Landscape Ecology

, Volume 23, Issue 2, pp 149–158

The sensitivity of dragonflies to landscape structure differs between life-history groups

Authors

    • Department of Ecosystem Studies, Institute of Agriculture and Life ScienceThe University of Tokyo
  • Shin-ichi Suda
    • Department of Ecosystem Studies, Institute of Agriculture and Life ScienceThe University of Tokyo
  • Yoshitaka Tsubaki
    • Center for Ecological ResearchKyoto University
  • Izumi Washitani
    • Department of Ecosystem Studies, Institute of Agriculture and Life ScienceThe University of Tokyo
Research Article

DOI: 10.1007/s10980-007-9151-1

Cite this article as:
Kadoya, T., Suda, S., Tsubaki, Y. et al. Landscape Ecol (2008) 23: 149. doi:10.1007/s10980-007-9151-1

Abstract

Contrasting life-history strategies of long versus short pre-reproductive phases are known in adult dragonflies (Odonata) of temperate regions. Because the long-phase species spend a longer time in terrestrial habitats such as grasslands or woodlands during their pre-reproductive phase, we hypothesized that long-phase species would be more sensitive to landscape structure than short-phase species. To test this hypothesis, we conducted periodic censuses of adult dragonflies at small man-made ponds. We compared the two above functional groups in terms of the degree to which species occurrence depended on landscape structure. The difference among the two groups was not significant, but occurrence of long-phase species tended to depend on landscape structure. Long-phase species responded to landscape structure at larger spatial scales and showed stronger spatial autocorrelation in their occurrence among sampling ponds than short-phase species.

Keywords

FragmentationFunctional groupOdonataComplex life cycle

Introduction

Habitat loss and fragmentation are the most important causes of the increase in species extinctions in recent decades (Fahrig 2003). Nevertheless, empirical evidence suggests that not all species decline following habitat fragmentation and that some species are more robust than others against fragmented landscapes (Driscoll and Weir 2005; Purtauf et al. 2005). This may be partly because individual taxa respond differently to the spatial pattern or dynamics of their environment. However, if species sharing similar traits (i.e., functional groups) respond similarly to changes in their environment, one may reduce the diversity of plant and animal species to a set of operational entities for the purpose of predicting their response to environmental change such as habitat fragmentation.

Several population and/or ecological features have been identified to explain the sensitivity of species to habitat fragmentation (Henle et al. 2004; Ewers and Didham 2006). However, although the life cycles of many animals have discrete stages, such as larval, juvenile and adult stages, each depending on different habitats at different locations (complex life cycles sensu Wilbur 1980), only a few studies have examined how sensitive the different life stages are to habitat fragmentation (Andrén et al. 1997; Pope et al. 2000). In a recent paper, Halpern et al. (2005) predicted that the length of different life stages partially explains which habitat type likely limits the abundance of species. Accordingly, species with a long adult stage are supposed to be more limited by the availability of adult habitat as compared with species with a short adult stage. This suggests that the relative duration of adult stages may be a relevant predictor for the sensitivity to destruction and fragmentation of adult habitats in animals with complex life cycles.

Dragonflies (Odonata) are common invertebrates in various aquatic ecosystems. They exhibit complex life cycles, shifting from aquatic to terrestrial habitats during their adult stage, thus crossing ecosystem boundaries (Knight et al. 2005). Most of species spend a much longer time at larval stage (aquatic habitats) than at adult stage (terrestrial habitats) (Corbet 1999), and thus local aquatic environments can be primarily factors limiting pecies presence (Kadoya et al. 2004). Landscape structure, however, can also affect their occurrences because adult dragonflies have to seek suitable habitats over the landscape (Corbet 1999). Their adult life stage consists of three successive phases: the pre-reproductive, reproductive and putative post-reproductive phases. During the pre-reproductive phase, individuals are compelled to move between two habitat types: From aquatic habitats to terrestrial habitats and back again. Woodlands or grasslands including paddy fields provide suitable terrestrial habitats for foraging, resting, and avoiding unsuitable (dry or hot) conditions (Morton 1977; Watanabe and Taguchi 1988; Fincke 1992; Corbet 1999; Watanabe et al. 2005). Therefore it is likely that, among successive adult phases, the duration of the pre-reproductive phase is a major factor determining the sensitivity of dragonflies to the fragmentation of terrestrial habitats.

