Breeding bird species diversity in the Negev: effects of ... - BES journal

6 downloads 0 Views 466KB Size Report
R. F. Desert finch (Rhodopechys obsoleta). R(N). F. Corn bunting (Miliaria calandra). R(N). S. JPE_667.fm Page 1147 Thursday, September 20, 2001 2:52 PM.
JPE_667.fm Page 1135 Thursday, September 20, 2001 2:52 PM

Journal of Applied Ecology 2001 38, 1135 –1147

Breeding bird species diversity in the Negev: effects of scrub fragmentation by planted forests

Blackwell Science Ltd

EYAL SHOCHAT, ZVIKA ABRAMSKY and BERRY PINSHOW Department of Life Sciences, Ben-Gurion University of the Negev, PO Box 653, Beer-Sheva, 84105, Israel; and Mitrani Department of Desert Ecology, Blaustein Institute for Desert Research, Sede-Boker, 84990 Israel

Summary 1. Afforestation of the Northern Negev, Israel, from 1956 resulted in patches of primarily coniferous trees that fragmented large scrubland areas. This alteration in landscape pattern was followed by immigration of mediterranean bird species to the Negev. 2. We counted breeding birds, and measured various environmental variables in scrubland and planted forest patches, to test whether bird assemblages were random subsets of the regional species pool, and whether area or habitat structure was the major correlate with species abundance and distribution. 3. Of 22 bird species recorded, only three appeared in both scrub and forest, showing that these two habitats were occupied by different species assemblages. In both habitats, species richness increased with area at a rate greater than that expected by random sampling. In the scrub this increase was related to area per se, while in the forest it was related to habitat diversity in terms of stand age and tree type. 4. The density of forest species was unaffected by area, but specialist scrubland species declined as area decreased. We suggest that edge effects might reduce species abundance in small scrubland patches. 5. Nested subset analysis indicated that, at the community level, species composition was not random. However, at the species level, the distribution of three forest-dwelling species appeared as random, as it was associated with habitat rather than with patch size. 6. Our results indicate that increased diversity of breeding birds in the Northern Negev will require scrub patches larger than 50 ha among the increasingly forested landscape. In contrast, increasing forest area would hardly increase species diversity in the whole landscape. Future forest management regimes should also aim to increase habitat diversity by adding foliage layers, especially in the understorey. Exotic coniferous forests support fewer species than deciduous forests in mediterranean zones around the world. The suggested management regime may improve such forests as habitat for species-rich bird communities. Key-words: afforestation programmes, area effects, bird communities, habitat structure, management, nested subsets, pine plantations. Journal of Applied Ecology (2001) 38, 1135–1147

Introduction Habitat fragmentation due to the conversion of natural habitats into agricultural, industrial or urbanized land has received much attention from conservation biologists (Saunders, Hobbs & Margules 1991; Haila,

© 2001 British Ecological Society

Present address and correspondence: Eyal Shochat, Center for Environmental Studies, Arizona State University, Tempe, AZ 85287 – 3211, USA (fax 480 9658087; e-mail [email protected]).

Saunders & Hobbs 1993; Fahrig & Merriam 1994; Murcia 1995; White et al. 1997). These studies have extended the view of habitat area loss per se to postulate that two more components might be important: (i) reductions in average habitat patch size and (ii) the increase in patch isolation, which turns habitat patches into islands (Andren 1994). Studies of habitat fragmentation generally emphasize the response of biological populations to changes in patch size, shape or configuration at various scales (Picket & Cadenasso 1995; Wiens 1995). In fragmented landscapes, reduction in patch size and

