Mapping the potential extinction debt of butterflies in a modern city: implications for conservation priorities in urban landscapes

Authors


Correspondence

Masashi Soga, Division of Environmental Resources, Graduate School of Agriculture, Hokkaido University, Nishi 9, Kita 9, Kita-ku, Sapporo 080-8589, Japan. Tel: +81 42 367 5630; Fax: +81 42 364 7812

Email: soga06154053@yahoo.co.jp

Abstract

Cities are expanding rapidly worldwide. Modern cities are expected to carry heavy extinction debts owing to their recent and drastic fragmentation histories. Therefore, detecting extinction debt and identifying species threatened by it in recently created cities are necessary to prevent future biodiversity losses. Here, we studied the relationship between the life-history traits of butterfly species and the extent of their extinction debts using two different methodological approaches in Tokyo, central Japan. First, we compared the effects of current and past landscape parameters on current species richness using generalized linear models. Second, we predicted species richness in unstable (i.e. high loss) habitats using a model developed for stable (i.e. low loss) habitats. The difference between predicted and observed species richness was used to estimate the extinction debt (the number of species expected to go extinct). We classified butterfly species as seasonal specialists or generalists and as habitat specialists or generalists based on their life-history traits. With both methods, we found significant extinction debts only for specialist species. Mapping the potential extinction debts within our study area indicated that currently large patches had relatively low extinction debts, whereas small patches often had high extinction debts. These results suggested that improving patch area, connectivity and especially quality, would have more significant impacts in small patches than in large ones. Extinction debt is an important concept for setting conservation priorities in highly fragmented landscapes, especially in urban areas.

Introduction

Human-driven fragmentation and loss of natural and semi-natural habitats have rapidly accelerated the extinction rates of many species worldwide in what is sometimes called the sixth major extinction (Wake & Vredenburg, 2008). In general, species extinctions following landscape change do not occur immediately, but rather occur over the ‘relaxation time’ (Diamond, 1972). Therefore, even if conditions for long-term persistence of species are not met, many species occupy fragmented landscapes as ‘living dead’, defined as species predicted to go locally extinct when the community reaches an equilibrium level of species richness based on the current landscape structure (Janzen, 1986). Populations of such species may be deterministically extinct, even without any further habitat loss or fragmentation (described as an extinction debt, Tilman et al., 1994). In this case, current species distributions are not necessarily at equilibrium with current landscape structures, but depend instead on past landscapes. Consequently, patch distributions of current species may overestimate the long-term species richness in current landscapes and underestimate the real threats to extinction (Hanski & Ovaskainen, 2002). Therefore, to prevent future biodiversity losses, conservation managers and agencies must take extinction debt (i.e. the number of living dead) into their planning. Rescuing the living dead is a challenging task for minimizing future biodiversity losses (Kuussaari et al., 2009).

Predicting as early as possible which species are living dead may allow us to facilitate their recoveries and prevent their ultimate extinctions (Hanski & Ovaskainen, 2002; Schrott, With & King, 2005). Because species respond to habitat fragmentation differently, depending on their specific life-history traits (e.g. Öckinger et al., 2010; Williams et al., 2010), the living dead may also be predicted by their life-history traits. Despite the sharp increase in attention to extinction debt in recent years (Helm, Hanski & Pärtel, 2006; Krauss et al., 2010; Sang et al., 2010; Cousins & Vanhoenacker, 2011), studies that test whether species with different life-history traits possess different time lags to extinction are still scarce. In general, patch specialists and species with low turnover rates are considered to have longer relaxation times (e.g. Kuussaari et al., 2009; Krauss et al., 2010), indicating they are more likely to remain inside focal patches as living dead. However, there is a general lack of empirical studies and substantial bias in the taxonomic scope of the studies; approximately half of the studies reviewed in Kuussaari et al. (2009) focused on vascular plants (e.g. Adriaens, Honnay & Hermy, 2006; Lindborg, 2007) and woods-inhabiting cryptogams. To comprehensively improve our knowledge about extinction debt, more empirical studies using various taxonomic groups with different degrees of mobility and turnover rates are needed.

