The Elwha River is a relatively short (72-km long), high gradient river flowing north out of the Olympic Mountains into the Strait of Juan de Fuca ( Figure 1 ). Two dams constructed on the Elwha River, Washington State, USA ( Figure 1B ), in the early 1900s trapped approximately 30 million tonnes (Mt) of sediment by 2011 (Ritchie et al., 2018). The staged removal of the 32-m tall Elwha Dam (at river kilometer 8) and the 64-m tall Glines Canyon Dam (at river kilometer 21), began in September 2011 and was completed by 2014. Sediment flux through the lower river and to the coast increased dramatically in response to dam removal, starting in March 2012, during the deconstruction of the lower Elwha Dam (Ritchie et al., 2018). The reservoir upstream of the Glines Canyon Dam, Lake Mills, started spilling sediment past the dam site in October of 2012. During the first two years of dam removal, between September 2011 and September 2013, over 10 Mt of sediment were eroded from the two reservoirs. Turbidity in the river downstream of the dams increased nearly three orders of magnitude above background levels during the same period (Magirl et al., 2015). In the subsequent three-year period (September 2013 to September 2016), an additional ~9 Mt of sediment eroded from the reservoirs, thereby resulting in almost 20 Mt of sediment eroded since the beginning of the project (Ritchie et al., 2018).

Roughly 90% of the sediment eroded from the reservoirs during dam removal was transported rapidly through the river to the dynamic coastal environment of the Strait of Juan de Fuca. The diurnal tidal range, the difference between mean lower low and mean higher high water is 2.1 meters, and the exchange of tides through the Strait of Juan de Fuca generates large tidal currents that regularly exceed 1 m/s near the Elwha River delta (Warrick and Stevens, 2011). Wave energy in the Strait of Juan de Fuca near the Elwha River delta is a mix of locally forced wind waves and swell waves propagating from the Pacific Ocean, with a mean annual significant wave height of ~0.4 m and a bimodal distribution of dominant period, with peaks at 4 and 10 seconds (Miller et al., 2011).

The large increase in sediment supply associated with the dam removals resulted in dramatic physical changes in the nearshore, including: 1) growth of the river mouth delta landform and modification of a variety of coastal habitats (Foley et al., 2017c; Warrick et al., 2012; Gelfenbaum et al., 2015; Warrick et al., 2019), 2) increased turbidity in coastal waters near the river mouth (Foley and Warrick, 2017; Glover et al., 2019), and 3) persistent and intermittent sedimentation on the seafloor offshore of the delta (Eidam et al., 2016; Foley and Warrick, 2017; Glover et al., 2019). These dam removal-mediated habitat alterations had a variety of implications. Rubin et al. (2017) documented > 75% decrease in brown algal density and decreased benthic invertebrate species diversity at sites near the Elwha River mouth during dam removal compared to before dam removal surveys. Changes in algal abundance during dam removal were associated with the deposition of sediment and reductions in light driven by increased turbidity (Rubin et al., 2017; Glover et al., 2019).

Seafloor grain size and topographic and bathymetric measurements were collected annually on and around the Elwha River delta, typically over 3-4 days during the summer (Gelfenbaum et al., 2015; Stevens et al., 2017). Intertidal topographic measurements were collected with differentially corrected GNSS systems, mounted on backpacks, and walked by surveyors along the beach and the backshore during low tides. The GNSS system records positions and elevations in an auto-by-interval mode, in which observations were collected automatically as the surveyors moved over the beach and backshore. Measurements were concentrated on cross-shore oriented transects with approximately 50-m alongshore spacing, but surveyors also collected observations along slope breaks and irregular features.

Bathymetric observations were made during the same surveys using personal watercraft equipped with GNSS systems coupled with depths measured using single beam echosounders. Vessel operators collected bathymetric data at high tide along the same set of cross-shore oriented transects used by walking surveyors at low tide, and generally collect observations between mean lower low water (0.12 m NAVD88 at Elwha) and offshore depths of 15-20 m. The combined topographic and bathymetric data are interpolated onto a uniform grid with a resolution of 5 m. The resulting digital elevation model (DEM) has an estimated vertical uncertainty of 12 cm (Stevens et al., 2017).

Surface sediment samples were collected with a petite ponar grab sampler (Gelfenbaum et al., 2015). Samples were obtained along a series of cross-shore transects at sites that, prior to dam removal, were subtidal (elevations between -1 and -12 m, NAVD88). As the bathymetry of the Elwha River delta changed during and after dam removal (Ritchie et al., 2018), the sampling locations stayed the same, such that some sites sampled from a boat in 2012 were sampled by hand from the beach in subsequent years. For samples collected with the grab sampler, up to three attempts were made to collect a sample with adequate mass (approximately 50 g) from the seafloor; if no sample could be collected the site was assumed to be coarse-grained. Suitable samples were transported to a U.S. Geological Survey (USGS) laboratory in Santa Cruz, California, USA, and analyzed to determine the disaggregated grain size distributions using a combination of sieves and a Beckman Coulter LS 13 320 particle size analyzer (Stevens et al., 2017).

