Description of Study Area

Two study sites in central Puget Sound, Washington (fig. 1) were selected to sample a suite of biological, hydrological, and geological variables. These sites, on the northern tip of Bainbridge Island (fig. 2), lie adjacent to each other, but have different land-use characteristics. The first study site, Fay Bainbridge Park, is a 6.8-hectare general-use Bainbridge Island Metropolitan Parks and Recreation District Park, with approximately 430 m of undeveloped shoreline. The coastline is an estuarine accretion beach with a backshore berm containing an accumulation of large woody debris, and the shoreline is not armored. The upland habitat is a lowland that contains a grass lawn, paved roadways and parking areas, and park shelters that abut a forested upland. The second study site, Point Monroe, lies immediately north of Fay Bainbridge Park. Point Monroe is an approximately 1-km long residentially developed, recurved spit constructed from fill (Williams and others, 2004) and an accretion beach seaward, and the site forms a tidal lagoon landward. The seaward-facing aspect of the spit is largely devoid of terrestrial vegetation, with shoreline armoring along most of the developed beachfront.

Benthic substrates at both study sites are classified as mixed-sediment, with a gravel foreshore and a sandy low-tide terrace (Pontee and others, 2004). Intertidal and subtidal vegetation observed at both sites includes primarily patchy and contiguous beds of the native eelgrass (Zostera marina L.), along with scattered patches of bull kelp (Nereocystis luetkeana), and seaweed (Ulva sp., Gracilariopsis sp.). At the time of survey, the non-native eelgrass Z. japonica was not directly observed at these sites, but it is possible that it could have been interspersed within or near the beds of Z. marina. Shoot densities of eelgrass beds at Fay Bainbridge Park (1,143±475 shoots per square meters [m²]) were about twofold greater than the eelgrass beds at Point Monroe (507±479 shoots per m²; R. Takesue, U.S. Geological Survey, written commun., 2012). These eelgrass-shoot densities are high but consistent with those in eelgrass beds at other central Puget Sound locations. Additionally, forage fish spawning surveys previously conducted near this study site indicate that these beaches could provide favorable spawning habitat for surf smelt (Penttila, 2001).

Bathymetric and Eelgrass Mapping

Detailed data on bathymetry and eelgrass distribution were simultaneously collected using a single beam echosounder system. The system consisted of a Biosonics 420 kilohertz (6° beam angle) transducer and deck unit, a data acquisition laptop PC, and Trimble real-time kinematic global positioning system (RTK-GPS). For a complete description of the acoustic survey system, see Stevens and others (2008). Cross-shore survey transect lines were spaced at 25-m intervals along the 3-km stretch of the study area and were followed from as close to shore as possible to a depth of 20 m. Cross-shore survey data were supplemented with several additional along-shore lines that roughly followed bathymetric contours.

Acoustic data were analyzed using a custom graphical user interface that implements a signal processing algorithm (Sabol and others, 2002) that is applied to each sonar sounding to extract the location of the bottom and presence of vegetation (Stevens and others, 2008; Depew and others, 2009). Individual acoustic returns along a survey line were grouped into packets of 10, and eelgrass percentage cover was calculated as the fractional percentage of acoustic returns that were classified as vegetated within each group, resulting in an estimate of eelgrass percentage cover every 4 to 5 m (depending on the vessel speed). Eelgrass percentage cover measurements were interpolated onto a grid with 5-m resolution using kriging (Davis, 2002).

Topographic elevation measurements were collected on foot during low tide using the RTK-GPS mounted on backpacks. Bathymetric and topographic data were merged, and linear interpolation was used to produce a representing the surface topography and nearshore bathymetry. The survey-grade global positioning system (GPS) equipment used in this project have manufacturer-reported root-mean-square accuracies of approximately ±3 centimeters (cm) + 2 parts per million of baseline length (typically 10 km or less) in the horizontal and approximately ±5 cm + 2 parts per m in the vertical while operating in real-time kinematic surveying mode (Trimble Navigation Limited, 1998). These reported accuracies are, however, additionally subject to multi-path, satellite obstructions, poor satellite geometry, and atmospheric conditions that can combine to cause a vertical GPS drift that can be as much as 10 cm (Sallenger and others, 2003). While repeatability tests and merges with topographic data collected with a backpack RTK-GPS suggest sub-decimeter vertical accuracy, significant variability in seawater temperature (about 10 degrees Celsius) and salinity can affect depth estimates by as much as 20 cm in 12 m of water. Therefore, a conservative estimate of the total vertical uncertainty for these nearshore bathymetry measurements is approximately 0.15 m.