In dragonflies, the duration of the pre-reproductive phase varies from a few days to several months (Corbet 1999). Because a long pre-reproductive phase strongly depends on terrestrial habitats, we assumed that species that have a longer pre-reproductive phase are more sensitive to landscape structure than species with a shorter pre-reproductive phase. To verify this hypothesis, the relationship between the duration of the pre-reproductive phase and the species’ sensitivity to changes in the landscape structure has to be estimated. We first conducted periodic censuses of adult dragonflies at small, newly created man-made ponds. We then compared the occurrence of species of two functional groups with pre-reproductive phases of different durations (long-phase vs. short-phase) among ponds in relation to landscape structure.

Methods

Classification of life history of pond-breeding dragonflies

Contrasting life-histories of dragonflies in temperate regions are recognized based on the duration of the adult pre-reproductive phase. Some species have a long pre-reproductive phase of more than one month after which they return to aquatic habitats for reproduction, whereas other species have a short pre-reproductive phase of less than ten days (Corbet 1999). The long-phase species spend a much longer time in terrestrial habitats such as woodlands and grasslands, including paddy fields (Watanabe and Taguchi 1988; Watanabe et al. 2004). We classified all species observed at the study site into two life-history groups. In Japan, species belonging to the genera Sympetrum and Lestes are long-phase species (Taguchi 1987; Corbet 1999; Sugimura et al. 1999). Species such as Indolestes peregrinus that overwinter at the adult stage have an exceptionally long pre-reproductive phase (ca. nine months). In contrast, species of the genus Ischnura have a short pre-reproductive phase (from one to five days; Corbet 1999). Although the precise duration of their pre-reproductive phase has not been reported, spring ephemerals such as Anax nigrofasciatus nigrofasciatus (Sugimura et al. 1999) or multivoltine pioneer species such as Orthetrum albistylum speciosum and Pantala flavescens (Sugimura et al. 1999) have an apparently shorter pre-reproductive phase. Cercion sieboldii, Anax parthenope julius and Crocothemis servilia mariannae are also suggested to be multivoltine (Yoshida et al. 1998) and are thus likely to have a shorter pre-reproductive phase as well.

Study site

We surveyed 48 small shallow ponds that were newly created between 2000 and 2002 around Lake Kasumigaura (36° N, 140° E) in the Ibaraki Prefecture, eastern Japan (median distance between nearest ponds: 4,805 m; Fig. 1). These ponds were created to promote the restoration of habitats for dragonflies and endangered aquatic plants in wetland restoration activities (Washitani 2003). The surface areas of ponds varied from 13 to 144 m2. The ponds had a maximum depth of about 40 cm, gently sloped borders and were sparsely fringed by aquatic plants such as Nymphoides peltata, Sparganium erectum, Scirpus fluviatilis and Marsilea quadrifolia. The ponds were supplied with water as needed to prevent them from drying out.
https://static-content.springer.com/image/art%3A10.1007%2Fs10980-007-9151-1/MediaObjects/10980_2007_9151_Fig1_HTML.gif
Fig. 1

Study site in the Ibaraki Prefecture, eastern Japan. Solid circles represent sampling ponds, and dark gray patches represent woodlands

Species occurrence data

We conducted periodic censuses of adult dragonflies at each of the 48 ponds during the season in 2003 and 2005, except for three ponds that were under reconstruction in 2005 (these ponds were surveyed only in 2003). To cover all phenological groups (from spring species to late summer and autumn species), we censused each pond once at sunny or slightly overcast days during three time periods in 2003 (15 May to 3 June, 27 July to 7 August, 29 September to 9 October) and during four time periods in 2005 (17 to 23 July, 17 to 24 August, 26 September to 1 October, 20 October to 2 November). Thus, we obtained seven census data for each species for 45 ponds and three census data for three ponds. During each census, we walked the perimeter of the pond and recorded the occurrence of all species encountered. Dragonflies were identified to species without capturing them. We only captured dragonflies using insect nets when it was difficult to identify taxa by sight.

