by cloud forest or shrublands.

The surrounding matrix is important because it affects resource availability, movements of pollinators, seed dispersers, and herbivores (Jules & Shahani 2003). If the edge contrast is low, the similarity in cover type and physiognomy between patch and matrix increases which enhances species exchange. Most likely, low edge contrast situations represent conditions of low disturbance by agriculture or mining in the surroundings of the jalca patches. We therefore assume that the plant composition in such patches best reflects the wide variation found in natural jalca patches. Contrary, in view of the encroaching agriculture, the jalca patches which show a high contrast with the surrounding matrix most likely represent jalca plant communities under threat by isolation, human disturbance and invasion by arable taxa. In such patches we would expect higher frequencies of species adapted to maintain viable populations under stress conditions (i.e., ferns like P. aquilinum or woody taxa like H. lanuginosa or B. naudinii). We would also expect a peak of fire resistant or species resistant to grazing activities (i.e., grass species like S. rosea). Our exploratory study suggests that potentially predictive information about the response of jalca communities to human stress can be obtained by measuring these plant traits in jalca patches as function of human disturbance, especially if conditions of soil pH, soil depth, substrate and altitude are controlled for.

Considerations for conservation measures

Our results show an increasing fragmentation process where a clear upward shift of agricultural practices has occurred over the last decades and similar tendencies have been reported for the Ecuadorian paramos (Hess 1990, López Sandoval 2004). The main drivers for this encroachment are most likely socio-economic in nature, such as an increasing population pressure, changes in land tenure, and economic resources stimulated by migration (López Sandoval 2004). Between 1950 and 1998, however, temperatures have increased throughout the tropical Andes (Vuille et al. 2003). It is therefore likely that this upward migration is favored by more suitable crop conditions. Future climate change will further increase temperatures, and is therefore expected to lead to higher fragmentation of the Andes in general (Young 2009), perhaps, reaching the threshold at which diversity index is affected by the fragmentation process. In addition, shifting cultivation in the jalca system, leading to the regeneration of jalca patches after agriculture, imposes a further hazard. The interaction of succession with fragmentation may influence extinction dynamics (Collins et al. 2009), and negatively affect the diversity and composition of species in the long run.

Our results also showed that not only the matrix surrounding each patch correlated with the abundance and persistence of some species (Ricketts 2001, Williams et al. 2006) but its temporal changes as well. Therefore, conservation efforts should consider matrix management with a long term perspective in this tropical alpine ecosystem. Despite the naturally fragmented condition of the jalca, human induced fragmentation in the last 20 yr showed an effect on species assemblages. Here, the history of the landscape may be even more important than the current landscape structure (Burel 1992).

Our approach might be replicable in other alpine areas and contribute to understand how landscape changes affect biodiversity. The methodology might be improved, however, by considering a longer period of time or more temporal resolution allowing for more detailed insights how jalca vegetation is affected by recent changes in the jalca landscape. In these, further evaluations how to apply CORE and PROX metrics are needed. More essentially, information about dynamic changes in the species composition and diversity of the jalca vegetation should be included, measured in permanent plots and analyzed through manipulative experiments. For this, initiatives like those of the GLORIA network (http://www.gloria.ac.at/?a=7) to set-up and maintain permanent sites of paramo vegetation and other high-mountain ecosystems worldwide, should be encouraged.

Conclusions

  1. Top of page
  2. AbstractResumen
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments
  8. Literature Cited
  9. Supporting Information

Comparative land use/cover maps indicate that the natural jalca vegetation has become increasingly more fragmented mainly by encroaching agriculture in the period from 1987 to 2007. In two decades jalca patches became smaller and more contrasting with the surrounding matrix. Vascular plant species diversity did not relate to metrics of changes in patch characteristics, neither to patch size or edge contrast, possibly because the degradation of the jalca patches was not strong enough or because species invaded patches out of the surrounding matrix. Metrics of changes in patch characteristics were associated to vascular plant species composition of the natural jalca vegetation, mainly through patch contrast change with the surrounding matrix, after controlling by soil and terrain characteristics. Conservation efforts should consider matrix management with a long term perspective as a strategy to protect the jalca vegetation. While our methods might be applicable to other alpine areas to improve management, more long term analysis would be important to fully understand the ecological processes creating the patterns observed.