Butterflies are among the most well-studied and species-rich taxonomic groups, and therefore, the relationship between life-history traits and sensitivity to habitat loss and fragmentation is relatively well known (e.g. Koh, Sodhi & Brook, 2004; Dover & Settele, 2009; Öckinger et al., 2010). Unlike many plants, butterflies have relatively rapid life cycles and moderate dispersal abilities; thus, they typically have been considered to quickly respond to landscape changes (Kuussaari et al., 2009; Krauss et al., 2010). For butterflies, patch-dependent species (hereafter called patch specialists) and species with long generation times (hereafter called seasonal specialists) are considered to remain in focal patches as living dead after fragmentation for the following reasons: (1) patch specialists typically have low permeability to the matrix because the habitat boundary is considered to be a ‘hard edge’ for them (e.g. Stamps, Buechner & Krishnan, 1987; Ries & Debinski, 2001); (2) the response of seasonal specialists to habitat fragmentation is typically slow (e.g. Kuussaari et al., 2009). From a conservation viewpoint, prioritizing only species is not sufficient for effective restoration, conservation and management; conservation managers also need to determine accurately how much unpaid extinction debt remains in reserve areas to identify high-priority conservation areas in need of urgent action. Although modeling and mapping of species distributions and diversity are widely acknowledged to be extremely useful tools for setting conservation priorities (Conroy & Noon, 1996; Rodríguez-Soto et al., 2011), mapping potential extinction debts has only been done in one national-scale study of African forest primates (Cowlishaw, 1999). Because modern cities are expected to carry a large number of extinction debts owing to their recent and drastic fragmentation histories (Hahs et al., 2009), mapping extinction debts in urban areas is crucial for rescuing many living dead species and preventing future biodiversity losses.

Tokyo, central Japan, is a so-called hypercity or megacity of more than 20 million inhabitants (Gaston, 2010). In western Tokyo, the landscape structure over the past 40 years has been highly dynamic, with reductions in deciduous forest cover to one-quarter of its 1971 area. While the overall forest area declined, the number of patches increased rapidly; today, more than 90% of patches are < 20 ha in size (Fig. 1). Such a recent fragmentation history, with dramatic declines in patch area and connectivity, provides an excellent model system to test for the presence of an extinction debt. Here, we tested for the presence of an extinction debt of butterflies in Tokyo using two widely accepted approaches (see also Kuussaari et al., 2009): (1) comparing the effect of current and past landscape parameters on current species richness (e.g. Krauss et al., 2010); (2) comparing species richness in unstable (i.e. highly impacted by fragmentation) habitats versus relatively stable habitats (e.g. Helm et al., 2006; Vellend et al., 2006). The goals of our study were: (1) to clarify whether an extinction debt was found only for patch- and seasonal-specialist butterfly species; (2) to map the potential extinction debt of butterflies in our study area. Finally, we discussed what conservation actions in urban forest remnants are needed and how to prioritize them to conserve living dead species and minimize their future losses in urban areas.

Figure 1.

Comparison of the forest habitat patches in Tokyo, Japan, in 1971 and 2011. (a) Map of forest cover in 2011. The 35 sampled patches are shown in black. (b) Percentages of forest patches in each of six size classes in 1971 and 2011. Patch sizes were estimated from aerial photographs. Total number of patches is shown above each column.

Methods

Study area and history of study region

The study region (approximately 150 km2) was located in the Tama area, southwestern Tokyo, central Japan (Fig. 1). The western boundary of the study region was formed by a mountain range dominated by Mount Takao (599 m a.s.l.). Deciduous forests were mostly created and maintained as part of the traditional agricultural system and were once widespread in the study area. However, since 1970, many of these deciduous forests have been entirely lost to urbanization. One of the most drastic urbanizations in Japan, the Tama New Town Development (Parthenon Tama, 1998), was designed as a new town in 1965 during the Japanese post-war economic miracle. The landscape structure of this area over the past 40 years has been highly dynamic, with forest cover declining to one-quarter of its original area. In 1971, 20% of the forests were >100 ha and 30% were <20 ha (Fig. 1). By 2011, smaller patches composed 90% of forests and the number of forest patches increased more than twofold (Fig. 1).