Explanatory variables

Sampling pond characteristics

In a previous study (Kadoya et al. 2004), we demonstrated that pond surface area and vegetation cover play an important role in determining the species composition of adult dragonflies. Therefore, we included these two pond characteristics as explanatory variables in the statistical models used. Vegetation cover (including any type of aquatic plant) was recorded using 10% increments during each census period. In addition, degree of openness also plays a role in determining the composition of adult dragonfly communities in lentic water habitats (Steyler and Samways 1995; Bernath et al. 2002). Therefore, we also included the open-sky-to-canopy ratio (hereafter “open-sky ratio”) in the statistical models. The open-sky ratio of each sampling pond was measured as follows. In the center of each pond, a 180° hemispherical fisheye photograph was taken above the pond vegetation using a super-wide-angle lens (UR-E6 and Fisheye Converter FC-E8; COOLPIX5000; Nikon, Tokyo). The open-sky ratio was calculated using the program CANOPON 2 (http://takenaka-akio.cool.ne.jp/etc/canopon2/) which calculates the ratio of the sky area to other area in the hemispherical solid angle (e.g., ponds surrounded by trees vs. ponds in open land). To reduce the effects of the year when a pond was created, we included the age of each pond in 2003 (hereafter “pond age”) as explanatory variable.

Landscape metrics

To model the occurrence of dragonfly species at sampling ponds, we selected the following three landscape variables: amount of (1) irrigation ponds and small reservoirs, (2) paddy fields and (3) woodlands including deciduous and evergreen forests. These are the dominant habitat types for larval or adult lentic dragonflies at the study site. The distribution of irrigation ponds was determined from a 1:25,000 digital topographic map (Geographical Survey Institute of Japan, Tsukuba). The distributions of woodlands and paddy fields were obtained from 1998 1:50,000 national land-cover data (Environment Agency of Japan, Tokyo). These variables were spatially mapped as raster grid (25 × 25 m) layers in ESRI ARCGIS 9.0 (Environmental Systems Research Institute, Redlands). Each cell in a layer was assigned a value of 1 or 0 according to the presence or absence of a particular variable.

For each variable and for each sampling pond, we calculated a set of distance-weighted metrics (Rhodes et al. 2006). These metrics are the weighted means of the values of the variables around each sampling pond with an exponential decline in weighting with distance from the sampling pond. For sampling pond i = 1, ... , M, the metric Xi was calculated as
$$ X_{i} = \frac{{{\sum\limits_{c = 1}^k {V_{c} \exp ( - \lambda d_{{ic}} )} }}} {{{\sum\limits_{c = 1}^k {\exp ( - \lambda d_{{ic}} )} }}} $$
(1)
where Vc is the value of the variable in cell c, dic is the distance between sampling pond i and the center of cell c and λ is the scale parameter of the negative-exponential function. The sum is over all cells in the landscape, c = 1,..., k. We considered all cells within the rage of 30 km from the nearest sampling pond.

The parameter λ controls how rapidly the influence (i.e., weighting) of a variable declines with distance. If λ is small, there is a slow decline in weighting with distance, and values of the variable both close to and far from each sampling pond determine the value of the metric. Dragonfly dispersal has been explained by a negative-exponential function (Conrad et al. 1999). However, the distances over which they travel vary from several dozens of meters to several kilometers according to species and landscape type (Watanabe and Taguchi 1988; Conrad et al. 1999; Corbet 1999; Knaus and Wildermuth 2002; Purse et al. 2003; Watanabe et al. 2004). Therefore, we employed an exploratory approach considering 30 different values of λ, corresponding to negative-exponential probability distributions with expected values equal to mean dispersal distances from 100 (λ = 10−2 m−1) to 3000 m (λ = 3 × 10−3 m−1) at 100 m intervals (i.e., 30 spatial scales for each metric).

To reduce correlations among the three metrics, we used principal component analysis (PCA) based on the correlation matrix to produce a single axis for each scale and used the first component of the PCA as a landscape metric at each scale.

