Effectiveness of nature reserves for bird conservation in urban parks in Tokyo

Kazuhiro Katoh ·Misako Matsuba

Abstract In some urban parks in Tokyo,semi-natural habitat patches are maintained as nature reserves for birds,called bird sanctuaries.Bird censuses and vegetation surveys were conducted in eight urban parks in Tokyo from December 2015 to July 2016 to determine the effectiveness of bird sanctuaries on avian species composition.The relationship between avian species composition and environmental conditions was analyzed by partial redundancy analysis (partial RDA) using vegetation variables,number of visitors,presence or absence of reserves within the same park,area of wooded parts,and normalized difference vegetation index(NDVI) in the surrounding area.The results of the partial RDA are as follows:(1) Area,lower vegetation cover,higher vegetation cover,tree species composition obtained from detrended correspondence analysis (DCA) on vegetation survey data,and NDVI in the surrounding area were considered as covariates and explained 17.4—33.6% of the total variation in bird species composition,(2) The presence or absence of sanctuary significantly explained bird species compositional variation regardless of season,indicating that the sanctuaries were beneficial for urban avoider species not only in the sanctuaries but also in the surrounding wooded area,and (3)Tree coverage within a 100 m range and leaf litter coverage also influenced avian species composition.We concluded that bird sanctuaries and other nature reserves can be beneficial to bird conservation,even within Tokyo.The positive effect of sanctuaries could be due to the limited influence of human presence and the developed vegetation within them.These reserves may play the role of core habitats within the studied urban parks.

Keywords Avian species composition·Leaf litter ·Nature reserve·Tree coverage·Urban biodiversity ·Vegetation structure

Introduction

Currently,biodiversity conservation in urban areas is a global issue (Hostetler et al.2011).Although previous studies have revealed that small habitats,such as private gardens and small vegetated areas,can also play a role as habitats in urban areas (Cannon 1999),large habitats are still important for maintaining biodiversity in urban areas (Beninde et al.2015),because of the biased species composition (i.e.,lack of urban avoider species) in small habitats (Katoh et al.2015).

However,large woodlands in urban areas,most of which are managed as parks,are not always suitable habitats for maintaining biodiversity.The vegetation structure of urban woodlands is often too simple to maintain high biodiversity because of social pressure demanding comfortable open spaces (Heyman 2010).Vegetation with a simple structure is widely attained by clearing understory (Hedblom and Söderström 2008).Vegetation with scarce understory is not suitable for some organisms because understory shrubs and herbs provide them with food,refuge,and nesting sites (Katoh 1996;Heyman 2010).Typical urban parks with lawns and tall trees usually contain few breeding bird species (Forman 2014).

Therefore,maintaining natural or semi-natural habitat patches within urban areas is expected to be a means to increase urban biodiversity (Harris et al.2018).Limitation of human use and suitable maintenance of vegetation for wildlife is essential to sustain natural or semi-natural habitat patches and to avoid the situation in most urban parks mentioned above.Presently,we have some semi-natural habitat patches within parks in Tokyo.They are protected and managed as nature reserves for birds,called bird sanctuaries.A representative one is the sanctuary founded in 1971 in the Yoyogi Park in Tokyo (Kanai 2018).Human activities are limited in these sanctuaries to conserve avian communities,to maintain suitable vegetation for animals,and to avoid human disturbances for wildlife.However,the effectiveness of such bird sanctuaries has not been well studied,except for a few examples of waterbird habitats (Kanai 1991;Yokohari and Amati 2005).In this study,we surveyed vegetation and avian communities in urban parks with or without nature reserves of woodland.The objectives of the study were as follows.

The first objective was to confirm the effectiveness of nature reserves which mainly comprise woodlands in maintaining forest bird communities.Effectiveness was evaluated in two ways.First,avian communities were compared inside and outside the sanctuaries from the viewpoint of species composition.Protected areas are usually considered to be effective for biodiversity conservation (Pimm et al.2001).Second,a similar comparison was made between parks with and without sanctuaries.If sanctuaries are good habitats for birds,avian communities in the surrounding woodland can be expected to have more forest species through the spillover effect (Loss et al.2009;Forman 2014).

The second objective was to detect environmental factors influencing the composition of avian species.In most urban parks,vegetation is managed considering amenities for visitors,which makes the vegetation structure simpler (Forman 2014) and leaf litter is removed from the ground.Leaf litter may provide essential resources for invertebrates and ground foraging birds (Aronson et al.2017).Furthermore,visitors may influence the distribution or behavior of animals (Bötsch et al.2018).Tree coverage within nearby areas was also considered because of the importance of surrounding tree cover for bird diversity (Fontana et al.2011).

