Richardson Bay is an approximately four km² embayment along the western edge of San Francisco Bay, 10 km north of the Golden Gate (Fig. 1). Waterfront land use is predominantly residential, commercial, and parkland with no industrial zones or commercial ports. The exact history of drill introduction to Richardson Bay is uncertain, but there was a fenced oyster bed in the eastern half of Richardson Bay in the vicinity of Hilarita Ave. (Fig. 1) sometime between 1851–1869 (Bonnot 1935; Barrett 1963). The Richardson Bay intertidal shoreline is approximately 16 km long and characterized by a combination of riprap, rocky substrate, salt marsh, sandy beach, and mudflats. Rocky substrate includes material from both natural and built surfaces, such as demolished seawalls and other structures. Eight sampling sites were chosen to represent Richardson Bay’s range of shoreline characteristics and drill densities. In pre-project site visits, Atlantic drills were not observed at the four sites in the western half of Richardson Bay but were observed at a variety of densities at the four sites in the eastern half of Richardson Bay (Fig. 1). Conversely, oysters were not observed in pre-project site visits at the four sites where drills were observed, but oysters were observed at the four sites were drills were not observed.

Figure 1. 

Study area. Map indicates the location of the eight study sites in Richardson Bay, California, an embayment of San Francisco Bay (inset). Atlantic oyster drills are present in the north and east sides of Richardson Bay at varying levels of abundance, but they are absent in the south and west sides. “(S)” indicates sites where salinity loggers were deployed.

We collected drill population data at all eight sites quarterly from June 12, 2017 to July 28, 2018 (N = 5 time points) at three tidal elevations where substrate was present: 15 cm (low intertidal), 46 cm (middle intertidal), and 76 cm (high intertidal) above MLLW (hereafter, +15 cm, +46 cm, and +76 cm. Pre-project exploratory surveys around Richardson Bay showed that +15, +46, and +76 cm correspond to lower, middle, and upper elevations at which hard substrate is present to support drills and their prey. At five sites, the low or middle elevations are situated on mud flat, which is typical of subtidal elevations in San Francisco Bay (Table 1). Drills spend little time outside of their feeding grounds except for short forays across mud or sandy bottom in response to chemoreception of prey (Pratt 1977; authors’ observations). Therefore, based on opportunistic and historical observations in Richardson Bay that found no drills in the mud flats, we assumed that drill density on mudflats was zero and did not perform time-consuming transect surveys in that environment.

Drill surveys consisted of placing a 0.5 × 0.5-m PVC quadrat at five randomly chosen points along either side of a 30-m shore-parallel transect, resulting in 10 quadrat samples per transect. On six occasions, we were unable to attain a transect survey due to the rising tide. This approach resulted in a total of 693 quadrats from 79 transect surveys. Each quadrat was searched exhaustively to generate count data for drills and oysters. We also measured surface cover of barnacles, which are among the strongest attractant for Atlantic drills in general and a primary prey source for drills in Richardson Bay (Pratt 1977; Rittschof and Gruber 1988). All of our field equipment and gear, especially boots, were thoroughly washed between site visits to prevent accidental introduction of drills to non-drill sites.

We estimated tidal elevation using tide predictions for Sausalito, CA from the National Oceanographic and Atmospheric Administration (www.tidesandcurrents.noaa.gov) and Tide Graph mobile app (Brainware, Long Beach, CA). At each study site, we deployed a stake at the water’s edge at the time that the tidal elevation was predicted to be +15 cm, +46 cm, and +76 cm. These stakes were left in place and used as transect start and end points for repeated visits to each site.