Spatial data

The spatial data set consisted of nine spatial variables calculated from the geographic coordinates of each sampling point using the following procedure. The x- and y-coordinates (x: longitude, y: latitude) of the sampling points were centered on their means, and the following synthetic variables were calculated: x, y, xy, x2, y2, x3, y3, x2y, and xy2. These variables were used to express a response variable as a nonlinear function of the geographic coordinates (x, y) of the sampling sites (trend surface analysis: Borcard et al. 1992; Legendre 1993). Usually, the cubic trend surface is appropriate for ecological data because ecological phenomena often have single-peaked patterns (Borcard et al. 1992).

Statistical modeling

We modeled the probability of occurrence of each dragonfly species using mixed-effect logistic regression, with an intercept random effect between census times. We used R 2.4.0 (package “lme4”; http://www.r-project.org/) to fit these models to presence and absence data by maximum likelihood. Species for which no combination of variables produced a more parsimonious model than the null model were excluded from analyses.

To avoid the problem of multicollinearity and to reduce the number of possible combinations of explanatory variables to a manageable level, we chose only one of the 30 spatial scales calculated for the landscape metric, namely the one with the lowest Akaike information criterion (AIC) in univariate models. To examine our hypothesis that sensitivity to landscape structure would be stronger in long-phase species than in short-phase species, we constructed a set of alternative models from all linear combinations of explanatory variables (i.e., pond surface area, vegetation cover, open-sky ratio, pond age and landscape metric at selected scale) and fitted each model to the presence/absence data of each species belonging either to long- or short-phase types. We then ranked these models by their AIC values and computed Akaike weights (Burnham and Anderson 2002) to weigh models.

All parameter estimates of each model were standardized by the standardized error, and the model-averaged parameter estimates were determined by Akaike weights (Johnson and Omland 2004). The relative influence of individual variables was assessed by summing the weights of those models in which each variable appeared. Comparing these variable weights helps to avoid the risk of discarding variables that help explain dragonfly distribution but do not appear in the best model (Johnson and Omland 2004).

To assess the discrimination ability of each logistic regression model, the area under the curve (AUC) was calculated. This parameter represents the predictive performance of a model evaluated using the receiver operating characteristics technique (Fielding and Bell 1997; Manel et al. 2001). The AUC was calculated for each model, and the weighted sum of AUCs by Akaike weights was then calculated for each species using R 2.4.0 (package “ROC”; http://www.r-project.org/).

We quantified spatial autocorrelation in random effects estimated for each sampling pond by trend surface analyses in which the spatial variables were used as explanatory variables. The random effects represent the deviations caused by unmeasured factors in occurrence probability of a species at each sampling pond. We constructed a set of alternative models from all linear combinations of nine spatial variables (i.e., x, y, xy, x2, y2, x3, y3, x2y and xy2) and fitted each model to the random effects using multiple regression. We then ranked these models by their AIC values and recorded the amount of variance explained (adjusted R2) by the model yielding the lowest AIC value. We conducted a likelihood-ratio test between the selected model and the null model. When a null model was selected as the most parsimonious one, the amount of explained variance in random effects for the species was regarded as zero.

Comparison between life-history groups

To compare the degree to which species occurrence depended on landscape structure between long- and short-phase species, we compared the absolute values of the model-averaged parameter estimates of landscape metrics between life-history groups. In addition, we compared the spatial scale of landscape metrics selected for each species between life-history groups. To compare the spatial dependence of species occurrence between life-history groups, we also compared the amount of variance in random effects explained by spatial variables between the two groups.

For each comparison above, we constructed a regression model with life-history group as an explanatory variable and conducted likelihood-ratio test with the null model using R 2.4.0 (http://www.r-project.org/).