The forest patches were dominated by deciduous forests of two oaks, Quercus serrata Thunb. ex Murray and Q. acutissima Carruthers, with Pinus densiflora Sieb. et Zucc., Abies firma Sieb. et Zucc, and evergreen oaks, Q. glauca Thunb. ex Murray and Q. acuta Thunb. ex Murray also present. The border of a forest patch was defined as any treeless belt with an open canopy of more than 10 m, following Kurosawa & Askins (2003). A total of 35 patches that ranged in area from 1.1 to 121.6 ha were selected to be studied in the field. To avoid pseudoreplication, forest patches were selected based on whether they were at least 600 m apart. The matrix consisted mostly of housing areas, which are considered to be unsuitable butterfly habitats. However, several plant species that are known host plants of some butterfly species are cultivated intensively in gardens, along roadsides and in semi-urban agricultural fields.

Current and past landscape structures

The year 1971 roughly corresponds to the beginning of rapid urbanization in the area (Parthenon Tama, 1998). As previously mentioned, the vegetation structure of each 2011 patch was similar, so patch area and connectivity were the key determinants of butterfly diversity Soga and Koike, (2012a). Therefore, in this study, current and past patch areas and current and past patch connectivities were used as landscape variables. For each time period (1971 and 2011), current and past areas of forest patches were calculated from detailed aerial photographs for those years (Geospatial Information Authority of Japan) with the ArcView geographic information system software (ver. 3.2, ESRI, Redlands, CA, USA). Past and current patch connectivities were described by the total forest area within a 1-km radius around a focal patch. Such a simple proportional index has been used in many previous studies (e.g. Brückmann, Krauss & Steffan-Dewenter, 2010). If patches are oddly shaped and relatively close together, such an index is useful (Winfree et al., 2005). Following previous studies in the same area (Soga and Koike, 2012a), we considered the buffer radius to be 1 km, because this distance was appropriate to the dispersal of several of the butterfly species (Honda & Kato, 2005).

Butterfly counts and species classification

Butterfly communities were monitored using the line transect method (Pollard, 1977). One 500-m transect was established in each patch, and counts were conducted once every 3 weeks between 09:30 am and 2:30 pm during the adult flight season (early April to early October 2011) if the weather was suitable. Butterflies within a 10-m radius of a position along each transect were recorded while walking at a steady pace (10 m min−1). Individuals that could not be identified to species by sight were caught using 8-m long insect nets, identified and released. We spent 50 min in each 500-m transect, and in total, we spent 300 min (six surveys) in each patch. To prevent us overlooking several species that establish territories (Artopoetes pryeri, Japonica lutea, J. saepestriata, Antigius attilia and Favonius orientalis), we also conducted beating surveys of trees using long insect nets. Because Pieris melete and P. rapae cannot be distinguished in the field, these two species were treated as Pieris spp. in all the analyses.

In this study, we classified all 52 butterfly species that were observed in the field surveys as either seasonal specialists or generalists, and as either patch specialists (i.e. species that depend on only forest patches) or generalists (i.e. species that could dwell also in the matrix). To classify butterflies as seasonal specialists or generalists, we used their generation times, following Soga & Koike (2012b). We defined species with one generation per year as seasonal specialists and species with two or more generations per year as seasonal generalists. Deciduous forests are the original habitat of this area, and species with woody larval host plants typically depend on forest patches (Soga & Koike, 2012b). In addition, species whose larval host plants do not exist in the matrix (e.g. gardens, roadside trees or agricultural areas) also depend on forest patches (Soga & Koike, 2012b). Therefore, to classify butterfly species as patch specialists or generalists, we used the following two classifications: (1) woody plant feeders versus herbaceous plant feeders, depending on their larval host plants; (2) patch-dependent species versus matrix-dwelling species, depending on whether their larval host plants existed in the matrix, following Soga & Koike (2012b). These classifications were based on data from Fukuda et al. (1982, 1983, 1984a, b ) and Soga & Koike (2012a, b ).