We measured substrate composition along each transect using point-intercepts within a 0.5 × 0.5 m PVC quadrat threaded with twine to make a 6 × 6 grid with 36 intersecting points. The quadrat was placed ten times at three-meter intervals on alternating sides of the transect, and the point on the beach immediately beneath each intercept was classified as according to a modified Wentworth scale of grain size: boulder ≥ 250 mm, cobble ≥ 65 mm and < 250 mm, pebble ≥ 2.5 mm and < 65 mm, sand (< 2.5 mm), or mud (saturated clay or clay-dominant mixture) (Bunte and Abt 2001). For statistical analysis, substrate classifications were grouped as either coarse (boulder or cobble, suitable habitat for drills) or fine (pebble, sand, or mud, unsuitable habitat for drills). We represented coarse substrate cover as the average amount of cobble or boulder (the requisite size class of substrate for Atlantic drill feeding and reproduction) per quadrat, as measured by psi (Ψ), the product of the sum of the number of points of boulder or cobble per quadrat and a multiplier, the base-two log of the average grain size of the boulder and cobble grain size classes, respectively (Bunte and Abt 2001). The grain-size multiplier accounts for the three-dimensional aspect of usable rocky surface per quadrat. Substrate composition was assumed, and informally observed, to be static over time, and therefore only measured once per elevation per site.

HOBO UA-002-64 Pendant Temperature/Light loggers (Onset Computer Corp., Bourne, MA) were deployed at each site on PVC stakes at +15 cm, +46 cm, and +76 cm from June 2017 until July 2018. These loggers recorded ambient air and water temperature at 15-minute intervals. Air and water temperatures were later separated using water level data from NOAA Tides and Currents (www.tidesandcurrents.noaa.gov). We summarized water temperature based on the percentage of logger readings above 26.5 °C (optimal temperature for drill growth) and 37.6 °C (critical mortality threshold) and below 10 °C (lower limit of normal activity) (Suppl. material 2: table S1).

We measured inundation (water depth relative to MLLW) at each site from July 2017 until December 2017 using a HOBO U20-001-01 (Onset Computer Corp., Bourne, MA) water level logger deployed at +15 cm taking absolute pressure readings at 15-minute intervals. Inundation was calculated using HOBOware Pro 3.7 software, which converts absolute pressure readings to water depth factoring for atmospheric pressure data, which was downloaded from NOAA Tides & Currents. Data were summarized as the average percentage of time per 24-hour period each survey elevation was out-of-water.

We measured salinity at three sites from December 2017 until June 2018 using Odyssey conductivity and temperature loggers (Dataflow Systems Ltd, Christchurch, NZ) deployed at +15 cm taking readings at 30-minute intervals. The three deployment sites were chosen to represent the geographic extremes of Richardson Bay based on their positions closest to and furthest from the ocean (Dunphy Park and Hilarita Ave., respectively) as well as the most sheltered study site (Cove Apartments) (Fig. 1). Conductivity was converted from specific conductance (SC, mS/cm) to practical salinity units (PSU) following Wagner et al. (2006).

In order to associate the effectively continuous measurements of temperature, salinity, and inundation to the temporally discrete measurements of drill abundance, we summarized these three factors by season to match the five survey timepoints. Seasons were defined as summer (June 1–August 31), fall (September 1–November 30), winter (December 1–February 28), and spring (March 1–May 31).

Using generalized linear mixed effects models (GLMM), we built two types of models to regress separately Urosalpinx population density and presence/absence on the four abiotic habitat factors described above. GLMMs were chosen for their flexibility in accounting for both fixed (environmental factors) and random (timepoint and site) effects and are suitable for analyzing data that contain repeated measurements over time, like those in this study, and exhibit spatial correlation (Bolker et al. 2009; Zuur et al. 2009). We used transects as the unit of replication in order to maintain the temporal association between our seasonally discrete drill surveys and the continuous temperature and salinity datasets. All models were developed in R (version 3.6.1: R Core Team 2019) using the glmmTMB package (Brooks et al. 2017).

To investigate possible associations between habitat and the overall presence or absence of drills, we used a GLMM with a binomial error distribution and logit link function. The unit of replication was the transect survey at each elevation and timepoint at all eight sites (N = 79). Transects were scored as drills present or absent based on whether drills were observed in any quadrat in the transect. We modeled separately drill density (average drills per quadrat) using the same unit of replication but only at the four sites where drills were present (N = 37) using a negative binomial error distribution and log link function. In both types of model, fixed effects included elevation as a categorical variable, the psi measure of coarse substrate per quadrat, and the above temporally corresponding metrics of water temperature and salinity. Site and season were included as random effects.