Results

Observed species

We recorded a total of 19 species (6 zygopterans and 13 anisopterans) at the 48 ponds studied (Table 1). We excluded tenerals (individuals with a nearly colorless, unhardened integument with weak, fluttering flight abilities; Corbet 1999), because they were recorded immediately after emergence and had not yet interacted with habitats beyond ponds.
Table 1

Observed dragonfly species, suborder category (aniso: Anisoptera; zygo: Zygoptera), life-history group based on the duration of pre-reproductive phase (short: short-phase; long: long-phase), voltinism (single: univoltine; multi; multivoltine; s/m: multivoltine depending on environmental conditions) and the number of ponds at which each species was recorded in 2003 and 2005 and in total

Species

Sub-order

Life history group

Voltinism

No. of ponds

2003

2005

Total

Sympetrum frequens

aniso

long

single

41

41

45

Sympetrum infuscatum

aniso

long

single

40

41

47

Sympetrum kunckeli

aniso

long

single

23

16

28

Sympetrum darwinianum

aniso

long

single

11

23

28

Indolestes peregrinus

zygo

long

single

12

0

12

Lestes temporalis

zygo

long

single

5

6

9

Sympetrum eroticum eroticum

aniso

long

single

3

0

3

Lestes sponsa

zygo

long

single

1

2

3

Ischnura asiatica

zygo

short

multi

43

28

44

Orthetrum albistylum speciosum

aniso

short

multi

33

37

42

Crocothemis servilia mariannae

aniso

short

s/m

22

25

30

Ischnura senegalensis

zygo

short

multi

17

4

17

Anax nigrofasciatus nigrofasciatus

aniso

short

single

12

1

12

Pantala flavescens

aniso

short

multi

3

7

8

Anax parthenope julius

aniso

short

s/m

2

7

8

Cercion sieboldii

zygo

short

s/m

5

2

7

Orthetrum triangulare melania

aniso

short

s/m

1

3

4

Pseudothemis zonata

aniso

short

single

1

0

1

Rhyothemis fuliginosa

aniso

short

single

0

2

2

Spatial scale of landscape metrics

The first component from the three landscape metrics (woodland, pond and paddy fields) produced by the PCA (hereafter “PC1”) accounted for 39.5% to 53.4% of variation according to spatial scale. At all spatial scale, PC1 was positively related to the amount of paddy fields (r = 0.606 to 0.894), while it was negatively related to the amount of woodlands (r = −0.769 to −0.621) and ponds (r = −0.705 to −0.441). The selected spatial scale of the PC1 based on the AIC of univariate models per species (Table 2) varied from 100 to 3,000 m.
Table 2

Results of mixed-effect logistic regressions using pond characteristics and landscape metrics (PC1) as fixed factors and amount of variance in random effects explained by spatial variables

Species

Life history group

AUC

Scale (m)

 

Fixed effect

Spatial autocorrelation in random effect

Area

Veg

Open-sky

Age

PC1

Explained variance (%)

P value

Spatial variable

S. frequens

long

0.56

2200

Estimate

0.21

−0.16

1.18

0.11

−0.75

4.5

0.122

x, x3

Weight

0.33

0.31

0.67

0.30

0.53

S. infuscatum

long

0.62

3000

Estimate

3.26

−0.15

−0.21

0.16

0.75

13.6

0.022

y, x2, x3, y3

Weight

0.99

0.30

0.31

0.30

0.51

S. kunckeli

long

0.78

100

Estimate

3.51

0.88

−0.91

0.86

2.78

5.3

0.058

x3

Weight

0.99

0.56

0.56

0.55

0.97

S. darwinianum

long

0.79

200

Estimate

0.82

0.06

−0.38

−0.01

0.53

27.0

<0.001

y,x2, y2, y3

Weight

0.56

0.30

0.40

0.30

0.46

I. peregrinus

long

0.74

1500

Estimate

0.10

−2.17

1.50

−0.12

−0.24

45.7

<0.001

y2, xy, x2y, xy2, x3, y3

Weight

0.29

0.92

0.76

0.29

0.33

L. temporalis

long

0.94

3,000

Estimate

0.28

−0.20

−1.20

1.32

0.84

8.7

0.039

x, x2y

Weight

0.36

0.32

0.74

0.84

0.62

I. asiatica

short

0.65

100

Estimate

3.26

0.25

2.61

0.43

0.02

4.2

0.081

y2

Weight

0.99

0.33

0.95

0.40

0.27

O. albistylum speciosum

short

0.65

200

Estimate

2.09

0.04

3.44

−0.20

0.10

8.1

0.069

x, xy, x3

Weight

0.87

0.27

0.99

0.31

0.28

C. servilia

short

0.76

100

Estimate

3.19

0.34

2.78

0.03

1.29

0.0

NA

Const.

