Top-down effects of predators and bottom-up effects of resources are important drivers of community structure and function in a wide array of ecosystems. Fertilization experiments impose variation in resource availability that can mediate the strength of predator impacts, but the prevalence of such interactions across natural productivity gradients is less clear. We studied the joint impacts of top-down and bottom-up factors in a tropical mangrove forest system, leveraging fine-grained patchiness in resource availability and primary productivity on coastal cays of Belize. We excluded birds from canopies of red mangrove (Rhizophoraceae: Rhizophora mangle) for 13 months in zones of phosphorus-limited, stunted dwarf mangroves, and in adjacent zones of vigorous mangroves that receive detrital subsidies. Birds decreased total arthropod densities by 62%, herbivore densities more than fivefold, and reduced rates of leaf and bud herbivory by 45% and 52%, respectively. Despite similar arthropod densities across both zones of productivity, leaf and bud damage were 2 and 4.3 times greater in productive stands. Detrital subsidies strongly impacted a suite of plant traits in productive stands, potentially making leaves more nutritious and vulnerable to damage. Despite consistently strong impacts on herbivory, we did not detect top-down forcing that impacted mangrove growth, which was similar with and without birds. Our results indicated that both top-down and bottom-up forces drive arthropod community dynamics, but attenuation at the plant-herbivore interface weakens top-down control by avian insectivores.

We established our field experiment on Twin Cays, Belize (16.82° N, 88.10° W), a 92-ha archipelago of peat-based mangrove islands, approximately 12 km from the mainland. Island vegetation is dominated by mangroves, the majority of which are Rhizophora mangle. Black mangroves (Avicennia germinans (L.)) are locally common in some areas, while white mangroves (Laguncularia racemosa (L.) C.F. Gaertn.) are scattered and relatively rare. A narrow 5-20m band of large, productive R. mangle trees (5-6m tall) lines the island perimeters, while interior ponds are inhabited by old-growth stands of dwarf R. mangle (<1.5m tall) that are short-statured primarily due to phosphorus (P) limitation.

Two main islands (“West island” and “East island”) constitute most of Twin Cays’ landmass, and we established one study site on each of these islands ("Dock" and "Boa" respectively). Site locations were chosen based on accessibility and on the co-occurrence of fast-growing floc-associated R. mangle on pond edges and dwarf R. mangle within ponds. The two sites we selected were approximately 0.7km apart and differed somewhat in their hydrological conditions and spatial patterns of floc deposits.

Experimental design

Experimental manipulations were established 7-12 January 2010. At each of two sites, we selected 20 experimental units: 10 R. mangle dwarf trees (low resource treatment) and 10 branches on different R. mangle trees near flocculent organic deposits (high resource treatment, hereafter “floc trees”). Vigorous floc trees were much larger than the dwarf R. mangle in interior ponds. Therefore, we manipulated only portions of the floc trees of similar size to dwarfs, in order to standardize the size of our experimental units. We built 1-m3 frames of ¾” PVC around each experimental unit and randomly selected half of the frames around dwarfs and floc trees to be covered in polypropylene netting (2x2cm openings; Dalen Deer-X netting) to exclude vertebrates but not arthropods. Thus, each combination of resource treatment (floc vs. dwarf tree) by bird access treatment (no net vs. net) had a total replication of n=10 (5 per site). Nets did not extend underwater to the sediment surface, rather they stopped at the height of high tide so that they allowed floating detritus movement but not bird access. Experimental units were chosen so that they were all at least 5m apart, had 10-20 terminal shoots and could be oriented within frames so that they would not contact net coverings. Three terminal stems were haphazardly chosen on each experimental unit and marked with aluminum tags. These stems were later used to quantify plant growth and rates of herbivory (see below). Nets and frames were checked and repaired if needed at 2, 4, and 8 months after initiation of treatments. After 13 months (10-16 February 2011), we measured all final response variables.

Measurements of flocculent detritus

The depth of the floc near the base of each experimental unit was measured repeatedly; once at the start of the experiment and again after 2, 4, and 8 months. At each time point, an extending ruler was used to measure the depth of floc at five locations within the base of each frame. All floc depth measures were averaged to create a single floc depth value for each tree.

At four sites (two on each island) where accumulations of floc were present and associated with bands of rapidly growing R. mangle, we collected three floc samples and three substrate (i.e., mostly peat) samples. Floc samples were collected every 5m along a pond edge where floc was accumulating, and peat was collected 10m toward the pond’s interior from each floc sample (where only trace amounts of floc were present). Substrate %P by mass was determined by placing a known mass (~2 mg) of dried, ground material in a muffle furnace at 550°C for two hours, followed by colorimetric analysis using the ammonium molybdate method.

