INTRODUCTION
In his classic study on the sociology of epiphytes, Went (Reference WENT1940) described the canopy by lying, face-up, with field glasses on a stretcher, while field-assistants transcribed his spoken observations. Plants fallen to the ground and local tree-climbers provided voucher collections. Although these techniques seem primitive compared with those used by modern canopy scientists, much of our knowledge of epiphyte distribution is still based on this type of data (Johansson Reference JOHANSSON1974, Ochsner Reference OCHSNER1935, Schimper Reference SCHIMPER1888, van Oye Reference VAN OYE1924).
The innovation of rope-climbing and other canopy-access techniques such as walkways, platforms, cranes and hot-air balloons to gain access to the forest canopy resulted in a burgeoning of interest in canopy research (Dial & Tobin Reference DIAL and TOBIN1994, Gottsberger & Döring Reference GOTTSBERGER and DÖRING1995, Laman Reference LAMAN1995, Mitchell et al. Reference MITCHELL, SECOY and JACKSON2002, Moffett Reference MOFFETT1993, Nadkarni & Parker Reference NADKARNI and PARKER1994, Parker et al. Reference PARKER, SMITH and HOGAN1992, Perry Reference PERRY1978, Whiteacre Reference WHITEACRE1981). In contrast, quantitative methods to sample and analyse vascular epiphyte vegetation in the canopy have received little attention, with a few exceptions (Barker & Pinard Reference BARKER and PINARD2001, Bergstrom & Tweedie Reference BERGSTROM and TWEEDIE1998, Gradstein et al. Reference GRADSTEIN, HIETZ, LÜCKING, LÜCKING, SIPMAN, VESTER, WOLF and GARDETTE1996, Reference GRADSTEIN, NADKARNI, KRÖMER, HOLZ and NÖSKE2003; ter Steege & Cornelissen Reference TER STEEGE, CORNELISSEN and Glime1988). Comparisons among inventories and datasets are often hampered because investigators used different sample units (e.g. branch segments, branches, trees, ground surface plots) or sample sizes (e.g. number of trees, number and size of plots) (Wolf & Flamenco-S. Reference WOLF and FLAMENCO-S.2003). Ideally, comparisons should be performed on trees of the same species and of the same size (Johansson Reference JOHANSSON1974). Such situations, however, are difficult to find in multiple localities in species-rich tropical forests. Our objective in this paper is to develop a universal, quantitatively rigorous sample design that permits comparisons of epiphyte richness and abundance among forest stands of differing structure.
We recognize three reasons for the lack of universal sampling protocols for epiphytes. First, epiphytes grow in an intricately structured three-dimensional habitat within which discrete branches are patchily distributed. Even within the same tree, two branch segments differ with respect to their history, topology and microclimate. Second, epiphyte communities are very rich in species, particularly in wet tropical areas, so minimal areas (defined as the area that contains an adequate sample of species of regular occurrence; Westhoff & van der Maarel Reference WESTHOFF and Van Der MAAREL1973) can exceed the surface area of individual branches or even whole trees (Johansson Reference JOHANSSON1974, Wolf Reference WOLF1993a, Reference WOLFb). Third, the distribution of epiphytes is non-random (Laube & Zotz Reference LAUBE and ZOTZ2006), being heterogeneous at a range of spatial and temporal scales (Bennett Reference BENNETT1986, Nieder et al. Reference NIEDER, ENGWALD, KLAWUN and BARTHLOTT2000, Wolf Reference WOLF1994, Reference WOLF2005). Nearby trees may have more similar epiphyte vegetation than those far away (Bader et al. Reference BADER, VAN DUNNÉ and STUIVER2000, Barkman Reference BARKMAN1958, Cascante-Marín et al. Reference CASCANTE-MARÍN, WOLF, OOSTERMEIJER, NIJS, SANAHUJA and DURÁN-APUY2006, Catling et al. Reference CATLING, BROWNELL and LEFKOVITCH1986, Hietz & Hietz-Seifert Reference HIETZ and HIETZ-SEIFERT1995, Madison Reference MADISON1979).
For epiphytes in humid tropical forests, the combination of high richness (α-diversity) and high species turnover (β-diversity) causes the number of species and their densities to rise continuously with increasing sample area (Annaselvam & Parthasarathy Reference ANNASELVAM and PARTHASARATHY2001). In most epiphyte community studies, sample plots have been 0.1 to 1 ha, which comprise only a portion of the local epiphyte flora. For example, in a 170-ha study area in Ecuador, Gentry & Dodson (Reference GENTRY and DODSON1987a, Reference GENTRY and DODSONb) found 127 out of 227 vascular epiphyte species in a 0.1-ha plot. On relatively small Annona trees, Nieder & Zotz (Reference NIEDER and ZOTZ1998) concluded that 30 trees must be sampled to include the 10 most abundant species with 50% certainty. If the same holds for large trees, a substantial sampling effort is needed. This problem may be handled by using a relatively scale-independent diversity measure such as parametric Fisher's alpha (Magurran Reference MAGURRAN1988). However, alpha is underestimated in communities in which the spatial arrangement of individuals is strongly clustered (Schultea et al. Reference SCHULTEA, EGBERT, LANTING and HAWKINS2005). Moreover, because many epiphytes are clonal, it is difficult to count discrete individuals.