Weight

0.98

0.36

0.96

0.28

0.69

I. senegalensis

short

0.81

1400

Estimate

1.00

0.67

2.20

0.75

0.34

0.0

NA

Const.

Weight

0.59

0.50

0.91

0.54

0.37

A. nigrofasciatus nigrofasciatus

short

0.71

300

Estimate

2.55

−0.02

0.05

0.38

0.55

0.0

NA

Const.

Weight

0.93

0.28

0.28

0.39

0.45

A. parthenope julius

short

0.74

600

Estimate

1.96

0.14

0.14

−1.25

−0.62

4.0

0.084

x2

Weight

0.83

0.30

0.30

0.66

0.46

C. sieboldii

short

0.79

1500

Estimate

3.01

0.03

−0.02

−0.78

0.15

6.7

0.094

y, x2y, x3

Weight

0.98

0.28

0.27

0.52

0.30

AUC: area under curve; Scale: selected spatial scale of landscape metric; Estimate: estimate for each fixed factor determined by model averaging; Weight: sum of Akaike weight; Area: pond surface area; Veg: vegetation cover; Open-sky: open-sky ratio; Age: pond age; PC1: landscape metric; x: longitude; y: latitude

Model selection

We analyzed the occurrence patterns of 13 species consisting of six long- and seven short-phase species (Table 2). Pond surface area was positively related with the occurrence of all species whereas the effects of vegetation cover differed among species. For example, Indolestes peregrinus was negatively whereas Sympetrum kunckeli and Ischnura senegalensis were positively related to vegetation cover. Open-sky ratio was positively related to all species categorized as short-phase types except Cercion sieboldii. On the other hand, more than half of the long-phase species (four out of six) responded negatively to open-sky ratio. Pond age had no strong effects (sum of Akaike weight <0.9) on any species, and the directions of effects were different among species irrespective of life-cycle type (Table 2). PC1 was positively related to most of the species, but was negatively related to S. frequens, I peregrinus and Anax parthenope julius (Table 2). In the trend surface analysis, the null model was selected as the most parsimonious one for three species belonging to the short-phase type: Crocothemis servilia mariannae, A. nigrofasciatus nigrofasciatus and I. senegalensis (Table 2).

Comparison between life-history groups

As shown in Fig. 2a, the absolute values of the standardized estimate for PC1 were on average larger in long-phase than in short-phase species, although the difference was not statistically significant (likelihood-ratio test; P = 0.158). In addition, selected spatial scales of PC1 were marginally larger in long-phase than in short-phase species (likelihood-ratio test; P = 0.052; Fig. 2b). Significantly larger amounts of variance in random effects were explained by spatial variables in long-phase as compared with short-phase species (likelihood ratio test; P = 0.020; Fig. 2c).
https://static-content.springer.com/image/art%3A10.1007%2Fs10980-007-9151-1/MediaObjects/10980_2007_9151_Fig2_HTML.gif
Fig. 2

Comparison of (a) the absolute values of standardized estimates of landscape metrics determined by model averaging, (b) the selected spatial scales of landscape metrics and (c) the amount of variance in random effects explained by spatial variables between long- and short-phase species