Statistical analyses

Effects of current and past landscape structures on current species richness

To test for an extinction debt, we examined the effect of past landscape structure on current butterfly species richness. Past landscape structures having stronger effects on current species richness than current ones is interpreted as evidence of an extinction debt (Kuussaari et al., 2009). Cases in which both current and past landscape structures have strong effects on current species richness are interpreted as evidence of partly paid extinction debts (Kuussaari et al., 2009). We used generalized linear models (GLMs), each with a Poisson distribution and a log link function, using current species richness as the response variable and current and past patch areas and current patch connectivity as explanatory variables. To select the best models from among the candidate models, we used the small-sample corrected version of Akaike's information criterion (AICc; Burnham & Anderson, 2002). The AICc for each model quantifies its parsimony (based on the trade-off between the model fit and the number of parameters included) relative to the other models. We used the ‘dredge’ function from the ‘MuMIn’ package (ver. 1.0.0) (Barton, 2009) to test models defined by all possible variable combinations and rank them by their AICc measures. In this analysis, several models were plausible based on the AICc (ΔAICc of several models were less than two; Supporting Information Appendix S1). To address this problem, we based our inferences on the entire set of models by computing a weighted average for each estimated parameter based on the Akaike weights. This approach is termed multimodel inference or model averaging; it provides unconditional model variances and more reliable parameter estimates for each predictor. To determine the reliability of the predictor estimates from averaging, we calculated the weighted unconditional standard error with its associated 95% confidence intervals (CI). To interpret the significance of our estimates for each parameter, we examined whether the 95% CI overlapped zero. Also, to assess the relative importance of each predictor, we calculated the relative variance importance (RVI). The RVI indicated the relative importance of each predictor, on a scale of 0–1.0. We determined which landscape parameters had the strongest effects on butterfly species richness using 95% CI and RVI (see also Table 2). GLMs and AICc values were calculated using the R software package (ver. 2.12.0, R Development Core Team).

Comparing species richness of stable and unstable patches

To estimate the extent of extinction debt (i.e. the number of species predicted to go locally extinct), we compared the butterfly species richness of ‘stable’ (less than 80% loss in area since 1971) and ‘unstable’ patches (more than 80% loss in area since 1971). The 80% loss limit was chosen to make our analysis comparable to other studies (Helm et al., 2006; Piqueray et al., 2011). We classified our 35 patches into 14 stable (median patch area = 9.2 ha) and 21 unstable patches (median patch area = 7.9 ha). Over the past 40 years, the average area losses of stable and unstable patches were 44 and 94%, respectively. We assumed that the levels of species richness remaining in the stable patches were at their long-term equilibria, because the patches themselves had lost relatively little area (Kuussaari et al., 2009). This assumption was supported by a significant species–area relationship between current species richness and current patch area, with the exception of herbaceous plant feeders (Table 1). Therefore, we first conducted simple regression models (species–area relationships in Table 1) using patch area to predict equilibrium species richness in stable patches. Second, the resulting parameter estimates in the regression models based on stable patches were used to predict species richness in each unstable patch. Then, the difference between the number of observed species and the predicted equilibrium species number (i.e. observed number of species – predicted number of species) in unstable patches was defined as the extent of extinction debt (Kuussaari et al., 2009). The statistical significance of the extinction debt was determined by Wilcoxon signed-rank test. Finally, to map the potential extinction debt of butterflies, we estimated the extinction debts in all 182 patches. In this case, the potential extinction debt was estimated as the difference between current species richness obtained from averaging in the GLMs and the predicted equilibrium species richness obtained from species–area relationships.

Table 1. Species–area relationships in stable patches (n = 14) and the estimated extinction debt of each species group
Species groupsSpecies-area relationshipa Extinction debtb
  1. a P-values are the significance of each single regression model.
  2. bExtinction debt was estimated as the difference between the numbers of predicted and observed butterfly species. P-values are the significance of Wicoxon test comparing the similarity of the numbers of predicted and observed species.
Seasonal groups  
Seasonal specialists S = 5.40 logA −1.60 P < 0.0012.3 (−0.1∼8.0) species P < 0.001
Seasonal generalists S = 4.08 logA + 19.57 P = 0.031.8 (−3.6∼7.6) species P = 0.08
Patch-matrix groups  
Patch-dependent species S = 7.68 logA + 5.88 P < 0.0013.8 (−1.1∼13.3) species P = 0.002
Matrix-dwelling species S = 1.80 logA + 12.09 P = 0.040.3 (−4.4∼3.4) species P = 0.55
Host plants groups  
Woody plant feeders S = 8.16 logA + 7.37 P < 0.0013.3 (−0.4∼8.7) species P = 0.01
Herbaceous plant feeders S = 1.32 logA + 10.60 P = 0.190.8 (−2.5∼6.3) species P = 0.12

