INTRODUCTION
Traps have been used in both studies of seed dispersal and fruit abundance (Chapman et al. Reference CHAPMAN, WRANGHAM and CHAPMAN1994, Hamrick & Godt Reference HAMRICK and GODT1996, Harms et al. Reference HARMS, WRIGHT, CALDERON, HERNANDEZ and HERRE2000, Hubbell et al. Reference HUBBELL, FOSTER, O'BRIEN, HARMS, CONDIT, WECHSLER, WRIGHT and DE LAO1999, Jordano & Godoy Reference JORDANO, GODOY, Levey, Silva and Galetti2002, Nathan & Muller-Landau Reference NATHAN and MULLER-LANDAU2000, Parrado-Rosselli et al. Reference PARRADO-ROSSELLI, MACHADO and PRIETO-LOPEZ2006, Stevenson et al. Reference STEVENSON, QUIÑONES and AHUMADA1998, Reference STEVENSON, LINK and RAMIREZ2005, Terborgh Reference TERBORGH1983, Terborgh et al. Reference TERBORGH, PITMAN, SILMAN, SCHICHTER, NUÑEZ, Levey, Silva and Galetti2002, Wright et al. Reference WRIGHT, CARRASCO, CALDERON and PATON1999, Zhang & Wang Reference ZHANG and WANG1995). In spite of the wide use of traps in important areas of tropical ecology, no published studies compare the results provided by different trap designs in tropical forests, even though it has been noted that trap design affects results in other ecosystems (Chabrerie & Alard Reference CHABRERIE and ALARD2005, Kollmann & Goetze Reference KOLLMANN and GOETZE1998, Page et al. Reference PAGE, NEWLANDS and EALES2002). Kollmann & Goetze (Reference KOLLMANN and GOETZE1998) found that the rate of seed removal by predators and wind are important parameters to consider when designing a seed trap. They also found that trap height and area affect the quantity of seeds they recovered for the species they monitored, mostly herbs, bushes and small trees.
The potential bias inherent to each design makes it difficult to compare the results between studies using different traps. In addition, there has been no justification for the number of traps used in past studies. Accordingly, the first purpose of this study was to provide an estimate of the number of traps necessary to estimate species richness of seeds, fruit production and the mass and number of seeds dispersed. The second purpose was to compare the results among five of the common designs used to quantify seed and fruit fall in terms of bouncing, wind effects, area effects and seed removal by predators.
METHODS
Fruit and seed traps
We compared and evaluated the effectiveness of five different traps. The effective area of collection differed among traps because of mechanical constraints (e.g. a large basin standing by a single support would not be stable).
(1) Delimited area on the ground. The simplest ‘trap’ design is a marked area on the ground where the seeds or fruits are counted (Izhaki & Walton Reference IZHAKI and WALTON1991). This procedure may lead to biased results due to the high potential for fruits and seeds to be removed by frugivores, predators, wind or water (Gibson Reference GIBSON2002, Kollmann & Goetze Reference KOLLMANN and GOETZE1998). We used a marked area of 1 m2 on the ground where leaf litter and seeds were cleared.
(2) Mesh with PVC support. This commonly used trap design in tropical rain forests consists of a frame of PVC tubing that supports a net (Harms et al. Reference HARMS, WRIGHT, CALDERON, HERNANDEZ and HERRE2000, Muller-Landau et al. Reference MULLER-LANDAU, WRIGHT, CALDERON, HUBBELL, FOSTER, Levey, Silva and Galetti2002, Wright et al. Reference WRIGHT, CARRASCO, CALDERON and PATON1999). We used a trap with a collection area of 0.64 m2 with a polyester mesh bag, with holes of less than 1 mm, which was supported by 4 PVC tubes at 0.8 m from the ground (Figure 1a).
(3) Hanging mesh trap. A similar trap design consists of a mesh or bag hanging from branches of plants in the understorey (Stevenson Reference STEVENSON2002, Terborgh Reference TERBORGH1983), and thus, the wind may flip the traps over. We used polyester mesh bag with a collection area of 0.64 m2. The corners of the trap were tied up with string to branches of the surrounding vegetation at an approximate height of 0.8 m (Figure 1b).
(4) Basin traps. This trap consisted of a basin 33 cm in diameter with an area of 0.085 m2. The basin was supported with a plastic screw to a single PVC tube just in the middle of the base (Figure 1c). Each basin had a considerable number of holes 2 mm in diameter to allow drainage. We placed a plastic bag full of leaf litter inside the basin (about 5 cm thick), to help cushion fruits landing in the trap.