Discussion

Sensitivity to landscape structure

We hypothesized that sensitivity to landscape structure could be partially predicted from life-cycle types categorized by the duration of the pre-reproductive phase. Although long-phase species tended to respond more sensitively to landscape structure than short-phase species, the dependency of species occurrence on landscape structure (i.e., absolute value of estimates for PC1) did not significantly differ between life-history groups. This may partly be because other factors affected on the magnitude of the sensitivity to landscape structure of dragonfly species. In fact, the absolute values of PC1 estimates seemed to be different between suborders (i.e., Zygoptera and Anisoptera). The occurrence of anisopterans tended to depend more strongly on landscape structure than zygopterans independent of life cycle stage although the difference was not statistically significant (likelihood-ratio test; P = 0.12). This tendency between suborders may be ascribed to differences in dispersal abilities, habitat requirements and/or use of habitats. However, to clarify the effects of life-cycle and taxonimal groups on sensitivity to landscape structure, further studies replicated at the landscape level are needed. In addition, the long-phase species were mainly comprised of Sympetrum, whereas the short-phase group had a wider selection of genera. This can potentially confound the effect of life-cycle with those of other characteristics shared by the first phylogenetic group and may limit the detection of effects of life-cycle types independent of phylogeny.

Response to environmental variables

Previous studies have indicated that within the same landscape, responses in abundance and/or occurrence to landscape structure differ among species in direction and spatial scale (Krawchuk and Taylor 2003; Van Buskirk 2005). In our study, dragonfly species also responded differently to local pond characteristics and landscape structure, which reflects ecological and behavioral differences among species.

The duration of the pre-reproductive phase was related to the spatial scale of the landscape metric (PC1) to which the species responded. The scales of landscape metrics were larger in long-phase species than in short-phase species. This was probably because long-phase species do longer interact with habitats at a broader spatial scale than short-phase species.

All species in long-phase species responded negatively to either open-sky ratio or PC1 axis, whereas five out of seven species in the short-phase group responded positively to both variables. This indicates that species with a longer pre-reproductive phase tended to occur at ponds surrounded by trees or ponds adjacent to wooded habitats and that short-phase species tended to occur at ponds within open landscapes.

The correlation between species occurrence and pond characteristics was in accordance with their known ecological and behavioral characteristics (Kadoya et al. 2004). The fact that pond surface area was positively related to occupancy in all species may suggest that the requirements of the larval stage are paramount in governing species’ responses to landscapes. Another possibility for this size dependency is that detection rate of newly created ponds for adult dragonfly species depended on pond size.

Since the pond system under study is relatively newly created, there were possibilities that the dragonfly communities had not yet reached equilibrium and thus the occurrence of particular species may depend on pond age. However, in the present study, pond age (a three year span) had no strong effects on any species. The results seem reasonable because all of dragonfly species have terrestrial stage in their life-cycle, and thus ponds can be re-colonized by the species every year irrespective of their age.

Interpretation of spatial autocorrelation

Spatial autocorrelation of species distribution data could be caused by spatially structured environmental factors such as landscape elements, microclimate, biological processes (such as dispersal) and biological interactions (Legendre 1993). The interpretation of spatial autocorrelation patterns may provide some information on factors or processes determining species distributions (Legendre 1993). In our study, the random effects of the occurrence of long-phase species among ponds had a larger spatial component of variance than in short-phase species. Although the composite effects of landscape structure, movements and other unmeasured variables on the observed autocorrelation pattern cannot be distinguished, we speculate that unmeasured, spatially structured landscape effects were dominant, because the trend of species occurrence across sampling ponds modeled by spatial variables per species seemed to be related to the distribution of woodlands, which long-phase species likely prefer as adult habitats at a larger scale. It is possible that ponds closer to one another had a similar background with respect to the amount of woodlands surrounding them. Hence, the abundance of long-phase species would be similar among closer ponds. To test the hypothesis, we would need replicates at the landscape scale in which the amount of woodlands vary.

Acknowledgments

We thank P.D. Taylor for critically reading the original manuscript and providing helpful comments and suggestions and Y. Yamaura for advice regarding statistics. We also thank G. Fujita, J. Nishihiro and three anonymous referees for helpful comments on earlier versions of the manuscript, A. Goto for assistance in the field and H. Iijima (NPO Asaza Fund) for encouragement during the course of this study. We especially thank the primary-school teachers and students in Ibaraki Prefecture for allowing us to study dragonflies in their ponds. This study was partly supported by a grant from the Japan Society for the Promotion of Science (JSPS 17-11543).

Copyright information

© Springer Science+Business Media B.V. 2007