We fit and compared multiple models of each type using a manual backwards stepwise model-selection approach that sequentially excluded variables from a baseline model that included elevation, coarse substrate, and temperature. Due to the asynchronous collection of salinity data, this term was included in a permutation of the baseline model covering the shorter time period of the salinity dataset.

Finally, we compared model outcomes using the Akaike information criterion (AIC). Any model that resulted in a convergence problem warning was eliminated. (Brooks et al. 2019). The best-fitting model was defined as the one with the lowest the AIC (Zuur et al. 2009) and competed against a null model.

As in the pre-project site visits, oysters but no drills were found during the study at the four sites along the southwestern portion of the Richardson Bay shoreline (from Dunphy Park to Strawberry Point); drills but no oysters were present at the rest of the sites along the northeastern portion of the study area (from Aramburu Island to Hilarita Ave) (Fig. 1). Live barnacles were found at all sites. Drill density was highest at Cove Apartments at 12.92 (SE ± 7.43) drills per quadrat on average, well above the drill density at all other sites, which ranged from 0.83 (SE ±0.34) drills per quadrat at Blackie’s Pasture to 3.04 (SE ±0.75) drills per quadrat at Aramburu Island (Fig. 2). Differences in drill densities between all sites (Kruskal-Wallis chi-squared = 263.03, df = 7, p < 0.001) and within the subset of sites where drills were present (Kruskal-Wallis chi-squared = 38.05, df = 3, p < 0.001) were significant.

Figure 2. 

Atlantic oyster drill abundance per quadrat based on 693 total quadrat samples at eight study sites in Richardson Bay, CA, surveyed quarterly from June 2017 to July 2018. The order of the sites from left to right corresponds to clockwise around Richardson Bay from southwest (Dunphy Park), to northeast (Hilarita Ave.). Surveys were not conducted in deep mud flats, which is represented by the absence of data in some instances. The horizontal lines at Dunphy Park, Bothin Marsh, Brickyard Park, and Strawberry Point represent zeroes; drills were never observed at these sites. One extreme outlier at Cove Apartments (74.3 drills per 0.25 m², third quarter of 2017) has been removed to present the data more clearly.

The dramatic difference between drill density at Cove Apartments and the other sites was driven by an extreme outlier that occurred during the fall 2017 survey at +46 cm, in which we observed over 300 drills in one quadrat at the middle elevation and seven of the nine highest drill counts per quadrat of the entire project. Overall, greater than 70 percent of drill counts per quadrat in the project were zero, and quadrats containing drills were the exception regardless of site, season, or location. This pattern of drill count results followed a negative binomial distribution (Suppl. material 1: fig. S1).

Sites were not uniform in terms of elevations that were composed of substrate other than deep mud. Table 1 summarizes where we conducted surveys at each site by elevation. The only sites at which hard substrate was present at +15 cm were along the southwestern shore of Richardson Bay where drills were not present: Dunphy Park, Brickyard Park, and Strawberry Point. At all other sites, +15 cm consisted entirely of deep (i.e. greater than 30 cm) mud. Bothin Marsh and Blackie’s Pasture were the only sites at which hard substrate was only present at one elevation (+76 cm); deep mud occupied the other elevations at those two sites. When substrate types were grouped into fine (mud, sand, pebble) and coarse (cobble and boulder) categories, fine material was found to be more abundant than coarse material at all sites and elevations (Suppl. material 1: fig. S2).

The average water temperature ranged from a low of 11.5 °C at Bothin Marsh during the 2017–2018 winter and a high of 22.0 °C at Cove Apartments during the summer of 2018. A spatial gradient was evident during hot periods: almost four times more water temperature readings above 26.5 °C per site were recorded from Aramburu Island to Blackie’s Pasture (the inner portion of northeast Richardson Bay, where drills are present) than from Dunphy Park to Strawberry Point (the southwest portion of Richardson Bay). Hilarita Ave. was an outlier as a cooler site at the edge of the northeastern part of Richardson Bay, and Bothin Marsh was an outlier as a warmer site in the southeastern part of Richardson Bay. During cold periods, there was a similar spatial gradient. With the exceptions of Bothin Marsh and Hilarita Ave., sites on the northeastern side of Richardson Bay experienced a greater proportion of hot periods than sites on the southwestern side.

