BIODIVERSITY RESEARCH: Nestedness for different reasons: the distributions of birds, lizards and small mammals on islands of an inundated lake

Abstract

Aim We examined whether the community compositions of birds, lizards and small mammals were nested in a fragmented landscape in the Thousand Island Lake, China. We also assessed whether the mechanisms influencing nestedness differed among these taxonomic groups.

Location Thousand Island Lake, China.

Methods Presence/absence matrices were compiled for birds (42 islands) and lizards (42 islands) using line‐transect methods, and for small mammals (14 islands) using live‐trapping methods from 2006 to 2009. Nestedness was analysed using BINMATNEST, and statistical significance was assessed using the conservative null model 3. We used Spearman rank correlations and partial Spearman rank correlations to examine associations of nestedness and habitat variables (area, isolation, habitat diversity and plant richness) as well as life‐history traits (body size, habitat specificity, geographical range size and area requirement) related to species extinction and immigration tendencies.

Results The community compositions of birds, lizards and small mammals were all significantly nested, but the causal factors underlying nestedness differed among taxonomic groups. For birds, island area, habitat specificity and area requirement were significantly correlated with nestedness after controlling for other independent variables. For lizards, habitat heterogeneity was the single best correlate of nestedness. For small mammals, island area, habitat heterogeneity and habitat specificity were significantly correlated with nestedness. The nested patterns of birds, lizards and small mammals were not attributable to passive sampling or selective colonization.

Main conclusions The processes influencing nested patterns differed among taxonomic groups. Nestedness of bird assemblages was driven by selective extinction, and lizard assemblage was caused by habitat nestedness, while nestedness of small mammals resulted from both selective extinction and habitat nestedness. Therefore, we should take taxonomic differences into account when analysing nestedness to develop conservation guidelines and refrain from using single taxa as surrogates for others.

Introduction

Nestedness is one of the most commonly observed properties of a regional collection of local biotas (Wright et al., 1998; Gaston & Blackburn, 2000). Nestedness occurs when species at species‐poor sites are proper subsets of the suite of species that occur at more species‐rich sites (Patterson & Atmar, 1986). Since early descriptions of nested patterns (Darlington, 1957), numerous studies have investigated nestedness and underlying processes (e.g. Patterson & Atmar, 1986; Bolger et al., 1991; Cook & Quinn, 1995; Wright et al., 1998; Davidar et al., 2002; Fischer & Lindenmayer, 2005; Schouten et al., 2007; Frick et al., 2009). Although these studies document that a wide range of taxa on islands and in fragmented habitats show nested patterns, few studies have compared data on multiple taxonomic groups (especially among vertebrate taxa) at the same locations (Wright et al., 1998). In view of their functional significance in ecosystems as pollinators, seed dispersers, arthropod predators, nutrient depositors and ecosystem engineers (Davic & Welsh, 2004; Sekercioglu, 2006; Zubaid et al., 2006), it is important to evaluate whether different vertebrate groups follow the same mechanisms underlying nestedness.

Four general hypotheses have been proposed to account for nested patterns of species distribution: (1) passive sampling, (2) selective extinction, (3) selective colonization and (4) habitat nestedness. Passive sampling could generate nestedness as an artefact of underlying stochastic principles, because rare species are less likely to be sampled in a given area than common species (Andrén, 1994; Cutler, 1994; Higgins et al., 2006). As passive sampling simply reflects a sampling effect, it is stressed that data should be tested for passive sampling prior to other hypotheses (Andrén, 1994; Worthen et al., 1996; Wright et al., 1998). The selective‐extinction hypothesis is based on the assumption that in systems experiencing species loss or ‘relaxation’, species would disappear from sites in a predictable sequence and thus lead to nestedness (Patterson, 1984; Simberloff & Levin, 1985). The selective‐extinction hypothesis predicts that area is the main factor explaining species nestedness because species with large minimum‐area requirements and small population size have higher extinction risks (Wright et al., 1998). According to the selective‐colonization hypothesis, habitat isolation would create nested subsets through dispersal limitation, as species differ in their ability to colonize distant sites (Darlington, 1957; Patterson, 1987). Finally, the habitat nestedness hypothesis considers the nestedness of species assemblages as a consequence of their close association with habitats which have a nested distribution (Wright et al., 1998; Calmé & Desrochers, 1999; Honnay et al., 1999).

Furthermore, species life‐history traits may also be useful for assessing the contribution of differential abilities of species to nestedness (Worthen et al., 1996; Meyer & Kalko, 2008; Frick et al., 2009), although this approach has received less attention. For example, if dispersal ability is a strong determinant of nested patterns (Cook & Quinn, 1995; Kadmon, 1995; Loo et al., 2002), then ecomorphological characters reflecting relative mobility of species may order species occurrence patterns (Frick et al., 2009). In contrast, if extinction susceptibility was the main driver of nested patterns, then life‐history traits linked to extinction proneness would point towards extinction as a major causal factor structuring composition patterns. However, although these environmental variables and species traits are intimately linked (Ulrich et al., 2009), few studies have combined them simultaneously to examine their roles in generating nestedness (Schouten et al., 2007).