JPE_667.fm Page 1136 Thursday, September 20, 2001 2:52 PM

1136 E. Shochat, Z. Abramsky & B. Pinshow

© 2001 British Ecological Society, Journal of Applied Ecology, 38, 1135–1147

increasing isolation enhance the negative effect of the reduction in total habitat area on population size (Andren 1994). Habitat fragmentation often reduces connectivity, which breaks continuous populations into metapopulations (Hanski & Gilpin 1991) or source – sink populations (Pulliam 1988). It may also increase negative edge effects from stochastic processes (reviewed by Simberloff 1994) or from predators or parasites that enter small patches from the core environment (Ambuel & Temple 1983; Wilcove 1985; Blake & Karr 1987; Robinson et al. 1995). The extirpation of a particular species from isolated fragments may be followed by recolonization by individuals of the same or different species according to the concept of dynamic equilibrium (MacArthur & Wilson 1967). Whether a species is re-established and remains in sink populations ( Pulliam 1988), or fails to recolonize a patch, may, in turn, influence species diversity. Prior to the development of landscape ecology, species abundance, distribution and diversity were generally related to habitat structure (MacArthur & MacArthur 1961; Pianka 1967; Cody 1975). Following the theory of island biogeography (MacArthur & Wilson 1967), ecologists have tried to separate the effects of area and habitat structure by various experimental and statistical methods (Schoener & Schoener 1981; Tonn & Magnuson 1982; Haila 1983; Haila, Jarvinen & Kuusela 1983; Kohn & Walsh 1994). A major process that both alters and fragments natural habitats is the creation of pine plantations. Exotic pine forests have become increasingly popular in afforestation programmes in mediterranean zones worldwide (Gordo & Gil 1990; Cal 1994; Barbero 1995). For example, pines have been planted in Spain in large areas where oak woodlands were initially cleared to create arable land (Gordo & Gil 1990; Diaz et al. 1998). Studies on bird communities suggested that the reforestation of these former arable lands with pines might not increase forest bird species diversity because suitable microhabitats for specialists are absent, especially in the understorey (Lopez & Moro 1997; Diaz et al. 1998). In the Northern Negev, Israel, thousands of hectares of scrubland have been afforested with exotic conifers (mostly Pinus halepensis Mill, Pinus pinea L., Pinus canariensis Smith and Cupressus sempervirens L.) since 1956. The mediterranean and semi-arid scrubs of Israel are characterized by low and thick perennials that are unique habitat for specialists such as the longbilled pipit Anthus similis (Shirihai 1995). While many studies on birds in fragmented habitats have involved natural forests fragmented by commercial clear-cutting (Ambuel & Temple 1983; Blake & Karr 1987; Moller 1987; Lynch & Saunders 1991; Robinson et al. 1995; Telleria & Santos 1997), the Negev provides an opportunity to assess effects where forest is the new habitat fragmenting open habitats. Over the past four decades, these forest islands have grown both in size and age, while the once continuous scrub area has become a

patchy habitat. These changes in landscape structure have been followed by dramatic changes in the local bird community composition. One conspicuous phenomenon was the immigration of several mediterranean bird species from northern and central Israel to the Negev and their establishment in the plantations (Shirihai 1996). We investigated whether bird abundance and distribution correlated with area or habitat structure in planted forest and scrubland patches in the Northern Negev. We assumed that two different processes shape bird assemblages in each habitat. In the fragmented scrub we assumed that the reduction in patch size leads to extinction of specialist species, whereas in the planted forests the observed bird assemblage would be the product of colonization. Because the spatial scale in our study was relatively small, we did not consider distance from source populations as an important factor affecting forest bird species diversity. Five of the species studied were long-distance migrants, and the spatial scale of our study is negligible compared with the distance to their wintering grounds. Rather, we assumed that bird species colonize forest patches according to the presence of their preferred habitat subtypes (in terms of tree types or age). We hypothesized that assemblages in both habitats are non-random sets of a regional species pool, and are not the result of an increase in species richness due to increasing area and sample size (Connor & McCoy 1979; Haila 1983) but rather are affected by habitat structure or edge effects (hereafter biological effects). We predicted that (i) in each habitat there will be an increase in both species richness and species diversity with an increase in patch size; (ii) in the fragmented scrub, where specialist species exist, area will be a major correlate with species diversity and population density; and (iii) in the forest, habitat structure will be the primary correlate with species diversity and population density, whereas area would be of minor importance.