Materials and methods

Study sites

In Tokyo,bird sanctuaries have been maintained in several urban parks.Among these,three parks with more than 5 ha of woodland area were selected as the study sites (Hikarigaoka Park,Yoyogi Park,and Kinuta Park).The areas of the bird sanctuaries in each of the three parks range from 1.5 to 3.0 ha.Another park (Akatsuka Park),which contains a series of off -limits reserve woodlands along a terrace scarp was also selected because the woodlands could be considered equivalent to a sanctuary.The total area of the off -limits reserves in Akatsuka Park is ca.4.2 ha.Hikarigaoka Park has off -limits reserve areas for conservation in addition to the sanctuary.Hereafter,the word sanctuary also represents these equivalent reserves in Akatsuka Park and Hikarigaoka Park.Four parks (Johoku-chuo Park,Toyama Park,Roka-Koshunen Park,and Rinshinomori Park) without sanctuaries and with more than 5 ha of woodland were also selected for comparison (Table 1).These eight parks are all located on the west side of the center of Tokyo and in a limited area(almost 16 km north to south,10 km east to west,Fig.1).

Table 1 A list of the studied urban parks

Six to ten study plots were located within each park.Sites with tree cover were selected as plots.As for the parks with sanctuaries,three or more of these plots were located in the sanctuary (hereafter,sanctuary plots).Each plot was circular with a radius of 20 m.The distance between plots was at least 100 m,following Katoh et al.(2 015),except for the three plots located in the sanctuary in Yoyogi Park,which were located closer to the nearest plot.Therefore,the centrally placed plot among the three sanctuary plots in Yoyogi Park was omitted in the statistical tests of avian data so that the distance between plots was at least 100 m for all plot pairs.In total,17 plots were located in the sanctuaries,21 plots were located outside the sanctuaries in the parks with sanctuaries (outside plots) and 33 plots were located in the parks without sanctuaries (no-sanctuary park plots).

Bird surveys

Bird censuses were conducted using the point count method from December 2015 to March 2016 (wintering season) and from April 2016 to July 2016 (breeding season).Each plot was visited three times in each season for bird censuses,following previous studies (e.g.,Katoh 1996).Throughout the study period,the bird surveys were conducted by a single researcher (KK) to avoid any variations caused by multiple observers.During each visit,all bird individuals within the plot were recorded according to the species for 15 min.The number of visitors that entered the plot was also recorded.Censuses were conducted from 9 a.m.to 1 p.m.Censuses were not conducted under bad weather;rainy or snowyconditions and high winds were avoided.Species composition data was constructed for each season by summing the records obtained through the three surveys.

Vegetation surveys

Tree census

A transect 40 m in length,which was equal to the diameter of the plots,was selected in each plot representing the vegetation features.Trees taller than 2 m located within 1 m from the transect were counted according to the species.Tree censuses were conducted once in the breeding season.Based on the results,the total number of tree individuals was calculated for each transect.Tree species composition data were analyzed by detrended correspondence analysis (DCA,Hill and Gauch 1980),and site scores of the first three axes were used for further analysis.Data on species that were recorded in six or fewer plots was removed before DCA to avoid the distorting effect of low-frequency species (Gauch 1982).DCA was calculated using PC-ORD 6.0 (McCune and Meff ord 2016).

Vegetation structure surveys

Along each 40 m transect,the observation points for vegetation structure surveys were set at 2 m intervals (21 points per transect).At each point,the existence of any plant material(alive or dead leaves,branches,stems,fruits,and flowers)was recorded at 12 heights namely 0 m,0.12 m,0.37 m,0.75 m,1.5 m,3 m,6 m,9 m,12 m,15 m,18 m,and 21 m.When any plant material was observed within a 30 cmradius horizontal disk from the observation point at each height,we considered that the material existed at that point at that height.At a height of 0 m,the existence of leaf litter was separately recorded from other plant material.Vegetation structure surveys were conducted in both the wintering and the breeding season at each plot,as vegetation structure may differ between the seasons because of defoliation.Based on the records,vegetation coverage at each height and leaf litter coverage was calculated for each plot for each season.The higher vegetation coverage was calculated for each plot by averaging the cover from 6 to 21 m.Similarly,lower vegetation coverage was calculated considering the cover from 0 to 3 m.