The salinity data we collected from December 23, 2017 – June 30, 2018 ranged from a high of 34.18 PSU at Cove Apartments to a low of 11.23 PSU at Hilarita Ave. during a rain event. Those low salinity readings, however, were outliers and only represent approximately one hour of exposure. Excluding the outliers, the low end of the salinity range was approximately 15 PSU, putting the overall salinity range for all of Richardson Bay within the tolerance range of Atlantic drills (Suppl. material 2: table S1). Salinity values tended to remain within 3–6 PSU of each other across sites at any given time point with Cove Apartments typically registering the highest salinity and Hilarita Ave. typically registering the lowest.

Unfortunately, numerous water level loggers failed due to damage or battery failure. As a result, we were not able to recover any data from one of the sites (Strawberry Point), and of the seven months during which we collected water level data (June – December 2017), there were too few timepoints during which all loggers were functioning simultaneously to make meaningful comparisons. We, therefore, excluded inundation from the models but were able to make some basic comparisons across sites. We observed slightly higher water levels at sites along the southwestern shore of Richardson Bay (Dunphy to Brickyard Park) than along the northeastern shore (Aramburu Island to Hilarita Ave.) during the summer and spring (Suppl. material 1: fig. S5). Average water level was 0.21 meters higher at these sites, suggesting the possibility of a pushed-up water effect or greater tidal amplitude. At the same time, differences between sites in terms of average percentage of the day exposed to air were not statistically significant (GLMM, p > 0.05, Suppl. material 1: fig. S6).

According to the presence/absence model results, none of the environmental factors that we measured was significantly related to the presence or absence of drills. The best-fitting (lowest AIC) variation of the drill population density model indicated that elevation and coarse substrate were significantly related to drill density (Suppl. material 2: table S2). The effects of elevation and coarse substrate on drill density according to the best-fitting model are pictured in marginal effects plots Fig. 3 and Fig. 4, respectively. These plots illustrate that the model predicted greater drill density at +46 cm than at +76 cm and a positive relationship with coarse substrate cover.

Figure 3. 

Marginal effects plot of elevation from the GLMM incorporating elevation and coarse substrate as fixed effects and all seasons of drill abundance data.

Figure 4. 

Marginal effects plot of coarse substrate (boulder and cobble) from the GLMM incorporating elevation and coarse substrate as fixed effects and all seasons of drill abundance data. The solid line indicates the predicted count of drills per quadrat and the shaded area is the 95% confidence interval.

There were no model permutations in which water temperature was a significant predictor of drill presence, whether by average water temperature or by the percentage of readings above or below the biologically important thresholds of 26.5 °C and 37.6 °C. Removing water temperature from the model always lowered the AIC, so it was dropped from all models. Nesting variables and variable interactions resulted in a model convergence warning, suggesting that insufficient replication was available to use such model constructions, so neither was used. The best fitting model for the available data, therefore, incorporated coarse substrate and elevation as fixed effects with site and season as random effects and did not include nesting or interactions.

Population density models including salinity as a fixed variable covered six months (December 2017 – May 2018) and only the three sites where salinity loggers were deployed (Dunphy Park, Cove Apt., Hilarita Ave). Salinity was not a significant predictor in either of any of these model variations (Suppl. material 2: table S3). Models of salinity alone, salinity and coarse substrate, or any variations including temperature did not converge or yield AIC or P values. Because the spatial and temporal parameters of the salinity dataset were different than those of the baseline model including all timepoint and sites, we did not consider the AIC score to be comparable. The model did not indicate that salinity was a significant factor.

There were no patterns in our Richardson Bay data that suggested that the habitat at sites without drills is currently unsuitable for Urosalpinx cinerea. The presence of a 19^th-century fenced oyster bed in the eastern half of Richardson Bay (and/or subsequent unrecorded oyster plantings) is a likely explanation for the presence of drills at the eastern sites (Aramburu Island, Cove Apartments, Blackie’s Pasture, and Hilarita Ave). The question remains why drills have not spread south or west of Aramburu Island. Our project ruled out five aspects of habitat suitability, so some other factor or combination of factors that we did not model is responsible for the absence of drills at Strawberry Point, Brickyard Park, Bothin Marsh, and Dunphy Park.