Considerable variation exists among taxa in life‐history traits such as vagility, habitat specificity, geographical range size and area requirement (Cook & Quinn, 1995; Wright et al., 1998; Davidar et al., 2002; Schouten et al., 2007). Variation in interactions between life‐history traits and habitat variables, such as area requirement and island size, dispersal ability and island isolation, is hypothesized to cause taxonomic differences in nestedness (Cook & Quinn, 1995; Wright et al., 1998). Nestedness can result from different area requirements by different species, such that species with large area requirements will go extinct first and be restricted to large islands, while species with small area requirements can be found everywhere (Wright et al., 1998). According to this hypothesis, taxa with large area requirements (e.g. birds) should be influenced more by island area than ones with small area requirements (e.g. lizards). Moreover, differences in dispersal ability among species may interact with island isolation to produce nestedness (Darlington, 1957). According to this hypothesis, poor dispersers would be present only on the closest or most diverse islands, while strong dispersers would be present on most islands because of frequent colonizations. If good colonization ability causes nestedness, then the more vagile taxa (e.g. birds) should be more nested than less vagile taxa (e.g. lizards). However, very few studies have explicitly tested these hypotheses and compared taxonomic differences in mechanisms underlying nestedness, especially among vertebrate assemblages.

In this study, we examined whether the community compositions of birds, lizards and small mammals were nested in a fragmented landscape in the Thousand Island Lake, China. In addition, we assessed whether the mechanisms influencing nestedness differed among these taxonomic groups. Understanding the mechanisms influencing nestedness and how they differ across taxonomic groups can have important implications for conservation and can be used to direct management efforts.

Methods

Study sites

The Thousand Island Lake (hereafter TIL) (29°22″– 29°50″ N, 118° 34″– 119° 15″ E) was created in 1959 by the damming of the Xinanjiang River in western Zhejiang Province for the purpose of generating hydroelectricity (Fig. 1). With construction of the Xinanjiang dam, an area of approximately 580 km² was inundated, creating 1078 islands (0.25–1320 ha) out of former hilltops when the water reached its final level (108 m) (Wang et al., 2009a). The major vegetation on the islands is the naturally secondary forest dominated by Pinus massoniana, which is mixed with a lot of broad‐leaved trees and shrub species, such as Cyclobalanopsis glauca, Castanopsis sclerophylla, Smilax davidiana, Grewia biloba and Loropetalum chinense. The climate is typical of the subtropical monsoon zone and is highly seasonal, with hot summers and cold winters. The average annual temperature is 17.0°C, ranging from −7.6°C in January to 41.8°C in July. Annual precipitation at the study sites is 1430 mm.

Figure 1

Map showing 46 sampling islands in the Thousand Island Lake and (inset) general location of the study area in Zhejiang Province, China (Highlighted in red are the locations of the 46 study islands, and islands are ranked in order of decreasing area). For the island numbers, refer to the information in Table 1.

Our study system provides an ideal opportunity to study nested subset patterns as it meets the three key conditions generally considered necessary for the development of nestedness (Patterson & Atmar, 1986; Patterson & Brown, 1991; Wright et al., 1998). First, the islands have a common biogeographic history. Second, the islands share an ancestral pool of species. Third, species inhabiting the islands are somewhat hierarchically ordered in terms of their niches and relatively complete species inventories are available (Zhang et al., 2008; Wang et al., 2009a; Zhao et al., 2009).

Sampling methods

We compiled presence/absence data for birds, lizards and small mammals across a set of 46 islands. These islands were selected to represent a range of areas and degrees of isolation (Table 1) (Terborgh et al., 1997; Wang et al., 2009a). To facilitate surveys, we cleared transect trails (ca. 20 cm wide) that traversed the mountain ridges on all the islands (Terborgh et al., 1997; Wang et al., 2009a). To account for the greater habitat variability associated with larger sites, sampling effort was roughly proportional to island area (log₁₀‐transformed) (Schoereder et al., 2004). Accordingly, eight transect trails were sampled on island 1 (the largest island, area > 1000 ha), four on islands 2 and 3 (island area >100 ha), two on five islands (10 < island area < 100 ha) and one on each of the remaining small islands (island area <10 ha; Table 1) (Wang et al., 2009a).

Table 1. Characteristics of 46 islands sampled for the presence of birds, lizards and small mammals in the Thousand Island Lake, China.

