Long-term resource addition to a detrital food web yields a pattern of responses more complex than pervasive bottom-up control

Experimental design

The experiment started two years after CW99 and ran from 1997 through 1999 (hereon designated Years 1, 2, and 3). Each experimental unit (20 in total) was a 2 × 2-m area of forest floor, separated from each other by at least 10 m. Experimental units were randomly assigned to one of two levels of a resource treatment and one of two levels of a fencing treatment, yielding five replicates of each of the four combinations of treatment levels. Thus, half of the units received a detrital supplement (Supplemented), the others none (Ambient), and half of the plots in each resource treatment were open to emigration and immigration (Open), while the others were enclosed with 35-cm aluminum flashing inserted 8 cm into the ground (Fenced). The fence was topped with a 15-cm horizontal strip of flashing that formed two lips coated on the underside with a tree-banding compound (Tanglefoot^®, Grand Rapids, MI, USA) to further retard movement of epigeic (ground-active) arthropods across the barrier. This design is more complex than that of CW99, which had no fenced plots but employed the same total number of experimental units: twenty 2 × 5-m open plots, half of which received a detrital supplement.

We employed the detritus-supplementation protocol of CW99, which has also been used in other experiments (Chen & Wise, 1997; Raub et al., 2014). Our goal was not to determine which components of the resource base (bacteria, fungi, and organic debris) were possibly limiting densities of microbivores and detritivores, as was the aim of Salamon et al. (2006). Rather, our goal was to follow the community response to long-term enhancement of a broadly defined resource base. This experimental approach, including that of Salamon et al. (2006) and others, involves adding artificial forms of organic matter and nutrients. Every two weeks from April through September we added chopped “fresh” (i.e., not dried) mushrooms and potatoes, and dry flakes of Drosophila medium (Carolina Biological Supply; Burlington, North Carolina, USA; Formula 4–24) to the Supplemented plots.

We decided initially to supplement at a rate approximately 1/3 that of CW99 because we hypothesized that the strong responses exhibited in the earlier experiment were due to a high level of detrital enhancement. We planned to continue this rate of supplementation in the following years, but decided to increase the rate because the increase in densities of most taxa in response to the detrital enhancement in Year 1 was much less (including no responses) than that observed by CW99. Therefore, in Years 2 and 3 we increased the rate of supplementation to a level similar to CW99. In Year 1 each Supplemented plot received 195 g (dry wt.) m⁻² of detritus (26 g m⁻², 79 g m⁻² and 90 g m⁻² of mushrooms, potatoes, and Drosophila medium, respectively). In Years 2 and 3 the rate of detritus supplementation was increased ca. 4×  (to 770 g m⁻² and 874 g m⁻² total dry wt., respectively). The slightly larger amount of detritus added during Year 3 reflects a slightly longer period of detrital addition than in Year 2. Biweekly rates were the same in Years 2 and 3. This change in the rate of resource addition after Year 1 meant that statistical analyses focusing on the Resource × Time interaction only included Years 2 and 3, when rates of resource addition were the same.

Adding artificial detritus can influence the structure of the leaf-litter layer, so it is instructive to compare amounts added with litter standing crops. In Year 1 the average (±SE) standing crop of litter (dry wt. m⁻²) in Ambient plots was 671 ± 43 g (n = 10). Thus, the dry weight of detritus added throughout Year 1 was ∼30% of the average standing crop of detritus, compared to an addition rate of ∼100% of the litter standing crop in CW99. In Years 2 and 3 each Supplemented plot received ∼136% and ∼122% of that year’s average standing crop of litter, respectively. By the end of Year 3 litter weight was only 13% higher in Supplemented than Ambient plots (F_1,16 = 4.93, P = .04), compared to a ∼30% increase at the end of CW99.

Due to our decision to increase the supplementation rate in Years 2 and 3 to ∼4× the rate of Year 1, differences in response patterns between Year 1 and the following two years can be attributed not only to differences in abiotic factors between years and time lags in the appearance of direct and indirect effects, but also to markedly different rates of detrital supplementation. In contrast, differences between Years 2 and 3 must have been due primarily to factors other than a difference in the rate of resource addition, since biweekly rates of supplementation in Years 2 and 3 were the same.