Study area and methods   Within an area of 1000 km2 north of Beer-Sheva, Israel (Fig. 1), we sampled all habitat patches (scrub or forest) larger than 50 ha. Patches smaller than 50 ha were selected according to their resolution level. That is, we selected patches with well-defined borders from the rest of the environment, so that patch size and habitat characteristics could be defined. Altogether we sampled 14 scrub patches that ranged from 2·5 to 2000 ha, and 10 forest patches between 4 and 3000 ha. The study area was located between the mediterranean zone in the north and the desert zone in the south. Average annual precipitation decreases from 350 mm in the north to 250 mm in the south. Settled and agricultural habitats exist, but semi-desert scrub and planted forests cover about 80% of the total area.

JPE_667.fm Page 1137 Thursday, September 20, 2001 2:52 PM

1137 Breeding bird species diversity in the Negev

Fig. 1. A map of the studied forest and scrub patches north of Beer-Sheva, Israel. Scrub patches larger than 50 ha are shown. For more information see text.

   

© 2001 British Ecological Society, Journal of Applied Ecology, 38, 1135 –1147

To sample bird densities we used line transects (Bibby & Burgess 1992) of 250 m that were surveyed for 15 min each. Generally, all 250 m fell within the plot. In four cases, however, the transect crossed plots of different habitat subtypes (e.g. young – old coniferous; pine – broadleaf plots) because plantations were too small. We counted pairs, singing males and active nests within 50 m to each side of the line. All bird counts were done by three observers throughout three consecutive breeding seasons, 1996–98. Birds flying overhead were not counted. As patch area increased, the number of transects increased from 1 to 10 (log number of transects = 0·35 log area – 0·55). This protocol allowed us to cover most of the habitat subtypes in the forest (e.g. different stand age, broadleaf and coniferous plots) because the number of habitat subtypes was not linearly correlated with area. Censuses lasted for 4 h from first light. Because summer birds arrive in late spring and breed later than resident birds, we repeated each transect three times during each breeding season (late March–mid-June). For the same reason, for each species we used the highest number observed per transect (see the Appendix for common and scientific names). We sampled habitat characteristics by measuring several variables in each line transect plot (250 × 100 m). In forest transects we randomly chose 20 trees and measured their height, diameter at base (basal area; BA), diameter at breast height (d.b.h.) and distance to the nearest neighbouring tree. Height of tall trees was

recorded by measuring the length of a tree shadow standardized against the shadow length of a 1-m pole. Within each line transect plot we measured natural vegetation cover along two 50-m transects (vegetation transect). Along each vegetation transect, we measured the total proportion of area covered by low perennial plants (mainly Sarcopoterium spinosum, Euphorbia hierosolymitana, Ballota undulata and Phlomis brachyodon; nomenclature after Danin 1970). The density of large bushes (above 1 m, mainly Thymelaea hirsuta, Rhamnus palaestinus and Ephedra campylopoda) was measured by counting all bushes in a line transect plot. We then randomly selected half of the bushes and measured maximum width and height to the nearest 10 cm. Because our study area was located on the border between two biogeographical zones (mediterranean and desert), we included altitude, longitude and latitude of each transect in our analysis. Some mediterranean species (e.g. Sardinian warbler, Eurasian jay, long-billed pipit and rock sparrow) were found only in the northern part of the study area; desert species were found only in the south (desert finch and desert lark: border with the Negev Desert) or east (desert lark and scrub warbler: border with the Judean Desert). To assess how environmental factors might affect the distribution of individual bird species, we performed detrended canonical correspondence analysis (DCCA) using  (ter Braak 1986, 1992). Densities of each species on a transect were first averaged over the three breeding seasons, 1996–98. Environmental data used in the analysis included both geographical

JPE_667.fm Page 1138 Thursday, September 20, 2001 2:52 PM

1138 E. Shochat, Z. Abramsky & B. Pinshow

Table 1. Results of PCA showing the first three axes (PC1, PC2 and PC3) of habitat structure and environmental variables at our Northern Negev research site (see text for details). The numbers represent correlation of each variable with each axis. Only significant loadings (P < 0·05) are shown Variable Area Altitude Longitude Latitude Tree height Tree density Basal area d.b.h. Perennial cover Total tree species Bush density Bush height Bush width % variance Total percentage

