Cost-effectiveness of plant and animal biodiversity indicators in tropical forest and agroforest habitats

Authors

  • Michael Kessler,

    Corresponding author
    1. Albrecht-von-Haller-Institute of Plant Sciences, University of Göttingen, Untere Karspüle 2, D-37073 Göttingen, Germany
    2. Systematic Botany, University of Zürich, Zollikerstrasse 107, CH-8008 Zürich, Switzerland
      Correspondence author. E-mail: michael.kessler@systbot.uzh.ch
    Search for more papers by this author
  • Stefan Abrahamczyk,

    1. Albrecht-von-Haller-Institute of Plant Sciences, University of Göttingen, Untere Karspüle 2, D-37073 Göttingen, Germany
    2. Systematic Botany, University of Zürich, Zollikerstrasse 107, CH-8008 Zürich, Switzerland
    3. Agroecology, University of Göttingen, Grisebachstr. 6, D-37077 Göttingen, Germany
    Search for more papers by this author
  • Merijn Bos,

    1. Agroecology, University of Göttingen, Grisebachstr. 6, D-37077 Göttingen, Germany
    2. Louis Bolk Institute, Hoofdstraat 24, 3972 LA Driebergen, the Netherlands
    Search for more papers by this author
  • Damayanti Buchori,

    1. Faculty of Biology, Bogor Agricultural University, Jalan Padjajaran, 16144 Bogor, Indonesia
    Search for more papers by this author
  • Dadang Dwi Putra,

    1. Celebes Bird Club, c/o Balai Penelitian dan Pengembangan Zoologi, Puslitbang Biologi – LIPI, Jl. Raya Bogor Jakarta Km 46, Cibinong 16911, Indonesia
    Search for more papers by this author
  • S. Robbert Gradstein,

    1. Albrecht-von-Haller-Institute of Plant Sciences, University of Göttingen, Untere Karspüle 2, D-37073 Göttingen, Germany
    Search for more papers by this author
  • Patrick Höhn,

    1. Agroecology, University of Göttingen, Grisebachstr. 6, D-37077 Göttingen, Germany
    Search for more papers by this author
  • Jürgen Kluge,

    1. Albrecht-von-Haller-Institute of Plant Sciences, University of Göttingen, Untere Karspüle 2, D-37073 Göttingen, Germany
    2. Faculty of Geography, University of Marburg, Deutschhausstraße 10, 35032 Marburg, Germany
    Search for more papers by this author
  • Friederike Orend,

    1. Albrecht-von-Haller-Institute of Plant Sciences, University of Göttingen, Untere Karspüle 2, D-37073 Göttingen, Germany
    Search for more papers by this author
  • Ramadhaniel Pitopang,

    1. Faculty of Agriculture, Tadulako University, Palu, Indonesia
    Search for more papers by this author
  • Shahabuddin Saleh,

    1. Faculty of Agriculture, Tadulako University, Palu, Indonesia
    Search for more papers by this author
  • Christian H. Schulze,

    1. Department of Animal Biodiversity, Faculty of Life Sciences, University of Vienna, Rennweg 14, A-1030 Vienna, Austria
    Search for more papers by this author
  • Simone G. Sporn,

    1. Albrecht-von-Haller-Institute of Plant Sciences, University of Göttingen, Untere Karspüle 2, D-37073 Göttingen, Germany
    Search for more papers by this author
  • Ingolf Steffan-Dewenter,

    1. Agroecology, University of Göttingen, Grisebachstr. 6, D-37077 Göttingen, Germany
    2. Department of Animal Ecology I, University of Bayreuth, Universitätsstrasse 30, 95440 Bayreuth, Germany
    Search for more papers by this author
  • Sri S. Tjitrosoedirdjo,

    1. Faculty of Biology, Bogor Agricultural University, Jalan Padjajaran, 16144 Bogor, Indonesia
    Search for more papers by this author
  • Teja Tscharntke

    1. Agroecology, University of Göttingen, Grisebachstr. 6, D-37077 Göttingen, Germany
    Search for more papers by this author

Correspondence author. E-mail: michael.kessler@systbot.uzh.ch

Summary

1. Biodiversity data are needed for conservation and management of tropical habitats, but the high diversity of these ecosystems makes comprehensive surveys prohibitively expensive and indicator taxa reflecting the biodiversity patterns of other taxa are frequently used. Few studies have produced the necessary comprehensive data sets to assess the quality of the indicator groups, however, and only one previous study has considered the monetary costs involved in sampling them.