(5) Funnel traps. These traps consisted of a plastic funnel, 22 cm in diameter, with a collection area of 0.038 m2 that was inserted into a PVC tube (8 cm diameter), which was buried in the ground (Figure 1d). At the funnel base a small polyester mesh bag was attached to collect the fruits and seeds. The PVC tube had several holes of 5 mm in diameter at the base for drainage. Funnel traps have been recommended as the best trap in open habitats in temperate zones (Chabrerie & Alard Reference CHABRERIE and ALARD2005, Kollmann & Goetze Reference KOLLMANN and GOETZE1998; but see Page et al. Reference PAGE, NEWLANDS and EALES2002).
Figure 1. Trap designs used in the study in Caparú Biological Station (Vaupés, Colombia) and under laboratory conditions: mesh with PVC support (a), hanging mesh (b), basin (c), and funnel (d). Basin traps were also used in Tinigua National Park to assess sampling effort.
Data collection and analyses
Sample size
The data were obtained in three phases. In the first phase, adequate sample size (number of traps) was assessed from data collected in Tinigua National Park, Colombia (2°40′N; 74°10′W) during the years 1990 and 1991. Tinigua National Park is situated in a tropical lowland forest at an altitude of 350–400 m above sea level with an annual precipitation of 2782 mm (Stevenson Reference STEVENSON2002).
To find out how many traps are necessary for a good estimate of the number of species producing fruits, total fruit production, and biomass and number of dispersed seeds, we used information from 300 basin traps. The traps were located randomly in 12 transects (each c. 450 m length), scattered over an area of about 3 km2. All seeds, fruits, and fruit parts were collected twice a month over 1 y, identified to species, and dried to constant weight (Stevenson et al. Reference STEVENSON, QUIÑONES and AHUMADA1998). Identification was based on the fruit guide of the study site (Stevenson et al. Reference STEVENSON, QUIÑONES and CASTELLANOS2000). Additionally, all trees above each trap were identified to assess the likelihood that a seed in the trap could be classified as dispersed. This analysis was restricted to species with fleshy fruits, excluding liana species, because it was difficult to assess the presence of a parental liana above the traps.
In order to assess the number of traps necessary to get stable estimates of the ecological parameters, we used the program EstimateS, version 7.5.0 (http://purl.oclc.org/estimates) to graph the cumulative number of species as a function of sample size. We also calculated the percentage of precision (Norton-Griffiths Reference NORTON-GRIFFITHS1975, which is a measure of variation that decreases as sample size increases. This coefficient was calculated as follows:
Then the percentage is proportional to the value for the 95% confidence interval of the parameter evaluated from a set of traps, divided by its average in that sample (NRC 1981). To assess the variation in precision as a function of sample size, we first calculated the percentage for a set of 10 randomly chosen traps and this procedure was replicated ten times for analyses of both fruit production and dispersed seeds. This calculation provided an average percentage of precision for 10 traps and the standard error. Then, the same calculations were performed for 20 traps, 30 traps, and so on until reaching the total of 300 traps. Minimum sampling effort is achieved when the precision of the estimates does not improve as more traps are added (i.e. when the slope of the curve approaches zero). In order to find the number of traps required to get stable estimates, we assessed deviations from horizontality using traditional linear regression analyses (testing for a slope < 0).
Bouncing and wind effects
The second phase was carried out in a laboratory. We evaluated the bouncing and wind effects for each of four different trap designs (mesh with PVC support, hanging mesh, basin and funnel). We did not test the ground trap because we assumed that the bouncing effect would be negligible (i.e. the probability that a seed will land in the square and bounce out of the square is the same as the probability that a seed will land outside and bounce in). To evaluate bouncing effects, we used seeds or fruits of species with different weight and consistency (Table 1). Throughout this paper we will refer to each species only by generic name. Six bouncing trials were conducted for each species. Each trial consisted in dropping 10 seeds/fruits from 14 m above the traps and we counted the number of seeds/fruits that bounced out of the traps.
Table 1. Characteristics of the species used in three different experimental trials: (a) bouncing effects, (b) wind effects and (c) removal by predators.
1Based on the total weight of 77 seeds
To evaluate wind effects we tested fruits and seeds belonging to five species with wind-dispersed seeds, and with different weights and shapes (Table 1). We measured the wind effects for a wind speed of 20 km h−1 generated by a Falcon Super Deluxe® fan, which is within the range of highest values reported for the Colombian Amazon in forested areas (http://www.ideam.gov.co/sectores/aero/climat/index44.htm). Tests were conducted with each of the four trap designs and each plant species studied. We placed 10 seeds of each species in each trap and turned the fan on. The fan was positioned horizontally 1 m from each trap. Six replicates were done for each test and we counted the number of seeds that blew out of the traps.
