B I O L O G I C A L C O N S E RVAT I O N
1 4 1 ( 2 0 0 8 ) 2 6 9 8 –2 7 0 7
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/biocon
Effects of age and intensity of urbanization on farmland bird communities Ondine Filippi-Codaccionia,*, Vincent Devictora, Jean Clobertb, Romain Julliarda a
Muse´um national d’Histoire naturelle, Conservation des Espe`ces, Restauration et Suivi des Populations, UMR 5173 MNHN-CNRS-UPMC, 55 rue Buffon, 75005 Paris, France b Station d’Ecologie Expe´rimentale du CNRS a` Moulis USR 2936, Moulis, 09200 Saint-Girons, France
A R T I C L E I N F O
A B S T R A C T
Article history:
Urban sprawl is now occurring worldwide and considered as a major large-scale perturba-
Received 6 February 2008
tion on ecosystems. Consequently, urban territory is replacing other habitats such as agri-
Received in revised form
cultural areas. As farmland biotic communities are already reported to be declining, it
11 July 2008
seems necessary to assess the urbanization impact on them. We conducted a bird survey
Accepted 7 August 2008
on 92 plots of 1 · 1 km chosen after stratification on the proportion of urban area and farm-
Available online 14 September 2008
land habitat (either 0%, 25%, 50%, 75%), focusing on farmland habitat. Two aspects of urbanization were studied: its intensity and its age. We found that farmland bird species
Keywords:
richness did not vary with increasing proportion of urbanized habitat. Non-farmland bird
Biotic homogenization
species richness increased from 0% to 25% classes and was constant for other classes.
Urbanization
No effect of the urbanization age on farmland bird species richness was found, whereas
Age
a positive one was found on the non-farmland birds’ species richness. Abundance of the
Spatial extent
most specialized farmland birds decreases with urbanization intensity and age. We also
Bird counts
found that, the more urbanized and the more recently urbanized the plots, the more similar bird communities. A strong difference in farmland bird’s communities’ compositions was found between 0% and 25% of urbanization, whereas no distinction was found between 50% and 75%. Altogether, our results suggest that to maintain for farmland birds, it is better to add new urban habitat in place where it already exist, rather than to spread it in small lots throughout the landscape. 2008 Elsevier Ltd. All rights reserved.
1.
Introduction
Among large-scale perturbations known to affect biotic communities’ fate, urbanization is considered the most severe one and is occurring worldwide (Vitousek et al., 1997; Pauchard et al., 2006). One of urban sprawl consequence is the replacement of any habitat by built-up features and the development of human-related infrastructures (e.g., roads, railways, gardens). Usually, two approaches are used to study urbanization effect on animal or vegetal communities. The first one investi-
gates changes of communities’ composition, species richness and diversity along a urban gradient (Blair and Launer, 1997; Ishitani et al., 2003; Clergeau et al., 2006) and the second one focuses on native habitats surrounded by different levels of urbanization (Collinge et al., 2003; Sadler et al., 2006; Scott, 2006; Veech, 2006). Studies using the first approach have generally concluded that (i) species abundance, functional diversity and evenness are declining with increasing urbanization, and (ii) that functional and taxonomic homogenization are increasing along this gradient due to a greater proportion of the same successful species inside cities. Studies using the
* Corresponding author: Tel.: +33 1 40 79 58 53; fax: +33 1 40 79 38 35. E-mail address:
[email protected] (O. Filippi-Codaccioni). 0006-3207/$ - see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.biocon.2008.08.006
B I O L O G I C A L C O N S E RVAT I O N
second approach have stressed the role of several factors to explain the consequences of urbanization, e.g., greater hydroperiod in urban wetlands for the loss of amphibian specialists species (Rubbo and Kiesecker, 2005), amplified direct effect of Argentine ant invasions in natural environments because of urban irrigation (Holway and Suarez, 2006), building and road density for fish assemblage composition and biotic homogenization (Scott, 2006). Despite the fact that some studies have stressed the importance of time factors on the present biotic communities’ composition, very few studies explicitly account for the temporal effect of urbanization (but see, Harding et al., 1998; Scott, 2006). Indeed, urbanization is generally studied only along a spatial dimension through the proportion of urban territory or the human population density (Collinge et al., 2003; Rubbo and Kiesecker, 2005). However, it seems interesting to investigate the