© 2001 British Ecological Society, Journal of Applied Ecology, 38, 1135–1147

PC1

PC2

PC3

– 0·523 – 0·600 – 0·625

– 0·554 – 0·752 – 0·663

– 0·530 – 0·722 – 0·743 19·1 59·6

0·562 0·531 0·542 17·9 77·5

– 0·301 0·942 0·880 0·961 0·942 – 0·875 0·856

40·5 40·5

number of individuals and species extracted from all three breeding seasons. The random sample hypothesis (Connor & McCoy 1979) stated that larger patches sample more individuals of a regional species pool, and therefore more species. The observed species– area relationship is therefore expected by chance. However, an increase in Fisher’s alpha with sample size indicates that species richness increases due to biological factors, such as habitat structure, edge effects or isolation (Rosenzweig 1995). We chose this index because other indices, such as rarefaction (Sanders 1968) or Simpson’s index (Simpson 1949), may, by themselves, be sensitive to sample size (Rosenzweig 1995). To test whether changes in species diversity are scale-dependent, we applied linear and higher-order polynomial regression analysis to the observed species diversity patterns (i.e. the change in Fisher’s alpha as patch size increases). Additional variables were included in the species diversity equations only if significant.

d.b.h., diameter at breast height.

  

and vegetation variables measured during spring 1997 (Table 1). Gradients in biogeography and vegetation structure among plots were described using principal component analysis (PCA; Morrison 1967) based on 13 environmental variables. We calculated component scores for each plot from the first three principal components (PC) extracted during the analysis. To assess how bird species respond to environmental variables, we used stepwise multiple regression (holding P to enter = 0·05 and P to remove = 0·1) in which bird densities were the dependent variable. We first regressed the abundance of each species against each environmental variable separately, starting with PC 1, 2 and 3 (following Blake & Karr 1987). When bird densities were significantly correlated with one or more of these components, we did not use individual environmental variables to explain bird abundance, but noted which environmental variable gave the best fit. When species densities were not correlated with any PC variable, we correlated densities against each true environmental variable. Only in cases where functions appeared non-linear, and where non-linear transformations yielded a better fit, did we apply these transformations to the model. We did either second-degree polynomial or exponential transformations and always selected the variable with the best fit (i.e. the one with the highest r 2).

In each habitat (scrub or forest), we tested whether the bird community had a nested subset pattern by using the randomization program 1 (Patterson & Atmar 1986). We found this program to be more conservative than the more modern program  (Atmar & Patterson 1993). 1 compares species distribution patterns on habitat islands (presence/ absence scores) of 1000 random assemblages to the observed community using the equivalent of the Student’s t-test. For this analysis we used all species that breed regularly in the forest patches, based on a longterm survey (Shirihai 1996). To test which species diverge from nestedness we developed a simple algorithm to calculate how many displacements-of-presence scores are needed to bring each species to perfect nestedness (Simberloff & Levin 1985). For this procedure, the null hypothesis (H0) was that species occurrences on the different patches were random. To test H0, the program randomized species presence scores within each species column to create a random species distribution, and compared the number of displacements required to achieve complete nestedness between the observed community and the random pattern. This process was repeated 10 000 times. We accepted H0 in cases where the number of transformations in the randomized patterns was equal to or smaller than that of the observed pattern in more than 500 iterations (i.e. P < 0·05).

   

Results

We plotted species–area curves for scrub and forest patches on a log – log scale (Arrhenius 1921) for all passerine species. To test whether the increase in species richness with area was due to random sampling or biological factors, we calculated Fisher’s alpha diversity index (Fisher, Corbet & Williams 1943) using the total

  Three PCs accounted for 76% of the variation in habitat structure (Table 1). The first component separated plots on the basis of vegetation structure, being positively correlated with tree density, tree maturity (e.g.

JPE_667.fm Page 1139 Thursday, September 20, 2001 2:52 PM

1139 Breeding bird species diversity in the Negev

tree height, BA and d.b.h.) and the number of tree species, but negatively correlated with perennial cover. The second axis contrasted plots according to patch size, altitude, longitude and bush structure (i.e. density, height and width of bushes). The third component was similar to the second factor, but separated bush density, height and width (Table 1).