Fungal biomass

Differences in fungal abundance between Supplemented and Ambient plots at the end of the experiment were estimated by assaying leaf litter for ergosterol (Appendix S1), a common sterol in fungal hyphae that is nearly absent from plants (Weete & Weber, 1980). The amount of ergosterol is correlated with both total hyphal mass and membrane content and likely assayed both living and dead hyphae (Mille-Lindblom, Wachenfeldt & Tranvik, 2004; Ruzicka et al., 2000; Zhao, Lin & Brookes, 2005).

Arthropods

Selected taxa of arthropods active within and on the surface of leaf litter were sampled three times in Year 1 (18 July, 15 August, 13 October), twice in Year 2 (20 June, 23 September) and twice in Year 3 (3 June, 5 September). Values for the first two dates in Year 1 were averaged, yielding two values (Summer and Fall) for each year. Densities of most taxa were estimated by litter extraction and/or litter sifting. Sticky traps were used to measure activity-densities of adult flies (Diptera) just above the litter layer.

Statistical analyses

Effects of resource addition and fencing were first analyzed for Year 1 to determine how the arthropod community responded to the lower rate of detrital supplementation. We then analyzed responses for Years 2 and 3 using a repeated-measures 2 × 2 × 2 design (Resource (Ambient, Supplemented)  ×  Fencing (Open, Fenced)  ×  Year (2 and 3)). All statistical modeling was done separately for summer and fall samples (rationale is explained below). The approach just described was used for both multivariate and univariate analyses. Methodological details of the statistical modeling appear in Appendix S2.

The central conceptual question addressed by our field experiment is: How does the response of the system to resource supplementation change over time? The statistic that directly addresses this question is the Resource  × Year interaction that includes only Years 2 and 3, when the rate of resource input did not vary between years. We evaluated this interaction by relying on a combination of multivariate and univariate statistical techniques. We first used multivariate techniques to determine how detrital supplementation changed arthropod community structure over time and which response variables were most closely linked to these changes. Multivariate analyses were performed first because if there is no multivariate effect, there is no justification for doing separate univariate analyses. Because multivariate effects were present we then conducted independent univariate analyses. This same approach was use to analyze the simple effect of resource supplementation in Year 1.

We postulated that the system would respond differently in summer and fall because summer samples had been exposed to detrital enhancement for fewer months than fall samples and because life histories of many taxa show pronounced seasonal patterns. Initial multivariate analyses of the interaction between (Resource  × Year) and Season confirmed this expectation (Tables S3.1, S3.2 and S4.1 in Appendices S3 and S4). Thus, we divided the community distance matrix into summer and fall subsets, yielding one distance matrix for each season. All subsequent multivariate and univariate analyses were done separately for summer and fall samples.

The central logistical question addressed by our field experiment is: How does the openness of the plots affect the observed pattern of responses to detrital supplementation? Thus, we first tested for an interaction between (Resource  × Year) and Fencing. If there was no evidence for an interaction (P value of the 3-way interaction > .10), the design was collapsed (Open and Fenced plots were pooled, yielding twice as many replicates per Resource level) to one involving only Resource and Year. If Fencing possibly influenced the Resource  × Year interaction, further analyses were done separately for Open and Fenced plots. Use of P > .10 (instead of the conventional P > .05 criterion) for inferring no interaction with Fencing made it less likely to ignore a weak effect of fencing (due to the relatively low statistical power of only five replicates for each Resource/Fencing combination) that could have influenced interpretation of the response patterns.

Multivariate analyses

Using all 18 response variables, we calculated a distance matrix using Gower’s similarity index (S15 of Legendre & Legendre (2012)). Gower’s index was employed because it is designed to accommodate different types of variables (Kempson, sifting, and sticky-trap samples; Fig. 1). We implemented S15 because this version of Gower’s measure is a symmetrical index that gives equal weight to double zeroes (absence/absence) and ++ (presence/presence), which is the type of distance measure philosophically appropriate for analyzing results of a field experiment (unlike the more commonly employed “Bray-Curtis” measure, which is less suited to the assumptions behind controlled manipulative experiments Legendre & Legendre (2012)). Further details are in Appendix S2.

