Willapa Bay is a shallow estuary on the coast of Washington State, USA. About half the bay’s area dries at low tide, and both oyster and clam aquaculture occur on these broad intertidal flats. The bay is a major contributor to U.S. shellfish production (Ruesink et al. 2006), primarily of non-native species (Pacific oysters, Magallana gigas; and Manila clams, Ruditapes philippinarum). The average depth is 3.2 m and mean tidal range is 1.9 m (Hickey and Banas 2003). Study sites were located on privately-owned commercial clam beds at elevations of 0.7 to 1.4 m relative to mean lower low water (Table 1). Crushed rock had been added as protection from predators. Study sites spanned 20 km along the western shore of the bay (Table 1, Fig. 1), a gradient that is typically warmer and less saline up-estuary (to the south) in summer (Ruesink et al. 2015). Despite this gradient, summer salinities exceed 25 PSU at all sites (Ruesink et al. 2015), whereas EGC can live and feed at much lower salinities (De Rivera et al. 2025). A northern and southern region are separated by a physical feature of the bay, known as the “fattening line,” beyond which the water residence time shifts from ~1 week to ~1 month (Banas and Hickey 2005); sites were evenly divided between the two regions. Recruitment monitoring in 2023 and 2024 demonstrated substantial natural recruitment of non-native clams, especially up-estuary (Table 1, Grason et al. 2024).

Three introduced species at two trophic levels were the focus of this study. Manila clams were inadvertently introduced with Pacific oysters before the 1950s and now dominate commercial clam aquaculture in Washington State (Decker 2015; Grason et al. 2024). Soft-shell clams (Mya arenaria) arrived as aquaculture hitchhikers from the eastern U.S. and were intentionally planted in the late 1800s and were briefly abundant in many U.S. west coast estuaries (Carlton 2023). The two clams have similar summertime reproductive periods in Willapa Bay (Grason et al. 2024). Over their lifespan, Manila clams develop sturdy shells and are limited to surface sediments by their short siphons, whereas large soft-shell clams can bury deeply. European green crabs (Carcinus maenas) spread northward after their unintentional introduction to San Francisco Bay in the 1980s (Wonham and Carlton 2005), with a brief appearance in Willapa Bay in the late 1990s and more recent rapid increase (Yamada et al. 2025).

Three trapping datasets were used to evaluate the patterns of relative abundance of predators across the Willapa Bay estuarine gradient. The first dataset was collected from a synoptic trapping effort conducted at the eight sites where clam recruitment (Table 1) and predator exclusion (small seeded plots, described below) took place. During this survey, predators were trapped with an overnight soak of baited traps from 2–3 August 2024. At each site, 10 traps were set during low tide, specifically five square Fukui traps (13 mm mesh) and five minnow traps (Gee-40, 50 mm diameter opening, and 6.35 mm mesh) alternating at 10-m intervals along a depth contour. All traps were baited using ~100–150 g Pacific mackerel (Scomber japonicus) and left on the mudflat overnight, for a 24-hour deployment period that included a night-time high tide. The traps were retrieved at low tide. During the trap retrieval, all organisms in the trap were identified to species and counted. For crabs other than EGC, size was measured as total carapace width at the widest point and sex determined on-site. Trap contents other than EGC were then released. Captured EGC were measured, sexed, and counted off-site, and they were subsequently euthanized.