Acknowledgments

  1. Top of page
  2. AbstractResumen
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments
  8. Literature Cited
  9. Supporting Information

We thank Fresia Chunga (CIPDER), the Instituto de Montaña, Mirella Gallardo (GTZ-Rural Cajamarca), Alicia Quispe (Proyecto Zonificacion Ecológica Económica de la región Cajamarca—Gobierno Regional de Cajamarca), Cesar Regalado (Agencia Agraria de Bambamarca), Eleuterio Mejía and Esteban Campos (Municipalidad Provincial Hualgayoc-Cajamarca) and Pedro Vásquez (CDC-UNALM) for the logistical support. We are also grateful to Edwin Cabrera, Segundo Sánchez, Germán Alcántara, Alindor Ordoñez, Wilson Salazar, Carlos Aguilar, Dagoberto Calvo, Ronal Regalado, Antero Zavaleta, Victor Briones, Laura Lucio, Manuel Peralvo, Juan Cusquisivan, Mara Deza and Nilton Deza for fieldwork assistance. David Rosario and Enrique Crisólogo kindly provided GIS information. Juan Montoya and Victor Campos assisted in species identification. This study received financial support of Proyecto Páramo Andino implemented by CONDESAN and the Universiteit van Amsterdam. Species were identified at the Herbarium of the Universidad Nacional de Cajamarca (Isidoro Sánchez-Vega), the Herbarium of Universidad Nacional Mayor de San Marcos (Oscar Tovar, Hamilton Beltrán, Raquel Gonzales), the Herbarium MOL of the Universidad Agraria La Molina (Mercedes Flores, Arturo Granda) and the Smithsonian Institute (Harold Robinson). We thank all of them for their invaluable help. Finally we thank Wouter Buytaert for his help in the statistical analysis using R.