Island Code	Island area (ha)	Isolation (m)	Number of habitat types (n)	Plant richness (n)	Nested matrix rank			Number of transects (n)	Total length of transects (m)
					Birds	Lizards	Small mammals
1	1289.23	897.41	7	198	1	1	1	8	3200
2	143.19	1415.09	6	99	2	2	−	4	1600
3	109.03	964.97	6	86	3	6	7	4	1600
4	55.08	953.95	5	59	6	3	6	2	800
5	46.37	729.80	5	51	8	11	8	2	800
6	35.64	2110.41	5	49	7	7	−	2	800
7	32.29	1936.95	5	57	4	8	11	2	800
8	12.02	1158.00	4	65	−	−	5	2	800
9	9.12	24.50	4	35	−	−	13	1	400
10	5.69	21.85	3	69	5	10	−	1	375
11	3.42	583.00	4	74	11	33	−	1	300
12	2.90	1785.30	3	85	14	25	2	1	275
13	2.83	1238.14	4	86	9	13	10	1	150
14	2.29	973.85	4	65	16	9	−	1	300
15	2.23	3261.96	3	53	21	33	−	1	400
16	2.00	1042.38	3	45	30	17	−	1	300
17	1.93	888.05	4	50	18	19	−	1	250
18	1.74	2293.25	3	100	39	16	−	1	300
19	1.67	570.00	3	80	−	−	3	1	200
20	1.54	711.04	3	88	23	27	9	1	375
21	1.52	849.88	3	40	22	30	−	1	250
22	1.52	2849.99	3	53	19	24	−	1	175
23	1.40	1760.34	3	49	10	26	−	1	375
24	1.26	54.86	3	65	12	28	−	1	200
25	1.20	657.72	3	56	29	18	−	1	225
26	1.20	2128.52	3	68	35	12	−	1	225
27	1.17	2453.37	3	69	38	14	−	1	250
28	1.15	847.12	3	33	36	33	−	1	275
29	1.08	1271.00	3	43	−	−	4	1	250
30	1.03	1458.81	3	36	34	29	−	1	250
31	1.01	2437.85	3	29	25	5	−	1	250
32	1.01	2103.85	3	36	15	23	12	1	250
33	0.97	938.85	3	70	13	33	14	1	200
34	0.96	3133.96	3	50	42	33	−	1	250
35	0.91	1339.71	4	50	40	33	−	1	275
36	0.86	2321.51	3	56	31	21	−	1	225
37	0.83	2298.50	3	50	32	22	−	1	275
38	0.83	1098.58	4	45	37	33	−	1	250
39	0.80	102.60	3	68	26	33	−	1	325
40	0.80	2097.52	2	80	20	32	−	1	300
41	0.73	1320.40	3	31	33	15	−	1	300
42	0.67	1139.87	3	39	41	20	−	1	325
43	0.59	640.53	3	42	27	33	−	1	225
44	0.59	1018.42	3	55	28	31	−	1	250
45	0.57	3712.31	3	47	24	33	−	1	200
46	0.30	1086.03	2	75	17	4	−	1	175

• − denotes islands not surveyed for specific taxon.

Bird sampling

We used the line‐transect method (Bibby et al., 2000) to determine bird occupancy and abundance on 42 islands during two breeding seasons (April–June) and two winter seasons (November–January) from 2006 to 2009. During the survey, the observers walked each transect at a constant speed (about 2.0 km h⁻¹). We recorded all bird species heard or seen within 50 m of the transect lines, but not high‐flying species that just passed over the island. Surveys were conducted between 0.5 h after dawn through to 11:00 h in the mornings and between 15:00 and 0.5 h before sunset in the afternoons. Bird activity is low in the mid‐day period, so surveys were not conducted in that interval. Surveys were also not conducted if there was more than a light sprinkling of rain, high wind or high temperatures (Robbins, 1981). We used global positioning system (GPS) (Unistrong Industrial Co., Ltd., Beijing, China) to record the length of each transect (Table 1). Each island was surveyed 60 times. To avoid possible systematic sampling bias owing to observer fatigue or weather conditions, a restricted random‐visitation ordering for survey transects was used (Mac Nally et al., 2002).

Lizard sampling

We used the line‐transect method (Jaeger, 1994) to assess lizard occupancy and abundance on 42 islands during two breeding seasons between 16 April and 30 July in 2007 and 2008. At each island, an observer walked each transect at a steady pace (10 m min⁻¹) searching the ground and tree boles with PANDA^® 10 × 40 roof prism binoculars (Yunnan North Optical Electron Group Co., Ltd., Kunming, Yunnan, China). Any lizards detected within 4 m of the transect lines were recorded. Once a lizard was detected, time spent in identification (if necessary) was excluded from the elapsed survey time; only individuals for whom confident identifications could be made (using criteria of Stebbins, 2003) were included in analyses (Germaine & Wakeling, 2001). We used GPS to record the length of each line transect (Table 1). Surveys were conducted from between 1 h after sunrise until 5 h after sunrise. Each island was surveyed 20 times. The order in which islands were surveyed and the direction in which the transects were walked were randomized and rotated each new census day to eliminate potential biases (Wang et al., 2009a). Censuses were not conducted during inclement weather such as strong winds or rains.