2. We surveyed four plant groups (herbs, liverworts, trees, lianas) and eight animal groups (ants, canopy and dung beetles, birds, butterflies, bees, wasps and the parasitoids of the latter two) in 15 plots of 50 × 50 m2 each, representing undisturbed rainforest and two types of cacao agroforest in Sulawesi, Indonesia. We calculated three biodiversity measures (α and β diversity; percentage of species indicative of habitat conditions), built simple and multiple regression models among species groups (single groups, combinations of 2–11 groups, averaged relative diversity of all 12 groups), and related these to three measures of survey cost (absolute costs and two approaches correcting for different sampling intensities).

3. Determination coefficients (R2 values) of diversity patterns between single study groups were generally low (<0·25), while the consideration of several study groups increased R2 values to up to 0·8 for combinations of four groups, and to almost 1·0 for combinations of 11 groups. Survey costs varied 10-fold between study groups, but their cost-effectiveness (indicator potential versus monetary cost) varied strongly depending on the biodiversity aspect taken into account (α or β diversity, single or multiple groups, etc.).

4.Synthesis and applications. We found that increasing the number of taxa resulted in best overall biodiversity indication. We thus propose that the most cost-efficient approach to general tropical biodiversity inventories is to increase taxonomic coverage by selecting taxa with the lowest survey costs.

Introduction

Biodiversity data are of great importance for decisions on design and location of reserves, management of natural habitats, monitoring of biodiversity changes, and the assessment of long-term sustainability of land-use systems. However, the field surveys needed for a full census of biodiversity of a prospected site or region exceed the budget of almost any project. One solution is to restrict a census to groups of organisms whose diversity patterns are believed to reflect overall biodiversity patterns or other important ecosystem parameters, and thus function as biodiversity surrogates (Pearson 1994; Prendergast 1997; Pharo, Beatti & Binns 1999; Rodrigues & Brooks 2007), although the search for such biodiversity indicators has not been particularly successful (Wolters, Bengtsson & Zaitsev 2006; Billeter et al. 2008; Kessler et al. 2009). The selection of surrogate study groups is furthermore a matter of deciding between study groups that are adequate surrogates and those that are easy, in terms of lower costs and time effort, to sample (Favreau et al. 2006; Wolters, Bengtsson & Zaitsev 2006; Rodrigues & Brooks 2007). ‘High-performance indicator study groups’ are those that fulfil both criteria (Gardner et al. 2008).

Recently, Gardner et al. (2008) presented a framework for assessing the performance of indicator study groups using a weighting technique to balance the costs and indicative value of different study groups. Because costs for the survey of different study groups typically differ by about an order of magnitude due to differences in species numbers, ecology, sampling methods, and time effort for identification (Lawton et al. 1998; Favreau et al. 2006; Wolters, Bengtsson & Zaitsev 2006; Rodrigues & Brooks 2007), Gardner et al. (2008) advocate the use of standardized survey costs, in which rarefaction is used to calculate the survey cost at a comparable level of survey completeness for the different study groups. This approach is intuitively appealing, but has the important potential drawback that whereas survey costs are readily standardized, the same is not true for biodiversity indication. Accordingly, Gardner et al. (2008) compared the standardized survey costs with non-standardized indicator performances. A further potential problem with their approach is that the standardization procedure relies on the accuracy of species richness estimators (Colwell & Coddington 1995) to calculate the sampling completeness of the respective taxa. However, there is a positive relationship between sampling completeness and the estimated total richness and, at low levels of sampling completeness, the estimators become unreliable (Herzog, Kessler & Cahill 2002; Walther & Moore 2005).

A more general shortcoming of many previous studies on biodiversity indicators is that often only a single taxon is evaluated (Wolters, Bengtsson & Zaitsev 2006; Billeter et al. 2008). Furthermore, most studies evaluate the indicator potential only among the sampled taxa, rather than relative to total biodiversity. While it is true that a full sampling of all taxa is not yet available from any site, it has been argued that averaging the relative diversity patterns of a larger number of taxa may give an indication of total biodiversity (Westphal et al. 2008).

In the present study we set out to explore solutions to the problems outlined above. We applied the approach of Gardner et al. (2008), but also propose an alternative to calculate the cost-effectiveness of different groups for biodiversity indication, based on the calculation of residuals of the indicator potential relative to absolute sampling costs, and compare this approach to that of Gardner et al. (2008). Further, we used multiple linear models to assess the performance of multiple predictor species groups used in combination. Finally, we evaluated our results at the level of estimated total biodiversity.