Permutational multivariate analysis of variance (perMANOVA) (Anderson, Gorley & Clarke, 2008) was then used to assess the multivariate Resource × Year interaction. Because our analyses revealed a Resource × Year interaction, we then assessed the changing impact of Resource on arthropod community structure in two ways. First, we plotted the location of each experimental unit (i.e., 2 × 2-m plot) on the first two axes of the Principal Coordinates Ordination (PCO) for each year along with the P value for the Resource effect each year. Secondly, we evaluated how different response variables contributed to the change in community structure by plotting both simple and multiple correlation vectors on constrained PCO ordinations (CAP; constrained by Resource and Fencing treatments) for each year.

Fungal density

Fungal hyphae, as measured by concentrations of ergosterol, were 3× denser in Supplemented than in Ambient plots (Appendix S1).

Relative abundance of taxa

The most abundant taxa were six families of microbi-detritivores (Collembola) collected by Kempson extraction, with the Hypogastruridae, Onychiuridae, and Entomobryidae being the three most numerous (Fig. 1A). Remaining response variables were similar in value to each other, differing by less than 50%. Centipedes (Chilopoda) were the exception. This predatory taxon had the fewest individuals sampled of all response variables (Fig. 1B).

Response of arthropods to detrital supplementation

Results of multivariate analyses are presented first because they constituted the critical first step in the analysis. Results of testing for interactions with Fencing are presented before summarizing evidence for changes in the presence and strength of bottom-up control processes over the three years of the experiment. In each section below, results for Year 1 are presented first, followed by the pattern of change between Years 2 and 3, when the rate of Resource supplementation was the same and was ∼4× higher than in Year 1.

Multivariate modeling, including univariate vector overlays

Fencing — Permutational multivariate analysis of variance (perMANOVA) uncovered no evidence of an interaction between the openness of a plot and the overall community response to the addition of detritus for either summer or fall (Appendix S3) over the course of the experiment. Therefore, Open and Fenced plots were pooled for multivariate analyses of the Resource effect in Year 1 and the Resource × Year interaction for Years 2 and 3.

Addition of detritus — In the summer of Year 1 there was no clear divergence in arthropod community structure between Ambient and Supplemented plots (Fig. 2A, Appendix S4). By fall the centroids appeared to have diverged slightly more between treatments, but evidence for a pronounced Resource effect was not strong (P = .035, Fig. 2B, Appendix S4).

In contrast, divergence in community structure due to the addition of detritus was clear for both seasons in Years 2 and 3, as indicated by the small degree of overlap between communities in Ambient and Resource treatments in ordination space (Fig. 2, Appendix S4). The impact of the Resource treatment on arthropod community structure in Year 3 differed from that of Year 2 (P(Pseudo-F_1,70(Resource × Year) = .012, perMANOVA; Table S3.2, Appendix S3). Lower dispersion among Ambient plots in the fall of Year 3 (Fig. 2B) is likely causing the Resource × Year interaction; this is the only time when within-treatment dispersions in ordination space differed (Fall, Year 3: Pseudo- F_1,18 = 8.85, P = .007, PERMDISP test (Anderson, Gorley & Clarke, 2008)).

Comparing simple Spearman and multiple (“partial”) correlations between response variables and the two axes of a constrained ordination (CAP) gives some insight into which taxa may have been the primary drivers of the observed divergence in community structure (Fig. 3). The identity of the response variables most highly correlated with the divergence changed over time. The pattern of these correlations also differed between seasons. Furthermore, several correlations were negative, indicating the community divergence was not due solely to simple bottom-up control of the response variables.

Univariate statistical modeling

Plots over time of all 18 response variables appear in Figs. S5.1–S5.18 in Appendix S5. Full results of the statistical modeling of the univariate responses, including a comparison with the vector patterns in Fig. 3, appear in Tables S6.1–S6.4 in Appendix S6. These analyses yield a complicated pattern of responses that is best summarized pictorially (Fig. 4A; note that Fig. 4A also includes some symbols for vector overlays, which have been excluded from the tallies in the following sections but will be treated in the Discussion).

