Gauging megadiversity with optimized and standardized sampling protocols: A case for tropical forest spiders

Abstract Characterizing and monitoring biodiversity and assessing its drivers require accurate and comparable data on species assemblages, which, in turn, should rely on efficient and standardized field collection. Unfortunately, protocols that follow such criteria remain scarce and it is unclear whether they can be applied to megadiverse communities, whose study can be particularly challenging. Here, we develop and evaluate the first optimized and standardized sampling protocol for megadiverse communities, using tropical forest spiders as a model taxon. We designed the protocol COBRA‐TF (Conservation Oriented Biodiversity Rapid Assessment for Tropical Forests) using a large dataset of semiquantitative field data from different continents. This protocol combines samples of different collecting methods to obtain as many species as possible with minimum effort (optimized) and widest applicability and comparability (standardized). We ran sampling simulations to assess the efficiency of COBRA‐TF (optimized, non‐site‐specific) and its reliability for estimating taxonomic, phylogenetic, and functional diversity, and community structure by comparing it with (1) commonly used expert‐based ad hoc protocols (nonoptimized, site‐specific) and (2) optimal protocols (optimized, site‐specific). We then tested the performance and feasibility of COBRA‐TF in the field. COBRA‐TF yielded similar results as ad hoc protocols for species (observed and estimated) and family richness, phylogenetic and functional diversity, and species abundance distribution. Optimal protocols detected more species than COBRA‐TF. Data from the field test showed high sampling completeness and yielded low numbers of singletons and doubletons. Optimized and standardized protocols can be as effective in sampling and studying megadiverse communities as traditional sampling, while allowing data comparison. Although our target taxa are spiders, COBRA‐TF can be modified to apply to any highly diverse taxon and habitat as long as multiple collecting techniques exist and the unit effort per sample is comparable. Protocols such as COBRA‐TF facilitate studying megadiverse communities and therefore may become essential tools for monitoring community changes in space and time, assessing the effects of disturbances and selecting conservation areas.

We find it useful to view the problem of designing standardized sampling protocols as comprised of seven complementary criteria (Cardoso, 2009;Chao & Jost, 2012). The first is efficiency: Does the protocol have high return (of important data) on investment (time and resources required to acquire the data)? The second is suitability for a specific taxon or problem: Sampling in cloud forests with high epiphytic biomass may require a different protocol compared to lowland dry forest or rain forest-and closed canopy tropical forests may be different from, say, savannas or Mediterranean ecosystems. Suitable for one problem may not be suitable for another. The third is comparability: If the second criterion, suitability, is the only priority, one may end up with results that cannot be compared between sites, habitat types, or biomes, even if broadly similar. Comparability ideally means that one protocol design can be applied everywhere, even if it is suboptimal at each site. The fourth is feasibility: The protocol must be doable given the available resources. The remaining three criteria are flexibility, transparency, and accountability, which mean that protocols should be adaptable to resources, clearly explained for replication and appropriate for evaluation (Cardoso, 2009). Here, we address each criterion and suggest a repeatable, flexible, transparent, and accountable method to design efficient, suitable, comparable, and feasible sampling protocols. Cardoso (2009) developed protocols to sample in an optimized and standardized way-to obtain the most information possible with a given level of effort-by applying an optimization algorithm to existing data and by selecting a combination of samples that is as efficient as possible across all sites. This approach for creating protocols is flexible: Different nested subprotocols with varying levels of effort may be defined to cope with different objectives or available resources (human, time, or financial). This algorithm is applicable to any taxon and method, and in any biome, as long as the effort per sample is comparable. This comparability can be ensured by measuring effort using person-hours as the unit of quantitative sampling, which has been applied to spiders in many habitats, for example temperate forests (Coddington, Young, & Coyle, 1996;Dobyns, 1997;Scharff, Coddington, Griswold, Hormiga, & de Bjørn, 2003); habitats with no tree cover (Carvalho et al., 2012;Toti, Coyle, & Miller, 2000); savanna (Muelelwa, Foord, Dippenaar-Schoeman, & Stam, 2010); and tropical forests (Coddington, Griswold, Silva Dávila, Peñaranda, & Larcher, 1991;Silva-Davila & Coddington, 1996). However, the question remains as to whether such protocolsand their development-are applicable effectively to extremely diverse communities.
Here, we show that optimized and standardized protocols for collecting quantitative data on communities (species composition and relative abundance) are not only easy to develop and more informative than traditional nonoptimized protocols, but also feasible even for megadiverse taxa, such as tropical spiders. Spiders are an excellent test group for developing effort-focused standard protocols because they are speciose and abundant both locally and worldwide in terrestrial ecosystems (Basset et al., 2012;World Spider Catalog, 2016).
With many small, cryptic, and locally rare species, it is unrealistic to compile complete species lists for most habitats, even more so in highdiversity habitats such as tropical forests.
From an ecological and conservation perspective, spiders provide valuable information as they are especially sensitive to habitat disturbance (Malumbres-Olarte, Vink, Ross, Cruickshank, & Paterson, 2013). Because they are usually dominant invertebrate predators, spiders are also potential indicators of trends in the populations of taxa-such as their prey-that may take longer to go extinct (Cardoso, Arnedo, Triantis, & Borges, 2010). Conservation studies often lack reliable invertebrate data, but the wide adoption of standard protocols can help to solve this problem (Cardoso, Pekár, Jocqué, & Coddington, 2011).
Our objectives are (1) to develop the first optimized and standardized sampling protocol for tropical spider communities using a large body of existing data; (2) to evaluate the efficiency of this protocol by comparing it with two alternatives: protocols tailored to each site following expert opinion (a commonly used approach and here termed ad hoc) and protocols that are statistically optimal for each individual site (termed optimal) (see definitions in Materials and Methods); and (3) to test the effectiveness and feasibility of our protocol in the field.