Two other trapping-based EGC monitoring efforts currently occur in Willapa Bay and include sites spanning from Stackpole near the mouth to up-estuary sites in the southern part of the bay (Fig. 1, Suppl. material 1: table SS1). We queried 2024 data from both of these programs to augment our interpretation of the spatial pattern of EGC. First, Washington Sea Grant (WSG) Crab Team is a regional participatory science monitoring program that engages community members and staff at partner organizations to conduct systematic trapping and molt surveys that target EGC (https://wsg.uw.edu/crabteam). Crab Team trapping protocols entail setting six traps at designated sites for a single overnight soak each month from April through September annually (Grason et al. 2018). Similar to the present study, this protocol includes two trap types, Fukui and minnow, set in alternating order at 10-m intervals, with the slight difference that minnow traps used in Crab Team surveys have a smaller opening (25 mm diameter). All WSG Crab Team monitoring sites in Willapa Bay occur in estuarine, stream-influenced, channels. Second, the Washington Department of Fish and Wildlife (WDFW) green crab management program initiated a similar monitoring effort for EGC in 2024. The WDFWEGC monitoring protocol utilizes three types of traps; in addition to Fukui and minnow (50 mm opening) traps, rigid box-style shrimp traps are used (Pro-Mar TR-244, 25 mm mesh size). Three traps of each type are arranged in a transect alternating trap type and placed 10-m apart. The six WDFW monitoring sites in this study consist of salt marsh channels or edges of emergent marsh vegetation. While WDFW monitors monthly from April through October, not all sites were sampled every month. Thus, for this exploration, WDFW monitoring data were only used from April, May, and August 2024. While both programs also monitor at sites to the north of the mouth of the bay, data from those sites were not included in this investigation because no comparable clam survival observations were collected. Both programs use bait and data collection procedures consistent with the present study. Though trap, seasonal, and habitat-related differences in protocols make it difficult to compare absolute catch rates across these three trapping efforts, within-effort comparisons will still yield valid site-to-site comparisons of spatial/geographic patterns of relative abundance of EGC during 2024.

Predator exclusion experiment - small seeded plots

Predator access to clams was manipulated in a field experiment conducted at the same eight sites as above. All experimental units consisted of window-screen mesh boxes (20 × 20 × 10 cm; Suppl. material 1: fig. S1), dug into the sediment, and then local sediment was placed inside the unit. Each unit was covered by sewing Vexar mesh (1 cm mesh) to two sides at the top. Once sediment was added to exclosures in the field, the top was fully closed by connecting to the two remaining sides with zip-ties. The Vexar mesh of control units had two large cuts, 3 cm × 10 cm, allowing predators to enter and exit the unit. Before the units were sealed in the field, five large local Manila clams, ~40 mm, and 20 small hatchery-raised Manila clams (6.27 mm, SD 2.41, N = 25) were placed into each unit. Large clams were pushed into the sediment, while small clams were covered by additional substrate. The experiment was deployed 11–12 July and collected 23–24 August 2024 for a total of 6 weeks. At collection, units were sieved (to 2 mm) on the tidal flats to reduce the volume of sediment necessary to carry offsite. The remaining substrate was then sorted by hand, identifying and measuring all live clams, including Manila, soft-shell, and other clams.

Predator exclusion experiment - large nets

Large nets were installed by anchoring with rebar and digging the edges into the sediment, a commercial aquaculture method, at two northern and two southern sites (Table 1, Fig. 1). One net (14 × 5 m) per site with 6.35 mm mesh was installed in November 2023, and another with 12.7 mm mesh in April 2024; this larger-mesh net varied in size depending on grower preference: 12 × 6 m at ST and MC, 6 × 6 m at OY, and two 6 × 6 m nets at WO. Sampling for clams occurred in November 2023 (prior to small mesh installation), April 2024 (prior to large mesh installation), and November/December 2024 by digging 0.0929 m² cores to 10 cm depth and sieving material to collect any clams >5 mm length. All Manila clams in each sample were measured for shell length (nearest mm), which demonstrated a highly bimodal size frequency and enabled us to divide the clams into two size categories – small (<2.5 cm) and large (>2.5 cm). Counts in each size category were determined at each of the four sites for samples under the small-mesh net (N = 3), samples under the large-mesh net (N = 3), and reference samples. Reference samples were collected in a grid across the entire clam bed at each site (ST: 9,608 m², OY: 9,931 m², MC: 20,112 m², WO: 30,000 m²). Each grid was divided into four quadrants, which we considered replicates of the treatment without nets, within which were 8–11 subsamples (on average, 9).