importance of the urbanization age on present communities’ compositions at the same time than its intensity. Investigations on the urbanization age effects could also reveal dynamic processes of colonization and explain present communities’ compositions. Here, we focused on farmland bird communities. To our knowledge, no study was conducted on farmland birds’ communities although urbanization disproportionably affects farmland over other habitats (Mason, 2006). The increasing need of relieving housing shortage, mainly in the vicinity of big cities, is entailing a massive conversion of agricultural areas into urbanized habitats (Mason, 2006). The world population increasing, it seems obvious that this mechanism is occurring worldwide and is not ready to be stopped. Moreover, it is already a fact that urban human populations often inhabit richly cultivated suburban habitats with a relatively high local floral and faunal diversity and/or abundance without awareness of the global impoverishment caused by urbanization (McKinney, 2006). As farmland birds’ fate itself is already endangered according to the massive decline of bird populations over the past three decades (Donald et al., 2001, 2006), investigating other pressures effects such as urbanization on those communities could be of great interest for their conservation. We specifically address three objectives (i) does urbanization induce a change in diversity and composition of bird communities in farmland-dominated landscapes, (ii) which aspect of urbanization (its intensity and/or its age) influences
1 4 1 ( 2 0 0 8 ) 2 6 9 8 –2 7 0 7
2699
taxonomic and functional homogenization of bird communities, (iii) which urbanization thresholds (for intensity and age) were differentiating bird communities’ composition. Specialist vulnerability to disturbance often relies on the following assumption: as habitat generalist species use various types of habitat in the matrix, they should be less affected by habitat degradation than specialist ones which are more dependent on one or few habitat types (Krauss et al., 2003). Consequently, we predict that the most specialized farmland birds should be more negatively affected by the proportion of urban territory in the landscape. These theoretical predictions were also supported by empirical findings showing that specialists species were more negatively affected by land-use changes (Devictor et al., 2007; Schweiger et al., 2007). Concerning the age of present urbanized areas, we predict that generalists’ abundances will be favored in more recently urbanized areas whereas specialist species’ abundance will be lower. Indeed, generalist species are known to have better colonizing abilities and to better cope with disturbance (Devictor et al., 2008a). Therefore, we predict that the abundance of generalist species will be greater among the most urbanized study plots as well as in the more recently urbanized ones while the reverse is predicted for the specialist birds.
2.
Methods
2.1.
Study design
The study was carried out from April to the middle of June 2007 in the Seine-et-Marne region (France). We first divided the entire territory (5915 km2) in 1 · 1 km squares and then calculated for every squares, the proportion of three coarse habitats: urban, farmland and natural habitat using the geographical information package ArcView 3.2 (ESRI, 2000). Landscape features were obtained from the MOS (IAURIF, 2003) which is a detailed regional geo-referenced database including the main habitats for the region in continuous polygons classified according to 83 categories. In this land-cover each polygon was classified in the above cited coarse habitats. The ‘‘Urban’’ habitat was selected as every artificial area (e.g. built-up features, roads and railway). The ‘‘farmland’’ habitat gathered arable areas, tree nurseries and orchards whereas ‘‘natural’’ habitat included forested areas, marshland and other fallow lands. Ninety-two squares were then chosen
Table 1 – Description of the 92 squares of 1 · 1 km chosen according to their varying proportion of urban-farmland habitat cover (mean ± SD and range) Classes
nb squares
0–100
30
25–75
30
50–50
17
75–25
15
%Urban 0.00 ± 0.00 (0–0) 25.24 ± 8.84 (23.62–26.82) 50.04 ± 1.82 (0.46–0.53) 72.98 ± 3.67 (64.61–78.22)
%Farm land
%Natural
%Culture
100.00 ± 0.00 (100–100) 70.47 ± 2.83 (65.40–74.28) 41.56 ± 4.76 (0.31–0.48) 19.84 ± 4.98 (13.68–29.82)
0.00 ± 0.00 (0–0) 4.27 ± 2.67 (0–9.37) 8.38 ± 4.40 (0.20–16.76) 7.17 ± 4.52 (0.41–15.64)
Mixt (23%)/BC(77%) Mixt (25%)/BC(75%) Mixt(29%)/BC(71%) Mixt (22%)/BC(78%)
Mixt = mixed farmland (grassland and cultures); BC = ‘‘Big’’ culture characterising openfield agricultural type (cereals, oleaginous, proteaginous and other mechanized and big scales cultures).