| MATERIALS AND METHODS
In this study, a "standardized protocol" is defined as a protocol that is designed to be applicable to all sites of the same habitat type and to provide data comparable across sites. In contrast, an "optimized protocol" distributes the number of samples among methods to obtain the theoretical maximum possible number of species, and provides as much information on species assemblages (composition and relative abundances) as possible with minimum effort. A protocol can be optimized either for only a specific site ("optimal protocol") or for multiple sites ("quasi-optimal protocol"). A quasi-optimal protocol is therefore standardized and may not be optimal for any specific site alone. Here, the quasi-optimal protocol for tropical forests is named COBRA-TF (Conservation Oriented Biodiversity Rapid Assessment for Tropical Forests) after Cardoso (2009).
Three protocols are compared: (1) quasi-optimal (COBRA-TF) (standardized, optimized); (2) ad hoc (not standardized, not optimized); and 3) optimal (not standardized, optimized). Ad hoc protocols are site specific and based on little or no quantitative analysis of sample datadesigned according to the best judgment of experienced collectors as to what combination of methods would provide the maximum number of species.

| Protocol design
The COBRA-TF was developed using data from semiquantitative spider inventories from tropical forests in South America and Africa (Table 1 and Figure 1). Nine studies where sampling was conducted in a one-hectare (100 × 100 m) plot were considered. Although the collecting teams had different numbers of collectors and the numbers of samples per collecting method were specifically designed for each site (ad hoc), individual samples were standardized to one hour to be comparable from site to site. Six sampling methods were used during the day (d) and night (n), and each combination was considered independent in the analyses, giving a total of 12 separate methods (Cardoso, 2009;Coddington, Agnarsson, Miller, Kuntner, & Hormiga, 2009;Sørensen, Coddington, & Scharff, 2002): Aerial hand collecting (aerial, Ad/An)-hand collecting from knee level to as high as one can reach. This method targets web-building and/ or free-living spiders on the foliage and stems of living or dead shrubs, high herbs, tree trunks, or lianas.
Ground hand collecting (ground, Gd/Gn)-hand collecting of spiders visible on (but not hiding in) the leaf litter and on the ground, low buttresses, logs, and the lowest vegetation. It covers the vegetation stratum from ground to knee level.
Cryptic searching (cryptic, Cd/Cn)-hand collecting of species hiding in cryptic habitats (e.g., within litter, small holes in trees or fallen logs, under logs, bark, stones, and moss) or litter sampling performed either by direct search, or search in unsifted/sifted litter.
Vegetation beating (beating, Bd/Bn)-beating the vegetation with a rigid stick while holding a beating tray or screen underneath, from which the spiders were collected .
This method collects spiders living in shrubs, high herb vegetation, bushes, and lower branches of trees.
Sweep netting (sweeping, Sd/Sn)-sweeping low, primarily herbaceous or shrubby vegetation using a sweep net. The net was emptied at regular intervals (after three to five sweeps) to avoid loss or destruction of the specimens.