Assessing growth and plant traits

We assessed the growth of experimental units throughout the experiment by quantifying the length and number of leaves present on three marked twigs, as well as recording the total number of live terminal stems present. As marked twigs branched, we summed the lengths of all shoots and shoot segments distal to a marked point to calculate length values. Measurements of total twig length and overall terminal shoot abundance were taken 0, 2, 4, 8, and 13 months after treatments were imposed, while twig leaf abundance was measured 2, 4, 8, and 13 months after treatments were imposed. To create a single value summarizing the rate of change in each of these variables, we subtracted the last collected measure from the earliest collected measure for each variable and then divided these differences by the number of months that elapsed between them. Data collection errors led 3/120 marked twigs to have earliest measures later than the start of the experiment while twig death led to 6/120 marked twigs to have their latest measures before the end of the experiment (three of these were on an experimental unit that died between months 8 and 13). After means for each experimental unit were calculated for marked twig values, we performed a principal components analysis (PCA) on the three growth rate variables, first centered and scaled to unit variance, to create a synthetic variable representing the overall growth of each plant (PCA axis 1). The first PCA axis captured 87.9% of the total variance in the growth-related measurements, and loadings were similar across measured variables (0.54-0.6).

To assess the toughness and nutrient content of leaves on our manipulated trees, we collected three leaves from different terminal stems which were growing in full sun and members of the newest fully expanded pair of leaves on each stem. We used a Wagner FT penetrometer (Wagner Instruments) to measure the amount of force required to punch through each leaf. The penetrometer was pushed through the leaf at the midpoint between a leaf edge and the midrib halfway down the length of the leaf. The leaves were oven-dried at 60°C until constant weight and then the three leaves from each plant were separately ground and homogenized. Leaf %P was determined using the protocol outlined above for substrate samples. We measured leaf %C and %N by packing tin cups with known masses of dried ground leaf material (~4mg) and using a Flash EA series 1112 NC soils analyzer. Internal lab standards measured along with experimental leaf samples were used to confirm the validity of nutrient analyses.

To quantify leaf morphology (leaf mass per area, or LMA), we collected single leaves from experimental units at the East island site only, due to logistical constraints. The leaves were growing in full sun and were part of the newest fully expanded pair of leaves on a terminal stem. We took digital photographs of each leaf, used ImageJ v.1.49 to measure their surface area, oven dried the leaves to constant mass at 60°C and measured leaf biomass using an electronic balance. We subsequently calculated LMA by dividing leaf biomass by dry leaf area (mg/cm2).

Quantifying arthropod density and herbivory

At the experiment’s end, we exhaustively searched each experimental unit, collected all arthropods that were discovered using handheld aspirators, and identified them to order. Half of the foliage and branches on each plant/branch were clipped into plastic bags and searched in the lab while the remaining plant parts were searched in the field. The number of terminal shoots on the branches collected from each plant were noted and their dry mass was measured to quantify the relationship between biomass and terminal shoot abundance. One floc-tree experimental unit was damaged in a storm and died between months 8 and 11 of the experiment, so this replicate, to which birds had access, did not contribute to the arthropod and herbivory datasets.

To quantify herbivory rates on leaves, one of the newest fully expanded leaves and one of the oldest fully expanded leaves from each of the three marked stems on all experimental units were digitally photographed. We used ImageJ v.1.49 to measure the area of each leaf that had been damaged by herbivores and averaged the percent damage values for the six leaves from each tree. Leaf lifespan in R. mangle is approximately 18 months, thus the oldest leaves on each stem likely originated before the start of our experiment and were exposed to the entire duration of our experiment treatments. The young leaves that we sampled, on the other hand, likely originated during our experimental treatments and were exposed to varying durations of our bird access treatments. Both young and old leaves were included in our measurements of herbivory, despite the possibility that old leaves may have been damaged before the experiment began, because the dominant folivore of R. mangle, Aratus pisonii, preferentially feeds on older leaves.

To quantify rates of herbivory on buds, we counted the number of buds with herbivore damage on each experimental unit as well as the number of leaf pairs in which both leaves were symmetrically damaged and/or deformed, indicating the damage occurred when they were rolled within a bud. We summed the number of damaged buds and symmetrically damaged leaf pairs as an index of bud herbivory. Dead buds were not counted as “damaged” if cause of death could not be determined.

All statistical analyses were performed in R v. 3.2.3.

Please refer to ReadMe file.