Materials and methods

Study Area and Site Selection

The study took place in an area of about 10 km2 at 850–1100 m a.s.l. at the western border of the Lore Lindu National Park in and around the village of Toro in the Kulawi Valley, Central Sulawesi, Indonesia. The region has an annual average (±SE) temperature of 24·0 (±0·16) °C and a mean monthly rainfall of 143·7 (±22·74) mm. There are no clear seasonal precipitation fluctuations. The natural vegetation of the area is submontane rainforest. For the present study, we selected 15 plots of 50 × 50 m2 each with a minimum distance of 300 m between them. The plots were situated in three habitat types: mature forest (4 plots), high-shade agroforests (7 plots; canopy closure >60%), and low-shade agroforests (4 plots; canopy closure <60%). A full description of the study area is available in Appendix S1 in Supporting Information.

Species Collection and Identification

Twelve species groups were chosen for the biodiversity assessment: trees, lianas, terrestrial herbs, understorey liverworts, birds, butterflies, lower canopy ants, lower canopy beetles, dung beetles, bees, wasps, and the parasitoids of the latter two. The organisms were sampled in each plot following a consistent sampling protocol and identified at least to morphospecies, when a specimen could not be named reliably. Depending on the study group, plots were sampled simultaneously or consecutively in random order, so that no confounding effects of temporal environmental variability are to be expected. Full data on species collection and determination is available in Appendix S1 in Supporting Information.

Data Analyses

Diversity estimates

For the analyses of α-diversity we used the number of species recorded in each plot. Because observed species richness values in field studies are typically an underestimate of the actual number of species occurring at a site (Colwell & Coddington 1995), we used the species richness estimator Chao2 (Chao 1987) as implemented in EstimateS 8 Windows (Colwell 2006) to estimate the actual total number of species per study group across all plots. We did not, however, correct the species numbers per plot, for two reasons. First, a previous analysis of the data has shown that between-taxon correlations using estimated values were similar to those obtained with the raw data (Kessler et al. 2009). Second, some of the analyses conducted here can only be made with raw data (e.g. the indicator group analysis) and for consistency it seems best to use the same data throughout. Our approach assumes that sampling completeness is consistent across habitat types for a given study group.

Biodiversity indication of individual groups

We used three measures of biodiversity indication, two based on the relationship of diversity patterns between study groups (α- and β-diversity) and one assessing the degree to which the species of a given study group can be used to evaluate habitat condition.

The relationship of diversity values between study groups was calculated using linear regression analyses (α-diversity) and Mantel analyses (β-diversity). Because our aim was to assess to what degree a group can be used to predict the biodiversity patterns of another group, we calculated determination coefficients (R2-values), while maintaining the original sign of the regression values to allow us to indicate negative relationships (Kessler et al. 2009). Mantel analyses are correlation tests between matrices consisting of pair-wise similarities or differences in a selected characteristic of samples or study sites (Legendre & Legendre 1998). In the first step of the analysis, pair-wise correlations were calculated between plot-wise similarity matrices of each study group pair, separately for each of the two diversity measures (α- and β-diversity). For each study group, the 11 correlation values were then averaged. Mantel analyses were conducted with PCOrd 4·5 (McCune & Mefford 1999), applying 9999 randomization runs.

To assess the degree of habitat fidelity of the species of a given study group, we used the indicator species analysis of Dufrene & Legendre (1997) as implemented in PCOrd 4·5, applying 999 randomization runs to test for significance. This analysis combines information on the concentration of species abundance and the faithfulness of occurrence of a species in a particular habitat. Indicator values range from zero (no indication) to 100 (perfect indication). For our analysis, we calculated the percentage of species of a given study group that have significant indicator values. To assess whether the percentage of species with significant indicator values depended on sampling completeness, we calculated a linear Spearman correlation between both parameters. Our intention here was not to differentiate between the three habitat types, which were visually easily distinguishable, but rather to quantify the degree of habitat specificity of the species of different study groups, following the reasoning that groups with higher habitat fidelity are better indicators of habitat conditions.

Biodiversity indication for combinations of groups

To assess the indicator potential of combinations of predictor taxa within multiple linear models, we created a list of all possible combinations of two to 11 taxa and calculated adjusted R2 values for each model, as well as mean adjusted R2 values of all models of each combination. The creation of linear models of all possible combinations of predictor species groups was automated using R (R Development Core Team 2008). The routine (set of command lines) to implement in R is given Appendix S2 in Supporting Information. For the calculation of adjusted R2 values for the β-diversity dataset, we used the R-function multRegress.R constructed by Legendre (2005), which computes a multiple regression and tests the coefficient of determination (R2) by permutation.