Each of the eight sites had an associated distance along the estuarine gradient based on their Northing geoposition, and an intertidal elevation determined by RTK-GPS (Table 1). Northing geoposition was adjusted to distance from the estuary mouth in km, so larger values represent distances further up-estuary. Separately, the sites were grouped into northern and southern regions.

Crab catch was compared across three model structures for each of two crab species separately. In each of the three model structures, the predictors represented different environmental gradients. Intertidal elevation represented a gradient of desiccation and access, given that Dungeness crabs (Metacarcinus magister) often move onto tidal flats to forage at high tide (Holsman et al. 2006). Distance from estuary mouth represented a suite of correlated thermal, hydrodynamic, and salinity changes characteristic of ocean-to-river gradients. North and South regions incorporated the residence time changes indicated by the “fattening line.” These models were compared with Akaike Information Criterion (AICc, adjusted for small sample size): 1) Distance into estuary (km) + Intertidal elevation (m), 2) Distance into estuary (km), 3) North vs. South region as a categorical predictor. Response variables were the number of each of two crab species captured in traps (EGC and native Dungeness crabs). These analyses only included data from the trapping effort in August 2024 that precisely matched the predator exclusion sites. In all models of crab catch, the 10 traps from each of the 8 sites were nested as replicates within the site. We assumed a tweedie data distribution, which accommodates data with many zeros and occasional very large non-zero values; residuals were examined and found to meet assumptions. Relative abundance of EGC across the estuary as surveyed in the WSG Crab Team and WDFW green crab monitoring efforts was evaluated visually to assess whether patterns were consistent with those found in synoptic sampling.

Analysis of the predation experiment (small-scale) was focused on determining if site-level EGC catch predicted the treatment effect of excluding predators. Inherently, this question addresses the EGC x Treatment interaction effect, rather than the main effects of EGC catch or predator exclusion treatment. This two-factor analysis was carried out on the response variable of live Manila clam count per experimental unit at the end of the experiment. As in the analysis of EGC spatial patterns, only EGC data from the synoptic trapping effort in August 2024 was used for this analysis. Site was considered a random effect to account for the physical set-up of the study design. Small and large clams were analyzed separately. Some of the final counts of small clams exceeded the number initially added, which reflects that clams were already present from the 2023 recruitment season when we added local sediment to the experimental units. Soft-shell clams, all of which would have been added with the local substrate at the time of experimental setup, were similarly analyzed for EGC x Treatment, with site as a random effect. Data were initially assumed to have a poisson distribution; however, gaussian provided better residuals in the analysis of small Manila clams.

Analysis of the predation experiment (large nets) was done separately for large (>2.5 cm) and small clams (<2.5 cm) for data collected in November/December 2024. (Earlier dates are visualized in Suppl. material 1: fig. S2.) Each of the four sites had two nets (except for one site with three: WO) and four quadrants of the clam bed sampled outside the net. This strategy provided two or three true replicates of predator exclusion and four open areas at each site. Manila clam counts were the response variable, and site, netting, and their interaction were fixed effects, with a group factor as a random effect to account for multiple subsamples per net or quadrant. Data fit assumptions with a negative binomial distribution for both size classes of clams.

All statistical models were built with the package glmmTMB (Brooks et al. 2017) and residuals tested with the package DHARMa (Hartig 2016) in R version 4.5.1 (R Core Team 2025).

In the synoptic baited trapping effort at the eight sites, catch rate of crabs differed across the estuarine gradient for EGC but not for Dungeness crab (Suppl. material 1: table S2). For EGC, highest catch on these mid-intertidal flats was in the northern region towards the mouth of Willapa Bay, with fewer in the southern region (Fig. 2). The continuous north-to-south distance predictor was statistically significant, but the model with north vs. south region provided a better fit. Therefore, the regional catch pattern for EGC was opposite the recruitment pattern for clams (Suppl. material 1: table S2, figs S3, S4). Dungeness crabs (Metacarcinus magister) were 87% less abundant than green crabs and showed no regional difference in density (Suppl. material 1: table S2). Elevation did not significantly improve model fit for either crab species over the range of elevations trapped.