A possible caveat for nestedness analyses for lizard assemblage is that the species lists were probably incomplete. However, we consider our data set complete and reliable mainly for three reasons. First, there are only six lizard species in Zhejiang Province (Huang, 1990). Indeed, we found all the species on the sampled islands except Takydromus sexlineatus, whose distribution on the nearby mainland is still controversial. Second, all the lizards are common species, and thus these species can be surveyed and identified with confidence. Finally, considering the high survey frequencies (20 times each island), the species lists are complete and reliable.

Small mammal sampling

Small mammal species were surveyed by live‐trapping (Lynam & Billick, 1999) in autumn 2007 and spring 2008. Since data collection for small mammals was much more labour‐consuming than for birds and lizards, small mammals were surveyed only on 14 islands (Table 1). Small mammals were captured with Sherman traps (7.5 × 8.75 × 22.5 cm) and wire‐mesh cage traps (25 × 14 × 14 cm) baited with banana and peanut butter‐covered coconut pieces. Traps were arranged along lines at each island at 5‐m intervals. Each trapline was sampled for five consecutive nights per season. Captured mammals were weighed, individually marked and released. To increase the likelihood of encountering all species present, traplines were oriented to sample all available habitat types at each island.

In our analyses, it is important to consider whether the observed species distributions represent primarily patterns of occupancy rather than patterns of differential habitat use or foraging behaviour. To assess this, we calculated species turnover between successive sampling years as T = (J + E)/(S₁ + S₂), where J is the number of species recorded in the second but not in the first year, E the number of species found in the first but not in the second year, and S₁ and S₂ the total number of species during both years (Aguirre et al., 2003; Meyer & Kalko, 2008). Turnover rates can vary between 0 (no turnover) and 1 (complete turnover). Overall species turnover between sampling years was very low for lizards (0), small mammals (0.1) and birds (0.276). The low species turnover indicates that our sampling intensity is adequate to fully disclose each island’s residents, and that the islands are sufficiently isolated for most species such that the presence/absence data primarily represents species occurrence.

Habitat variables

For each island, we selected four habitat variables (area, isolation, habitat diversity and plant richness) that were commonly hypothesized to influence nestedness (Patterson & Brown, 1991; Wright et al., 1998; Honnay et al., 1999). Island size was measured by polar planimetry as the total island area in hectares. We used the distance from the nearest mainland as a measure of isolation (Meyer & Kalko, 2008; Wang et al., 2009b) and estimated it from a map at a scale of 1:10,000.

Habitat diversity was studied by noting the number of habitat types on each island (Table 1). The different habitat types could be easily identified as vegetation composition of the region was relatively simple. Photographs were taken as a record during the intensive surveys between April and November in 2007. Considering the requirements of birds, lizards and small mammals, all habitat types encountered on each island were identified and classified as follows: (1) conifer forest, (2) broadleaf forest, (3) coniferous‐broad mixed forests, (4) bamboo groves, (5) shrubs, (6) grasses and (7) farmlands (Appendix S4) (Zhang et al., 2008; Zhao et al., 2009).

The presence or absence of vascular plant species was recorded on all 46 islands (Table 1). The number of vascular plant species on each island was obtained by intensive searches between April and November in 2007. During the survey, we collected plant species along the same transects where birds, lizards and small mammals were surveyed. Some plant species were identified immediately, while others that could not be identified with confidence were collected and then identified according to Flora of Zhejiang (Zheng, 2005a).

Life‐history traits

For each taxon, we selected four commonly cited life‐history traits (Appendices S6, S7 and S8), i.e. body size, habitat specificity, geographical range size and area requirement, which reflect species extinction and immigration tendencies. Among the life‐history traits, geographical range size and area requirements are two key traits associated with extinction‐proneness (Davidar et al., 2002; Feeley et al., 2007), habitat specificity is associated with both extinction‐proneness (Fischer & Lindenmayer, 2005; Feeley et al., 2007) and habitat nestedness (Schouten et al., 2007; Ulrich et al., 2009), while body size is usually linked to dispersal ability (Schoener & Schoener, 1984; Cook & Quinn, 1995). Body size has been shown to be positively correlated with flotation and swimming endurance in reptiles and mammals (Schoener & Schoener, 1984), and with probability of arrival and frequency of occurrence of small mammals and birds on nearshore islands (Lomolino, 1986; Cook & Quinn, 1995). We used body length (mm) to represent body size, and the data were obtained from Huang (1990) and Zhuge (1990a,b). Habitat specificity was based on the incidence or number of habitats used by a given species (Feeley et al., 2007; Schouten et al., 2007), i.e. if a species used or occurred in only one habitat type, it was considered highly specific, and its habitat specificity was attributed a value of 1. Following Jones et al. (2003), geographic range size (km²) was obtained from the most recent available published species range maps by digitizing the area into a Geographic Information System (ArcView 3.2, Environmental Systems Research Institute Inc., CA, USA). Where no range maps were available, the minimum area convex polygon of published point data was calculated excluding areas of water. The data on geographic range size were obtained from Huang (1990), Zhuge (1990a,b) and Zheng (2005b). The minimum area requirement of each species was estimated as the area of the smallest island occupied by each species (Yiming et al., 1998; Davidar et al., 2002).