These obstacles have hampered the development of a standard sampling method and diversity measure for epiphyte inventories. Gradstein et al. (Reference GRADSTEIN, NADKARNI, KRÖMER, HOLZ and NÖSKE2003) developed a useful protocol, Rapid and Representative Analysis of Epiphyte Diversity (RRED). Based on empirical evidence they suggested that sampling eight large trees – standing well apart and being representative of local variation in bark and canopy structure – and all shrubs and treelets in a 20 × 20-m area around each tree should represent the species richness of vascular epiphytes in 1 ha of tropical forest. Trees are subdivided in five vertical zones according to Johansson (Reference JOHANSSON1974); presence–absence of epiphyte species is scored in each tree zone and on the shrubs and treelets.
Here, we present a protocol to sample and analyse vascular epiphytes that addresses these issues at the community level: Sampling of Vascular Epiphyte Richness and Abundance (SVERA). SVERA compares the richness and abundance of epiphytes in forest stands that differ in structure (as a result, for example, from disturbance). SVERA adds three critical activities to the RRED-protocol. First, SVERA estimates the true abundance of each species, in terms of species biomass. Although per cent branch cover provides a good estimate of biomass for non-vascular epiphytes (McCune Reference MCCUNE1990), it is not a good measure for vascular ones because many species extend beyond the branch surface area. Using the number of individuals is also suspect because large plants use more resources than small ones (Benavides et al. Reference BENAVIDES, DUQUE, DUIVENVOORDEN, VASCO and CALLEJAS2005, Zotz Reference ZOTZ2007a, Reference ZOTZb). Distinguishing true individuals (genets) in a mass of ramets is also difficult, especially in montane forests. Rare species that occur only once or twice, however, may be counted with confidence because the number of occurrences is likely the same as the number of individuals. SVERA uses the number of individuals in rare species to estimate total number of species in the inventories (SChao). In lowland forests, spatially separated stands of epiphytes have been counted (Sanford Reference SANFORD1968) instead of individuals, but this ignores the fact that epiphytes can have hidden connections within canopy soil or on undersides of branches. Estimating biomass may overcome the problem of counting individuals but is more time-consuming. Destructive sampling can incur long-term damage (Muir et al. Reference MUIR, NORMAN and SIKES2006). Nevertheless, dry weight from recorded plant sizes in a number of size-cohorts provides meaningful results (Benavides et al. Reference BENAVIDES, WOLF and DUIVENVOORDEN2006, Wolf Reference WOLF2005).
Second, SVERA incorporates tree basal area, tree density and trunk diameter frequency distributions so that epiphyte abundance data can be extrapolated to a forest area. The diameter frequency distribution can reveal the history and conservation status of the forest, and discern whether epiphyte biomass values at the plot level relate to the biomass per tree or to the density of (large) host trees. Third, SVERA samples a similar number of host trees of all sizes, which enables covariate comparison of epiphyte richness and biomass between inventories, and controls for differences in tree size. The covariate analysis assumes a strong positive relationship between epiphyte richness or biomass and tree size. This has been documented (Bartareau & Skull Reference BARTAREAU and SKULL1994, Burns & Dawson Reference BURNS and DAWSON2005, Dunn Reference DUNN2000, Flores-Palacios & Garcia-Franco Reference FLORES-PALACIOS and GARCIA-FRANCO2006, Hietz Reference HIETZ2005, Hietz-Seifert et al. Reference HIETZ-SEIFERT, HIETZ and GUEVARA1996, Hsu et al. Reference HSU, HORNG and KUO2002, Werner et al. Reference WERNER, HOMEIER and GRADSTEIN2005, Wolf Reference WOLF2005, Wolf & Konings Reference WOLF and KONINGS2001, Zimmerman & Olmsted Reference ZIMMERMAN and OLMSTED1992, Zotz & Vollrath Reference ZOTZ and VOLLRATH2003, Zotz et al. Reference ZOTZ, BERMEJO and DIETZ1999) independent of tree architecture, bark structure and microclimatic conditions. Larger trees offer more branch surface area and have been available for establishment by epiphytes over longer periods of time (Gradstein et al. Reference GRADSTEIN, NADKARNI, KRÖMER, HOLZ and NÖSKE2003). Exceptions, however, occur, e.g. in a Costa Rican oak forest more epiphytic bryophytes occur on young, c. 40-y-old oak trees than on much larger, c. 200-y-old ones (Holz & Gradstein Reference HOLZ and GRADSTEIN2005). Canopy closure and forest microclimate were the main factors determining epiphyte richness, not diameter or tree size. The different relationship between tree size and epiphyte species richness in this forest would be detected by SVERA. The positive correlation between vascular epiphyte abundance and tree size is smaller at more disturbed sites (Bartareau & Skull Reference BARTAREAU and SKULL1994, Wolf Reference WOLF2005), which suggests that in disturbed forests, all suitable niches are not yet occupied. Accordingly, the square of the Pearson product moment correlation coefficient (r2) has been used to assess whether a population is near its carrying capacity.