Literature Cited

  1. Top of page
  2. AbstractResumen
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments
  8. Literature Cited
  9. Supporting Information
  • Andrén, H. 1994. Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable habitat: a review. Oikos 71: 355366.
  • Arroyo-Rodríguez, V., E. Pineda, F. Escobar, and J. Benítez-Malvido. 2009. Value of small patches in the conservation of plant-species diversity in highly fragmented rainforest. Conserv. Biol. 23: 729739.
  • Baatz, M., and A. Schäpe. 2000. Multiresolution segmentation – an optimization approach for high quality multi-scale image segmentation. In J. Strobl, T. Blaschke, and G. Griesebner (Eds.). Angewandte Geographische Informationsverarbeitung XII, pp. 1223. Wichmann Verlag, Salzburg.
  • Benz, U., P. Hofmann, G. Willhauck, I. Lingenfelder, and M. Heynen. 2004. Multi-resolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready information. ISPRS J. Photogramm. Remote Sens. 58: 239258.
  • Borcard, D., P. Legendre, and P. Drapeau. 1992. Partialling out the spatial component of ecological variation. Ecology 73: 10451055.
  • ter Braak, C. J. F., and P. Smilauer. 2002. CANOCO Reference manual and CanoDraw for Windows User's guide: Software for Canonical Community Ordination (version 4.5) Microcomputer Power. New York.
  • Bruun, H. H. 2000. Patterns of species richness in dry grassland patches in an agricultural landscape. Ecography 23: 641650.
  • Burel, F. 1992. Effect of landscape structure and dynamics on species diversity in hedgerow networks. Landsc. Ecol. 6: 161174.
  • Burnham, K. P., and D. R. Anderson. 2002. Model selection and multimodel inference. A practical information-theoretic approach (2nd Edition). Springer-Verlag, New York.
  • Buytaert, W., R. Célleri, B. De Bièvre, F. Cisneros, G. Wyseure, J. Deckers, and R. Hofstede. 2006. Human impact on the hydrology of the Andean páramos. Earth Sci. Rev. 79: 5372.
  • Collins, C. D., R. D. Holt, and B. L. Foster. 2009. Patch size effects on plant species decline in an experimentally fragmented landscape. Ecology 90: 25772588.
  • Congalton, R. G. 1991. A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens. Environ. 37: 3546.
  • Cooper, H. D., E. C. Wolf, C. Colson, W. Vering, A. Granda, and M. Meyer. 2010. Alpine peatlands of the Andes, Cajamarca, Peru. Arct. Antarct. Alp. Res. 42: 1933.
  • Definiens, A. G. 2007. Definiens Developer, München, Germany: Definiens A.G. Available at: http://www.definiens.com (accessed 4 December 2008).
  • Dingle Robertson, L., and D. King. 2011. Comparison of pixel- and object-based classification in land cover change mapping. Int. J. Remote Sens. 32: 15051529.
  • Dirnböck, T., S. Dullinger, and G. Grabherr. 2003. A regional impact assessment of climate and land-use change on alpine vegetation. J. Biogeogr. 30: 401417.
  • Eriksson, A., O. Eriksson, and H. Berglund. 1995. Species abundance patterns of plants in Swedish semi-natural pastures. Ecography 18: 310317.
  • Ewers, R. M., and R. K. Didham. 2006. Confounding factors in the detection of species responses to habitat fragmentation. Biol. Rev. 81: 117142.
  • Fahrig, L. 2003. Effects of habitat fragmentation on biodiversity. Annu. Rev. Ecol. Evol. Syst. 34: 487515.
  • Guirado, M., J. Pino, and F. Roda. 2007. Comparing the role of site disturbance and landscape properties on understory species richness in fragmented periurban Mediterranean forests. Landsc. Ecol. 22: 117129.
  • Gustafson, E. J. 1998. Quantifying landscape spatial pattern: what is the state of the art? Ecosystems 1: 143156.
  • Gustafson, E. J., and G. R. Parker. 1992. Relationships between landcover proportion and indices of landscape spatial pattern. Landsc. Ecol. 7: 101110.
  • van der Hammen, T. 1979. Historia y tolerancia de ecosistemas parameros. In M. L. Salgado-Labouriau (Ed.). El Medio Ambiente Páramo, pp. 5566. Ediciones Centro de Estudios Avanzados, Mérida, Colombia.
  • Helm, A., I. Hanski, and M. Partel. 2006. Slow response of plant species richness to habitat loss and fragmentation. Ecol. Lett. 9: 7277.
  • Hess, C. 1990. “Moving up-Moving down”: agro-Pastoral land-use patterns in the Ecuadorian Paramos. Mt. Res. Dev. 10: 333342.
  • Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, and A. Jarvis. 2005. Very high resolution interpolated climate surfaces for global land areas. Int. J. Climatol. 25: 19651978.
  • Hodgson, J., A. Moilanen, and C. D. Thomas. 2009. Metapopulation responses to patch connectivity and quality are masked by successional habitat dynamics. Ecology 90: 16081619.
  • Hooghiemstra, H., and T. van der Hammen. 2004. Quaternary Ice-Age dynamics in the Colombian Andes: developing an understanding of our legacy. Philos. Trans. R. Soc. B 359: 173181.
  • Janzen, D. H. 1973. Rate of regeneration after a tropical high elevation fire. Biotropica 5: 117122.
  • Jules, E. S., and P. Shahani. 2003. A broader ecological context to habitat fragmentation: why matrix habitat is more important than we thought. J. Veg. Sci. 14: 459464.
  • Laurance, W. F., L. V. Ferreira, J. M. Rankin-de Merona, and S. G. Laurance. 1998. Rain forest fragmentation and the dynamics of Amazonian tree communities. Ecology 79: 20322040.
  • Laurance, W. F., D. Pérez-Salicrup, P. Delamonica, P. Fearnside, S. D'Angelo, A. Jerozolinski, L. Pohl, and T. E. Lovejoy. 2001. Rain forest fragmentation and the structure of Amazonian liana communities. Ecology 82: 105116.
  • Legendre, P., and L. Legendre. 1998. Numerical ecology (2nd Edition). Elservier, Amsterdam, The Netherlands.