Assessment of sampling costs

Sampling costs refer to two main resources: monetary costs and time effort. Monetary costs included all collecting materials that could be used only once or a few times, and could therefore not be shared between study group surveyors. We also included site costs, as they are typical for field surveys (e.g. transport to field sites, overnight fees), but excluded transport costs from home country to country of survey. Costs of non-perishable equipment were not included in the analysis, following Gardner et al. (2008). Additional costs (‘capital’ and ‘hidden’ costs) were also excluded, since they differ for each specific study and may distort comparisons of survey costs. These additional costs are defined as not directly related to the diversity survey (e.g. building and maintaining reference collections for vouchers, costs for project planning, data analyses and reporting). Time effort was the largest contributor to the overall costs. In order to make direct comparisons between study groups possible, we standardized the salary requirements for five different groups of workers: 100 Euro/day for MSc students, PhD students and postdoctoral researchers and 20 Euro/day for field assistants and other workers. Time effort comprised the total effort needed in field and laboratory (identification), based on 5 person-days per week with 8-h working days.

Based on the above values, we used two measures of costs. First, we used the absolute survey costs of the sampling. This measure may be misleading because sampling intensity differed between the study groups. For this reason, we also calculated the standardized survey costs following Gardner et al. (2008). In this approach, individual-based rarefaction curves were constructed for each study group, followed by a recalibration of the y-axis so as to represent the proportion of the estimated total number of species, based on the species richness estimator Chao2 (Chao 1987) as implemented in EstimateS 8 Windows (Colwell 2006). Then, survey cost was calculated per species and the x-axis recalibrated to represent survey costs. Finally, the survey cost of each study group was rarefied to equate to the point at which the sample representation is equivalent to that of the least effectively sampled study group (Fig. 1). A problem arises when standardized survey costs are related to regression values as measures of biodiversity indication values (see Discussion), as the latter are not based on rarefied data. Because this approach may thus overestimate the relative cost-effectiveness of study groups for which standardization results in a high loss of information, we here introduce a third measure, residual survey costs, which corrects the regression values relative to the survey costs. This approach is based on the assumption that an increase of monetary input leads to a continuous increase of regression values. This relationship is not linear because (i) at low levels of input, a rapid increase of output can be achieved which subsequently decelerates at higher levels of input, and (ii) the regression value reaches its maximum at 1·0. We therefore assumed a logarithmic relation of the form a + b * ln(costs) and calculated the residual regression values by measuring the positive and negative deviations of the observed regression values from the best fitting logarithmic model curve by iteratively adjusting the coefficients a and b. Positive residuals indicate that a given taxon has a higher regression value at a given cost level than would be expected based on the average cost-indication relationship of all taxa, negative residuals the opposite. These residuals were then plotted against absolute survey costs to assess which groups provide the highest relative indicator value at a given cost level.

Figure 1.

 Rarefaction curves of the 12 study groups relative to absolute survey cost. The peak of the rarefaction curve for each study group is determined by the sampling completeness of that group. Standardized survey costs were derived from the point at which the rarefaction curve reaches the sampling completeness of the least completely sampled study group (beetles: 29%).

To compare the cost-effectiveness of biodiversity indication by different study groups, we calculated regression coefficients and plotted our three indicator measures (regression relationships of α- and β-diversity; percentages of indicator species) against the cost measures for visual comparison.

Results

In total, we recorded 863 species, with total species richness per study group ranging from 9 cavity-nesting bee species to 198 canopy beetle species (Table 1).

Table 1.   Biodiversity indication values for 12 study groups along a gradient of land-use intensification in Sulawesi
GroupNumber of individualsRecorded number of speciesEstimated number of species (Chao 2)Sampling completeness (%)Mean pair-wise indication (R2-values)Residuals of mean pair-wise indication (R²-values)Indicator species (%)Indication of average diversity of all 12 groups (R2-values)
αβαβMature forestsHigh-shade agro-forestsLow-shade agro-forestsαβ
Herbs11,12016326162·50·020·370·020·021130·090·64
Lianas351357646·10·050·260·05−0·099000·080·40
Understorey liverworts704375863·8−0·05−0·03−0·05−0·27000−0·060·01
Trees1,41618524874·6−0·190·34−0·20−0·2310110·090·54
Ants3,153447856·40·080·160·08−0·055000·160·20
Bees43991181·80·150·190·15−0·0411000·630·30
Birds9318710880·6−0·150·34−0·150·261615−0·010·36
Butterflies680387848·7−0·030·24−0·030·133011−0·060·30
Canopy Beetles61319867929·20·080·370·080·121220·270·54
Dung Beetles928252986·2−0·060·27−0·06−0·232404−0·070·18
Parasitoids666183060·00·060·420·060·120660·250·66
Wasps8,575242788·90·070·460·070·1204210·300·50