The size of both species of crabs varied predictably across the estuarine gradient, but in opposite directions (Suppl. material 1: fig. S5). Notably, we observed clear size differentiation of Dungeness crabs at the two sites where they were abundant; crabs were generally 70–120 mm carapace width at the site closest to the mouth of the estuary, but ranged from 40–80 mm at the furthest up-estuary site. By contrast, EGC were smaller near the mouth of the estuary; most small crabs (<60 mm) were caught at northern sites, and most large crabs (>60 mm) were caught at southern sites.

Two other trapping programs also showed spatial variability in the catch of EGC (Fig. 3). Sites further up-estuary than our southernmost had very low catch rates, only rarely trapping EGC. Sites near Nahcotta, which grouped with the southern part of the bay in our analyses, had a high EGC catch rate in the WSG Crab Team data, but lower catch rates compared to northern sites in the WDFW monitoring.

Although 20 small Manila clams were added to each experimental unit, exclosures in the south part of the bay had more than 20 recovered after six weeks, which we interpret to mean that small clams were already present in the local sediment added to each unit. In the experimental analysis, the statistical interaction of EGC catch and exclosure treatment was not significant for small clams (Z(1,74)=0.24, P = 0.81). This means that the effect of predation at a site did not depend on how many EGC were trapped there (Fig. 4). The main effect of EGC was significant (Z=-3.7, P < 0.001), which simply reflects an inverse relationship between small Manila clams and crab density. The treatment effect was also significant, with open treatments on average having 8 fewer clams than in exclosures, or a survival of 60% of the original 20 clams (Fig. 4A, Z=-4.5, P < 0.001). By contrast, no factors were significant for large clams (Suppl. material 1: table S3). Large clams were found at the end of six weeks at essentially equivalent densities to the start of the experiment, indicating that they were not susceptible to mortality sources in summer in these graveled clam beds (Fig. 4B). Soft-shell clams, which were not deliberately added, were found at low densities in both open and exclosure units, mostly <1 clam per unit except at the two sites near Nahcotta with 3 clams per unit, but no factors were significant (Suppl. material 1: table S3).

Densities of large Manila clams were enhanced by about 50% by netting for 8–12 months at all four sites, with a non-significant trend for this enhancement to be less at the site farthest up-estuary (Woody’s; Fig. 5B, Suppl. material 1: table S4). The results for small clams were much more variable across sites, with two showing order-of-magnitude increases in small clam densities under nets – at the extreme ends of the estuarine gradient (Stackpole, Woody’s) – and little effect in the middle (Fig. 5A; Suppl. material 1: table S4). The small clams that were enhanced by netting were collected at a minimum size of 5 mm; these clams must have settled in 2023 several months prior to net placement, since the 2024 cohort would still be <2 mm by December (Ruesink et al. 2014a).

The estuarine gradient helped explain distributions of non-native clams and EGC, for which larval ecology rather than physical factors per se (i.e. temperature, salinity) provides an underlying mechanism for their inverse relationship. Clam larvae are likely produced and retained within the bay, therefore settling disproportionately in the southern region of long water residence time (Banas and Hickey 2005). By contrast, at the present stage of the ongoing invasion of Willapa Bay, the majority of EGC larvae likely originate in other bays along the coast where green crabs are currently more abundant (e.g. Oregon, California and British Columbia, Canada) and transit coastal ocean waters before settling in Willapa Bay (Banas et al. 2009), hence showing higher catches in the north region of the bay, which has strong ocean influence (Banas and Hickey 2005). It is expected that as EGC continue to expand their population into the southern region of Willapa Bay, this pattern will shift, as, similar to clams, long water residence time will favor retention of crab larvae released by adults further into the estuary. Thus, the current gradient in relative abundance of EGC could be a transient phenomenon.