Surf Smelt Spawning Surveys

Sea Level Rise Modeling

The effect of SLR on shorelines and tidal habitats within our study area was modeled using SLAMM, a flexible modeling tool used for predicting SLR-induced changes to coastal ecosystems based upon local topography and user-defined climate change and SLR predictions (Clough and others, 2010). SLAMM uses local land-cover data, which are converted into “SLAMM classes” that are habitat classes used in the model for predicting changes in habitat through time. Our objectives were to utilize SLAMM to model three climate change scenarios to predict the effects of SLR on the coastal habitats of our study area.

We used SLAMM with the A1B-mean SLR scenario derived from a suite of the A1B climate projections computed by the IPCC based upon current and future greenhouse gas emission scenarios (Meehl and others, 2007). The A1B-mean scenario is a conservative estimate that predicts a 0.40 m global SLR by 2100. Recent generally accepted climate change and SLR models indicate variations in SLR projections (Hieronymus and Kalén, 2020), with suites of models such as the AR5 (Church and others, 2013), RCP2.6, and RCP4.5 (IPCC, 2019) predicting SLR in the likely range from 0.3 to 2.6 m. The use of constantly evolving long-range SLR estimates for regional-scale modeling, however, becomes impractical, and introduces a greater amount of variability, with results that quickly become obsolete. Therefore, to account for additional elevated SLR predictions, SLAMM was also used with more general 1-m and 2-m rise in global sea level by 2100 scenarios.

The SLAMM program utilizes a suite of habitat, land-cover, and topography data sources for computing SLR predictions. The digital elevation model used in this study was obtained from a 2000 U.S. Geological Survey National Elevation Dataset. Land-cover categories were derived from a combination of visual observations recorded during the 2010 field visit and from a National Wetlands Inventory data layer with a photo date of 1977. Although SLAMM does not model an eelgrass habitat type, a combination of SLR predictions, bathymetry data, and eelgrass distribution were used to estimate eelgrass habitat change based on the three SLR scenarios. The historical SLR trend was estimated at 2.06 millimeters per year (mm/yr) (±0.17 mm/yr) using the Seattle, Washington, National Oceanic and Atmospheric Administration (NOAA, 2020) Center for Operational Oceanographic Products and Services (n.d.) tidal gage (station no. 9447130). The rate of SLR for this station is greater than the global average for the last 100 years (approximately 1.7 mm/yr). The tidal range at this gage is 3.462 m. Tidal flat erosion rates were set to 0.1 meters per year based upon Keuler (1988). Data for bathymetry and eelgrass distributions were recorded during the field visit. Additional study site parameter settings for the SLAMM model are as described in Glick and others (2007). To graphically display habitat changes through time, predictions for 2050 were included in the analysis.

Bathymetry and Distribution of Eelgrass

A total of 25 km of acoustic survey lines were collected via boat, and to supplement the bathymetry data for areas too shallow to access by boat, an additional 35 km of longshore topographic measurements were collected on foot. The bathymetric grid produced by the acoustic survey data (fig. 3) indicated a varied morphology within the study area, with changes occurring from east to west. Between the southern terminus of the study area and the northern tip of Point Monroe, the low-tide terrace is relatively uniform and spans a distance from 100 to 150 m from shore. Along the northwest facing shoreline, the low-tide terrace is narrow (about 60 m), with a steep slope near the shoreline. Within the area west of Point Monroe, the low-tide terrace is broad and shallow, as much as about 350 m wide, and approximately -1 m MLLW deep.

Analyses and interpretation of acoustic survey data indicate that eelgrass is present along the entire shoreline of the Point Monroe and Fay Bainbridge study area (fig. 4). Both the total area of eelgrass and percentage eelgrass cover appears to be limited by water depth, with eelgrass observed mostly between 0.5 and -5 m MLLW, though a small quantity was observed in both shallower and deeper water. The most abundant and highest percentage cover of eelgrass were at water depths between -1 and -4 m MLLW. These findings are nearly identical to those determined in a similar Puget Sound eelgrass mapping survey (Stevens and others, 2008).