Determination coefficients (R2-values) of α-diversity per site between the different study groups ranged from an average of R2 = –0·19 for trees to R2 = 0·15 for bees (Table 1). A similar analysis for correspondence of patterns of β-diversity between study groups based on Mantel analyses obtained values ranging from R2 = −0·03 for understorey liverworts to R2 = 0·46 for wasps.

Absolute survey costs varied among study groups due to differences in numbers of sampled individuals and time required for handling and identification, from 4830 Euro for birds to 22,690 Euro for wasps (Fig. 1, Table 2). The proportion of salary costs was between 82 and 96% for most groups, except for dung beetles, herbs, and lianas, where it dropped below 80% due to relatively low sampling effort in the field. Rarefaction curves varied between taxa, with many groups reaching the cutoff value of 29% sampling completeness with less than 20% of the total sampling intensity (Fig. 1). Accordingly, standardized survey costs varied even more strongly between study groups than absolute costs, from 300 Euro for birds to 16,960 Euro for beetles, the least completely sampled study group (Tables 1 and 2).

Table 2.   Survey costs for 12 study groups along a gradient of land-use intensification in Sulawesi
GroupPostdoc (days)Field assistant (days)Materials (Euro)Extra Costs (Euro)Total expend (Euro)Standardized survey costs (Euro)
fwhdofwhdo
  1. Survey time is given in days for field work (fw), handling and determination (hd), and other (o; databank preparation, etc.); Materials are non-durable items for field work and collecting material; extra costs are generalized costs for transport and overnight fees at 100 Euro per week (only field work), based on a 5-day working week and 8-hour working day, except for field assistants.

Herbs2040102040 200220010600730
Lianas2040102040 2002200106001050
Understorey liverworts42985423 310840165503900
Trees52702010460 2503600213303500
Ants22607566  300440177602750
Bees7624 16812 2501520153701100
Birds322 32  1506404830300
Butterflies4030 40  25080088501650
Canopy beetles2260756635503004401946016960
Dung beetles5340105320 250300015010550
Parasitoids7747 17018 2501540179502250
Wasps7890 17232 25015602269010500

Comparison of the mean R2-values of α- and β-diversity of each study group to absolute survey costs showed no clear visual patterns (Fig. 2) and there were no significant relationships between the parameters (Spearman rank correlations, rs-values between 0·13 and 0·25, P > 0·44 in all cases). Study groups with the lowest sampling costs (birds, butterflies) by default had the highest cost-effectiveness. Plotting the mean R2-values of α- and β-diversity of each study group against standardized survey costs resulted in different patterns, with the majority of study groups having relatively low sampling costs (<3000 Euro) but widely diverging R2-values. Again, there were no significant relationships (Spearman rank correlations, rs-values between 0·28 and 0·30, P > 0·35 in both cases). In this case, taxa with low survey costs and high R2-values had the highest cost-effectiveness. For α-diversity this included ants, bees, herbs, lianas and parasitoids, and for β-diversity birds, dung beetles, herbs, parasitoids and wasps. Finally, evaluating the residuals of the R2-values of α-diversity, there was no relationship between the residuals and absolute sampling costs (Spearman rank correlation, rs = 0·21, P = 0·52), with ants, bees and canopy beetles having the highest positive residuals. As to β-diversity, there was no relationship of the residuals with increasing absolute sampling costs (Spearman rank correlation, rs = 0·04, P = 0·91), and highest positive residuals were obtained for birds, canopy beetles, herbs, parasitoids and wasps.

Figure 2.

 Relationship of mean pair-wise determination coefficients (R2-values) relative to the other 11 study groups, for different diversity measures (a–c: α diversity; d–f: β diversity), plotted against three measures of survey costs (a, d: absolute survey costs; b, e: costs standardized after Gardner et al. 2008; c, f: residual R2-values from logarithmic models). Abbreviations for study groups: AN: Ants; BE: Bees; BI: Birds; BT: Butterflies; CB: Canopy Beetles; DB: Dung Beetles; HE: Herbs; LI: Lianas; LW: Understorey liverworts; PA: Parasitoids; TR: Trees; WA: Wasps.