SVERA's primary objective is to provide a protocol for sampling of epiphyte richness and abundance. Since dispersal at least partially drives epiphyte community structure and leads to spatial aggregation (Barkman Reference BARKMAN1958, Cascante-Marín Reference CASCANTE-MARÍN2006, Zotz & Schultz Reference ZOTZ and SCHULTZ2008), spatial components must be incorporated in epiphyte community studies. Therefore, SVERA also records geographic position of trees in the forest and of forest stands in the landscape. To assess the performance of SVERA, we applied this technique on montane epiphyte communities in oak forests in Colombia and southern Mexico. We also tested for the occurrence of spatial dependence in these epiphyte communities.
METHODS
Study area
Epiphyte vegetation was analysed on oak trees in Colombian and several Mexican oak forests (2300–2600 m). The Colombian forest was located c. 25 km west of the capital Bogotá D.C. and the Mexican forests are all near the municipality of San Cristóbal de Las Casas (16°42′N, 92°37′W) in the highlands of Chiapas. The Colombian forest was dominated by Quercus humboldtii Bonpl., whereas the Mexican forests are dominated by a cohort of several oak species, mainly Q. crassifolia H. & B., Q. crispipilis Trel., Q. laurina H. & B. and Q. rugosa Nee. With an annual rainfall of approximately 750 mm (Colombia) and 1000 mm (Mexico), associated with a pronounced dry season, the climate in all oak forests is considerably drier than in wet tropical forest. There was no evidence of anthropogenic disturbance in the Colombian forest. In the Mexican forest, we sought the least disturbed forests, but found some evidence of past disturbances such as coppiced oaks and dead stumps (Wolf Reference WOLF2005).
Field sampling
To analyse forest structure, we randomly selected a 30 × 30-m plot in the forest interior and measured all individual trees (dbh > 5 cm), which we then identified to species. Epiphyte sampling was done on a plotless basis, i.e. based on the presence of epiphytes on individual trees. We used this approach to explore the relationship between epiphyte richness or abundance and the size of the host tree, using the whole tree as sampling unit. When within-tree distribution is of interest, the host tree may be subdivided in trunk and branch segments following Ochsner (Reference OCHSNER1935) or Johansson (Reference JOHANSSON1974). In RRED-analysis, trees varying in bark and canopy structure are deliberately included. In contrast, the SVERA protocol calls for the exclusion of epiphyte-poor trees such as ant trees or those with smooth-, hard- or sloughing barks because they mask the correlation between tree size and species richness or biomass.
Following the selection of host tree species, we randomly selected and sampled a first tree, after which we sampled its nearest neighbour, etc. Although this may slightly bias the sample (Clark & Evans Reference CLARK and EVANS1954), it requires less time than truly random choice and is practical for mapping purposes (Cottam & Curtis Reference COTTAM and CURTIS1956). We determined the geographic position of each tree by recording the distance and compass bearing to the previously sampled neighbour tree. Other measured parameters included tree height, trunk diameter (dbh) and the number of branching points (forks with a side-branch diameter > 5 cm). Field trials on climbed trees showed that those variables were estimated with little error and varied little amongst observers, as opposed to estimates for crown width, crown height and branch surface area. Tree height of the larger trees, which were climbed using rope-climbing techniques (Mitchell et al. Reference MITCHELL, SECOY and JACKSON2002), was estimated with a measuring rope. We sampled a total of 35 trees: 10 large trees (dbh > 30 cm) and 25 trees equally divided over five diameter cohorts (5–10, 10.1–15, 15.1–20, 20.1–25, 25.1–30 cm). The total number of sampled trees was similar as in RRED-analysis but with SVERA we tried to attain an even distribution of tree sizes (the covariate). Thus, smaller trees are well represented as recommended by Zotz (Reference ZOTZ2007a). To avoid duplicate sampling of branches we tracked branch systems with sketches.
For each tree, we non-destructively estimated the dry weight of all epiphyte species by assessing the number of leaves, fronds or rosettes, or the lengths thereof, depending on the growth form of the species. For bromeliads, the number of rosettes in three size classes (5–20, 20–50, > 50 cm) was often satisfactory. Bromeliad seedlings (< 5 cm) were not included because they are difficult to identify, have high mortality rates, and contribute little to total biomass. For ferns, orchids and species of Peperomia, the number of leaves provided a good quantitative measure. Cacti were best quantified by the length of their leaves; creeping ferns and aroids by the length of their rhizome or stem. For species that invest highly in their inflorescences (e.g. bromeliads) these organs were quantified separately. Dry weight was computed from the mean weight of 10 specimens per cohort. Unattached dead organic matter was removed from the plants before weighing. Vouchers of each new species were collected, including sterile plants that were subsequently cultivated to enable identification. We also assembled a field herbarium.
Analysis
We estimated epiphyte biomass on a ground area basis in two ways. First, we multiplied mean epiphyte biomass on individual trees in the diameter cohorts by the number of trees in that cohort per hectare. Second, we extrapolated epiphyte biomass of all trees per diameter cohort. Note that this does not consider the dispersion around the fitted regression line and is only possible when the regression yields a significant relationship. When the size of the 10 trees in the largest size cohort (dbh > 30 cm) varies greatly, the second approach is preferable.