Satellite image analyses

The WorldView-2 satellite image (taken on May 9,2013,provided by Space Imaging Inc.) with a resolution of about 50 cm was analyzed to measure the area of continual wooded parts covered by tree crowns (hereafter,wooded patch) to which each study plot belonged.ArcGIS 10 (ESRI Inc.2011) was used to analyze satellite image.In the Akatsuka Park and the Toyama Park,the park area was divided into two or three parts by artificial structures that may prevent movement of birds,such as residential areas (Katoh and Kamiyama 2014) and roads (Forman and Alexander 1998).In these cases,the wooded part of each separated park area was considered as an individual wooded patch.Yoyogi Park is situated next to the Meiji Shrine,which is covered with mature forest with an area of more than 50 ha;therefore,the wooded patch area of Yoyogi Park was calculated including the shrine forest.

Tree coverage within a 100 m range from the center of each plot was also calculated as the ratio of the part covered by tree crowns using the same satellite image.The parts covered by tree crowns outside the parks were included in this calculation.

The normalized difference vegetation index (NDVI,Rouse et al.1974) was calculated using the same satellite image to understand the vegetation abundance in the surrounding area (Leveau et al.2018).From the outside boundary of the wooded patch,buffer areas with certain widths were created.The buffer width varied from 100 m to 2000 m with intervals of 100 m.The average NDVI values within the buffer area were calculated for each buffer size for each wooded patch.

Figure 2 indicates the relationship between the buffer sizes and the average of the obtained NDVI values for the 11 wooded patches.For most patches,narrower buffers resulted in larger NDVI values because parks and other wooded areas tended to be distributed in clumps in the study area.NDVI values appeared to be almost stable when the buffer width was more than 1000 m.Therefore,average NDVI values within a 1100 m buffer outside the boundary of the wooded patch were used as an indication of the level of urbanization around each woodland.

Fig.2 Relationship of average NDVI (normalized difference vegetation index) within the buffer area around the woodland patches and buffer size.The average value calculated for the 11 wooded patches belonging to the eight parks is shown

Data analyses

Basic statistics

The values of the environmental variables obtained for each study plot by the vegetation surveys and satellite image analyses were compared among the following three groups of the study plots:(1) the plots located within the sanctuaries(sanctuary plots),(2) the plots located outside the sanctuaries (outside plots),and (3) the plots located in the parks with no sanctuary (no-sanctuary park plots).Avian species richness,urban avoider bird species richness,and urban avoider bird abundance were also compared among these three groups.We classified bird species into two groups(urban avoider species and other species,after McKinney(2002)) based on the reported tolerance to urbanization(Katoh 2009).As the values of the variables did not follow a normal distribution,median and quartile ranges were used to show the magnitude and distribution range of the variables.The Steel—Dwass nonparametric test was applied to compare the values among the groups.JMP 14.0 (SAS Institute Inc.2018) was used for the calculations.

Constrained ordination

First,a subset of the bird species composition data was prepared for each season by removing the sanctuary plots from the original data set.The new set included 54 study plots located outside the sanctuaries (i.e.,outside plots and nosanctuary park plots).Hereafter,we refer to the new data set as the "without sanctuary data set",which was prepared to investigate the effect of the sanctuaries on surrounding areas.The analysis of the original data set may not detect the difference between parks with and without sanctuaries because the difference between sanctuary plots and others would be quite distinct.Data on species that were recorded in six or fewer plots was removed from the following ordination analysis.In the without sanctuary data set,species that were recorded in three or fewer plots were removed so that the remaining species were the same as those in the full data set after the removal of the species with low frequency.

The bird species composition data was analyzed by constrained ordination using the vegetation survey data and satellite image analysis results as environmental data to obtain the most relevant ordination considering the environmental variables.First,DCA was performed on the bird data to determine the length of the community gradient.It was 2.6 SD in the wintering season and 2.4 SD in the breeding season for the full data set,and 2.4 SD in the wintering season and 2.3 SD in the breeding season for the without sanctuary data set.Please note that "SD" is a unit for measuring the length of ordination axes analogous to standard deviation.The length ref lects species turnover along the axes (Lepš and Šmilauer 2003).These small values indicated that the length of the community gradient was rather short,and that the use of the ordination methods considering the linear relationship between variables was suitable(Lepš and Šmilauer 2003).Therefore,we used redundancy analysis (RDA).To increase the linearity between variables(i.e.,number of individuals of each species),the number of individuals was log-transformed and the value log10(n+1)was used for the following analysis,where "n" is the number of individuals.RDA was performed using CANOCO ver.5.0(ter Braak and Šmilauer 2012).