Pitfall trapping (pitfall, Pt)-pitfalls 9 cm in diameter, partly filled with preservative solution and a few drops of liquid soap to break the surface tension, and sheltered by lids on stilts 2-3 cm above ground. Traps were left in the field for 5-8 days. Groups of five pitfall samples were pooled to reduce variation in the abundance of adult spiders between samples and to make pitfall samples comparable to one person-hour effort-it takes around an hour to dig and fill five traps and to collect them after the sampling period.
All samples (except pitfall traps) comprised of one hour of continuous active sampling, measured with a stopwatch. Activity not directly involved in sampling was excluded by pausing the stopwatch (e.g., travel time to a different area within the plot, logistical problems).
Aspirators were generally used to transfer small specimens to vials. All putatively adult spiders seen were collected and transferred to vials with ethanol.
All adult specimens were identified at least to family level and sorted to species or morphospecies (as the majority of the species were undescribed) by examination of genitalia (Oliver & Beattie, 1996).
Somatic features, co-occurrence, and relative abundance were used to match sexes. Once the data were obtained, four steps were followed: 1. Assessment of the source data. It is critical to have comprehensive and robust data to develop the protocol and to understand how to select them. Data must come from exhaustive sampling-they must have enough samples per method, and high enough sampling intensity  and completeness (Scharff et al., 2003;Sørensen et al., 2002). Sampling completeness was compared using randomized accumulation curves of observed as well as estimated species richness as calculated with Chao1, Chao2, and first-and second-order Jackknife. The "final" slopes of Chao1 curves were used to determine whether an asymptote was reached (Cardoso, Pekár et al., 2011) and was calculated as: where S * a = estimated total number of species; S * a−1 = number of estimated species after adding the next to last sample; na = total number of individuals; na−1 = number of individuals after adding the next to last sample. Slopes below 0.01 were considered asymptotic. Likewise, the percentages of singletons and doubletons (species represented by one and two specimens, respectively), and the accumulation curves of singletons and doubletons Scharff et al., 2003;Sørensen et al., 2002) were assessed.

2.
Optimization of the effort per collecting method. The number of samples per method that maximized the species richness for the overall number of samples per site was determined using only the data from sites selected in the previous step. To achieve this, an iterative procedure was followed to maximize the slope of the accumulation curve for any total number or combination of samples as samples were successively added (Cardoso, 2009). This procedure was carried out for all selected sites simultaneously by running 10,000 simulations using an algorithm (function "optim.alpha") included in the R package BAT (Cardoso, Rigal, & Carvalho, 2015). The minimum number of samples per method was set equal to that used in the 24-sample protocol for Mediterranean forest (Cardoso, 2009) to ensure comparability with sites worldwide using the original COBRA protocol, a process termed constrained optimization    (Cardoso, 2009;Coddington et al., 2009). The total number of samples per protocol was set to a multiple of six because this is the maximum number of samples one can collect per day without a decline in sample quality due to fatigue .

4.
Standardization of the protocol. Finally, it was checked that the resulting combinations of samples were feasible and practical (e.g., exclusive night sampling is not practical logistically).