As noted above, Atlantic drill dispersal is limited by their life history, so it is possible that drills simply have never arrived at the four sites where we never observed them. This lack of propagule pressure could be a combination of both their failure to spread after their initial introduction to eastern Richardson Bay as well as an absence of subsequent human activity that would further assist in their dispersal, such as transport via recreational boats (Wasson et al. 2001). There are also non-human means of introduction that must be considered. For example, floating among juveniles and rafting (e.g. on driftwood) has been observed among the direct-developing gastropod Batillaria cumingi in small embayments (Adachi and Wada 1999), the potential for which depends on prevailing currents and wave intensity. It is possible that the drill-occurrence boundary somewhere between Strawberry Point and Aramburu Island is related to tidal currents, eddy circulation, and other hydrodynamic patterns that restrict the movement of water-borne drills.

It is also possible that drills have arrived in the western half of Richardson Bay, but they did not establish due to a very low number of arriving individuals (Allee effects) or if habitat conditions in the past were not as favorable as they are today. Carriker (1955) speculated that drills sometimes fail to colonize due to insufficient numbers, although it is possible that even a single gravid individual would suffice for establishment of a colony (Johannesson 1988). Colonization followed by localized extinction has occurred in several Atlantic drill populations on the West Coast (Carlton 1979; Wasson et al. 2001) and may have also occurred in the western half of Richardson Bay.

Finally, biotic resistance (i.e. one or more site characteristics that prevent the establishment of drill colonies even when they do arrive in sufficient number and healthy physiological condition) is a possible explanation for the lack of drills in half of Richardson Bay. For example, predation by native cancrid crabs (e.g. Cancer productus, Metacarcinus magister, Metacarcinus [Cancer] gracilis) on mollusks, including drills, can have profound trophic effects in intertidal communities (Kimbro et al. 2009; Cheng and Grosholz 2016; Grason and Buhle 2016). In nearby Tomales Bay, California (approximately 40 km northwest of Richardson Bay), crabs consume native (Acanthinucella spirata) and invasive (U. cinerea) drills with indirect positive effects on oysters, but temperature and salinity gradients restrict crab distribution, thus mediating this trophic benefit (Cheng and Grosholz 2016). Likewise, crab dispersal in Richardson Bay could be limited by environmental heterogeneities, such as the broad, shallow mud flat at +15 cm that is present at all study sites that host drills but not present at Dunphy Park, Brickyard Park, or Strawberry Point, sites where drills are absent. Shallow mud flats could contribute to a warm-water or other barrier preventing the entrance of rock crabs to eastern Richardson Bay where they might otherwise limit drill density through competition or predation. Future surveys of the area should enumerate crab populations to test this hypothesis. Whatever the causes, over the 12 months of our study the drill and non-drill sites never changed, demonstrating a limited capacity of drills to invade new areas over the course of a year as well as drill colonies’ ability to persist over the same timescale.

Consideration of the limits on drill dispersal should give some comfort to oyster restorationists. An uncolonized site with suitable drill habitat that is relatively isolated from drill-populated areas (for example, due to surrounding broad mud flats) might be considered low-risk for near-term future invasion and, therefore, would be a more desirable target for restoration. Care should always be taken not to introduce drills accidentally (for example in boot treads) to such a site. Additionally, given the uncertainty around limitations to drill dispersal, protective measures, such as fine-mesh netting enclosures (<1 mm diameter opening) might be appropriate in a project’s early years, as natural introduction of drills cannot be ruled out.

Our quantification of the relationship of drill density to coarse substrate and elevation could be used to inform oyster restoration project site selection and design. Although the presence of Atlantic drills generally makes a site less favorable for oyster restoration (Wasson et al. 2015), it may be increasingly necessary to consider utilizing sites where drills are present when restoration opportunities are limited, as is the case in a highly developed urban estuary like San Francisco Bay. To that end, restoration designers could: 1) select sites where environmental conditions limit drill density to levels where drills and oysters can coexist; 2) design oyster restoration substrates to provide refuge from drills; or 3) adopt a diversified approach by siting projects across a range of elevations and shoreline conditions.