    Altogether, 36 common birds (25 passerines and 11 non-passerines) breed in the scrubland and forest in the Northern Negev (see the Appendix). Ten species breed only in scrub, 17 only in forest and nine in both habitats. We applied statistical analysis to 20 species of passerines we detected in our line transects, omitting all non-passerine species and five passerines with < 5% of the total records. Of the 20 passerines, 15 were resident or nomadic within the study area, while five were long-distance migrants (see the Appendix). Eighteen of these species were found in all 3 years and two (scrub warbler and rock sparrow) in 1 year.

© 2001 British Ecological Society, Journal of Applied Ecology, 38, 1135 –1147

Table 2. Results of DCCA of 20 bird species in 67 plots in the Northern Negev. The total unconstrained eigenvalue was 2·64. Species ordination diagram is shown in Fig. 2 Axis 1

Axis 2

Canonical eigenvalues 0·87 0·16 Species variance accounted by axes (%) 33 6 Species variance accounted by arrows (%) 63·9 55·2 Species – environment correlations 0·984 0·790 Monte Carlo simulation, P-values < 0·01 < 0·01

In DCCA (Table 2), the first ordination axis consisted of vegetation structure and separated species according to the vegetation with which they were associated (tree age and perennial cover). The second axis consisted of geographical or physical variables such as area (patch size), latitude, longitude and altitude, and further separated each bird assemblage (Fig. 2). Among scrub-dwelling species, three species that occasionally also bred in the forest (crested lark, graceful prinia and woodchat shrike) were grouped together. Among the forest-dwelling species, the five new immigrant species (great tit, spotted flycatcher, blackbird,

Fig. 2. Ordination diagram of the first two axes of detrended canonical correspondence analysis for bird species and environmental variables in forest and scrub in the Northern Negev, Israel. Axis 1 and axis 2 accounted for 33% and 6% of the variance in the species data. Arrows represent directions of greatest change of environmental variables. The location of a species’ score relative to the arrows indicates the environmental preferences of that species. Open squares, species breeding in both scrub and forest; filled squares, local woodland dwelling species; open circles, new immigrant forest species; filled circles, scrubland species. The locations of two widespread woodland generalist non-passerine species (collared dove and turtle dove) are also shown (triangles).

JPE_667.fm Page 1140 Thursday, September 20, 2001 2:52 PM

1140 E. Shochat, Z. Abramsky & B. Pinshow

Fig. 3. Passerine species – area relationships and Fisher’s alpha index for species diversity in forest and scrub patches in the Northern Negev study area. (a) In forest patches all three parameters of the Fisher’s alpha curve (i.e. area, area 2 and area3) were significant. (b) In scrub patches high second- and third-order polynomial regressions did not yield significant curves.

jay and Sardinian warbler) that were associated with large and old forests were also grouped, whereas desert finch was separated (Fig. 2).

    

© 2001 British Ecological Society, Journal of Applied Ecology, 38, 1135–1147

We first tested area effects on species richness and diversity. In the forest, a multiple regression model yielded a significant third-order polynomial regression for the species diversity pattern (Fig. 3a). The decrease in Fisher’s alpha between the two smaller patches and the decrease among the three to four largest forest patches were both significant, indicating that at these spatial scales species richness increased due to a random sample (Fig. 3a). In contrast, Fisher’s alpha increased significantly within the four to five mid-sized patches, indicating that, at this scale, species richness increased as a result of a biological effect (Fig. 3a). Scrubland species richness increased linearly with area (Fig. 3b). suggesting that species accumulation was due to biological factors over the whole range of scrub patch sizes. To test whether habitat structure may be the factor affecting the species diversity, we evaluated habitat diversity for forest and scrubland patches by calculat-