Results

A total of 3471 butterfly individuals of 52 species were recorded in the surveyed forest patches. Ten and 42 species were classified as seasonal specialists and seasonal generalists, respectively; 30 were patch-dependent and 22 were matrix-dwelling; and 33 and 19 species were woody and herbaceous plant feeders, respectively (Supporting Information Appendix S2).

Effects of current and past landscape structures on current species richness

For seasonal specialists and patch-dependent species, past patch area had significant positive effects on current butterfly species richness, whereas the effects of current patch area were not significant (Table 2, Fig. 2). Thus, the RVIs of past patch area for seasonal specialists and patch-dependent species were higher than those of current patch area (Table 2). For woody plant feeders, both current and past patch areas had significant positive effects, and the RVIs of both current and past patch areas were 1.0 (Table 2). Neither current nor past patch area had a significant effect on current species richness of seasonal generalists, matrix-dwelling species or herbaceous plant feeders (Table 2, Fig. 2). Current and past patch connectivities had no effects on the species richness of any group (Table 2).

Figure 2.

Relationships between current patch area and current butterfly species richness (top panels), and past patch area and current butterfly species richness (bottom panels) for three classifications. Filled circles indicate (a) seasonal specialists; (b) patch-dependent species; (c) woody plant feeders. Open circles indicate (a) seasonal generalists; (b) matrix-dwelling species; (c) herbaceous plant feeders. Solid lines show significant averaged models of specialist butterflies, respectively (details in Table 2).

Table 2. Parameters (β), unconditional SE, 95% CI and the RVI of (a) patch, (b) seasonal and (c) feeding groups estimated by model averaging (see also Methods)
Species groupsVariables β SE95% CIRVI
  1. Asterisks indicate significance of each parameter (those of 95% CI are not overlapping 0).
  2. SE, standard error; CI, confidence interval; RVI, relative variance importance.
Seasonal groups     
Seasonal specialistsIntercept−0.300.73−1.72 to 1.13
Current area0.280.25−0.22 to 0.770.69
Past area0.760.200.37 to 1.15*1.00
Current connectivity0.240.25−0.26 to 0.730.62
Past connectivity−0.130.26−0.63 to 0.370.30
Seasonal generalistsIntercept2.950.202.56 to 3.34*
Current area0.120.11−0.06 to 0.330.70
Past area0.050.07−0.09 to 0.200.51
Current connectivity0.000.02−0.03 to 0.040.17
Past connectivity−0.020.06−0.14 to 0.090.22
Patch-matrix groups     
Patch-dependent speciesIntercept2.140.421.32 to 2.96*
Current area0.150.14−0.13 to 0.420.67
Past area0.320.100.12 to 0.52*1.00
Current connectivity0.040.07−0.11 to 0.180.33
Past connectivity−0.100.17−0.43 to 0.230.38
Matrix-dwelling speciesIntercept2.390.221.97 to 2.81
Current area0.080.11−0.13 to 0.300.48
Past area0.020.04−0.06 to 0.100.27
Current connectivity0.010.03−0.05 to 0.060.19
Past connectivity0.000.04−0.08 to 0.070.18
Host plants groups     
Woody plant feedersIntercept2.050.251.56 to 2.55*
Current area0.310.110.09 to 0.52*1.00
Past area0.220.090.05 to 0.40*1.00
Current connectivity0.010.03−0.05 to 0.070.16
Past connectivity−0.020.06−0.15 to 0.100.20
Herbaceous plant feedersIntercept2.410.251.93 to 2.89*
Current area0.020.05−0.07 to 0.110.25
Past area0.040.07−0.09 to 0.170.37
Current connectivity0.010.03−0.05 to 0.060.19
Past connectivity−0.030.07−0.16 to 0.110.21

Comparing species richness of stable and unstable patches

Extinction debts were found for all specialist butterfly groups (i.e. seasonal specialists, patch-dependent species and woody plant feeders), because the levels of observed butterfly species richness of these groups were significantly higher than predicted (Wilcoxon test, seasonal specialists: P < 0.001; patch-dependent species: P = 0.002; woody plant feeders: P = 0.01), but there were no significant differences between observed and predicted species richness for seasonal generalists, matrix-dwelling species and herbaceous plant feeders (Wilcoxon test, seasonal generalists: P = 0.08, patch-dependent species: P = 0.55; woody plant feeders: P = 0.12) (Table 1, Fig 3).