The percentage of significant indicator species ranged from 0% in understorey liverworts to 28% in dung beetles (Table 1). These values were positively correlated to the sampling completeness of the respective study groups (Spearman rank correlation, rs = 0·68, P = 0·015). Comparing the three habitat types, mature forests and low-shade agroforests had conspicuously higher percentages of indicator species (6·7 and 4·4%, respectively) than high-shade agroforests (1·3%), but this difference was not significant (two-tailed, one-way anova, F2,33 = 2·68, P = 0·084). Relating the percentage of indicator species to absolute survey costs did not result in any clear overall pattern (Spearman rank correlation, rs = −0·07, P = 0·83; Fig. 3). Birds and butterflies had the most cost-efficient combination of mean R2-values and low survey costs, whereas dung beetles had even higher R2-values but also higher costs. When survey costs were standardized, there was a tendency towards lower percentages at higher survey costs, but this was not significant (Spearman rank correlation, rs = −0·47, P = 0·13). In this situation, dung beetles had the highest cost-effectiveness. The residuals of the percentage of species with significant indicator values also showed a tendency towards a negative relationship with absolute sampling cost (Spearman rank correlation, rs = −0·31, P = 0·32), with highest positive residuals for birds, butterflies and wasps.

Figure 3.

 Relationship of three measures of survey costs for each study group (a, d: absolute survey costs; b, e: costs standardized after Gardner et al. 2008; c, f: residual R2-values from logarithmic models) plotted against the percentage of indicator species. Abbreviations for study groups: AN: Ants; BE: Bees; BI: Birds; BT: Butterflies; CB: Canopy Beetles; DB: Dung Beetles; HE: Herbs; LI: Lianas; LW: Liverworts; PA: Parasitoids; TR: Trees; WA: Wasps.

In the multiple linear models of all combinations of two to 11 predictor species groups, mean R2-values increased with the number of indicator taxa included, from 0·18 for two taxa to 0·72 for 11 taxa (Fig. 4). Within categories of a certain number of study groups, R2-values tended to increase with increased survey costs for combinations of two to five study groups, but decreased when more taxa where included. On the other hand, the variance of the R2-values was higher for α than for β diversity, so that even with 11 study groups single R2-values for specific combinations of study groups were close to zero (see Figs S1 and S2 in Supporting Information).

Figure 4.

 Relationship of R2-values of the α- (left) and β-diversity (right) of combinations of 2 to 11 study groups relative to other, single study groups, plotted against survey costs. Grey circles indicate the mean values for a number of combined groups, the lines the trend of the relationship of R²-values relative to survey costs. Note that for a given cost level, trend lines for higher numbers of combined groups generally lie above those for combinations with fewer taxa. Graphs showing all single values are provided in Fig. S1 in Supporting Information.

Considering the determination coefficients of the individual study groups against the averaged diversity patterns of all 12 study taxa, the results were qualitatively similar to those of the pair-wise comparisons (Fig. 5): There was no clear relationship between R2-values and survey costs, and the most cost-efficient groups were the same as in the pair-wise comparisons. In contrast, the results of the multiple regression models of combinations of taxa against the averaged diversity patterns differed conspicuously from those of the multiple regressions with individual groups as dependent variables (Fig. 6). For both α and β diversity, R2-values reached 0·8 when 5–7 groups were included and approximated 1·0 for 11 taxa.

Figure 5.

 Relationship of three measures of survey costs for each study group (a, d: absolute survey costs; b, e: costs standardized after Gardner et al. 2008; c, f: residual R2-values from logarithmic models) plotted against the R2-values relative to the mean biodiversity pattern of all 12 study groups, for different diversity measures (a–c: α diversity; d–f: β diversity). Abbreviations for study groups: AN: Ants; BE: Bees; BI: Birds; BT: Butterflies; CB: Canopy Beetles; DB: Dung Beetles; HE: Herbs; LI: Lianas; LW: Understory liverworts; PA: Parasitoids; TR: Trees; WA: Wasps.

Figure 6.

 Relationship of R2-values of the α- (left) and β-diversity (right) of combinations of 1 to 11 study groups relative to the averaged diversity patterns of all study groups plotted against survey costs. Grey circles indicate the mean values for a number of combined groups, the lines the trend of the relationship of R2-values relative to survey costs. Graphs showing all single values are provided in Fig. S2 in Supporting Information.