To avoid overf itting caused by the use of too many explanatory variables,the interactive forward selection procedure was used.In this procedure,the environmental variable that can improve the result most effectively was selected and added to the model.This process was repeated until no more variables could signif ciantly improve the model,i.e.,all the remaining variables showedp>0.05.Partial RDA was used to remove the inf ulence of the factors of which the effects are already known.It can "partial out" the inf ulence of the factors designated as covariates (Lepš and Šmilauer 2003).The wooded patch area,average NDVI within the surrounding area,lower vegetation coverage,higher vegetation coverage,and the tree DCA Axis 1 score were preliminarily inputs as covariates.These covariates ref lect environmental factors such as habitat area (Blake and Karr 1987),vegetation richness in the surrounding landscape(Mörtberg 2001),vegetation structure (Katoh 1996;Heyman 2010),and plant species composition (Osborne 1984),which are known to inf ulence avian communities.

The environmental variables used in the RDA are listed in Table 2.To make the distribution of values closer to that of a normal distribution,the variables for which the minimum value was zero and the maximum value was unlimited (i.e.,numbers and area) were log transformed.The variables for which the minimum value was zero and the maximum value was 1 (i.e.,cover values) were square-root arcsine transformed.As NDVI values lie between -1 and 1,we used the following transformation:

Transformed NDVI=arcsin (((NDVI+1)/2)1/2).

The sanctuary variable indicated whether a sanctuary or an equivalent off -limits reserve is located within the same wooded patch.If such an area was located,the value was 1;if not,the value was 0.None of the transformed variables showed high correlations (|r|>0.7,Dormann et al.2013)with each other in any season or in any data set,which suggested the lack of multicollinearity in the data.

Results

Summary of bird and vegetation surveys

During the bird censuses in the 71 study plots belonging to the eight parks,28 and 29 species were recorded in thewintering and the breeding seasons,respectively.The list of species is shown in Table 3.The median and quartile ranges of the avian species richness,urban avoider richness,and abundance are shown in Table 4.The sanctuary plots showed significantly (p<0.05) higher values than the outside plots and the no-sanctuary park plots in most cases,except for species richness in the breeding season.The outside plots showed significantly higher values than the no-sanctuary park plots,except for species richness in both seasons.

Table 2 The environmental variables used in the redundancy analysis (RDA) and their abbreviations

Ninety-six tree species were recorded during the tree census.Among them,24 frequent species were considered in the DCA,which revealed that the tree species composition varied from a dominance of natural deciduous species such asQuercus serrataMurray andQuercus acutissimaCarruth.to planted species such asGinkgo bilobaL.,Platanus×acerifolia(Aiton) Willd.,Zelkova serrata(Thunb.)Makino,Cinnamomum camphora(L.) J.Presl,and cherry trees such asCerasus×yedoensis(Matsum.) Masam.&Suzuki ‘Somei-yoshino’.This trend was presented along the first axis,with larger values indicating the dominance ofQ.serrataandQ.acutissima,which suggested that the first DCA axis was an indicator of tree species compositional change from natural species (larger values) to planted species (lower values).The second and further DCA axes showed no apparent tendency of tree species composition.TheR2values between ordination distances and distances in the original species composition data measured using the Bray—Curtis index (McCune et al.,2002) were 0.166 (the first axis only),0.219 (ordination distances considering the first two axes),and 0.305 (ordination distances considering the first three axes),which suggested that the first axis had an importance of twice or more than that of the lower axes.Therefore,only the DCA first axis (DCA Axis 1) score was considered in the following analysis.

Basic statistics

The basic statistics of the environmental variables and the results of the comparison among the plot groups are shown in Table 5.The number of visitors and lower vegetation coverage were apparently different between the sanctuary plots and the other plots in both seasons.The difference in lower vegetation coverage indicated that vegetation structure was more complex with developed lower vegetation in the sanctuary plots.Higher vegetation coverage and leaf litter coverage also showed significant differences between the sanctuary plots and the other plots.The number of tree individuals and tree coverage within 100 m did not show significant differences among them.The DCA Axis 1 scores on tree species composition showed significant differences between the sanctuary plots and outside plots,which indicates that planted trees tended to be dominant in the outside plots and in the no-sanctuary park plots.