Sediment fluxes in the Elwha River for the period between September 2011 and September 2016 were estimated using the methods of Magirl et al. (2015) as calculated and tabulated by Ritchie et al. (2018) for the USGS gaging station 12046260 on the Elwha River ( Figure 1 ), which captures the sediment inputs from both reservoirs upstream. Suspended sediment concentrations (SSC) were derived from a combination of standard flow-weighted suspended sediment samples, automated point-sampler daily composite samples, and three sediment-surrogate instruments—two optical turbidimeters and one acoustic Doppler velocity meter (ADVM)—to generate 15-min interval records of SSC during the 2012 to 2016 water years. These estimates include the proportions of the suspended sediment discharge within sand grain sizes (0.063 to 2 mm) and silt and clay grain sizes (less than 0.063 mm), which we termed fine suspended sediment load (FSSL). Sediment discharge was calculated by the product of SSC and measured river discharge, which were summarized into daily and annual sediment discharge values.

Turbidity measurements in the Elwha River were discontinued after the 2016 water year, so sediment fluxes after that time were estimated using a rating curve derived from measurements of 2016 water year average daily river discharge (Q, m3/s) and daily suspended-sediment discharge (Qs, t/d) using power-law formulations. Sediment rating curves were generated from least-squares regression for both sand grain-size fractions (Q_s = 0.000188 Q^3.3181; r^{2 =} 0.77) and silt-to-clay grain size fractions (Q_s = 0.000786 Q^3.1366; r^{2 =} 0.76) and applied to daily average discharge records (Warrick et al., 2012). Our computations assume that the sediment supply rates to the river continued to be dictated by river flow and that the sediment availability stayed relatively constant with time. These assumptions likely result in overestimates of sediment discharge because they do account for the observed decreased sediment availability over time after dam removal ended (Ritchie et al., 2018).

Subtidal marine community surveys were conducted annually between mid-July and early September at 13 sites near the Elwha River delta, and at two additional control sites at Green Point, approximately 20 km east in the Strait of Juan de Fuca ( Figure 1 ). Before dam removal, each site was surveyed at least once between 2008 and 2011 (Rubin et al., 2017). Since 2012 all sites have been surveyed annually except the two control sites, which were not surveyed in 2018. Sites vary in depth between 5 and 17 m, and (excepting the control sites) are distributed up to 7.5 km east and 2.5 km west of the Elwha River mouth. Substrate type at sites is a mix of sand, gravel, cobbles, and boulders (Rubin et al., 2017) and has not changed measurably except where persistent deposition occurred.

Each site was marked on the seafloor with a stainless steel post, which identifies the center of two 50-m long transects, one trending east and one trending west, marked on either end by a cement block. Along each transect, divers enumerated benthic invertebrates greater than 2.5 cm along any dimension and kelp and acid kelp (Desmarestia sp) greater than 24 cm in length, within a 30-m long and 1-m wide swath. Divers also enumerated fish of any size within a 30-m long by 2-m wide by 2-m high tube. Additionally, a modified uniform point contact methodology was used to characterize the substrate type and algal cover at 0.5 m intervals along the transect (Rubin et al., 2017).

Kelp, invertebrates, and fish were analyzed separately. Species were grouped to the lowest taxonomic level practicable for consistency among observers and across years. Many fish were small, cryptic, and mobile, and are generally aggregated to taxonomic levels higher than those for kelp or invertebrates. Common rather that scientific names were used for fish. Data were averaged over transects within each site-year before analysis. All data collected between 2008 and 2011, before the start of dam removal, were consolidated and are presented here as belonging to 2011 (see Rubin et al., 2017 for full details). All data used in analyses are publicly available (Rubin et al., 2023).

Sites were categorized into three “site groups” prior to analysis: buried (4 sites), unburied (9 sites), and control (2 sites; Figure 1 ). Buried and unburied site groups were separated because previous analyses (Rubin et al., 2017) indicated fundamental differences in biological response between sites buried under persistent sediment deposits and unburied sites. We classified a site as buried if it experienced deposition greater than 15 cm based on analysis of bathymetry survey data or measurements of a post installed at each site that allowed divers to estimate changes in elevation unless the depth of deposition exceeded post height (Rubin et al., 2017). Sites classified as buried first became buried in 2013 or 2014 and remained buried for the duration of our study; unburied sites did not experience persistent burial.

Community analyses compared trajectories of change among communities at buried, unburied, and control sites. A difference in trajectory for the control sites compared to the other two site groups would suggest that change at the buried and unburied sites was due to dam removal. A trajectory difference between buried and unburied sites would suggest that communities were differentially affected by dam removal at buried compared to unburied sites. Community analyses addressed hypotheses 1, 2, and 3.

Multivariate community analyses were conducted using species that were present in at least five site-years. Analyses were performed on Bray-Curtis similarity matrices computed from square root-transformed species densities (Clarke et al., 2014). Analyses were conducted in Primer 7 (Clarke and Gorley, 2015).