ing a coefficient of variance (CV) of vegetation structure corrected for sample size (Sokal & Rohlf 1981). In the forest we calculated CV for BA, whereas in the scrub CV was calculated for perennial cover density. We then plotted Fisher’s alpha against CV for each habitat and calculated the Spearman correlation. In the forest, bird species diversity increased significantly with habitat diversity as BA (Fig. 4a), with one outlier patch (Fig. 3a; Dixon’s test P < 0·05; Sokal & Rohlf 1981). This 150-ha patch comprised coniferous subunits only, planted mostly during the 1960s with 16 ha of conifers planted in 1988. Thus, while this patch was less diverse in tree structure than other large patches, its unique pattern gave a CV of BA that was much higher. In the scrub, perennial cover CV and Fisher’s alpha were not correlated (Fig. 4b), suggesting that habitat diversity is not the factor that affects species diversity. We further examined area vs. habitat structure effects by testing their influence on bird densities using . Area may affect population density due to the extinction of habitat-specialist species from small patches. For this analysis we used stand age in the forest and perennial cover in the scrub as the habitat structure variables.

JPE_667.fm Page 1141 Thursday, September 20, 2001 2:52 PM

1141 Breeding bird species diversity in the Negev

Fig. 4. The relationship between habitat diversity and bird species diversity. (a) In forest bird species, diversity increases according to patch habitat diversity (as measured by coefficient of variance of tree basal area). The correlation was calculated after omitting one outlier (square) from the analysis (for all the data r 2 = 0·648, P = 0·005). (b) In the scrub, no correlation was found between bird species diversity and habitat diversity (as measured by coefficient of variance of perennial cover).

© 2001 British Ecological Society, Journal of Applied Ecology, 38, 1135 –1147

We first tested whether there were differences in bird densities between years, and found that densities were significantly different in the forest (Wilk’s λ, F34,176 = 1·721, P = 0·013) but not in the scrub (Wilk’s λ, F14,142 = 0·801, P = 0·667). Therefore we analysed the effect of area and habitat structure on forest species abundance using the data set for each year separately. In the forest there was a significant correlation between total bird density and patch area in 1996, but not in 1997 or in 1998 (Table 3). However, of all forest species found in 1996, only one (great tit) had lower density in small patches than in large ones, while 11 other species showed no differences in densities. In

contrast, bird density was related to stand age in all 3 years of study (Table 3). Three species increased in density in old-growth plots (blackbird, great tit and spotted flycatcher), while two (rufous bush-robin and the graceful prinia) were significantly more abundant in young stands. There was no significant interaction between the effect of area and habitat structure on forest bird densities (Table 3). In the scrub, species density was lower in small patches due to the low density of three scrub specialists in small patches: desert lark, long-billed pipit and spectacled warbler. Habitat structure did not have a significant effect on total bird density (Table 3).

JPE_667.fm Page 1142 Thursday, September 20, 2001 2:52 PM

1142 E. Shochat, Z. Abramsky & B. Pinshow

Table 3.  of habitat structure and area effects on total bird densities. Forest bird densities varied significantly between years, and therefore are shown for each year separately. Habitat structure significantly affected bird densities in forest but not in scrub. In contrast, area affected bird densities in scrub, whereas in forest it had a significant effect on bird densities only in 1996. Interactions between habitat structure and area were not significant in both scrub and forest, and are not shown. Num, numerator; Den, denominator; d.f., degrees of freedom

Forest Age of forest 1996 1997 1998 Log patch area 1996 1997 1998 Scrubland All years Factor Log area % Perennial

Wilk’s λ

F-value

( Num)d.f.

(Den)d.f.

P-value

0·337 0·212 0·244

2·789 2·733 2·656

12 15 14

17 11 12

0·026 0·049 0·049

0·313 0·259 0·282

3·109 2·099 2·177

12 15 14

17 11 12

0·016 0·110 0·092

0·526 0·866

9·131 1·576

7 7

71 71

0·0001 0·157

Table 4. Stepwise multiple regression equations (multivariate habitat models) relating bird species densities in different groups to (log) environmental factors. Variables in each equation are listed from most to least important. PC1 (a measure of vegetation structure) is a primary factor in the forest. Area is the only factor that enters the equation of scrub specialist density

© 2001 British Ecological Society, Journal of Applied Ecology, 38, 1135–1147

Group

Equation

R2

P