Discussion

This is one of very few studies that quantitatively evaluate the cost-effectiveness of different potential biodiversity indicator groups along a tropical land-use gradient, and only the second to consider standardized sampling costs (Gardner et al. 2008). At the most basic level, we obtained different results when considering different aspects of biodiversity indication (α–diversity, β-diversity, indication of habitat condition), and different measures of survey costs. These differences have important practical implications and, as developed in detail below, lead us to conclude that the best approach for generalized tropical biodiversity inventories might simply be to include as many study groups as logistically and financially feasible.

Evaluation of indicator potentials

Overall, the indicator potential (measured as determination coefficients) of single study groups was low for all four biodiversity aspects that we considered. For α-diversity, on average not more than 4% (R2 = 0·04) of the variability of species richness of one study group could be predicted by another group. Indication of β-diversity was only slightly higher, with R2-values of up to 0·05. The idiosyncrasy of diversity patterns of different taxa is a well known phenomenon, both at the local scale, where habitats in a given area are compared (e.g. Lawton et al. 1998; Schulze et al. 2004; Barlow et al. 2007; Kessler et al. 2009), and at the regional scale (e.g. Prendergast 1997; Duque et al. 2005; Tuomisto & Ruokolainen 2005). This remains one of the main challenges in successfully applying biodiversity indication (Wolters et al. 2006; Rodrigues & Brooks 2007; Billeter et al. 2008).

Focusing on the capability of species to discriminate between habitats, i.e. on their utility as indicators of habitat condition, the values were also low. On average, only 12% of the species of a given study group were significantly associated with a given habitat type, although such low values may actually provide important information (McGeogh, van Rensburg & Botes 2002). The degree to which a study group included significant indicator species was significantly correlated with the sampling completeness of the group. This relationship has a simple statistical background: the probability of recovering significant associations of species with habitats is a direct function of the number of records of those species (Dufrene & Legendre 1997). More fully sampled study groups simply have more repeatedly sampled species for which the statistical analysis can recover significant habitat relations. Within this statistically determined relationship, there appears to be some variation that may have a biological background. Thus, while no study group had a high percentage of indicator species at low sampling completeness, the reverse situation was found for several study groups. In particular, ants, herbs and understorey liverworts all had low percentages of indicator species (<5%) despite levels of sampling completeness above 50%, suggesting that species of these taxa may overall have low indicator potential.

The use of several indicator groups in combination raised the regression values relative to both the diversity of other single study groups and to that of all 12 study groups combined. Even more strikingly, mean R2-values for higher numbers of groups were consistently better at the same costs than those for lower numbers of taxa (Figs 4 and 6). In these analyses, in order to assess the indicator potential of the individual taxa relative to the overall biodiversity, we averaged the diversity patterns of all study groups to obtain generalized patterns. This approach is potentially statistically debatable because the groups that are used as indicators are also among the 12 groups used to calculate the averaged diversity pattern. However, in the absence of full biodiversity surveys, averaging all available groups has been advocated as the best alternative (Westphal et al. 2008). Furthermore, this approach reflects the real life situation, where any study group is indeed part of overall diversity. Using this approach, we found that the indicator potential for ‘overall’ biodiversity is, on average, higher than for individual groups, and that it increases strongly with a higher number of study groups. We conclude that the indicator potential for biodiversity surveys across a tropical land-use gradient such as ours is limited with single taxa and that surveys should aim to include as many different taxa as financially and logistically possible.

Evaluation of survey costs and cost-effectiveness

In our study, absolute survey costs for different study groups varied by almost an order of magnitude, from 4830 Euro for birds to 22 690 Euro for wasps. These differences are partly related to the species richness of the taxa and the difficulty of sampling and identifying them, but they also are related to the fact that the taxa were sampled to different levels of completeness. We therefore applied two measures which relativize regression values versus costs: the standardized survey cost as calculated by Gardner et al. (2008) and the residual regression values introduced here. Both of these approaches have potential drawbacks. In particular, while survey costs are readily standardized, the same is not true for the biodiversity indication. The reason is that the calculation of indicator values with rarefied data requires the selection of individuals that are retained for the analysis, and unless there is a chronological account of the sampled individuals, there is no straightforward way to conduct this selection. For example, in our study dung beetles had the highest percentage of significant indicator species (28%), but were at the same time the group with the seventh-highest absolute survey cost (15 010 Euro). Here, standardization following Gardner et al. (2008) reduced the sampling cost by about 96·3% (Table 2). If we had also eliminated over 96% of all sampled individuals, then both the total number of sampled species and the percentage of significant indicator species would certainly also have been considerably reduced. This standardization approach is therefore misleading, because it suggests that taxa with low standardized sampling costs have high cost-effectiveness, when in fact their high percentage of indicator species can only be achieved as a result of high absolute sampling costs. The use of residual information contents avoids the above problems but, at least in our case, the results of the analysis are qualitatively similar to those using the absolute costs, suggesting that the only limited correction of different sampling intensities was possible.