It is not surprising that coarse substrate is positively correlated with drill density given our knowledge of Atlantic drill life history in which cobble, boulders, and other hard surfaces play an essential role in prey availability and reproduction. There does not, however, seem to be a simple linear relationship between coarse substrate and drill density. A plot of coarse substrate, elevation, and drill density suggests a positive relationship between these variables with a threshold (around 100 Ψ per 0.25 m²) above which a higher Ψ value (i.e. more cobble and boulder) was not associated with higher drill density. Our observation of a possible upper limit to this positive relationship suggests that increasingly rocky shorelines do not necessarily pose additional risk to oysters in the form of correspondingly higher drill densities.

Below these thresholds, moreover, drill density appears to be limited to a level thought to be compatible with oyster populations (Fig. 5). In Tomales Bay, Cheng and Grosholz (2016) documented Olympia oysters co-existing with non-native oyster drills (Atlantic and Japanese) when average drill density was approximately one drill or less per 0.25 m². In Richardson Bay, both Hilarita Ave. and Blackie’s Pasture support an average drill density around one drill per 0.25 m². These two sites might, therefore, be well-suited for experiments testing the viability of oyster restoration amid a low-density drill population. Restorationists looking for more sites that are limited in their potential to support drill populations could target areas with shoreline conditions in this zone of suboptimal drill habitat. Additionally, there are shoreline areas in southern San Francisco Bay where mature oysters coexist with drills of varying population density (authors’ unpublished data). Fine-scale substrate measurements at these and other potential restoration sites could help refine our habitat suitability model and reveal additional unexpected opportunities for Olympia oyster restoration.

Figure 5. 

Drill abundance relative to elevation and cover of coarse substrate (boulder and cobble). Each point represents a quarterly transect survey (N = 37) at one of the four sites where drills were present, and points are jittered to reduce overlap. The three clusters of points surrounded by an oval correspond to sites with an average drill density below one drill per 0.25 m², a level at or below which Olympia oyster and Atlantic drill populations have been observed in coexistence. The two solid ovals are Hilarita Ave., and the dashed oval is Blackie’s Pasture.

Our finding that drill density was negatively correlated with elevation is consistent with data from other San Francisco Bay studies (Boyer et al. 2017; Zabin et al. 2024) and could be translated to the design of restoration structures. Borrowing from aquaculture methods in which oyster racks are raised above the benthos, oyster restoration substrates that are raised in some fashion above drills’ preferred elevation may provide a buffer from drill predation, possibly in conjunction with hand-removal of drills (Cheng et al. 2022; Zabin et al. 2024). Although drills can easily climb vertical surfaces (authors’ personal observations), they are highly opportunistic and may be satisfied to concentrate around the proverbial low-hanging fruit, allowing higher-elevation oysters to reach maturity. At the same time, higher elevations place additional heat and desiccation stress on oysters, which leaves restoration designers facing a difficult tradeoff between providing optimal oyster habitat and avoiding predators that favor the same space. Diversifying the locations of restoration substrate over a range of elevations could provide a measure of bet-hedging against drills (Zabin et al. 2016; Zabin et al. 2022). Other drill avoidance measures could include placement of restoration substrate offshore from fringing marshes that lack rocky shoreline or even further offshore in the subtidal zone at a distance that drills may be unlikely to traverse if shoreline prey is abundant. At the same time, there are many site-specific factors which may interact with elevation to influence the vertical zonation of drill habitat. For example, slope, bathymetry, and fine-scale variations in tidal amplitude (which impacts inundation and exposure-to-air time) are physical characteristics that may affect habitat suitability.