| Evaluation of protocol efficiency
The COBRA-TF was evaluated by comparing it with (1)  Species richness was estimated using Chao1 and the abundancebased coverage estimator (ACE) (Chao & Lee, 1992;Magurran, 2004), two of the most widely used bias-corrected estimators. Phylogenetic diversity was calculated using a phylogenetic family tree with equal tree branch lengths generated using the latest spider phylogenies (Coddington & Levi, 2005;Dimitrov et al., 2016;Garrison et al., 2016).
A functional tree classifying spider families into predatory guilds built using UPGMA with Gower distance (Cardoso, Pekár et al., 2011) was used to compute functional diversity. The structures of the communities were compared by looking at the species abundance distributions obtained using the Gambin model (Matthews et al., 2014;Ugland et al., 2007). Gambin outperforms other models, such as log-series or lognormal, and its single variable α describes the "dimensionality" of the communities.
For each site, 1,000 sampling simulations were ran, and the confi-

| Protocol field test
The COBRA-TF protocol was tested in a tropical montane forest in This modification was needed to make these results comparable to the extensively used COBRA protocol for Mediterranean ecosystems ) in order to allow comparisons of spider assemblages at larger scales (results will be reported elsewhere). Guyana, Madagascar (two), and Tanzania (Table 1). These sites showed high sampling intensity (above 15) and completeness (above 80%) and relatively low proportions of singletons (below 30%). The accumulation curves for the Chao1 estimator mostly reached an asymptote (slope < 0.008 in all cases), and the singleton curves reached an asymptote or decreased and approached the doubleton curves by the end of the accumulation process ( Figure 1). Sweeping was excluded from the analyses of accumulation curves because only a few samples were taken at some sites.

2.
Optimization of the effort per method. The curves of the optimal protocols were steeper (more efficient) than the curves of the ad hoc protocols at all the sites (Figure 2). Most methods in isolation had less steep slopes than any combination of methods; however, in some cases, aerial night searching and both day and night beating also produced steep curves, especially at the beginning ( Figure 2).

Definition of the overall effort (stop-rules).
For the four analyzed areas, 36 samples provided between 41.5% and 63.1% of the species observed in each area and between 33.1% and 57.5% of the estimated species richness (sampling completeness) ( Table 2).
Being a multiple of six, 36 samples is a pragmatic goal that is possible to attain by a team of two collectors in 3 days or a team of three in 2 days.

Standardization of the protocol.
In general, optimal protocols were highly biased toward aerial night and beating day sampling, with some samples of either day or night cryptic and ground sampling (Table 2). Pitfall trapping was seldom chosen by the algorithm. The ideal compromise protocol across all sites includes a large proportion of aerial night and beating day sampling and small proportions of both day and night ground searching.
Using the Mediterranean forest COBRA as the base for COBRA-TF and optimizing all sites simultaneously resulted in a different combination of samples, mainly due to the inclusion of pitfall traps.
Numbers were kept even to make them easier to apply in the field.
Also, ground hand collecting and cryptic searching were combined in a single sampling method that can be adjusted to the features of the ground-depending on the ground cover and the amount of dead logs or rocks, more time may be spent using either method. The final combination of samples was 8 An + 6 Bd + 2 Bn + 2 Gd/Cd + 2 Gn/ Cn + 2 Sd + 2 Sn + 12 Pt.

| Evaluation of protocol efficiency
Optimal protocols yielded more observed species than the COBRA-TF and the ad hoc protocols (Figure 3

| Protocol field test
In the test plots 1 and 2, 756 and 1,252 adult specimens belonging to 125 and 92 morphospecies, respectively (Table 3), were collected.
The final slopes of Chao1 estimators were above 0.03 ( Figure 6 and Table 3) and sampling completeness was 65.9% and 78.6% (Table 3), although these numbers are probably overestimates ( Figure 6). The percentage of singletons was 37% and 27% and the percentage of doubletons 12% for both plots. Sampling intensity was relatively low (Table 3)