2700
B I O L O G I C A L C O N S E RVAT I O N
to represent a urban-farmland gradient with increasing urban cover: 0% (n = 30), 25% (n = 30), 50% (n = 17) and 75% (n = 15). A various but negligible part of natural habitat remained in squares (see Table 1 for more details). The proportion of urban area was further related as URBAN. In order to investigate correlation between the surrounding landscape composition and the within square composition, we calculated the former within a 1 km radius buffer area around each square using Patch analyst extension on ArcView 3.2. (McGarigal and Marks,1995; Elkie et al., 1999). Habitat structure and composition of the surrounding landscape of squares were highly correlated to the within squares landscape (r2 = 60%, n = 92). Hence, we did not consider the surrounding landscape composition in our study.
2.2.
Urbanization age variable construction
The age of urbanization of the landscape was assessed using the EVOLUMOS data base (IAURIF, 2003). The EVOLUMOS is similar to the MOS but describe the land-use change of each polygon for each of the following year (1982, 1987, 1990, 1994, 1999 and 2003). We calculated the age of artificial area of each surveyed polygon. We then considered the average of each polygon age weighted by its surface as a measure of the age of one square. This calculation was not available for the 30 squares without artificial area. We used this continuous variable for regression models to investigate the age effect on species richness, and as well as on functional and taxonomical similarity.
2.3.
The species specialization index (SSI)
In order to measure the specialization of each species, we used the species specialization index (SSI) proposed by Julliard et al. (2006). It was assumed that a given species is more specialized to certain habitat classes if its density there is higher than elsewhere. Conversely, a species which density varies little across habitats can be considered as more generalist. This index quantifies the degree of habitat specialization for a species as the coefficient of variation (SD/mean) of its densities across 18 habitat classes grouped from categories recorded by observers at each point count of the FBBS (French Breeding Bird Survey) on the French territory (Julliard and Jiguet, 2002). It has been calculated for the 100 most frequent terrestrial bird species using all BBS plots surveyed at least once between 2001 and 2004 (n = 1022, i.e., 10,220 point counts). For more details see Julliard et al. (2006). This measure allows the ordering of species from specialists (occurring in few habitat classes), to generalists (occurring in many habitat classes). The highest values indicate the more specialists and the lowest ones indicate the more generalists. We used log-transformed values of SSI in our study as it produces more gradual variation of specialization among species. Previous analyses have shown that the specialization measure (SSI) was not biased by taxonomic autocorrelation (Devictor et al., 2008a) and that biases due to small sample size were negligible for FBBS data (Devictor et al., 2008b). We also classified species as farmland (20 species) and non-farmland (20 species). Farmland species were defined as species more abundant in farmland habitat than in non-
1 4 1 ( 2 0 0 8 ) 2 6 9 8 –2 7 0 7
farmland one according to the FBBS data (Julliard and Jiguet, 2002).
2.4.
Bird abundance estimation
We sampled birds using the point count method. Each 1 · 1 km square was sampled with 5 point counts. In each one, every birds seen or heard were recorded during 5 min exactly. Each point was visited twice (one before and one after the 8th May separated by 4 weeks) in order to cover the peak activity period of most breeding birds. Counts were done during the first 4 h after sunrise. For a given species on a given point, the final considered abundance was the maximum counted individuals for the two visits at each point (Julliard and Jiguet, 2002). We only considered individuals contacted at less than 100 m from the observer to reduce heterogeneity of species detectability, (more distant contacts being the most likely to induce detectability biases). The great majority of records were produced by singing birds (majority of passerine species), reducing heterogeneity detectability due to visual openness among points. We assume that for a given species counts were proportional to density. As our aim was to sample bird communities in farmland within more or less urbanized landscapes, the point counts were located within the ‘‘farmland’’ habitat of the 1 · 1 km squares defined previously. Points were located from 50 to 1000 m of built-up areas, but there was no difference in distances between point counts and built-up areas among urbanization classes (URBAN) (distance = 422.22 ± 309.48, F3,454 = 0.53, P = 0.66). The region is dominated by openfield type of farmlands, with a few meadows of small area at some points. The agricultural dominant type was described at each point count within a list of six types (Ploughed meadow, unploughed meadow, mixed farmland, openfield, permanent crop, and other types of cultures). Among sample points of the study, only two kinds of agricultural type were found (mixed farmland (25%) and openfield (75%)) and were in similar proportion among urbanization classes (F3,88 = 0.12, P = 0.94). The total number of detected species was 94. Aquatic species such as the Black-headed Gull Larus ridibundus, and the Grey Heron Ardea cinerea were also excluded. Species present in less than 10% of the squares were removed from analyses in order to avoid spurious effects of occasional species reducing the number of species to 40 (see Appendix). However, in order to see if the rarest species follow the same trend than the commonest ones, we built four groups within the rarest species: rare specialist farmland species, rare generalist farmland species, rare generalist non-farmland species and rare specialist non-farmland species. For each group, we calculated the average SSI and further split species as below and above this average. We then pooled abundances of species within groups in order to test whether their responses to urbanization age and intensity followed the prediction based on the commonest species.