Statistical analyses

Quantification of nestedness

To quantify the level of nestedness, a variety of metrics have been proposed, each with different limitations (Wright et al., 1998; Ulrich et al., 2009). Among these metrics, the Nestedness Temperature Calculator (NTC; Atmar & Patterson, 1993) is one of the most widely used. However, the NTC has problems relating to the definition of the isocline of perfect order, the way of matrix reorganization, the robustness of the packing algorithm and choice of an appropriate null model (Fischer & Lindenmayer, 2002; Rodríguez‐Gironés & Santamaría, 2006). We used the recently developed program BINMATNEST (Rodríguez‐Gironés & Santamaría, 2006), which solves these problems. BINMATNEST uses a genetic algorithm to maximally pack the binary presence/absence matrix, calculates a uniquely defined isocline that minimizes nestedness temperature and provides three alternative null models to assess the statistical significance of matrix temperature. Following the authors’ recommendations, we used the null model 3 to randomly generate 1000 null communities and then evaluated whether assemblages of birds, lizards and small mammals were significantly nested. In null model 3, the probability of each cell in the species‐by‐site matrix being occupied is the average of its row and column occupancy probabilities. This model has also been shown to be associated with the smallest type I error. For all the other parameters, we used the recommended default settings of program BINMATNEST (Rodríguez‐Gironés & Santamaría, 2006).

The calculated temperature of a matrix is affected by its size and fill, but the probability that such a temperature is obtained by chance is not (Rodríguez‐Gironés & Santamaría, 2006). Therefore, we restricted our comparisons of matrices to interpretation of the P values from the Monte Carlo simulations and do not directly compare nestedness temperatures among bird, lizard and mammal matrices.

To ensure that our results were not biased owing to metric choice, we also examined the nestedness of assemblages using the various nestedness metrics in the program NESTEDNESS (Ulrich, 2006) and the metric NODF recently proposed by Almeida‐Neto et al. (2008). The results were qualitatively similar regardless of the choice of metric, so here we only report the results using the BINMATNEST method.

Determinants of nestedness

The order in which sites and species are sorted by BINMATNEST can be compared with numerous possible independent variables to evaluate their contributions to the nested pattern (Patterson & Atmar, 2000). For each taxon, to test the hypothesis that assemblages are nested as a result of hierarchical ordering among island characteristics, we performed Spearman rank correlations between the island ranks in the maximally packed matrix and ranked physical attributes of the islands (area, isolation, habitat heterogeneity and plant richness). Similarly, to examine whether variation in species life‐history traits explains nestedness for each taxon, we used Spearman rank correlations between the species ranks in the maximally packed matrix and ranked species traits (body size, habitat specificity, geographical range size and area requirement). Because collinearities exist among these variables (Appendices S9 and S10), we computed partial Spearman rank correlations to separate out the independent effect of these variables on nestedness (Shipley, 2000; Azeria & Kolasa, 2008; Frick et al., 2009). Spearman rank and partial rank correlation analyses were conducted with PROC CORR in SAS version 9.1 (SAS Institute, Cary, NC, USA).

In a maximally nested matrix, sites are largely ranked in decreasing order of species richness and species are ranked by decreasing incidence, such that species‐rich sites and high‐incidence species rank high (small digits), and vice versa. To compare the orders in which sites are sorted by BINMATNEST with habitat variables, we give the highest rank (1) to the islands with largest measures of area, isolation, habitat heterogeneity and plant richness (Table 1). Similarly, for species life‐history traits, we give the highest rank (1) to species with largest measures of body size, habitat specificity, geographical range size and area requirement (Appendices S6, S7 and S8). Islands with equivalent compositions or species with same incidences were given tied ranks.

To directly assess the importance of habitat specificity in generating nestedness, we first arranged species‐by‐habitat matrices for each taxon (Appendix S5) in a way similar to plant–animal mutualistic interaction matrices (Bascompte et al., 2003). We then used BINMATNEST and null model 3 to evaluate whether species‐by‐habitat matrices were significantly nested. If species‐by‐habitat matrix is nested, then habitat specificity may play a role in generating species nestedness.