| DISCUSSION
Optimized and standardized protocols are preferable to ad hoc protocols for two major reasons. First, they are equally efficient as measured in person-hours of work needed to achieve a given sampling completeness or to provide data on the phylogenetic and functional diversity, and the species abundance distribution in a community.
Second, they enable comparison of data from other studies and areas without having to account for sampling effort in the analyses. Not only does this help answer questions at larger scales but it allows reusing the data beyond the objectives of each particular project. The data resulting from the new protocol can now be easily and fully compared with previous data collected in different regions for multiple purposes such as inventorying ), conservation (Crespo, Silva, Borges, & Cardoso, 2013), or biogeography (Carvalho et al., 2011a(Carvalho et al., ,b, 2012. Our results show that a protocol combining day and night sampling is necessary to ensure an efficient coverage of the spider diversity at tropical forest sites. We propose that the combination of samples of COBRA-TF be the minimum so that if a more comprehensive protocol is applied, the resulting data are still comparable after subsampling from the minimum common denominator. Surprisingly, pitfall traps, usually deployed in the field during relatively long periods, are not particularly efficient in tropical forests, contrary to what has been found in temperate areas Cardoso, Scharff et al., 2008;Cardoso et al. 2009 (Moffett, 1994).
Canopy fogging, or other methods targeting the high vegetation, is essential to obtain a complete representation of spider species in tropical forests (Fannes, De Bakker, Loosveldt, & Jocqué, 2008;Sørensen, 2004). However, fogging is logistically demanding and as it is not time based, it may be difficult to combine with the quantifiable methods used in the analyses.
Based on our field test of COBRA-TF, we provide a number of recommendations adaptable to taxa other than spiders. First, we recommend limiting the number of one-hour samples per collector to six to avoid reduced sampling quality due to fatigue (Cardoso, 2009;Coddington et al., 1996). Second, we suggest concurrent collecting by multiple teams in different plots when logistics and available resources allow it. While requiring more resources, this will save time overall, and will reduce the effects of species phenology, rapid changes in community composition or capture efficiency due to varying weather conditions. Third, we suggest reducing sampling area from the originally proposed 1 ha to 0.25 ha (50 × 50 m) because finding multiple suitable areas of 1 ha can be a difficult task in topographically complex and fragmented habitats of montane tropical forests. The reduction in area should not cause any measurable effect on the number of species captured with low-effort sampling, as effort appears to be overwhelmingly more important than area (Coddington et al., 2009;Sørensen et al., 2002). Fourth, collecting should be conducted when vegetation is dry, when most collecting methods are more efficient. Although collecting in the rainy season may yield more species than in the dry season (Azevedo et al., 2014), it may be more time-consuming and therefore eventually become more expensive-in our field test, two to three collectors needed 4-5 days to collect 24 hand-collected samples because of intermittent rain during the light rainy season. A possible compromise and our recommendation for maximum efficiency is to collect shortly after the rainy season. Finally, we warn about the potential effects of simultaneous or continuous sampling in the same small area even with low sampling effort, that is, the number and identity of the spec- T A B L E 4 Number of species and percentages of total species (%) collected per method in the field test of the COBRA-TF protocol needed. In tropical forests and for megadiverse groups, accurate estimates require a large sampling effort, involving several collectors in a single site during one or more weeks (Coddington et al., 2009), and such effort, although possible, is rarely made in most studies.
Estimates are often required when comparing assemblages using incomplete species lists, but this need may be avoided if optimized and standardized sampling is used.
Here, we prove that optimized and standardized protocols such as COBRA-TF can be developed for megadiverse taxa and that their use is not restricted to less diverse and, in principle, more easily quantifiable communities, such as those in Mediterranean and temperate habitats. COBRA-TF is as good as traditional ad hoc protocols at collecting data despite the fact that COBRA-TF is based on data from only four sites. Designing a perfectly universal sampling protocol for tropical forest would require data from dozens of sites from all over the world, but such data are not available yet. Nevertheless, our field test proves that the COBRA-TF performs adequately in other tropical forests and is a sound protocol that may be refined through constant testing and addition of data from new sites.
Comparability, short duration and easy application make the COBRA-TF protocol extremely useful. But above all, adaptability is the key characteristic: The approach of COBRA-TF-the process of optimization and standardization-is applicable to most megadiverse taxa, habitats, and sampling methods. For instance, following the optimization and standardization steps outlined above and using If we are to record, quantify, and assess some of the most diverse and unique communities in the world, we must apply efficient, widely applicable, and standardized tools, such as COBRA-TF. Not only will they allow comparing communities in order to understand the ecological processes behind their assembly, they will also facilitate monitoring and assessing megadiverse communities, their changes in space and time caused by climatic changes or human disturbances, and selecting the areas optimizing their conservation.