Understanding the habitat conditions for co-existence of Olympia oysters and Atlantic drills will also require more study and may vary by site and over time. For example, getting to a temporal stage of oyster and drill coexistence could be an initial hurdle to restoration success. During the early stages (e.g. the first year) of a restoration project, a newly settled oyster population is more vulnerable to drill predation than a mature community, as drills exhibit a bias towards young, thin-shelled oysters (Federighi 1931; Buhle and Ruesink 2009; Lord and Whitlach 2013; Cheng 2014). As the settlement community on new substrate matures, however, this vulnerability may decline, at least until a large mortality event occurs and resets the age distribution of oysters. Such mortality events may occur during low salinity events associated with extreme precipitation patterns, which are expected to become more frequent as an effect of climate change (Cheng et al. 2016; Margulies 2023). Olympia oysters are more tolerant than drills to low salinity, which could extend the geographic range of possible restoration sites higher upstream in the estuary where low-salinity events are more common than Richardson Bay or lower San Francisco Bay and which is thought to preclude drill populations or could potentially limit their density. Finally, changes in prey availability could affect the fine-scale number, density, or distribution of oyster drills around a single estuary. In Richardson Bay, we frequently measured higher barnacle cover at sites without drills than with drills, suggesting that food availability was unlikely to be constraint on drill distribution there. Further study of community composition at nearby sites with and without drills, however, might reveal how changes in prey population could influence changes in drill abundance and even mediate known physical determinants, such as elevation.

Our study confirms that there is suitable drill habitat in Richardson Bay, and most likely in San Francisco Bay at large, at sites where drills are not currently found. To avoid the introduction of drills to new locales, anyone traveling from areas where drills are present to uninvaded areas should exercise extreme caution to avoid accidentally transporting drills, which may get lodged in the tread of shoes, boots, or tires, or attach to watercraft hulls, for example. Popular sites where drills have been observed, such as shoreline parks and marinas, should have signs posted alerting visitors to the presence of oyster drills and describing ways to prevent their spread.

We also recommend more extensive surveys to map the entire distribution of Atlantic drills in San Francisco Bay along with a broader set of environmental data. Intertidal surveys are highly time-constrained, and it would have been impossible to sample the entire vertical extent of the beach, but some exploratory surveys we performed at the intermediate elevations +24 cm (low-middle intertidal) and +61 cm (middle-high intertidal) suggested that these elevations might also be favorable for drills. Future surveys, therefore, along transects perpendicular to the shore that incorporate elevations in between this study’s target elevations might help fine-tune our understanding of elevation and Atlantic drill habitat. Additionally, there are many biotic and abiotic factors we did not analyze that could potentially influence drill density, distribution, and potential for future range expansion or contraction. These factors include ecological characteristics, such as prey availability, predation, and competition; and physical characteristics, such as pH, dissolved oxygen, hydrodynamic circulation patterns, and wave intensity. Making these data spatially explicit would allow analysts to produce GIS-based habitat suitability models, which could be incorporated into existing broad-scale shoreline classification datasets of San Francisco Bay (San Francisco Estuary Institute 2016).

Environmental datasets that extend spatial and temporal coverage to areas with more heterogeneous environmental profiles may offer additional insight into how applicable the importance of elevation and substrate are outside of Richardson Bay. These additional data could also be used to test our observation that there may be an ideal intersection of coarse substrate cover and elevation and whether or how salinity and inundation influence drill abundance outside of our study area. Broadening the geographical reach of our habitat study is especially important in large estuarine environments, like San Francisco Bay, where these factors may vary widely based on distance from the ocean. As such, our results should not be interpreted as definitive ecological niche measurements but rather should guide our understanding of two key environmental factors that, possibly in combination with other variables, influence variations in Atlantic drill abundance. Inundation measurements should also continue to be a part of these environmental datasets as more pronounced differences in tidal amplitude may be seen at sites further apart.

Finally, more frequent or continuous monitoring of the San Francisco Bay shoreline for drills is necessary to determine whether their range is static, expanding, or contracting. This knowledge is required for accurate habitat suitability modeling, as such models typically assume a level of pseudo-equilibrium between a species and its environment without which the model could produce biased results (Guisan et al. 2002). Understanding whether and how drill range is changing will be crucial in assessing what constitutes a safe distance for oysters from zones occupied by drills. As Olympia oyster restoration efforts expand in San Francisco Bay and throughout the West Coast, understanding how Atlantic drills fit into these landscapes will be an important component of successful restoration project outcomes.