We used the random placement model (Coleman, 1981; Coleman et al., 1982) to determine whether passive sampling from species abundance distributions was sufficient to explain nestedness patterns on the islands (Andrén, 1994; Cutler, 1994; Higgins et al., 2006). Under the random placement model, the number of species S_(α) to be found residing in a given region depends on the region’s relative area, , and the overall abundances n₁, n₂, …, n_s of the S species represented in C (cf. Coleman, 1981): . The hypothesis of random distribution should be rejected if more than one‐third of the points lie outside one standard deviation of the expected curve, and if the points are not evenly distributed about it (Coleman et al., 1982).

Among all bird species detected on islands, the owls which were nocturnal and waterbirds that were dependent on particular habitats, such as water bodies, were excluded from analyses (Davidar et al., 2002). Accordingly, a total of 93 bird species were analysed (see Appendix S1). All lizard species and mammal species found on the islands were analysed (Appendices S2 and S3).

Results

Nested patterns of birds, lizards and small mammals in the Thousand Island Lake

Nestedness temperatures for birds, lizards and small mammals were significantly lower than the means of randomly generated matrices under the null model 3, which indicates that the species compositions of all three taxa have nested structure (Table 2).

Table 2. Summary of results obtained from calculation of nestedness temperature T (°C) with BINMATNEST for birds, lizards and small mammals in the Thousand Island Lake, China. Given are observed matrix temperatures (T_obs), expected nestedness temperatures (T_exp) and Monte Carlo‐derived probabilities that the matrix was randomly generated under null model 3.

	Number of species	Number of islands	Fill (%)	T obs	T exp	P
Birds	93	42	30.54	18.29	48.23	< 0.001
Lizards	5	32	39.38	15.58	34.60	< 0.001
Small mammals	11	14	36.36	9.94	29.53	< 0.001

Mechanisms determining nestedness

The nested distribution patterns of birds, lizards and small mammals were not due to passive sampling (Fig. 2). For birds and small mammals, none of the observed data points lay within ±1 SD of the expected species‐area curves computed from the random placement models (Fig. 2a,c). For lizards, only six of the 32 observed data points lay within ±1 SD of the expected species‐area curve (Fig. 2b).

Figure 2

Comparison of observed data to expected values under the random placement models for (a) birds, (b) lizards and (c) small mammals in the Thousand Island Lake, China. Expected values (solid line) and associated standard deviations (±1 SD; dashed lines) are shown. Filled circles represent observed species richness.

There was no evidence that the nested distribution patterns of birds, lizards and small mammals resulted from selective colonization. Nestedness was not significantly correlated with island isolation or species body size as a measure of dispersal ability for birds, lizards and small mammals (Table 3).

Table 3. Relationships between rank orders of sites and species using BINMATNEST and orders of sites and species after rearranging the matrix according to each explanatory variable.

	Habitat variables				Species life‐history traits
	Island area (ha)	Isolation (m)	Number of habitat types (n)	Plant richness (n)	Body length (mm)	Habitat specificity	Geographical range size (km^2)	Area requirement (ha)
Birds	0.441 ^**	−0.212	0.086	0.215	−0.106	−0.297^**	0.147	−0.846^***
Lizards	0.034	−0.205	0.405 ^*	0.196	−0.300	−0.632	0.300	−0.051
Small mammals	0.659 ^*	−0.187	0.646 ^*	0.434	−0.025	−0.874^**	0.148	−0.251

• Values are partial Spearman’s rank correlations. Boldface indicates significant results. *P < 0.05, **P < 0.01, ***P < 0.001.

In contrast, the nested distribution patterns of birds and small mammals (but not lizards) were consistent with the selective extinction hypothesis. Nestedness was significantly positively correlated with island area, indicating that species found on small islands were a subset of those on large islands. Nestedness was significantly negatively correlated with species traits linked to extinction tendencies (i.e. habitat specificity and area requirement) (Table 3), such that species with high habitat specificity and small area requirements were subsets of species with low habitat specificity and large area requirements.

Habitat nestedness also played an important role in the development of species nestedness. Species nestedness was significantly positively correlated with habitat diversity (Table 3) and the habitat matrices estimated by BINMATNEST were highly nested (T_obs = 3.79 < T_exp = 29.03, P < 0.001) (Appendix S4). Furthermore, the species‐by‐habitat matrices (Appendix S5) estimated by BINMATNEST were significantly nested (all P_s < 0.05) for all three taxa, which provided further support for the habitat nestedness hypothesis.