Constrained ordination

The results of the RDA are summarized in Table 6,which indicates that the environmental variables considered in the study explained 28.4% to 42.8% of bird species composition variation observed in the study.The variables that were included in the analysis as covariates,i.e.,wooded patch area,average NDVI within the surrounding area,lower vegetation coverage,higher vegetation coverage,and the tree DCA Axis 1 score,explained 17.4% to 33.6% of the total variation.This result indicates the importance of these variables in explaining the variation in avian species composition.The remaining variation was largely explained by the sanctuary variable (3.4% to 7.2%) and tree coverage within 100 m(2.1% to 3.8%).In the breeding season analysis including the sanctuary plots,the sanctuary variable was not selected,and the number of visitors was selected instead,explaining 2.0% of the variation.This result indicates that the difference in avian composition between the sanctuary plots and the other plots can be explained by the number of visitors in the breeding season,if species compositional change caused by the number of visitors corresponded to the difference between the sanctuary plots and the other plots.We conf irm this result later.

Table 4 Median and quartile range of the avian species richness,urban avoider richness,and abundance

Leaf litter coverage was also included in the model in the breeding season (5.0% to 7.4% of total variation),but not in the wintering season.However,when the values recorded in the breeding season were used in the analysis instead of those recorded in the winter,the variable was selected and explained as much as 7.7% to 10.8% of the total variance.The other variables were not selected as significant variables in the RDA models.The contribution of each axis indicates that the first axis explained a large part of the variation and that the contribution of the third axis was relatively small.Therefore,biplots using the first and the second axis are shown for each data set to show the relationship between the environmental variables selected as significant and the bird species (Fig.3 a—d).

Three trends were detected regarding the relationship between the environmental variables and bird species.First,some urban avoider species such asZosterops japonicusTemminck &Schlegel,Dendrocopos kizuki(Temminck) (in winter),Cettia diphone(Kittlits) (in winter),Turdus pallidusGmelin (in winter),Coccothraustes coccothraustes(Linnaeus) (in winter),andAegithalos caudatus(Linnaeus) (in breeding season) tended to be recorded more abundantly in the plots located in the parks with sanctuary.In the breeding season,we intentionally input the sanctuary variable to the RDA model by removing the number of visitors from the original data set,to test if species compositional change caused by the number of visitors corresponded to the difference between the sanctuary plots and the other plots.The new RDA model indicated that the number of visitors showed an opposite trend against "sanctuary" (not shown).This suggested that the effect of sanctuaries could be derived from the exclusion of visitors from the sanctuaries.

Second,Passer montanus(Linnaeus),Columba liviaGmelin,Hirundo rusticaLinnaeus,Chloris sinica(Linnaeus),andCyanopica cyanus(Pallas) tended to occur in the plots where the leaf litter coverage was low in the breeding season.In this study,the wintering bird composition was explained by leaf litter coverage in the breeding (not wintering) season.In winter,new litter was continuously added to the ground from the trees and surrounding ground surface after ordinary management activities.As a result,most parts within the woodland patches were covered by litter (Table 5),which could mask the potential influences caused by leaf litter directly or indirectly.

Third,tree coverage tended to increase the abundance of some urban avoider species such asD.kizukiin the breeding season.In the winter,these species tended to be observed at the spots located in the parks with sanctuaries.Streptopelia orientalis(Latham),Corvus macrorhynchosWagler,andPsittacula krameri(Scopoli) also showed a positive correlation with tree abundance or tree coverage,but to a weaker extent.

Discussion

Effectiveness of bird sanctuaries

Avian communities were different between the sanctuary plots and the other plots (i.e.,outside plots and no-sanctuary park plots).In the sanctuary plots,vegetation cover was more developed and human presence was lesser than in the other plots.Tree density and tree species composition (Tree DCA Axis 1) seemed to be different between them;native trees and planted trees tended to be dominant in the sanctuary plots and in the other plots,respectively (Table 5).Many of the planted trees were introduced species such asGinkgo bilobaandPlatanus×acerifoliain the studied parks.These differences correspond to the difference in urban avoider species richness and abundance between the sanctuary plots and the other plots (Table 4).The importance of understory vegetation (Katoh 1996;Heyman 2010) and native plant species (Threlfall et al.2016) for some forest bird species have been emphasized in previous studies,which is consistent with our interpretation.

Table 6 A summary of the redundancy analysis (RDA) results

In this study,RDA was carried out as partial RDA considering patch area,vegetation in surrounding area,vegetation structure,and composition within the plot as covariables.The results indicated that the difference between the plots located in the parks with and without sanctuaries could be detected in both wintering and breeding seasons.They strongly suggested that the presence of visitors influences the avian composition in the sanctuaries.Previous studies have indicated that visitors may negatively influence animal (including bird) distribution and behavior(Bötsch et al.2018;Patten and Burger 2018).Limiting human entry within the sanctuaries would be beneficial to some urban avoider species.