Figure 3.

Top panels: the relationship between current patch area and extinction debt (i.e. the difference between observed and predicted numbers of species) for seasonal specialists, patch-dependent species and woody plant feeders. Bottom panels: observed species richness plotted against predicted species richness in 21 unstable patches. The solid line indicates the one-to-one relationship (the normal regression line; i.e. slope = 1) between observed and predicted species richness. Filled circles indicate (a) seasonal specialists; (b) patch-dependent species; (c) woody plant feeders. Open circles denote (a) seasonal generalists; (b) matrix-dwelling species; (c) herbaceous plant feeders. P-values indicate the significance level of differences between observed and predicted species richness by Wicoxon test.

Mean extinction debt per unstable patch was 2.3 (range: −0.1 to 8.0) species for seasonal specialists, 3.8 (range: −1.1 to 13.3) species for patch-dependent butterflies and 3.3 (range: −0.4 to 8.7) species for woody plant feeders (Table 1), and significantly different from zero (Wilcoxon test, all three groups: P < 0.001). There was a negative relationship between estimated extinction debt and current patch area only for seasonal specialists (P < 0.001), whereas the relationships for other groups were not significant (patch-dependent species: P = 0.13; woody plant feeders: P = 0.71) (Fig. 3). Spatial patterns of potential extinction debt in our study area (Fig. 4) indicated two general patterns: (1) regardless of past patch area, currently large patches had relatively little extinction debt; (2) for currently small patches, those that had shrunk drastically in size had high extinction debt.

Figure 4.

Top panels: the distribution of the potential extinction debts (i.e. the number of living dead species) in Tokyo, central Japan (according to the generalized linear models and species–area relationships). Extinction debt was measured as the estimated number of species predicted to go extinct at a long-term equilibrium. Bottom panels: reduced forest area (ha) in each remnant patch. The extent of reduced forest area was measured in hectares lost since 1971. Each patch was ranked by the magnitude of extinction debt and reduced area.

Discussion

Evidence of an extinction debt for specialist butterflies

As hypothesized, two different methodological approaches showed that an extinction debt was only evident in seasonal and patch-specialist butterflies. Despite the drastic increase in attention to extinction debt in recent years, a crucial question remains unanswered: which species and/or functional groups incur extinction debt (Kuussaari et al., 2009)? Our results revealed that the living dead were mostly seasonal or patch specialists. In general, specialist butterflies are considered to be sensitive to habitat fragmentation and hence more prone to extinction in fragmented landscapes (Davies et al., 2004), because they typically require the largest and most well-connected patches (e.g. Öckinger et al., 2010; Soga & Koike, 2012b). However, we found that even many specialist species could occupy fragmented landscapes as living dead for extended periods, and their relaxation times may be relatively long.

Our method using stable and unstable patches probably underestimated extinction debt, because some extinction debt may exist in stable patches as well as unstable ones. Despite this, we estimated that the extinction debt was more than 20% of specialist species in unstable patches (Fig. 3). Considering the long relaxation times for specialists, irrespective of past landscape configurations, the current patch occupancy of specialist species may overestimate long-term species richness. In our study, because patch occupancy patterns of several specialist species were positively correlated to past patch area (M. Soga, unpubl. data), many specialist species can probably not maintain relatively large populations in currently small patches, and are considered to be deterministically extinct even without further fragmentation.