Fencing — Although the multivariate analyses uncovered no interactions with Fencing, ∼30% of the univariate tests that revealed an effect of Resource (10/34; only Chilopoda, the rarest taxon, never responded) exhibited an interaction with the Fencing treatment (Fig. 4A, Appendix S6). For most groups the effect was sporadic, i.e., it occurred in only one of the six sampling periods. The two clear exceptions were larval Lepidoptera and pseudoscorpions (Fig. 4A). In both groups an interaction with fencing occurred in three or four sampling periods and always was associated with a decrease in numbers in response to detrital supplementation. The negative Resource effect occurred four times in Open plots, three times in Fenced plots

Addition of detritus — The pattern of response changed over the three years (Fig. 4A). In Year 1 eleven response variables showed a positive response to detrital addition in at least one season; the only exception was the lower number of Lepidoptera larvae in Open plots in the summer (arrows for vector overlays are not included in this and following tallies). In Year 2 eleven response variables also showed a positive response in one or two seasons, but only six of these had displayed a positive response the earlier year. In Year 3 only five variables showed a positive response in summer and/or fall (all had exhibited a positive response the previous year). In Year 2 three variables decreased in response to detrital addition in at least one season and in Year 3 four taxa responded negatively.

Thus, evidence of predominantly bottom-up control (i.e., a positive response to adding detritus) by the fall of Year 1 was transformed the following two years into a mixture of continued positive responses to detrital addition, disappearance of the early positive responses, appearance of new positive responses, and appearance of negative effects of adding detritus. Individual patterns are presented below by trophic categories.

Raw Data

The data on which analyses are based can be accessed in Appendix S7, which contains a description of both response and design variables, and in Appendix S8, which contains the raw data and values of the design variables.

Comparison with experiment of Chen & Wise (1999) (CW99)

Fencing and statistical power— Fencing half of our plots meant that we had 10 open plots compared to the 20 open plots of CW99. Thus, when the statistical model included an interaction with Fencing, the number of replicates per Resource treatment was only 5 instead of 10 as in CW99. This reduced statistical power, which occurred in ∼30% of the univariate models, would have made it more difficult to detect responses to detrital supplementation that were comparable in magnitude to CW99. However, in the remaining 70% of the univariate models the absence of an interaction with Fencing meant that the statistical power for detecting a Resource effect was comparable to that of CW99. The three instances of a fencing interaction for the microbi-detritivore taxa shared with CW99 (one sampling period each for Hypogastruridae, Tomoceridae and Smithuridae; Fig. 4A) confirmed our expectation that fencing can increase the evidence for bottom-up control, since the positive Resource effect occurred only in the Fenced plots. The barrier may have yielded a greater positive response by reducing emigration, since Collembola are not likely to climb the type of barrier we constructed. The interaction with fencing for Collembola, however, was relatively infrequent; and the direction of the interaction brought the overall pattern of Collembola responses in our experiment closer to those of CW99.

Among the primary consumers two other taxa also exhibited a similar interaction with Fencing on one sampling period (Thysanoptera and adult Diptera). Lepidoptera larvae, which were not sampled in CW99, showed a surprising pattern of weak negative responses across the entire experiment peppered with contrasting effects of Fencing (Fig. 4A). The appearance of a negative response at the very start of the experiment, and the fact that no other response variable exhibited a negative response in Year 1, suggests that the pattern is due to a sampling bias that was a consequence of only five replicates in each of the four treatments.

Among predators the interaction between Fencing and Resource was more frequent and variable (i.e., the Resource effect occurred only in Open or Fenced plots with similar frequency). The more variable pattern is likely due to the lower abundances of predators compared to other trophic categories, which would lead to higher overall sampling variation.

In summary: Interactions of Fencing with the Resource treatment were present, but showed no consistent pattern for the less-abundant taxa, and for the more abundant microbi-detritivores, the consequence was to bring the overall evidence for bottom-up control closer to that of CW99.

Plot size — Plot area was 2.5 larger in CW99 than in our experiment (10 m² versus 4 m²) but the effect of this difference on the possible swamping effect of migration is smaller than the 2.5 ratio might suggest, for three reasons: (1) the rectangular plot shape of CW99 (2 × 5 m) means that the perimeter-to-area ratio of our plots was only 1.4× greater than that of CW99 [(8/4) / (14/10); (2) microbi-detritivores showed a strong response to resource supplementation in open 1-m ² plots in an earlier experiment in the same forest (Chen & Wise, 1997); and (3) the absence of consistently strong Fencing × Resource or Fencing × (Resource × Year) interactions suggests that migration across open plot boundaries did not strongly dilute or increase the responses of most taxa to effects of detrital supplementation.