The RDA results obtained from the reduced data sets provided a chance to compare avian community composition between the outside plots and the no-sanctuary park plots.The results indicated that the existence of a sanctuary within the same patch produced avian compositional variation (Fig.3) and increased urban avoider species richness (Table 4).This can be explained as the spillover effect(Loss et al.2009;Forman 2014) from sanctuaries.Beier and Noss (1998) indicated that habitat connectivity may influence biodiversity in habitats through individual bird immigration and increased access to food resources.Dale(2018) indicated that green spaces with the presence of native forest were the most important for sensitive ecological groups of birds.Our study demonstrated that the sanctuaries could play the role of core habitats for urban birds like the native forest discussed in Dale (2018).Based on the results,we suggest that sanctuaries (or,in general,urban protected areas) can contribute to maintaining biodiversity not only within themselves but also in the surrounding wooded areas.

Fig.3 Biplot of the RDA (redundancy analysis) results showing species scores (f ine arrows) and environmental variables (bold dashed arrows) Dotted line arrows indicate suburban adapter species.Solid line arrows indicate urban avoider species.Abbreviations of the environmental variables and species names are as follows:(Environmental variables) Litter:Leaf litter coverage,C_Tree:Tree coverage within 100 m range,Visitor:number of visitors.(Bird species)STRori:Streptopelia orientalis,DENkiz:Dendrocopos kizuki,CYAcya:Cyanopica cyanus,CORmac:Corvus macrorhynchos,POEvar:Poecile varius,PARmin:Parus minor,HIRrus:Hirundo rustica,HYPama:Hypsipetes amaurotis,CETdip:Cettia diphone,AEGcau:Aegithalos caudatus,ZOSjap:Zosterops japonicus,SPOcin:Spodiopsar cineraceus,TURpal:Turdus pallidus,TURnau:Turdus naumanni,PASmon:Passer montanus,MOTalb:Motacilla alba,CHLsin:Chloris sinica,COCcoc:Coccothraustes coccothraustes,EMBspo:Emberiza spodocephala,COLliv:Columba livia,PSIkra:Psittacula krameri

Other environmental factors

The RDA results suggest that leaf litter on the ground could influence bird species composition.Leaf litter on the ground tends to be removed in most urban parks,although it is an important material providing food resources for invertebrates and indirectly the ground foraging birds (Aronson et al.2017).Stagoll et al.(2010) reported that the presence of litter cover increased woodland bird species richness by 35% in the Australian Capital Territory.However,it should be borne in mind that the observed effect of leaf litter may be an indirect one due to the correspondence with soil development and the intensity of management and human use.Low leaf litter coverage may suggest that these plots are frequently cleaned up to remove leaf litter and that the need for clean-up is relatively strong,one of the reasons being for human use.In most urban parks,vegetation is managed to improve the amenities for visitors (Forman 2014).

Understory vegetation comprising shrubs,grasses,and herbs tends to be removed,although it is important for some woodland species (Hedblom and Söderström 2008;Heyman 2010).Litter is also removed through vegetation management activities.Protected areas such as bird sanctuaries in the present study can help us maintain vegetation with a developed structure by inhibiting vegetation management for human uses.

Tree coverage around the plot was also detected as a factor influencing avian species composition.A larger tree coverage indicated a more continuous woody habitat around the plot (within 100 m in this study) without encroachment of buildings or open lands such as lawns and pavements.Woodland species tend to avoid buildings and artificially sealed open lands (Fontana et al.2011).

We conclude that bird sanctuaries and other nature reserves can be beneficial to avian diversity conservation even within Tokyo.They are natural or semi-natural habitat patches maintained within urban parks.The limitation of visitors would be the key feature of such sanctuaries;it can have a positive influence on avian communities directly by removing human activities,and also indirectly through vegetation development.Sanctuaries are beneficial to biodiversity not only within themselves but also in the surrounding areas.We would also like to state the importance of suitable vegetation management considering understories and leaf litter,which is realized in the studied sanctuaries.The ecological significance of tree cover connectivity should be further studied considering not only habitat area but also habitat shape and habitat mosaic features.

AcknowledgementsWe are grateful to the Tokyo Metropolitan Park Association for their support in our field surveys.