Area effects and seed removal by predators
The third phase was carried out at the Mosiro Itajura Biological Station (1°05′S; 69°31′W), Vaupés, Colombia, formerly known as Caparú. The biological station is situated in a tropical rain forest, at an altitude of 400 m asl with an annual precipitation between 3000 and 4000 mm (Defler & Defler Reference DEFLER and DEFLER1996). We monitored seed rain and removal effect by predators from February to June 2005. We evaluated the five trap designs to determine the extent of seed removal by predators from traps in three species that produced fruits during the study period (Table 1).
To assess the effect of trap area and seed removal by predators, we used six parental trees of both Byrsonima and Osteophloeum, and five Virola trees. All of the five trap designs were tested under each parental tree. Two traps for each design were placed under each tree and results were averaged to be used as independent samples in further analyses. Parental trees of each species were at least 30 m apart. Traps were checked every 7 d and during each visit we counted the number of seeds deposited in each trap.
In order to assess an area effect for each trap design, we compared the total number of seeds collected by trap area. Furthermore, we made a second comparison using a correction factor that takes into account the bouncing effect inherent to each trap design. We included only data from Byrsonima and Osteophloeum in this analysis, due to the low seed representation under Virola trees for basin and funnel traps.
To determine the seed removal by predators from traps, we placed marked seeds into the traps used above. We added a specific number of marked seeds to each trap, simulating the natural seed rain of each tree species (10 seeds for Byrsonima, 3 for Osteophloeum and 4 for Virola). Byrsonima and Osteophloeum seeds were marked with water resistant ink, and Virola seeds were marked with a thread inserted through the seed base with a needle. On each weekly visit we counted the number of marked seeds that were removed from the traps and in case of removal, the seeds were replaced. A replicate group of traps – of the same designs and quantities – were placed far away from parental trees at a minimum distance of 50 m. The replicates were used to assess differential removal between traps under parental trees and traps away from parents. Replicate traps were distributed at random and in similar distribution to those under parental trees.
Statistical analyses
All the analyses were done with Statistix 8.0 (2003). Bouncing data were log transformed (log(x+1)) and were analysed with an ANOVA. A posteriori Tukey tests were also performed. Differences among trap designs for wind and removal by predators were analysed using Kruskal–Wallis non-parametric tests.
RESULTS
Sample size
The cumulative number of species recorded as a function of sample size did not reach a stabilization point for the total number of traps used (Figure 2a). The percentage of precision for fruit production and the mass of dispersed seeds reached stabilization using 230 and 250 traps respectively (Figure 2b–c). The percentage of precision for number of seeds dispersed did not show a clear stabilization point (Figure 2d). In most cases we observed a rapid change in variation estimates with a small sample size (c. 100 traps) and afterwards the variation changed more gradually. The overall variation in the estimates of fruit production was lower than for both quantifications of seed dispersal, but in all cases there were large variations among traps. For instance, the exclusion of a single trap located under a howler monkey sleeping site changed drastically the patterns of variation in the mass of dispersed seeds (Figure 2c). The highest variation was observed for the number of dispersed seeds (Figure 2d).
Figure 2. Analyses of the sample size (i.e. number of basin traps) necessary to estimate different ecological parameters in Tinigua National Park, Colombia. Sample size required to get adequate estimates of species richness represented in fruit production (a). Percentage of precision in the estimates of fruit production (b). Percentage of precision of mass of seeds dispersed as a function of cumulative sampling size (c). Percentage of precision of number of dispersed seeds as a function of cumulative sampling size (d). The points represent averages from ten simulations and the standard error (dotted lines or bars) associated with the mean. In the third panel filled points represent an analysis subtracting a single trap below a howler monkey sleeping tree.
Bouncing and wind effects
We found highly significant differences for fruit and seed bouncing between trap designs (F = 8.44, N = 24, P = 0.002; Figure 3a). A posteriori comparisons showed differences between two groups: Funnel and basin traps vs. mesh traps with PVC support, with less bouncing effects for the later. Hanging mesh traps showed intermediate bouncing effects. We found highly significant differences in bouncing among the species used (F = 7.05, N = 24, P = 0.001, Figure 3b). A posteriori tests showed differences between Citrus vs. Carica, Physalis, Mucuna and Ormosia. There were also differences between Malus with Carica and Physalis. The heaviest fruits, Citrus and Malus, showed the greatest percentage of bouncing, while Carica and Physalis bounced the least.