Variation in rainfall— Biweekly rates of detrital supplementation were the same in Years 2 and 3, yet the patterns of response in both years differed markedly from each other. Furthermore, Year 3 patterns differed markedly from CW99, even though rates of detrital addition in the last two years of our experiment were close to those of CW99. One factor possibly contributing to these differences is year-to-year variation in rainfall, which during Year 2 of our experiment was close to normal but was ∼50% below normal during Year 3 ( Lawrence, 2000). During CW99 rainfall was 35% above the long-term average. (In Year 1 rainfall was close to normal ( Lawrence, 2000), which may explain why several taxa responded positively to the Resource treatment even though the rate of supplementation was only 1/3 that of CW99.) Higher-than-normal rainfall in the short-term experiment of CW99 may have accelerated fungal growth in the leaf litter, intensifying the effect of the detrital subsidy compared to all three years of our experiment. Furthermore, the much-lower rainfall in Year 3 could explain why densities of four Collembola families—Onychiuridae, Isotomidae, Tomoceridae and Sminthuridae—had declined to near zero in both Ambient and Supplemented plots by the fall of Year 3 (Figs. S5.2, S5.4–S5.6 in Appendix S5) (Christiansen, 1964; Hopkin, 1997; Petersen, 2002). In contrast to this pattern, however, both hypogastrurid and entomobryiid Collembola responded strongly and positively to Resource supplementation in the fall of Year 3, similarly to the fall of Year 2. Possibly these strong responses occurred despite the low rainfall because of release from competition with the other four Collembola families. Several long-term studies suggest that strong biotic interactions influence Collembola densities, particularly in forests (Chernova & Kuznetsova, 2000; Kampichler & Geissen, 2005; Kuznetsova, 2006; Takeda, 1987; Van Straalen, 1985; Vegter, 1987; Wolters, 1998). However, direct experimental evidence is lacking for resource competition among Collembola families or apparent competition due to shifting pressures from predation.

Comparison with other experiments testing for bottom-up control in terrestrial detritus-based food webs

Other experiments usually employed plot sizes comparable to, or smaller than, those we used, and they were conducted for less time. Two exceptions are grassland studies (Fountain et al. 2008; Patrick, Kershner & Fraser, 2012) in which inorganic nutrients were added for 4 years to very large (240 or 314 sq. m, respectively) unfenced plots. Fountain et al. (2008) found that densities of isotomid Collembola and three spider families increased, but densities of two other spider families decreased, in response to nutrient enhancement. In contrast, Patrick, Kershner & Fraser (2012) observed only positive responses by families of web-building and cursorial spiders. In another grassland study that used much smaller (1 × 1-m) open plots, Hoekman et al. (2011) added midge (Diptera: Chironomidae) carcasses for two years or over a single year starting at different times. As in our study, Collembola, larval Diptera, and Coleoptera exhibited varying positive responses. Spiders, however, showed no treatment effect, in contrast to our Year 1 and CW99. Because their plots were just 25% the area of our experimental units and 10% that of CW99, movement across plot boundaries could have diluted positive responses by spiders to increased prey. In addition, increased cannibalism and intraguild predation could have contributed to the absence of an effect on total spiders. Oelbermann, Langel & Scheu (2008) directly manipulated energy input to a grassland detrital web in fenced 5-m² plots for five months. Many Collembola families responded positively as did total spider numbers, which were ca. 2 × higher in the detritus-enhanced plots. Some spider taxa, however, exhibited no numerical response even though stable-isotope analysis revealed that all sampled spiders relied more heavily on the decomposer food web in the detritus-addition plots. Oelbermann, Langel & Scheu (2008) suggested that even though the secondary-consumer trophic level was strongly linked to the detritus base, individual groups of predators showed inconsistent evidence of bottom-up control due to increased cannibalism and intraguild predation in the plots with additional detritus. Similar interactions could have weakened bottom-up control among predators in Year 1 (with delayed effects in Years 2 and 3), but one still has to explain the strong bottom-up control among predators in CW99.