  • López Sandoval, M. F. 2004. Agricultural and Settlement frontiers in the tropical andes: the paramo belt of northern Ecuador, 1960–1990. Institut für Geographie and der Universität Regensburg, Regensburg, Germany.
  • Luteyn, J. L. 1992. Paramos: why study them? In H. Balslev, and J. L. Luteyn (Eds.). Paramo: an andean ecosystem under human influence, pp. 114. Academic Press Limited, U.K.
  • Luteyn, J. L. 1999. Paramos: a checklist of plant diversity, geographical distribution, and botanical literature. New York Botanical Garden Press, The Bronx.
  • Martinuzzi, S., W. A. Gould, and O. M. Ramos González. 2007. Creating cloud-free Landsat ETM+ Data sets in tropical landscapes: cloud and cloud-shadow removal. United States Department of Agriculture, Portland, Oregon.
  • McGarigal, K., and S. A. Cushman. 2002. Comparative evaluation of experimental approaches to the study of habitat fragmentation effects. Ecol. Appl. 12: 335345.
  • McGarigal, K., and B. J. Marks. 1995. FRAGSTATS: spatial pattern analysis program for quantifying landscape structure. Pacific Northwest Research Station, Portland, Oregon.
  • McGarigal, K., and W. C. McComb. 1995. Relationships between landscape structure and breeding birds in the Oregon Coast Range. Ecol. Monogr. 65: 235260.
  • Melcher, I., F. Bouman, and A. Cleef. 2000. Seed dispersal in páramo plants: epizoochorous and hydrochorous taxa. Plant Biol. 2: 4052.
  • Peterson, D. L., E. G. Schreiner, and N. M. Buckingham. 1997. Gradients, vegetation and climate: spatial and temporal dynamics in the Olympic Mountains, U.S.A. Glob. Ecol. Biogeogr. 6: 717.
  • Prugh, L., K. Hodges, A. Sinclair, and J. Brashares. 2008. Effect of habitat area and isolation on fragmented animal populations. PNAS 105: 2077020775.
  • R Development Core Team. 2008. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Available at: http://www.R-project.org (accessed 15 September 2010).
  • Ramsay, P. M., and E. R. B. Oxley. 2001. An assessment of aboveground net primary productivity in Andean grasslands of central Ecuador. Mt. Res. Dev. 21: 161167.
  • Ricketts, T. H. 2001. The matrix matters: effective isolation in fragmented landscapes. Am. Nat. 158: 8799.
  • Sanchez, P. E. 2003. Perú. In R. Hofstede, P. Segarra, and P. Mena (Eds.). Los páramos del mundo. Proyecto Atlas Mundial de los Páramos, pp. 159204. Global Peatlands Initiative/NC-IUCN/EcoCiencia, Quito, Ecuador.
  • Sánchez-Vega, I., M. Cabanillas, A. Miranda, W. Poma, J. Díaz, F. Terrones, and H. Bazán. 2005. La Jalca, el ecosistema frío del noroeste peruano – Fundamentos biológicos y Ecológicos (1st Edition). Yanacocha – Geográfica EIRL, Lima, Perú.
  • Shannon, C. E. 1948. A mathematical theory of communication. Bell Syst. Tech. J. 27: 379423.
  • Tilman, D., R. M. May, C. Lehmann, and M. A. Nowak. 1994. Habitat destruction and the extinction debt. Nature 371: 6566.
  • Turner, M. G. 2005. Landscape Ecology: what is the state of the Science? Annu. Rev. Ecol. Evol. Syst. 36: 319344.
  • Venables, W. N. 1998. Exegeses on Linear Models. Paper presented at S-PLUS User's Conference, pp. 126, Washington, DC.
  • Vuille, M., R. Bradley, M. Werner, and F. M. Keimig. 2003. 20th century climate change in the Tropical Andes: observations and model results. Clim. Change 59: 7599.
  • Walter, V. 2004. Object-based classification for remote sensing data for change detection. ISPRS J. Photogramm. Remote Sens. 58: 225238.
  • Weberbauer, A. 1945. El mundo vegetal de los Andes peruanos. UNALM, Ministerio de Agricultura, Lima, Peru.
  • Weigend, M. 2002. Observations on the biogeography of the Amotape-Huancabamba zone in northern Peru. Bot. Rev. 68: 3854.
  • Whittingham, M. J., P. A. Stephens, R. B. Bradbury, and R. P. Freckleton. 2006. Why do we still use stepwise modelling in ecology and behaviour? J. Anim. Ecol. 75: 11821189.
  • Whittingham, M. J., R. D. Swetnam, J. D. Wilson, D. E. Chamberlain, and R. P. Freckleton. 2005. Habitat selection by yellowhammers Emberiza citrinella on lowland farmland at two spatial scales: implications for conservation management. J. Appl. Ecol. 42: 270280.
  • Williams, N., J. Morgan, M. McCarhy, and M. McDonnell. 2006. Local extinction of grassland plants: the landscape matrix is more important than patch attributes. Ecology 87: 30003006.
  • Young, K. R. 2009. Andean land use and biodiversity: humanized landscapes in a time of change. Ann. Missouri Bot. Gard. 96: 492507.

Supporting Information

  1. Top of page
  2. AbstractResumen
  3. Methods
  4. Results
  5. Discussion
  6. Conclusions
  7. Acknowledgments
  8. Literature Cited
  9. Supporting Information
FilenameFormatSizeDescription
btp820-sup-0001-TableS1.docxWord document48KTABLE S1. Spearman correlation between variables.
btp820-sup-0002-TableS2.docxWord document25KTABLE S2. Stepwise regression results for each of the four sets of variables defined to explain diversity.
btp820-sup-0003-FigureS1.docxWord document13KFIGURE S1. Study area in the northern Andes of Peru.
btp820-sup-0004-FigureS2.docxWord document13KFIGURE S2. CCA ordination diagram showing how the four patch metrics relate to the species composition in the plots after controlling for soil pH, spatial variables and elevation.
btp820-sup-0005-f2.tifTIFF image529K 
btp820-sup-0006-AppendixS1.docxWord document50KAPPENDIX S1. Species list and species scores of the canonical correspondence analysis.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.