Selection of indicator taxa

Depending on which indicator parameters and cost measurements were applied, different study groups emerged as the most cost-efficient groups in the various analyses. This, along with the low overall R2-values obtained for single taxa, and the shifting group combinations in multiple group selections, should caution against a strong reliance on a few selected indicator taxa. Despite this caveat, some of our study groups came up more often as high quality indicator groups than others. In our study, if we scan over the results of the different analyses, birds, butterflies and wasps most often showed high R2-values, while canopy beetles, dung beetles and parasitoids did so less frequently, and ants, bees, herbs, lianas, understorey liverworts and trees least often. While these results are non-independent and study-specific, there is a certain degree of concordance with previous studies. For example, Gardner et al. (2008) found birds and dung beetles to be particularly good indicator taxa, with butterflies among the better third of the 15 taxa studied by them (wasps were not included). Birds, dung beetles and butterflies have often been advocated as good indicator taxa for tropical biodiversity surveys (Bibby et al. 1992; Davis et al. 2001; McGeoch, van Rensburg & Botes 2002; Horner-Devine, Daily & Ehrlich 2003; Cleary 2004), and also bees, wasps and their parasitoids have been shown to be effective bioindicators (Tscharntke, Gathmann & Steffan-Dewenter 1998; Tylianakis, Klein & Tscharntke 2005). The reasons for this in many cases appear to be the relative ease and effectiveness of quantitative survey techniques for these taxa in combination with manageable species numbers and a fairly well known taxonomy, rather than their biological indicator potential as such. Thus, the selection of these taxa as high quality biodiversity indicators is primarily driven by their low sampling costs. Accordingly, if one was to take this reasoning to its extreme, one might argue that indicator taxa can be selected exclusively on the basis of a simple cost calculation: how many different taxa can one sample with the funds available? Indeed, our study suggests that this approach might result in the best possible selection of biodiversity indicator taxa for tropical habitats. Our multiple regression analyses show that the increase of the information content is primarily the result of the inclusion of more taxa, rather than of more expensively sampled taxa.

A sampling approach that maximizes the number of sample taxa will also circumvent three further problems. First, the quality of single indicator groups cannot be extrapolated from one study to other geographical regions or habitats. Thus, time- and fund-consuming preliminary studies along the lines of our study would have to be conducted prior to each actual survey, in order to select the most cost-efficient taxa for a given situation. Second, the level of taxonomic knowledge and available expertise varies geographically and a flexible approach to the selection of taxa would easily take this into consideration. Third, even if a larger number of studies such as ours might in the future identify a limited number of taxa that generally outperform others – which may or may not be the case – an exclusive focus on a few high quality indicator groups would further increase the already existing taxonomic bias in tropical biodiversity inventories (Pawar 2003; Gardner et al. 2008). Eventually, such a one-sided approach in tropical biodiversity inventories might lead to a biased view of tropical biodiversity patterns and misinterpretations of how, for example, trophic webs in tropical forests and land-use systems are built up (Clough et al. 2007; Tylianakis, Tscharntke & Lewis 2007). Thus, if the aim of a given study is to assess overall levels of biodiversity, we strongly advocate that tropical biodiversity inventories should include as many study groups as feasible with the funds available. This approach is useful if the aim of a survey is to compare the overall species richness and community distinctness of different habitats or different sites within a given habitat, e.g. for the selection of conservation priorities, or to monitor changes over time. Other, more specific applications of biodiversity indication might be better conducted with specific taxa about which additional information, e.g. on their ecology, is available (Grantham et al. 2008; Cowling et al. 2009; Moran, Lacock & White 2010).

Acknowledgements

This study was funded by the German Research Foundation (DFG), grant SFB-552 STORMA (Stability of Tropical Rainforest Margins; http://www.storma.uni-goettingen.de). We thank Pak Mann, Arifin, Daniel Stietenroth, Adam Malik, Wolfram Lorenz, Surya Tarigan, and all plantation owners for their help in this work, and Toby Gardner, two anonymous reviewers and the editors for valuable comments on earlier versions of the manuscript.

Ancillary