The other comparable experiments on detritus-based systems have been conducted in forests. Scheu & Schaefer (1998) and Maraun et al. (2001) increased microbial growth in the litter layer of fenced 1-m² plots by adding glucose and nitrogen for 15 months with no effect on Collembola or centipedes of the litter layer. In contrast, Collembola and centipedes of the lower soil horizon responded negatively, most likely due to indirect effects of increased earthworm densities (Scheu & Schaefer, 1998; Maraun et al., 2001). We did not sample lower soil layers and in our forest earthworms appeared to be much less abundant than in the German beech forest on limestone (Göttinger Wald) where these studies were conducted (D Wise, pers. obs., 1984). Also, centipedes were rare in our study, in contrast to the much higher abundance just a few years earlier in CW99—a difference for which there is no obvious explanation. One might speculate that the rarity of a major predatory group could have altered the pattern of bottom-up control among other predators, but a possible mechanism is illusive.

Another explanation for why responses observed by Scheu & Schaefer (1998) and Maraun et al. (2001) differed from ours is the quite different type of resource enhancement they employed. However, Salamon et al. (2006), who used a similar type of resource enhancement in 1-sq. m fenced plots for 17 months, found that several Collembola families did respond positively with effect sizes similar to those observed in our experiment. Unlike our Year 1, though, spiders did not respond, but the absence of a response by pseudoscorpions does resemble our findings.

The forest-floor experiment of Raub et al. (2014) is the closest to ours in technique, as every two weeks for 3 months they added to 1.5 × 1.5-m unfenced plots the same type of artificial detritus used in CW99 and our study. Collembola numbers responded positively to resource enhancement, but total predators (spiders, pseudoscorpions, and centipedes combined) did not increase.

It is difficult to discern a pattern from such disparate studies, except that the general absence of bottom-up control among predators is similar to our results for Years 2 and 3 (but not for Year 1 nor CW99). Many studies did not employ the same degree of taxonomic resolution employed by us and CW99, which means that the absence of a response for an entire grouping (i.e., Collembola, predators) does not imply that smaller taxonomic categories did not respond positively and/or negatively. We avoided this problem by first using a multivariate approach to examine changes in community structure for a reasonably large number of response variables. Variation between studies in abiotic factors, such as rainfall, could explain some of the differences in outcomes, but that data is generally lacking, or the experiment was usually not conducted for a long enough time for rainfall differences to appear. The openness of the plots, especially for those that were smaller than those in CW99, could explain some of the variable responses. Our 3-year experiment, with open and fenced plots and the ability to compare directly with CW99, overcomes many of these challenges to interpretation.

Towards a deeper understanding of bottom-up control in detrital food webs

Some of the weakening of bottom-up control between Years 2 and 3 in our experiment can be explained by 50% less rainfall in Year 3. In detrital food webs a weak positive response by microbi-detritivores to resource addition is expected if drought impedes fungal growth. However, lower rainfall, although a likely contributing factor, cannot entirely explain the weaker bottom-up control in Year 3. Fungal hyphae were still 3× more abundant in Supplemented than Ambient plots in Year 3 (Appendix S1), and the total amount of detrital material (leaf litter plus added detritus) at the end of Year 3 was only 13% greater in Supplemented than Ambient plots, compared to a ∼30% difference for CW99, when rainfall was higher than any year of our experiment. Thus, fungal growth may still have been substantial in Year 3. Although drier conditions in the litter could explain why densities of four Collembola families were close to zero in all plots in the fall (and hence failed to exhibit bottom-up control), not all primary consumers exhibited a fall crash in Year 3 (Appendix S5). In addition, examination of changes in densities of Collembola families during the summer reveals no correlation with changes in rainfall over the three years of the experiment. Most critically, declining rainfall cannot readily explain the negative responses to detrital supplementation. Why was one major Collembola family, the Tomoceridae, less abundant in Supplementation than Ambient plots in Year 3? In this forest predation by cursorial spiders depresses tomocerid densities (Wise, 2004), but cursorial spiders were less, not more, abundant in the Supplementation treatment in Year 3. In fact, the widespread negative responses of predators to detrital addition the last two years was surprising when compared with CW99 and other studies. It is unclear how all these negative responses could result from shifting intensities of intraguild predation and cannibalism in response to increased abundances of microbi-detritivores. Possibly contributing factors are declining numbers of high-quality prey (Tomoceridae) (Toft & Wise, 1999b) and increased densities of possibly toxic isotomid (Toft & Wise, 1999a; Toft & Wise, 1999b) and hypogastrurid prey (Messer et al., 2000; Negri, 2004) in Years 2 and/or 3 in the Supplemented plots. It is difficult to evaluate the solidity of such post-hoc theorizing without additional information.