Taxonomic differences in nestedness mechanisms

Correlations of nestedness with site variables and species life‐history traits varied among taxonomic groups (Table 3). For birds, island area was the most important site variable influencing nestedness, and habitat specificity and area requirement were the most important life‐history traits determining nestedness after controlling for other independent variables. For lizards, habitat heterogeneity was the most important site variable influencing nestedness, while none of the life‐history traits was significantly correlated with nestedness. For small mammals, island area and habitat heterogeneity were the most important site variables influencing nestedness, and habitat specificity was the single best life‐history trait influencing nestedness after controlling for other independent variables (Table 3). These results as a whole indicated that nested pattern of bird assemblage was driven by selective extinction, and lizard assemblage was caused by habitat nestedness, while nestedness of small mammals resulted from both selective extinction and habitat nestedness.

Discussion

Our study is among the first to evaluate whether different vertebrate groups follow the same mechanisms underlying nestedness at the same locations. To date, although a wide range of taxa have been shown to present nested patterns, very few studies have compared data on multiple taxonomic groups, especially among vertebrate taxa (Wright et al., 1998). Our study on birds, lizards and small mammals thus fills in a significant gap, contributes to the ecological generality of nestedness across a range of vertebrate taxa and furthers our understanding on nestedness mechanisms for multiple vertebrate taxa.

Mechanisms determining nestedness

We found that the community compositions of birds, lizards and small mammals were all significantly nested in the TIL. Our study system meets all the three key conditions thought to be important for the development of nested assemblages (Patterson & Brown, 1991; Wright et al., 1998). First, the islands have a common biogeographic history. Second, the islands shared an ancestral pool of species prior to fragmentation. Third, species inhabiting the islands are somewhat hierarchically ordered in terms of their niches and relatively complete species inventories had been conducted (Zhang et al., 2008; Wang et al., 2009a; Zhao et al., 2009).

The nested distribution patterns of birds, lizards and small mammals were not attributable to passive sampling. Several studies have shown that nestedness could arise from random samples of species differing in their relative abundances (Andrén, 1994; Cutler, 1994; Worthen et al., 1998; Higgins et al., 2006). Although it is stressed that the data should be tested for passive sampling prior to other hypotheses, the sampling effect has rarely been examined probably because of the difficulties of collecting species abundance data (Andrén, 1994; Cutler, 1994; Wright et al., 1998; but see Calmé & Desrochers, 1999). Our study indicated that passive sampling played little part in the development of nested assemblages, which is consistent with the results of Worthen et al. (1998).

The nested distribution patterns of birds, lizards and small mammals also did not appear to result from selective colonization, as nestedness was not correlated with island isolation or species body size as a measure of dispersal ability. Several factors may explain why these correlations were weak. For lizards and small mammals, colonizations probably have rarely occurred in our system due to their poor dispersal ability. In contrast, frequent dispersal of birds may dilute and obscure the effect of differential colonization in generating nestedness by reducing extinction (Cook & Quinn, 1995; Yiming et al., 1998; Fleishman et al., 2002). Another possible reason for the lack of a significant effect of isolation in relation to nestedness is that the biologically meaningful quantification of isolation is notoriously difficult (Lomolino, 1996). We know little, if anything, about the relative dispersal ability of most birds, lizards and small mammals, which may preclude strong inferences about selective colonization. In addition, our use of body size to estimate dispersal ability may inadvertently include signals of extinction risk; Brown (1971) and Patterson (1984) have demonstrated body size‐extinction effects in montane mammal systems. Finally, it could also be that our measure of island isolation – distance to the nearest mainland – is not the most appropriate one for quantifying isolation given the spatial configuration of the islands in our study system (Fig. 1).

The nested distribution patterns of birds and small mammals (but not lizards) are consistent with the selective extinction hypothesis, because nestedness was significantly correlated with island area and species traits that are linked to extinction tendencies (i.e. habitat specificity and area requirement). Selective extinction is a key driver of nestedness, particularly in land‐bridge archipelagos and in habitat fragments that are experiencing species loss or ‘relaxation’ (Patterson, 1987; Bolger et al., 1991; Yiming et al., 1998). In our case, selective extinction may cause nestedness because species with large area requirements, high habitat specificity and small population density have higher extinction risks and will go extinct first (Yiming et al., 1998; Schouten et al., 2007; Wang et al., 2009a).

Our results showed that habitat nestedness also played an important role in the development of species nestedness. Compared with other mechanisms such as colonization or extinction, habitat nestedness is the least questionable process to explain species nestedness because it ignores population dynamics or life history of species, but rather points to associations between species and their habitats (Calmé & Desrochers, 1999). However, despite its theoretical and practical interest, very few studies have explicitly examined the relationship between habitat nestedness and species nestedness. In accord with previous findings (Calmé & Desrochers, 1999; Hylander et al., 2005; Azeria et al., 2009), our study provides further evidence for the role of habitat nestedness in generating species nestedness.