Surf Smelt Spawning-Area Surveys

Surf smelt eggs were observed in greater quantity and within more segments along the southern Fay Bainbridge Park beach than were observed in samples collected from the more northerly Point Monroe. Eggs were present at 9 of the 14 (64.3 percent) segments that were sampled (table 1; fig.5), and eggs per sample were significantly higher at Fay Bainbridge Park than at Point Monroe (ANOVA; F_{1, 166} = 9.08, P =0.003; fig. 5). Every beach segment along Fay Bainbridge Park contained eggs, with eggs observed in 64 of the 84 (76.2 percent) samples. In total, 630 eggs were observed within the Fay Bainbridge Park sediments, comprising 540 live eggs and 90 dead eggs (14.3 percent mortality). The number of surf smelt eggs per segment ranged from 46 to 197, with a mean of 90 eggs per segment. Eggs were observed on 2 of the 7 segments along Point Monroe, but only on those segments located less than 100 m from Fay Bainbridge Park and that had shoreline armoring that was greater than 3.84 m above MLLW (fig. 5; table 1). Eggs were observed in 11 of the 84 (13.1 percent) Point Monroe samples. In total, 134 eggs were observed within the Point Monroe sediments, comprising 132 live eggs and 2 dead eggs (1.5 percent mortality). At Point Monroe, the number of surf smelt eggs per segment ranged from 0 to 116, with an average of 19 eggs per segment.

Characteristics of bed sediment and habitat variables differed between Fay Bainbridge Park and Point Monroe (table 2). The mean sizes of sediment in samples collected from beach segments at the Fay Bainbridge Park were larger than the mean sizes of sediments collected from Point Monroe beach segments, and the differences were significant (ANOVA; F_{1, 166} = 24.43, P <0.001). The sorting of sediment grain sizes between the two sites were not significantly different, however (ANOVA; F_{1, 166} = 0.95, P =0.331), with the Point Monroe sediments being slightly less sorted than Fay Bainbridge Park sediments. Additionally, the percentage composition of sediment size classifications differed between the two sample sites. Fay Bainbridge Park sediments consisted primarily of gravel (80.30 percent), and the remainder was sand (19.69 percent). Most of the Point Monroe sediments also were gravels (58.73 percent), but the sand fraction was more than double (41.26 percent) that measured in the Fay Bainbridge Park sediments. The percentage occurrence of both gravel and sand were significantly different between sites (ANOVA; F_{1, 166} = 26.91, P <0.001 and F_{1, 166} = 26.92, P<0.001, respectively). Mud was virtually absent from both sample locations, comprising less than 1 percent of the total samples, and was not significantly different between sites (ANOVA; F_{1, 166} = 0.65, P =0.421). Shore slope and shoreline armoring also varied by sample location. The mean beach slope was 9.94 percent at Fay Bainbridge Park and a steeper 12.44 percent at Point Monroe and was significantly different between sites (ANOVA; F_{1, 166} = 222.70, P <0.001). In addition to differences in beach slope, no armoring was observed along the entirety of Fay Bainbridge Park, whereas at least a part of 6 of the 7 beach segments on Point Monroe were armored. The elevations at which samples were collected were significantly different between the two sites (ANOVA; F_{1, 166} = 21.26, P <0.001), with the mean elevation at Fay Bainbridge Park being higher on the beach than the Point Monroe mean elevation owing to armoring restricting the area of the upper beach on Point Monroe.

Characteristics of beach sediment and habitat variables where surf smelt eggs were present also showed some differences between Fay Bainbridge Park and Point Monroe (table 2). The mean sediment sizes in samples containing eggs from Fay Bainbridge Park beach segments were again larger than the mean sizes in sediment samples collected from the Point Monroe beach segments that contained eggs, and the differences were significant (ANOVA; F_{1, 73} = 6.69, P =0.012). The sorting of sediment grain sizes between the two sites were not significantly different, however (ANOVA; F_{1, 73} = 1.0, P =0.319), with the Fay Bainbridge Park sediments being slightly more sorted than Point Monroe sediments. The percentage composition of sediment size classifications of sample locations containing surf smelt eggs was similar to those observed in all sample locations. Fay Bainbridge Park sediments consisted primarily of gravel (78.82 percent), and the remainder was sand (21.28 percent). Most of the Point Monroe sediments were also gravels (59.72 percent), and the remainder sand. The percentage occurrence of both gravel and sand were significantly different between sites (ANOVA; F_{1, 73} = 7.02, P =0.01 and F_{1, 73} = 7.02, P =0.01, respectively). Again, mud was virtually absent in samples from both locations that contained surf smelt eggs and was not significantly different between sites (ANOVA; F_{1, 73} = 0.34, P =0.561). The mean beach slope for sample locations containing surf smelt eggs was 9.88 percent at Fay Bainbridge Park and a steeper 12.88 percent at Point Monroe and was significantly different between sites (ANOVA; F_{1, 73} = 44.66, P <0.001). Elevations of samples containing surf smelt eggs were similar at both Fay Bainbridge Park and Point Monroe and were not significantly different between sites (ANOVA; F_{1, 73} < 0.01, P =0.991).