We used Chao's non-parametric diversity estimator to estimate the overall richness of the samples (Chao Reference CHAO1984). Her estimator (SChao1) enables a comparison between unequal-sized samples and has a relatively low sensitivity to varying species richness (Colwell & Coddington Reference COLWELL and CODDINGTON1994). For clonal epiphytes, SChao1 is particularly attractive because it does not require counts of the number of individuals in those common species that form a mass of ramets. If no singletons or doubletons are present in the inventory, a version of the estimator may be computed using EstimateS.
We evaluated and compared epiphyte richness and biomass between sites with scaled linear regressions on trees that varied in size. To visualize the regression lines, we combined dbh, tree height (Height) and the number of branching points (BP) into a single value (Tree Size). Tree Size, computed as Standardized (dbh × Height) + Standardized (BP), is a dimensionless estimate of the total bark surface area. We achieved standardization by subtracting the population mean from an individual raw score and then dividing the difference by the population standard deviation.
The regression is characterized by its slope (the regression coefficient), its elevation (the y-intercept) and an error term. When the slopes between two sites are the same, we inferred that relationships between tree size and epiphyte richness or biomass were similar. The elevation of the regression is an indicator of species richness or biomass at a site, i.e. the theoretical number of species or biomass on a tree with size 0. The advantage of this scaled approach is that elevation can be analysed statistically in an analysis of covariance (ANCOVA) that controls for Tree Size. Moreover, this analysis does not require a representative sample of the local epiphyte flora.
In ANCOVA, we compared richness or biomass values between sites pair-wise with an a priori contrast analysis (t-test). ANCOVA assumes that the slopes are equal at all sites (Sokal & Rohlf Reference SOKAL and ROHLF1981). To test this homogeneity of covariate regression coefficient assumption, we determined the influence of the interaction term Tree Size × Site on a model with the predictor variables Tree Size and Site (Keppel Reference KEPPEL1991). A significant influence of the interaction term signifies that the null hypothesis of equal slopes must be rejected. Because ANCOVA remains robust when the difference in slope is not large and group sizes are equal (Keppel Reference KEPPEL1991), we used P = 0.001 as threshold. Since hemi-epiphytes may differ from holo-epiphytes in terms of dependence on tree size (Benavides et al. Reference BENAVIDES, WOLF and DUIVENVOORDEN2006, Burns & Dawson Reference BURNS and DAWSON2005), the two groups may be analysed separately. To attain homogeneity of variance and a normal distribution, it may be necessary to transform the dependent variable.
When the homogeneity assumption test shows that the slopes are not parallel, homogeneity may be achieved by elimination of sites on an individual basis. Visual inspections of the regression lines can help to select the most aberrant sites. Although omitted sites are not included in the following ANCOVA, they are of interest for ecological reasons. For example, unusually gentle regression slopes may relate to little canopy closure and the establishment of canopy species in the understorey (Holz & Gradstein Reference HOLZ and GRADSTEIN2005).
Two non-parallel regression lines will inevitably cross at some point. Differences between elevations will increase with distance from the point of intersection. As an alternative to ANCOVA, in these cases the Johnson–Neyman procedure, testing at what distance from the intersection the elevations of two regression lines are significantly different, may be applied (Aiken & West Reference AIKEN and WEST1991, D'Alonzo Reference D'ALONZO2004, Huitema Reference HUITEMA1980, Johnson & Neyman Reference JOHNSON and NEYMAN1936). Unfortunately, this procedure is not a standard feature of any statistical software package. We used syntax from the Statistical Package for the Social Sciences (SPSS) Online Technical Support webpage (based on Aiken & West Reference AIKEN and WEST1991).
For epiphyte spatial analysis (optional in SVERA) we assigned spatial descriptors of sampled trees and sites. Tree maps were generated based on compass readings and distances between individual trees (Reichard Reference REICHARD2002). Site position was determined with the Global Positioning System (GPS). The x-y coordinates (trees) and GPS readings (sites) allowed for computing a Euclidean distance matrix, by applying the Macintosh R-package for multivariate and spatial analysis. We explored for spatial dependence in the data by correlating the geographic distance matrix with a species distance matrix in a Mantel test (Legendre & Legendre Reference LEGENDRE and LEGENDRE1998). For the species distance matrix we used the probabilistic coefficient of Raup and Crick as the similarity index. This coefficient is a measure of how close the arrangement of species over the samples is to a random assignment. The Mantel test yields a correlation coefficient and a probability. For the correlation, Pearson's product moment linear correlation coefficient (i.e. Mantel's r), or Spearman's r may be used. We visualized spatial dependence in a correlogram, depicting the correlation coefficient against geographical distance. To this purpose, the geographic distance matrix was transformed into a distance class matrix on the basis of Sturge's rule (number of classes = 1 + 3.3log(n), where n = number of forest plots or trees).