Our study indicates that habitat types are nested in the TIL. One possible reason for habitat nestedness is that soil types upon which vegetation depends may be nested in our system. Soil types have been found to exhibit nested distributions in several other studies (Wright et al., 1998). Another possibility for habitat nestedness in our system may result from the underlying non‐random hydrologic processes, which are often changed in a stepwise fashion as area increases (Calmé & Desrochers, 1999). However, as we currently have no data on the soil and hydrology, their roles in generating habitat nestedness warrants further study.

Taxonomic differences in nestedness mechanisms

Although we found that birds, lizards and small mammals were all significantly nested, these taxa showed different patterns of correlations to habitat variables and species life‐history traits. For birds, island area, habitat specificity and area requirement were significantly correlated with nestedness. For lizards, habitat heterogeneity was the single best variable influencing nestedness. For small mammals, island area, habitat heterogeneity and habitat specificity were significantly correlated with nestedness. These results suggest that selective extinction was the main driver of nestedness for birds and small mammals, but not for lizards; and that habitat nestedness was the key determinant of nestedness for lizards and small mammals, but not for birds.

The taxonomic differences in mechanisms influencing nestedness were possibly as a result of the large variations in life‐history traits among taxa (Wright et al., 1998; Schouten et al., 2007). Differences in area requirements may be one possible reason for the taxonomic differences in the effects of selective extinction on nestedness. The average area requirement of lizards (0.586 ± 0.135 ha, Mean ± SE) (Appendix S7) was much smaller than that of birds (157.465 ± 43.275 ha) (Appendix S6) and small mammals (235.577 ± 157.069 ha) (Appendix S8), which may decrease the extinction probability of lizards (Wright et al., 1998; Yiming et al., 1998). Accordingly, selective extinction was the main driver of nestedness for birds and small mammals, but not for lizards.

Moreover, our results indicated that habitat nestedness was the key determinant of nestedness for lizards and small mammals, but not for birds. The taxonomic differences in the effects of habitat nestedness on species nestedness may possibly be attributable to the taxonomic differences in dispersal ability. The poor dispersal ability of lizards and small mammals may limit or even prevent their dispersal among islands (Cook & Quinn, 1995; Yiming et al., 1998), which may strengthen the relationships between habitat nestedness and species nestedness. In contrast, as the most mobile taxon (Cook & Quinn, 1995), birds can frequently disperse among islands, thus reducing the effects of habitat nestedness on species nestedness. The taxonomic differences in dispersal ability are further supported by our species turnover analyses, which showed that overall species turnover between sampling years was much higher for birds (0.276) than for lizards (0) and mammals (0.1).

Conservation implications

In the context of habitat fragmentation, understanding the mechanisms influencing nestedness and how they differ across taxonomic groups can have important implications for conservation and can be used to direct management efforts. Our study showed that the mechanisms influencing nestedness differed among birds, lizards and small mammals. Our results have two general conservation implications. First, as the three assemblages displayed limited congruence, we urge caution in using birds, lizards or small mammals as surrogates for each other in conservation planning. Second, conservation of species assemblages that have limited congruence will require a multiple criteria approach that incorporates taxonomic differences into reserve planning (Fleishman et al., 2002; Azeria et al., 2009).

Our study also provides some specific conservation guidelines for birds, lizards and small mammals in our system. For birds, we found that island area, habitat specificity and area requirement were significantly correlated with nestedness. Therefore, to preserve bird assemblages effectively, large islands and species with high habitat specificity and large area requirements should be given conservation priority. For lizards, as habitat heterogeneity was the single best variable influencing nestedness, we should pay more attention to islands with diverse habitats. For small mammals, island area, habitat heterogeneity and habitat specificity were significantly correlated with nestedness. Accordingly, large islands, islands with diverse habitats and species with high habitat specificity are important to conserve small mammal assemblages.

Acknowledgements

We thank Ermias Azeria, Christoph Meyer, Bruce Patterson and Fangliang He for valuable comments and English improvement on an earlier version of the manuscript. We are grateful to Jingcheng Zhang, Peng Li, Meng Zhang, Zhifeng Ding, Qingyang Zhao, Bo Sun, Longlong Zhang and Zhiyuan Hu for field assistance, and Chunan Forestry Bureau for permits necessary to conduct the research in the Thousand Island Lake. This study was supported by National Natural Science Foundation of China (No. 30670344), the China Postdoctoral Science Foundation (No. 20070411192), the Fundamental Research Funds for the Central Universities and Zhejiang Provincial Natural Science Foundation (Nos. R506054, Y507080).

Biosketch

Yanping Wang is a postdoctoral fellow at Zhejiang University. He works in the fields of biodiversity, conservation biology, community ecology, biological invasions and life‐history theory. Most of his work has focused on birds, amphibians and reptiles in fragmented landscapes.

Author contributions: Y.W. and P.D. conceived the ideas; Y.W., Y.B., M.Y. and G. X. collected the data; and Y.W. analysed the data and led the writing.

Editor: Bruce Patterson