Figure 3. Number of seeds and fruits bouncing out of the traps under laboratory conditions. Comparing different trap designs (including data from all plant species) (a) and comparing plant species (1. Carica, 2. Physalis, 3. Ormosia, 4. Mucuna, 5. Malus, 6. Citrus) (b). The box indicates the 25th and 75th percentiles from six replicates, the line inside the box is the median, the capped bars indicate the 10th and 90th percentiles, and the outlying symbols are the extreme values.
We found significant differences in fruit and seed removal by wind among trap designs (H = 8.17, N = 20, P = 0.042; Figure 4a), with the hanging mesh trap showing the highest losses. We also found significant differences in fruit/seed removal by wind among the species (H = 8.73, P = 0.06; all N = 20; Figure 4b). In all cases, we observed that small Tillandsia seeds had the greatest removal and the heavy Machaerium samaras had the lowest.
Figure 4. Number of fruits or seeds blowing out of the traps under laboratory conditions. Comparing trap designs (including data from all plant species) (a) and comparing plant species (1. Adenocalymma, 2. Cydista, 3. Machaerium, 4. Pterocarpus, 5. Tillandsia) (b). The box indicates the 25th and 75th percentiles from six replicates, the line inside the box is the median, the capped bars indicate the 10th and 90th percentiles, and the outlying symbols are the extreme values.
Area effects and seed removal by predators
We found significant differences in the number of seeds according to area between trap designs in Byrsonima (H = 10.3, P = 0.03, df = 4, N= 30); but not in Osteophloeum (H = 5.69, P = 0.23, df = 4 N = 30). However, the comparison taking into account the bouncing effect rendered no significant differences between trap designs with different area (Byrsonima: H = 9.3, P = 0.052, N = 30; Osteophloeum: H = 5.8, P = 0.21, df = 4, N= 30).
We did not find significant differences in the mean numbers of seeds removed under parental trees and far from parental trees (Byrsonima: H = 6.39, N = 60, P = 0.84; Virola: H = 1.64, N = 50, P = 0.99; Osteophloeum: H = 3.03, N = 60, P = 0.99). Thus, we pooled data for traps located under and away from parental trees for the remaining analyses, to assess differences among trap designs. In all three species studied, we found significant differences between trap designs for seed removal by predators (Byrsonima: H = 33.6, N = 60, P < 0.001; Virola: H = 41.5, N = 50, P < 0.001; Osteophloeum: H = 51.8, N= 60, P < 0.001). In all cases, a posteriori tests showed that seed removal from the ground was significantly greater than other traps (Figure 5), but we did not find differences in seed removal among traps (P > 0.05). The mean proportion of seeds removed from traps between each revision was 0.5% (SD = 0.5) for the three species, while the proportion removed from the ground was 38.0% (SD = 19.1).
Figure 5. Number of seeds removed by predators from each trap design and for three different species studied in Caparú Biological Station. Byrsonima (number of marked seeds placed in each trap N = 20) (a), Virola (N = 8) (b), and Osteophloeum (N = 6) (c). The box indicates the 25th and 75th percentiles, the line inside the box is the median, the capped bars indicate the 10th and 90th percentiles, and the outlying symbols are the extreme values.
DISCUSSION
Sample size
Our sample size of 300 traps is above the average for studies of fruit production and seed dispersal, e.g. 300 in Chapman et al. (Reference CHAPMAN, WRANGHAM and CHAPMAN1994), 120 in Jackson (Reference JACKSON1981), 40 in Silman (Reference SILMAN1996), 75 in Smythe (Reference SMYTHE1970), 100 in Terborgh (Reference TERBORGH1983), 200 in Wright et al. (Reference WRIGHT, CARRASCO, CALDERON and PATON1999). However, the parameters estimated in this study, in particular species richness did not completely stabilize with such large sampling effort and the relatively small basin traps. Similarly, we did not reach stabilization in the percentage of precision of the number of dispersed seeds, even though the degree of variation may be underestimated since a large number of small seeds might have been washed away through the drainage holes. Our analyses indicate that a large number of traps (>230) are necessary to obtain stability in the variance estimates of fruit production. All these results are mainly caused by the clumped nature in the spatial patterns of both fruit production and seed dispersal (Schupp et al. Reference SCHUPP, MILLERON, RUSSO, Levey, Silva and Galetti2002), and therefore, we recommend the use of more than 300 traps. However, when logistic problems might limit the construction of a large number of traps, this number should not be less than 100 traps.