Spatial variation may be attributed to spatially structured environmental gradients or to other factors that may have a spatial component, such as competition or dispersal. A partial Mantel test reveals whether the spatial dependence is sustained when the variation due to environmental gradients has been accounted for. For this test, we entered a Euclidean environmental distance matrix as a covariable. Canonical correspondence analysis (CANOCO, ter Braak Reference TER BRAAK1986, Reference TER BRAAK1988) was also applied (Borcard et al. Reference BORCARD, LEGENDRE and DRAPEAU1992), which shows what proportion of the variation in the data is explained by the geographic position of the sites versus the environment. It also allows Monte Carlo significance testing of the relationship of environmental and spatial variables to community structure (Jongman et al. Reference JONGMAN, TER BRAAK and VAN TONGEREN1987). Environmental variables include altitude, climate, forest management, forest structure and geographical position. We first submitted the explanatory variables to the CANOCO procedure of forward selection of variables. We retained only those variables that contributed a significant amount (P < 0.05), using Monte Carlo tests based on 1000 permutations, and evaluated the statistical significance of the first canonical axes with Monte Carlo tests. Geographical coordinates may be used to derive two types of spatial descriptors of the sites (Borcard & Legendre Reference BORCARD and LEGENDRE2002, Borcard et al. Reference BORCARD, LEGENDRE and DRAPEAU1992, Reference BORCARD, LEGENDRE, AVOIS-JACQUET and TUOMISTO2004; Legendre & Legendre Reference LEGENDRE and LEGENDRE1998). Traditionally, the terms of a cubic trend-surface equation that is derived from the geographical coordinates of the sites have been used. A new method is based on the principal coordinates of neighbour matrices (PCNM, Borcard & Legendre Reference BORCARD and LEGENDRE2002, Dray et al. Reference DRAY, LEGENDRE and PERES-NETO2006). The spatial structure is described by the eigenvectors – using positive eigenvalues only – of a principal coordinate analysis of a truncated Euclidean distance matrix or neighbour matrix. For epiphytes, cubic trend-surface analysis and PCNM analysis yield similar results (Wolf Reference WOLF2005).
Some ecological questions require the evaluation of the epiphyte community on the smaller spatial scale of a host tree. Environmental variables entered in the analysis depend on the research question and may include tree species, bark type, deciduousness, tree architecture, tree growth rate, animal activity, canopy soil accumulation, abundance of non-vascular epiphytes and distance to forest edge. The partitioning of the variation between environmental and spatial variables is complex because epiphyte assemblages among large trees are generally more similar than those on small trees. Since the probability that a large tree grows close to a small tree is higher than for two large trees, trees at larger distances are thus likely to be more similar in their epiphyte community. The resulting spatial pattern is not due to historical or biological processes and therefore of little interest. Hence, for autocorrelation analysis it is advisable to exclude the smaller trees. To eliminate further tree-size-related bias, the abundance of each epiphyte species on a single tree should be estimated relative to the total epiphyte biomass on that tree.
For the analysis of species richness and abundance, we used SPSS for Windows version 11.0, Microsoft Excel and EstimateS version 8.0 (R.K. Colwell, persistent URL <purl.oclc.org/estimates>). For the optional multivariate analysis of the distribution of epiphytes in relation to environmental and spatial variables we used CANOCO for Windows version 4.0 (ter Braak Reference TER BRAAK1988), PACE (Reichard Reference REICHARD2002), The R package version 4.0 d6 and Spacemaker2 (the latter two MacOS programmes are available on http://www.bio.umontreal.ca/legendre/; Windows versions are in development).
RESULTS
It took two persons approximately 20 d of fieldwork to complete an inventory. SVERA yielded results on two scales of observation: on the forest scale (Figure 1, Appendix 1) and on the host-tree scale (Table 1, Appendix 2). The raw data may be downloaded from the Canopy Databank (http://canopy.evergreen.edu/research_databank.asp?Id=2). We encourage other researchers to also make their data available online.
Figure 1. Number of oak trees per ha in various trunk diameter classes (dbh) at the sites, based on 900-m2 inventories. Col. = Colombia; Mex. = Mexico.
Table 1. Epiphyte species richness and biomass (dry weight) at the study sites. The estimated total number of species in the forest (SChao1) is calculated from the number of observed species and the number of singletons and doubletons in the sample. Paired SChao1 differences between sites are always significant (t-test, P < 0.05), except between Basom-1 and San Antonio and between Basom-1 and Chilil. Epiphyte biomass per unit stand area is calculated from the average tree load in several tree trunk diameter frequency classes. A complete species list is given in Appendix 2.
Epiphyte species richness and biomass
In the Colombian oak forest, we encountered 14 epiphyte species on the 35 sampled oak trees, and 23–35 species in the Mexican oak forests sample areas (Appendix 2, Table 1). The Colombian forest shared only two epiphyte species with the Mexican ones, but at the family level the forests were similar. Bromeliads dominated in all sites where they always comprised more than c. 90% of total epiphytic biomass. In Colombia, epiphyte biomass on the 35 host trees (87.1 kg) fell within the range of that observed in Mexico (74.7–243.9 kg).