Other predators are likely to be contributing to the clam mortality observed along the entire estuarine gradient, such that we could not predict clam losses from EGC alone. Nevertheless, clam loss did not vary predictably with Dungeness crabs, which were the other crab predator abundant at our trapping elevations. Like EGC, Dungeness crabs also have an extended pelagic development phase, resulting in densities that decline away from estuary mouths for both intertidal recruitment and 1+-year crabs in channels (Armstrong et al. 2003; Dumbauld et al. 2021). Therefore, the pattern observed from intertidal traps, in which Dungeness crab densities were elevated at both ends of the estuarine gradient, was unexpected. Other possible drivers include recent trapping, distance to channel, and foraging opportunities (Armstrong et al. 2003; Holsman et al. 2006; Lewis et al. 2021), but trapping sites with few Dungeness crabs were not markedly distant from channels (pers. obs.). Although uncommon in the traps in this study, native red rock crabs (Cancer productus) were caught at low intertidal elevations where large outplanted clams suffered high mortality in previous field experiments in Willapa Bay (Grason et al. 2024). Overall, our results reinforce the challenge of detecting novel predator effects within a complex food web in which clams across life stages are eaten by many taxa.

EGC abundance was not a significant predictor of the predation effect on clams in our study. This conclusion is complicated by a generally negative relationship between Manila clams and crabs in the small-plot experiment, which we interpret as having added small clams along with locally sourced sediment. The key outcome is the lack of statistical interaction between predation treatment (open vs. exclosure) and EGC relative abundance. Open units lost small clams during the experiment, suggesting top-down control in aggregate; however, this predation effect did not increase with greater EGC catch. Similarly, larger nets improved densities of large clams regardless of position along the estuarine gradient, while small clam densities increased dramatically under nets only at the two ends of the gradient.

These field experiments at multiple sites caution against generalizing field impacts of EGC from their demonstrated capacity to consume clams. Laboratory feeding trials, along with enclosures of EGC with specific prey, show their predatory potential, especially when the crabs are large relative to prey. EGC were less likely to damage 40-mm Manila clams than juvenile oysters when held in tanks (Anaya et al. 2025), and EGC recruiting into bags of clams were associated with reduced survival of seed clams, while adult EGC were able to consume >36-mm Manila clams (Grosholz et al. 2011). Additionally, previous diet analysis from Willapa Bay showed that Manila clam consumption by EGC was extremely rare (Fisher et al. 2024), supporting the inference that EGC are not yet exerting a significant population-level impact on Manila clams. Similar to our study, in five estuaries to the south of Willapa Bay, no correlation arose between predation intensity and local EGC density, although in that study predation varied while EGC catch was unrelated to salinity (de Rivera et al. 2025). While the role of EGC in top-down effects on clams currently appears small, substantial evidence has accumulated that native crabs limit clam production in the northeast Pacific region by consuming juvenile clams (Dudas et al. 2005; Dethier et al. 2019). In comparison to laboratory feeding trials, field studies allow EGC to encounter impediments with buried clams, as well as alternative prey.

Several possible explanations exist for a lack of detectable predation impact of EGC in the current study. First, abundances of EGC in Willapa Bay may not yet be high enough to yield detectable effects even on small clams; for instance, catch per trap was below levels that have previously been suggested to result in top-down effects on clams (Grosholz et al. 2011). Second, substrate at all sites had been augmented by gravel, an effective commercial intervention for clam growers (Ruesink et al. 2014a) that might have conferred protection on small clams from EGC even in open treatments. Third, it is also possible that crab sampling, though synoptic and consistently implemented, did not sufficiently characterize EGC abundance across the 6-week predator exclosure deployment. For example, if additional population suppression trapping occurred shortly before or after our predator assessment at one or several of the sites, it could have biased our estimates of relative abundance with respect to what clams experienced in the experiment. The consistency of the spatial pattern in EGC catch among the three efforts reported here makes this last issue less likely. Nevertheless, the invasion appears still to be in the growth phase in Willapa Bay, and it therefore seems likely that predation by EGC will become increasingly important and costly.