Selection and Fit of Spawning Habitat Models

The influence of spatial autocorrelation on the presence of surf smelt spawning was non-significant (p = 0.57, r = -0.007), so additional adjustments for autocorrelation were not required. Simulation results showed that the best explanatory value occurred once covariates grain sorting and gravel were removed from the full model (as indicated by AIC; 174.74; table 3), with a deviance explained of 70.5 percent, an adjusted r² of 0.35, and regression between the observed and fitted values that was indistinguishable from 1 (1.088). This model included variables such as elevation (p =0.06) and beach slope (p = 0.10), but the covariates that contributed the greatest impacts were grain size (p = 0.04), sand (p = 0.002), and armoring absence (p =0.02). Graphing of these variables show that as beach slope and grain size increase, the observations of spawning decrease (figs. 6A and 6B; respectively). Eggs were more abundant at higher elevations on the beach (fig. 6C), and there were fewer eggs observed as the proportion of sand increases (fig. 6D). The sign of the estimate (negative; -1.37) for armoring presence indicates that the presence of armoring negatively affects the presence of surf smelt eggs.

Sea Level Rise Modeling

Under all SLR scenarios, the SLAMM simulations predicted changes in habitat types from 2010 to 2100 within our study area (table 4). As expected, simulation of the most conservative scenario (A1B Mean) predicted the least change, followed by the 1-m scenario, and the 2-m scenario predicted the greatest amount of conversion of beach and upland habitat to open water. For all scenarios, the simulations predicted that swamps, tidal flats, and vegetated tidal flats were vulnerable to loss, but the initial area occupied by these habitat types was small. Under the A1B-mean scenario, both developed and undeveloped dry land lost 10 percent of their extent and were converted to transitional saltwater marsh and regularly flooded marsh habitat categories (fig. 7). The predictions of the 1-m SLR scenario were similar to the A1B-mean scenario except for estuarine beaches, which were reduced 6 percent in size (fig. 7). As expected, the greatest amount of habitat change occurred using the 2-m SLR scenario. The area of undeveloped dry land in the 2-m SLR scenario remained similar to the other scenarios, but as much as 53 percent of the developed dry land could be subjected to inundation and converted to transitional salt marsh (fig. 7). Additionally, this scenario predicted the conversion of 13 percent of the estuarine beach to open water. The loss of developed dry land and estuarine beach is predicted to occur primarily at Point Monroe.

When SLAMM sea level predictions were overlaid on existing eelgrass habitat, the potential available habitat for eelgrass increased with increasing SLR (table 5). The model predicts that eelgrass habitat will be lost at the lower depth margins owing to light attenuation at increased water depth, but an increase in the inundation of the current intertidal beach adjacent to the existing eelgrass beds will allow for landward expansion to compensate for that loss. The least amount of eelgrass habitat gain occurred under the A1B Mean scenario (10.2 percent), followed by 22.1 percent increase under the 1-m SLR scenario, while the 2-m SLR scenario had the greatest increase in potential eelgrass habitat (37.6 percent).

Surf smelt use habitat falls within the SLAMM estuarine beach habitat category for deposition of their eggs during spawning. Under all three SLR scenarios, the area of potential surf smelt spawning habitat was reduced (table 5). The smallest amount of estuarine beach habitat loss occurred under the A1B-mean scenario (1.8 percent), followed by a 6.1 percent loss under the 1-m SLR scenario, and the 2-m SLR scenario had the greatest loss of potential surf smelt spawning habitat (12.8 percent). The beaches around Point Monroe experienced the greatest loss of estuarine beach habitat (fig. 7), primarily because shoreline armoring prevented beach formation from moving landward, as was possible along Fay Bainbridge Park beaches where armoring was absent.