In this study, we identified areas where high numbers of living dead could exist (i.e. high-priority conservation areas). The map of potential extinction debt (Fig. 4) showed that especially small unstable patches had high amounts of unpaid extinction debt. These results indicate that, in modern cities, small patches may have more declining populations that are expected to go extinct in the near future; thus, the conservation of small patches is more urgent than that of large patches. Therefore, in fragmented landscapes, we should pay more attention to specialist species in small unstable forests. Our map also shows that, using only current landscape configurations, high-priority patches cannot be distinguished from other patches in the same area. Therefore, to accurately assess the real conservation importance of each patch, conservation managers need to take into consideration not only current landscape configurations and land use patterns, but also those of past landscapes.

Currently, small patches typically have shorter relaxation times than large patches (Brooks, Pimm & Oyugi, 1999; Kuussaari et al., 2009), and Cousins (2009) also indicated that high losses of habitat in landscapes will lead to rapid payoffs of extinction debts. In this study, fragmentation history of our study area is relatively short compared with other previous studies, for example 70 years (Helm et al., 2006); 75 years (Sang et al., 2010); 80 years (Piqueray et al., 2011); ∼1000 years (Vellend et al., 2006). Because such short payment periods may prevent patches to pay off an extinction debt completely, we could detect an unpaid extinction debt even in small patches. Moreover, in our study area, there are 25% habitats left in timescale of c. 40 years. An extinction debt has been predicted to exist in landscapes more than 10% habitats remains from past to present (Adriaens et al., 2006; Cousins & Vanhoenacker, 2011). Therefore, such high amounts of habitat remnants in our study area may rescue populations of some specialist butterfly species as refuges (‘rescue effects’ as defined in Brown & Kodric-Brown, 1977). Based on these facts, we face a serious situation: extinction debt may now be only partly paid, with a rapid payoff looming in the near future. In our study, especially in woody plant feeders, current specialist species richness was explained by both current and past patch area; this can be interpreted as evidence of a partly paid extinction debt (Kuussaari et al., 2009).

Preventing future biodiversity loss in urban area

Preserving current habitat patches and connectivity may not be sufficient to conserve the living dead, because even without further habitat fragmentation or loss, local populations of some species are likely to go extinct. In addition, the most serious extinction debts in our study area were evident in small patches, which typically had shorter relaxation times (Kuussaari et al., 2009), indicating that most living dead in our study area required urgent conservation actions.

What actions are required to preserve such living dead species in urban areas? To prevent future extinctions of specialist species, expanding habitat areas may be a straightforward solution, because large patches are considered to be crucial for maintaining relatively large population, as shown by the relationship with past patch area. Although we could not detect any statistically robust effects of habitat connectivity on butterfly species richness, several previous studies (e.g. Öckinger and Smith 2006) reported positive relationships between these parameters. Therefore, to facilitate dispersal in fragmented landscapes, improving habitat connectivity would also be important (Janin et al., 2009). However, in urban areas, preserving or restoring large and well-connected reserve areas is usually difficult, because of the high costs involved. Therefore, instead of increasing habitat area and connectivity, improving habitat quality (e.g. habitat heterogeneity and resource abundance) may be the most effective and realistic means for preserving biodiversity in urban areas. Recently, Huth & Possingham (2011) demonstrated that, in small patches, increasing patch quality may reap more benefits than increasing patch area, because small patches are typically degraded in quality. Although we did not measure parameters of patch quality in this study, our previous study Soga & Koike, (2012a) demonstrated that nectar-plant abundance (i.e. patch quality) had positive effects on butterfly species richness, and several small patches maintained too many species for their sizes. Improving habitat quality of small patches in urban areas may be a relatively easy and realistic way to prevent future losses of living dead.

In Tokyo, the population is expected to peak in 2015, after which the population is expected to start declining (Statics of Tokyo, 2012). This suggests that the rate of habitat fragmentation in Tokyo will be moderate, and the restoration of woodlands for biodiversity conservation should be a relatively easy task. In the case of Tokyo, our extinction debt mapping will help conservation managers to set restoration priorities. To restore biodiversity in urban areas efficiently, the networks between patches must be restored and maintained, with an emphasis on the patches that contain many living dead.

Acknowledgements

We are grateful to members of the Forest Science Program, the Tokyo University of Agriculture and Technology, and S. Sugiura, Y. Yamaura and three anonymous referees for their invaluable comments on the early versions of the manuscript. This study was funded by the Urban Green Tech Japan, and Fuji Film Green Foundation.

Ancillary