At all sites, there was a strong linear dependence (P < 0.001) of epiphyte species richness and biomass on tree size (Tree Size) of the host tree (Figures 2 and 3). Epiphyte biomass also tightly correlated with tree size (P < 0.001), using trunk diameter as the tree size variable. Therefore, the regressions at each site may be used to estimate epiphyte biomass on oaks per unit stand area. In Colombia, biomass was lower than in any of the Mexican forests (Table 1), both on the basis of single tree or diameter-size class calculations. The latter approach yielded slightly higher values per ha, but differences were not significant (paired-samples t-test, P = 0.10).
Figure 2. Scatterplot illustrating the relation between tree size and epiphyte species richness. Tree Size = Standardized(Height × dbh) + Standardized(Number of branching points). El Chivero: √richness = 0.4(Tree Size) + 2.9, r2 = 0.62, P < 0.001; La Florecilla: √richness = 0.3(Tree Size) + 3.1, r2 = 0.60, P < 0.001; Chilil: √richness = 0.3(Tree Size) + 2.7, r2 = 0.52, P < 0.001; Basom-2: √richness = 0.2(Tree Size) + 2.7, r2 = 0.62, P < 0.001; Basom-1: √richness = 0.2(Tree Size) + 2.8, r2 = 0.61, P < 0.001; San Antonio: √richness = 0.2(Tree Size) + 2.6, r2 = 0.54, P < 0.001; Macanal, COL: √richness = 0.2(Tree Size) + 2.3, r2 = 0.54, P < 0.001. All sites in Mexico, except Macanal, COL, which is in Colombia.
Figure 3. Scatterplot illustrating the relation between tree size and epiphyte biomass. Tree Size = Standardized(Height × dbh) + Standardized(Number of branching points). Dotted lines have a significantly different slope from the solid lines (ANCOVA, Tree Size × Site, P < 0.001). For La Florecilla, the line has a significantly higher elevation than the cohort of the homogeneous slopes (solid lines) at Tree Size > −1.30 and for Basom-2 at Tree Size > 1.08 (Johnson–Neyman procedure at P = 0.05). La Florecilla: √g biomass = 24.9(Tree Size) + 66.8, r2 = 0.93, P < 0.001; Basom-2: √g biomass = 18.3(Tree Size) + 43.8, r2 = 0.95, P < 0.001; Basom-1: √g biomass = 14.0(Tree Size) + 45.6, r2 = 0.86, P < 0.001; Macanal, COL: √g biomass = 14.5(Tree Size) + 36.0, r2 = 0.65, P < 0.001; El Chivero: √g biomass = 11.1(Tree Size) + 47.4, r2 = 0.81, P < 0.001; San Antonio: √g biomass = 10.8(Tree Size) + 47.5, r2 = 0.68, P < 0.001; Chilil: √g biomass = 11.3(Tree Size) + 37.1, r2 = 0.58, P < 0.001. All sites in Mexico, except Macanal, COL, which is in Colombia.
For species richness, visual inspection of the regression lines (Figure 2) indicated that all slopes were roughly parallel, which was corroborated by the homogeneity assumption test. In contrast, the ANCOVA showed significant differences in the elevations of the slopes (Table 2). The trees at the Colombian site supported significantly (P < 0.05) fewer species than in any of the Mexican sites, except at San Antonio. Species richness at La Florecilla was significantly higher than at most other sites (except at Basom-1 and El Chivero).
Table 2. Pairwise comparisons between the study sites of epiphyte species richness and epiphyte biomass, based on ANCOVA (Bonferroni-adjusted) that controlled for Tree Size. Cell values are the differences in the adjusted means. Richness differences are shown in the upper triangle (row subtracted from column) and biomass differences in the lower triangle (column subtracted from row). Site effect on species richness: F6,237 = 8.80, P < 0.001, and on epiphyte biomass: F4,169 = 4.62, P < 0.01. Species richness and biomass (kg dry weight) were square root-transformed. Note that the biomass values of La Florecilla and Basom-2 were not compared (excl.) with the other sites because both sites showed deviating regression coefficients. Significant differences in bold: *0.05 < P < 0.01, **0.01 < P < 0.001, ***P < 0.001.
For biomass (Figure 3), the homogeneity assumption test showed that the regression coefficients were heterogeneous (ANCOVA, Tree Size × Site F6,231 = 15.7, P < 0.001). Exclusion of the two sites with the most aberrant slopes (La Florecilla and Basom-2) generated a remaining cohort of five parallel slopes (ANCOVA, Tree Size × Site, F4,165 = 1.63, P > 0.05). Contrast analysis in the subsequent ANCOVA showed that the elevation of the regression line at the Colombian site was significantly (P < 0.05) lower than at San Antonio and El Chivero in Mexico (Table 2).
Visual inspection of the aberrant slopes at La Florecilla and Basom-2 suggested that the biomass at these sites was considerably higher than in any of the other sites (Figure 3). The Johnson–Neyman procedure confirmed that the elevation at La Florecilla was significantly higher than the averaged elevations of the sites in the cohort with the homogeneous slopes for most trees (Tree Size > −1.30, P = 0.05). The elevation at La Florecilla was also higher than at Basom-2 for most trees (Tree Size > −1.90). Basom-2 showed no significantly different elevation from the slopes in the cohort for [−1.46 < Tree Size < 1.08].