2.5.
Statistical analyses
We investigated the linear or quadratic effect of urbanization in time and space on estimated species richness of every
B I O L O G I C A L C O N S E RVAT I O N
farmland and non-farmland species. Analyses were conducted on all non-aquatic species. Estimations of species richness for each single square, based on capture–recapture models, were obtained from the program COMDYN (Hines et al., 1999) using the five spatial replicates. This method accounts for heterogeneity in species detectability in using the jackknife estimator Burnham and Overton (1979); Boulinier et al. (1998). In order to investigate the effect of urbanization in space and time we performed the following models for each species: Model(space): Species abundance URBAN. Model(age): Species abundance URBAN + AGEA. Note that AGEA was not available for the 100% farmland squares (0% urban squares) since there was no artificial area in these squares. Consequently, we did not test the effects of URBAN adjusted to AGEA. There was no correlation between AGEA and URBAN (Spearman’s rank correlation test: P = 0.97). We used generalized least square (GLS) models to check for autocorrelation structure on residuals of the 2 models. Only 5 from the 40 species presented a better fit of the data when taking autocorrelation into account for the model (space) and only 1 from 40 for the model (age). Thus, we did not consider further spatial autocorrelation in this analysis. We used generalized linear models (GLM) assuming Poisson’s distribution of the number of individuals and using a log-link function. In a second step, we modelled the various species responses (the slope of the relationship between species abundance and age/intensity) taken from the preceding models according to their SSI using the following model: Model2: Species response SSI + T + T:SSI, weight = 1/SE2. Species response being the estimate from the first model, SE the associated standard error. In case of overdispersion (residual deviance > residual df), SE2 was corrected by the variance inflation factor (residual deviance/residual df) (e.g. Julliard et al., 2004). T being the species type (farmland or non-farmland). This two-step approach was used rather than treating all the species in the same model (e.g. GLMM) to properly account for overdispersion (e.g. social species vs territorial species). If treated in the same model, species would contribute proportionally to their deviance and thus, giving more weight to overdispersed species. We investigated the taxonomic similarities of farmland and non-farmland communities at two levels using the Jaccard’s index: among main habitat classes using distancebased Redundancy Analysis (dbRDA) (Legendre and Legendre, 1998) and within habitat classes using Within Principal Coordinate Analysis (within-PCA) (Bouroche, 1975). The Jaccard’s Index of similarity measures percent overlap in species composition between two sites (Magguran, 1998). In dbRDA, Principal Coordinate Analysis (PcoA) is used to extract the principal coordinates of a calculated matrix of ecological distances (Jaccard’s distances here). ANOVAs were performed to test variable (urbanization and age) significance on similarity data. Double principal coordinate analysis (dpcoa) and distance matrix computation were used to calculate distances between communities living at different urbanization intensity and age (Pavoine et al., 2004). We used the ‘‘dist’’ function (R statistical software) that computes and returns the distance
1 4 1 ( 2 0 0 8 ) 2 6 9 8 –2 7 0 7
2701
matrix based on the Jaccard’s index to compute the distances between the rows of a data matrix. The shorter the distances between communities, the more similar the communities. Within-PCA was used to test taxonomical homogeneity of bird communities (Bouroche, 1975). This multivariate analysis PP P is based on the equality: 1=n ðX i X moy Þ2 ¼ i j 1=2n2 ðX i X j Þ2 , which says that the average distances between the medium point (Xmoy) and the other points (i.e. variance) is equal to the average distance between all the points of the multivariate space. Within-PCA calculates these distances between centres and points within each class. It is then possible to compare those distances between classes with analyses of variance (ANOVAs) and post hoc tests. When the data were not normally distributed, Kruskal–Wallis test was used. If m1 is the mean distance between plots of the class 1 and m2 the mean distance between plots of the class 2, and if m1 < m2, then class 1 is more homogenous than class 2. All calculations were performed using the statistical software R 2.5.1 (R Development Core Team, 2005) with the packages ade4 and vegan (multivariate analyses).
3.
Results
3.1.