Bouncing and wind effects
The results of the study showed astonishing differences in bouncing effects among trap designs. Overall, 41.9% and 36.6% of the seeds falling into funnel and basin traps were not retained. In terms of biomass (dry weight in g), this turns out to be a 67.7% and 52.6% of loss, respectively (Table 2). The percentage of number of seeds and biomass lost in other trap designs was not as high, but still their values considerably underestimate fruit production and seed fall (Table 2). Thus, the funnel and basin traps showed the highest bouncing effects, and this result was expected for the funnel trap because the plastic material had little cushioning for either fruits or seeds that fell into them. It was surprising that the cushion in the basin trap was not very effective at avoiding bouncing of large fruits and seeds. The mesh traps, both the hanging and on PVC frame, showed the smallest bouncing effect. In particular, the concave shape of the mesh on PVC frame allowed even heavy fruits to remain against the frame of the trap. The bias was more pronounced for heavy seeds and fruits than for light ones. Although the consistency of fruit and seeds influenced bouncing probabilities, there were no differences between large fruits of different consistencies. This indicates that there is a greater influence of weight than consistency and since fruit productivity may depend on a higher degree from large fruits, then the choice of a particular fruit trap design is relevant to provide rigorous estimates of fruit production.
Table 2. Mean percentage of loss by bouncing and wind effects in terms of number of seeds or fruits lost and biomass (dry weight, g).
The highest removal of seeds by wind from the hanging trap (26% of seeds lost and 16% of biomass lost) was due to trap instability (i.e. traps turned upside down). This problem of flipping was avoided in the traps with PVC frame (2.6% of seeds lost and 0.02% of biomass lost). Nevertheless, in the wind experiments, the small Tillandsia seeds were more frequently removed than the heavier Machaerium seeds. Fruit production estimates were more biased for the effects of bouncing than wind. For example, for our study, on average, the percentage of loss from wind effects was 19.2 for number of seeds/fruits lost, 5.2% for biomass, and from bouncing effect was 27.8% for number of seeds/fruits lost and 49.9% for biomass (Table 2).
Area effects and removal by predators
Contrary to the suggestion of Page et al. (Reference PAGE, NEWLANDS and EALES2002), we found no significant differences in the number of seeds per area among traps, when the effect of bouncing was included. Therefore, for the trap designs included in this study, the effective collection area does not seem to have an influence on the performance of the trap. However, it is possible that very small or large traps should affect the seed rain representation. An analysis determining the efficiency of trap area in terms of number of species that can be collected in each trap design remains to be done.
In all comparisons we found that the ground traps had the greatest removal of marked seeds by predators and this removal was significantly different from all the other trap designs. Therefore, any estimates of fruit production using marked areas on the ground may be strongly biased, and the use of traps would help to reduce the problem of predator removal, unless there is an additional mechanism to prevent rodent removal (Au et al. Reference AU, CORLETT and HAU2006).
Which is the most adequate trap design?
As was mentioned above, the greatest influence on the effectiveness of the traps in our study was the bouncing effect, which strongly affected the estimates from funnel and basin traps. In addition, the funnel trap tended to lean and fall more frequently than the other designs, especially in sandy soils. On some occasions the basin trap also tended to lean because of the weight of the basin. The ground trap showed the greatest removal by frugivores and predators. The hanging mesh traps showed the highest removal rates by wind, though this is not as problematic when estimating fruit production. Therefore, according to the results of this study, the mesh traps on PVC frame seem to be the most appropriate design among the traps evaluated for studies of fruit production and seed dispersal in tropical forests. At the same time, this trap design would be preferred for long-term investigations because the PVC frame is strong and resistant to degradation. However, an additional consideration is that traps on PVC frame are the most expensive and the most time-consuming to install. When these economic and practical points are considered, the hanging mesh trap is a good second choice for studies of seed dispersal or frugivory, especially in forests where wind speeds are not strong. Additional consideration should be taken into account for specific sites (e.g. flooding regimes, particular behaviours of predators, mechanical damage by megafauna, etc.).
ACKNOWLEDGEMENTS
We thank Marcela Quiñones, Jorge Ahumada, and Beatriz Ramírez who collected trap data in the field. Nicole Gibson and Daniel Cadena corrected the manuscript and provided helpful comments. The study at Tinigua National Park was made possible by the logistic support from the Center of Ecological Investigations La Macarena (CIEM). The study in Caparú was made possible by the logistic help of Conservation International Colombia, and we are grateful to Erwin Palacios. This study was funded by Conservation International Colombia, Banco de La República and Universidad de Los Andes, Bogotá.