Epiphyte spatial dependence
A Mantel test for spatial dependence on the 20 largest sampled host trees in each forest site revealed that only at El Chivero was there a significant linear correlation between the distance between trees and the composition of the epiphyte community (Spearman r = 0.24, P < 0.05). The correlation remained significant after controlling for environmental variables in a partial Mantel test. Environmental variables entered were tree dbh, height, number of branching points and species of host tree. Correlation between tree distance and epiphyte similarity switched from positive in nearby trees to significantly negative at the largest distance (Figure 4). Spatial descriptors explained more variation (26%) than the above-mentioned environmental variables (16%). Approximately 12% of this variation was shared by environmental and spatial variables, leaving 14% being explained by space alone (Figure 5).
Figure 4. Mantel correlogram of epiphyte vegetation on large trees at the El Chivero (Spearman r = 0.27, P < 0.05). The species similarity is based on the Raup and Crick probabilistic coefficient, using the biomass of each species relative to the total epiphyte biomass on that tree. Filled symbols indicate a significant value of Spearman r (P < 0.05).
Figure 5. Variation partitioning for the distribution of epiphytes over large trees at El Chivero between environmental and spatial variables, following Borcard et al. (Reference BORCARD, LEGENDRE and DRAPEAU1992). For the spatial descriptors of the trees, the terms of a cubic trend-surface equation that is derived from the x-y coordinates of the trees were used. The environmental and spatial explanatory variables were first submitted to the CANOCO procedure of forward selection of variables in two separate runs. Only the variables (2) that contributed a significant amount (Monte Carlo test, P < 0.05) were retained (i.e. the term of y2 and trunk dbh). The first canonical axes were all significant (Monte Carlo test, P < 0.05).
DISCUSSION
Sampling
With approximately 20 d of fieldwork to complete an inventory in montane oak forest, SVERA takes about twice as long as RRED analysis. However, in addition to species richness assessed by RRED, SVERA provides data on forest structure, position and size of trees, and abundance of each epiphyte species. In structurally more complex large-tree lowland rain forests more time will be required. Larger trees signify that a representative sample to assess forest structure must be larger than the 30 × 30-m plot we used. Similarly, more large trees must be sampled for epiphytes, e.g. by adding more diameter size-cohorts at the upper end, to avoid the presence of outlier trees on the independent axis of the regression. Visual inspection of the regression lines will guide the number of trees that must be sampled. In forests with a higher diversity of host trees, more trees than the 35 used in this study will probably have to be sampled. On the basis of previous studies we are confident that in most cases a positive correlation may be found, even in structurally complex lowland rain forest (Zotz & Schultz Reference ZOTZ and SCHULTZ2008).
Forest structure
Tree height and diameter data may be used in two different ways. First, the diameter frequency distributions allow for estimation of total epiphyte biomass in a forest stand. Second, these data allow for comparisons of the structure of the forest between sites. For example, our data showed that the Colombian oak forest fell within the range of the Mexican stands in terms of height and total basal area. In all forests, oaks dominated (except at Basom-2 where pines contributed most to total basal area, which may have reflected selective logging and other anthropogenic disturbances, González-Espinosa et al. Reference GONZÁLEZ-ESPINOSA, QUINTANA-ASCENCIO, RAMÍREZ-MARCIAL and GAYATÁN-GUZMÁN1991).
Epiphyte biomass and richness
In epiphyte studies, biomass is most often calculated either on a per tree or per unit stand area basis (reviewed in Wolf & Flamenco-S. Reference WOLF and FLAMENCO-S.2003). SVERA provides both types of information. Per unit stand area, however, SVERA only yields a minimum value for epiphyte biomass since epiphyte-poor host tree species are not sampled. Accidental individuals on ‘hostile’ trees (e.g. pines in Mexico) contribute little to epiphyte richness and biomass in the forest and they may well correspond to a sink population that for its survival depends on a continuous supply of seeds from ‘epiphyte-friendly’ trees. Nevertheless, especially in areas where the density of epiphyte-poor trees is high, SVERA will underestimate total biomass on an areal basis.
Biomass values per unit stand area calculated on the basis of diameter-size class were slightly higher compared with those based on single tree occupancy. We attribute the higher values to larger sample trees having less influence on the regression slope than on the calculated mean biomass in the large-diameter-size cohort.
In the case study, epiphyte biomass per unit stand area in the Colombian site was smaller than in most of the Mexican sites. This is unexpected since the epiphyte biomass on the 35 sampled trees (87.1 kg) is comparable to that in Mexico, whereas the number of oak trees ha−1, especially of large oaks, is higher at the Colombian site. The apparent discrepancy may be because larger trees were sampled in Colombia (despite an effort to sample trees of identical sizes). Epiphyte biomass per unit stand area results from the biomass per tree, which in turn depends on tree size, and number of trees therein. This indicates that epiphyte biomass data on a per tree basis are more informative than per unit stand area.