In addition to the appearance in the last two years of negative responses to detrital supplementation by predators and tomocerids, variation in community structure between Supplemented plots was greater than between Ambient plots the fall of Year 3—the only sampling period that displayed this difference (Fig. 2). This increased variation in community structure after three years, and all the unexpected contrasts of the univariate analyses with CW99, suggest that that long-term resource supplementation will lead to a community configuration that cannot be predicted from initial short-term population responses. On the other hand, could such negative responses in long-term experiments simply be random vagaries emerging from the low replication in most field experiments? Or might they be the surprising results that ecologists often fail to report (Doak et al., 2008)—the unexpected observations that can help build stronger theory? Additional long-term experiments will help resolve this dilemma.

How long is long enough? The behavior of mathematical models of long-term press perturbations suggests that many generations will be required to reach a new equilibrium (Bender, Case & Gilpin, 1984; Raffaelli & Moller, 2000; Yodzis, 1988), which coupled with environmental noise, such as variation in rainfall, might make the prospects of a “long-enough” field experiment seem hopeless. However, after reviewing short- and long-term experiments (2–31 months) in the intertidal, Menge (1997) concluded that “. . . community dynamics may be more predictable than expected . . .” because most indirect effects had appeared half-way through the experiment. “Long enough” is at least the number of generations sufficient to reveal indirect effects. An even longer time would reveal how close the community may be to a new equilibrium. This longer time will never be achieved with certainty, but snapshots yielded by short-term experiments cannot answer the question.

Simply advocating experiments of longer duration is not enough. Even in long-term experiments we need to reduce the impact of unpredictable and unknown variation in other factors. Furthermore, we must have a deeper understanding of which interaction pathways were altered in response to resource supplementation. How might these two challenges be met?

The challenge of accounting for uncontrollable yearly variation in abiotic and biotic factors can be addressed by expanding the experimental design to include initiation of resource additions in new sets of experimental units in each subsequent year after the experiment has started, along with continuing the experimental treatment in the plots from the first year. Phased temporal replication of the experimental perturbation offers the surest way of clearly separating effects due to time lags from those due to changing levels of unknown but influential abiotic and biotic factors. Hoekman et al. (2011) employed a preliminary but elegant version of this approach by combining two yearly pulse treatments of detritus addition with a two-year press treatment.

The second challenge can be met in several ways. One is the use of path analysis and structural equation modeling to help separate direct and indirect effects among interaction pathways between taxa in major trophic groupings in the food web (Grace, 2006; Wootton, 1994). However, this statistical modeling approach requires numerous replicates, more than are possible with most types of detrital enhancement experiments performed to date. Tradeoffs between too-many response variables, too-few replicates, and experimental units too small to be realistic present daunting challenges. Another approach is to measure more variables than just the densities of primary consumers and their predators. Molecular techniques are available to measure different functional categories of fungi (Nguyen et al., 2016; Shokralla et al., 2012). Shifts in major trophic pathways can be revealed by techniques such as stable isotopes and fatty acid analysis (Halaj & Wise, 2002; Ruess et al., 2004), and PCR of prey DNA in guts of predators such as spiders and carabids (Harper et al., 2005; Symondson, 2002).

Some of the above suggestions already have been incorporated into short-term field experiments testing for bottom-up control in detritus-based terrestrial food webs. Utilizing all of them in a single study is a challenge. Nevertheless, paying heed to their advantages when planning future long-term research would help reveal likely causes of the type of variability found in our 3-year experiment.