Species richness: spatial effect
No significant relationship was found between farmland birds’ species richness and the proportion of urban area (F1,90 = 1.67, P = 0.19) (Fig. 1a). No quadratic relationship was found (F1,89 = 0.46, P = 0.49) for farmland bird species richness but a quadratic effect was found to explain the relationship between non-farmland birds’ species richness and the proportion of urbanization (linear: F1,90 = 35.34, P < 0.001, r2 = 28%; quadratic: F2,89 = 48.02, P < 0.001, r2 = 52%) (Fig. 1b). TuckeyHSD test shows that there were differences of nonfarmland birds’ species richness between the class 0% of urbanization and the other classes (25%, 50% and 75%) (P < 0.001), whereas there were no differences of non-farmland birds’species richness among other classes 25%, 50% and 75% (25–50: P = 0.91, 50–75: P = 0.99, 25–75: P = 0.93).
3.2.
Species richness: age effect
Neither linear nor quadratic effect of urbanization age was found on farmland birds’ species richness (Linear: F1,60 = 1.47, P = 0.23; quadratic: F2,59 = 0.95, P = 0.39) (Fig. 1c) as well as on non-farmland birds’ species richness (Linear: F1,60 = 2.45, P = 0.12; quadratic: F2,59 = 2.82, P = 0.06) (Fig. 1d). However, non-farmland species richness was different between classes Y (Young) and M (Middle) (TuckeyHSD: P = 0.04), whereas no significant difference was found between classes M (Middle) and O (Old) (TuckeyHSD: P = 0.63) and between Y and O classes (TuckeyHSD: P = 0.16).
3.3. Specialist versus generalist species responses to urbanization The effect of proportion of urbanized area on abundance depended on species status (farmland or non-farmland), the specialization level of the species (SSI) and on the interaction between status and specialization (Table 2). On average,
2702
Estimated species richness
B I O L O G I C A L C O N S E RVAT I O N
35
35
a
30
1 4 1 ( 2 0 0 8 ) 2 6 9 8 –2 7 0 7
b
30
25
25
20
20
15
15
10
10
5
5
0
0 0
20
40
60
80
0
20
Estimated species richness
Proportion of urbanized area (%) 35
70
c
30
50
20
40
15
30
10
20
5
10
0
0 5
10
15
20
25
60
80
d
60
25
0
40
Proportion of urbanized area (%)
30
0
10
5
Urbanization age
15
20
25
30
Urbanization age
Fig. 1 – Farmland and non-farmland birds’ species richness (no transformed) variations along a gradient of increasing proportion of urbanized area and urbanization age; (a) farmland birds’ species richness along a gradient of increasing urbanized area proportion, (b) non-farmland birds’ species richness along a gradient of increasing urbanized area proportion, (c) farmland birds’ species richness along a gradient of increasing urbanization age, and (d) non-farmland birds’ species richness along a gradient of increasing urbanization age.
farmland birds were more negatively affected than non-farmland ones. We found linear negative relationships between the farmland bird species responses to the proportion of urbanized area (from 0% to 75%) and the SSI (F1,18 = 19.30, P < 0.001, r2 = 52%) (Fig. 2). No significant relationship between non-farmland birds’ species responses and the SSI was found (F1,18 = 0.89, P = 0.35) (Fig. 2). The effect of the age of urbanized areas on abundance depended on the species status (farmland or non-farmland) and the interaction between status and SSI (Table 2). A significant negative relationship was found between the farmland bird species responses to the age of the urbanized area and the SSI (F1,18 = 4.87, P = 0.04, r2 = 22%) (Fig. 2). No significant relationship was found between non-farmland birds’ species’ re-
sponses to the age of the urbanized area and the SSI (F1,18 = 1.23, P = 0.28, r2 = 6%) (Fig. 2).
3.4.
Taxonomic similarity
We found that the proportion of urbanized area had significant effect on similarities of farmland birds’ communities (F1,90 = 3.27, P < 0.001, explained inertia = 10%) (Table 3). A strong difference of community composition between class 0 and other classes was detected, whereas we found no significant difference between classes 50 and 75 (F1,30 = 0.75, P = 0.81) (Table 3 and Fig. 3). The urbanization age was also found to have a significant effect on farmland communities’ similarities (F1,60 = 1.74, P = 0.002, explained inertia = 5%)
Table 2 – Effect of the specialization level, species status and their interaction on the responses of species abundance to urbanization intensity and age Variables
df
F
P
Proportion of urban area SSI T SSI:T
1,36 1,36 1,36
18.686 4.297 10.627