To factor out a tree size effect, SVERA-analyses rely on the presence of a linear dependence of biomass or richness on tree size. To eliminate spurious correlations due to outliers, SVERA distributes trees evenly on the independent size axis. At all sites, strong positive correlations were found between the square root of biomass and richness with relative tree size.
ANCOVAs of epiphyte biomass revealed that the regression coefficients of the covariate Tree Size were uniform at five of the seven sites. In five sites, the same linear relationship existed between tree size and epiphyte biomass. In the two other sites (La Florecilla, Basom-2), slopes were steeper and large trees supported significantly more biomass than smaller ones. At La Florecilla the unusually steep slope may have been due to dominance of the clonal Tillandsia vicentina, which has dense rosette packing.
ANCOVA also revealed that epiphyte biomass at the Colombian site, after controlling for tree size, was significantly lower than at the Mexican sites. Thus, trees of the same size supported less biomass in Colombia than in Mexico. In contrast, if we would have considered the biomass on the 35 trees alone, i.e. without controlling for tree size, we would have concluded that biomass between sites was similar. The covariate approach thus gave valuable new insights.
Similarly, the covariate approach showed that species richness was significantly higher at the least-disturbed site (La Florecilla) than at San Antonio, Basom-2 and Chilil. This result contradicted simple species counts, which showed 27 species at La Florecilla and 29, 24 and 35 species at the other sites, respectively. ANCOVA also showed that the number of epiphyte species in the Colombian forest was significantly smaller than in any of the other sites, except at San Antonio. We might have deduced this from epiphyte species lists, but in the latter case, the conclusion could have been challenged due to the likelihood of unequal sample sizes and the lack of statistical analysis. A final advantage of the covariate approach was that differences between sites could be attributed to the number of species in the background (i.e. slope elevation) instead of tree size-richness relationship (i.e. regression coefficients).
Epiphyte spatial dependence
SVERA can test for autocorrelation, using data on the geographical position and the epiphyte floristic composition of forest stands in the landscape and of host trees in the forest (Wolf Reference WOLF2005). In the latter study, a Mantel test showed a strong spatial dependence of epiphyte inventories (16) within an approximately 100-km2 area. Spatial variables explained more variance than environmental variables of the forests that were distributed over a gradient of anthropogenic disturbance. We used SVERA-analysis to test for spatial autocorrelation at the tree level, which was significant at El Chivero. The correlation remained significant after controlling for environmental tree variables in a partial Mantel test. Moreover, space alone explained more variation than the tree variables, justifying SVERA's inclusion of the geographic position of the trees as an exploratory variable.
CONCLUSIONS
Our case study showed that in comparison with conventional methods, SVERA is a useful protocol for community ecologists. The forest structure data helped to evaluate the logging history of the forest, permitted the calculation of a minimum estimate of epiphyte biomass per unit stand area, and quantitatively described the factors affecting distribution of epiphytes in the landscape. The plotless epiphyte inventories on trees facilitated a statistical comparison of epiphyte biomass and species richness between sites, while controlling for differences in tree size. These comparisons yielded different results from more conventional plot-based sampling. To evaluate the distribution of epiphytes in the landscape and in the forest, optional in SVERA, the explicit inclusion of spatial descriptors was useful. The variation partitioning was based on CANOCO. It is best viewed as an exploratory type of data analysis that generates hypotheses that may be tested experimentally (Jongman et al. Reference JONGMAN, TER BRAAK and VAN TONGEREN1987).
This study took place in relatively simply-structured montane oak forests, and forests that are structurally complex and diverse need testing with this technique. A study in Panama indicated the potential value of SVERA, because epiphyte assemblages were significantly influenced by tree size, tree species and space (Zotz & Schultz Reference ZOTZ and SCHULTZ2008). We encourage other epiphyte community ecology researchers to use this protocol to ease comparisons of data that are difficult to collect.
ACKNOWLEDGEMENTS
We thank Teresa Santiago-V., Mariana T. Toledo-A. and Henry E. Castañeda-O. for their assistance during fieldwork in Mexico. The Colombian data were kindly provided by Diego Higuera and Eliana Martínez of the Corporación Sentido Natural. The support of the Stichting Het Kronendak to the Miss Dr. Jakoba Ruinen Chair Canopy Sciences, held by the first author, is gratefully acknowledged. Funds were also provided by El Colegio de la Frontera Sur (ECOSUR) and the Comisión Nacional para el Conocimiento y Uso de la Biodiversidad (CONABIO) in Mexico, grants B060 and L050. Support was also provided by the German Research Foundation (grants to SRGr), a National Science Foundation (NSF) grant to NMN (DEB-05–42130), and by the Helen R. Whiteley Center at the Froday Harbor Laboratories, University of Washington. Database support came from a NSF grant to NMN and Judith Cushing (BDI 04–17311).
Appendix 1. Characteristics of the study sites. Forest height is the average of the five tallest trees in the inventory. Tree density corresponds to individuals with trunk dbh > 5 cm. NP means not present. Large oaks refer to oaks with dbh > 45 cm. Rainfall in Mexico is the average amount (1978–1995) at San Cristóbal de Las Casas (Wolf Reference WOLF2005).
Appendix 2. List of epiphyte species recorded on 35 host trees at the study sites.