Two-way analysis of similarity (ANOSIM) with factors survey year and site was used to investigate which major taxon (kelp, invertebrates, or fish)-by-site group combinations showed significant community change across years (Clarke et al., 2014). Site was included as a factor to account for (i.e., factor out) variation due to spatial differences among sites. Taxon-by-site group combinations that did not exhibit significant community change across years were not included in subsequent analyses, except for testing the effect of the density of the sea star Pycnopodia on community composition within the control and unburied site groups.

Two-way analysis of similarity percentages (SIMPER) with factors survey year and site was used to investigate which species contributed most to community change across years within each site group (Clarke et al., 2014). SIMPER generated percent contribution of species contributing most to the difference between each year pair, up to a cumulative contribution of 70%. We computed a weighted average of a species’ contribution to each year pair difference where the weights were the average community dissimilarity between each year pair. This gave a single mean percent contribution to community differences among years for each species. Species contributing most to differences over time were highlighted in subsequent analyses.

Nonmetric multidimensional scaling (NMDS) analysis was conducted on densities averaged across sites within site groups and was used to investigate when community change occurred and the trajectory of change (e.g., unidirectional with different start and end states, circular with similar start and end states). Species vectors, indicating the strength and direction of correlations between species densities and each NMDS axis were superimposed on NMDS plots (Clarke et al., 2014; Clarke and Gorley, 2015).

Species analyses (Somerfield and Clarke, 2013) were used to identify groups of species that changed similarly across years. Analyses were conducted on densities averaged over sites within site groups. Densities were standardized by total within species (density from each year was divided by total density across years). Then cluster analysis was performed on Bray-Curtis similarity matrices computed from standardized densities (no square root transformation). Species that did not contribute to community change between at least one pair of years according to SIMPER analysis were excluded. Species that did contribute to community change but only occurred in one year were also excluded.

Analyses of mechanisms driving change addressed whether communities or individual species were affected by variables directly or indirectly related to dam removal or other stressors. Variables considered were fine suspended sediment load (FSSL), algal cover, and Pycnopodia density. Mechanism analyses primarily addressed hypotheses 3 and 4.

We used regression approaches to test the three mechanisms, with FSSL, percent algal cover, and Pycnopodia density as independent variables. We chose FSSL in the river as the independent variable representing suspended sediment because suspended sediment in the nearshore was linked to FSSL in the river (Glover et al., 2019). We used cumulative FSSL between February 1 and July 31 because our observations of the system suggest that elevated suspended sediment during that period could suppress algal recruitment and growth in our study area.

Distance-based redundancy analysis (dbRDA) was used to investigate whether community change was related to FSSL, percent algal cover, or Pycnopodia density (McArdle and Anderson, 2001). dbRDA is akin to multivariate regression and is appropriate when the dependent variables, in our case densities of species, are represented by a Bray-Curtis similarity matrix as ours were. Site was included in dbRDAs as an independent variable (as a fixed effect because dbRDAs do not allow random effects) to account for spatial effects. dbRDAs were conducted in Permanova+ (Anderson et al., 2008).

Univariate regression analysis was used to investigate whether densities of single species were related to FSSL, algal cover, or Pycnopodia density. Species identified by SIMPER as contributing most to community change across years at Elwha unburied sites were chosen for this analysis. Species density was log-transformed before analysis to increase normality and decrease heteroskedasticity. Sites where the species did not occur in any year were excluded. Analyses were conducted in R (R Core Team, 2022). We used General Additive Models (GAM) to test which shape of a curved relationship fit best, or whether a linear relationship fit better than any curve, and also to test which of four error structures fit best: 1) including site as a random effect, 2) including an auto-correlation term (AR1) to account for any non-independence due to consecutive years of data, 3) including both site as a random effect and an auto-correlation term, and 4) including neither site nor an auto-correlation term (R package mgcv; calls gam or gamm; Wood, 2017). The model with the lowest AIC value was selected as the best fit. A final step was to test whether adding to the best fitting model a term allowing residual spread to vary as a function of the independent variable (VarExp variance structure; Zuur et al., 2009) improved model fit further (Pinheiro et al., 2023).

Between water years 2011 and 2015 the Elwha River discharged roughly 14 Mt of suspended sediment to the coastal waters, composed of ~58% silt and clay, and ~42% sand, and roughly 5 Mt of sand and gravel bedload. After dam removal ended in 2014 sediment discharge declined rapidly: A total of ~3 Mt of suspended sediment were discharged between water years 2016 and 2022, composed of ~62% silt and clay and 38% sand ( Figure 2A ). The seasonality of suspended discharge also changed during and after dam removal. Most of the suspended-sediment discharge occurred during winter high flow events both during and after dam removal. Suspended sediment fluxes during the lower flows of summer, though, decreased rapidly after dam removal ended as the availability of erodible sediment decreased ( Figure 2B ). These summer season decreases in sediment discharge were especially apparent after dam removal was completed (WY2015-2022; Figure 2B ), whereas during dam removal (WY2012-2014), the suspended-sediment discharge during the summer was several orders of magnitude higher. For example, by early August (day of year 200) when many of the scuba surveys were conducted, the Elwha River was discharging 20 to 400 t/d during active dam removal (WY2012-2014) but only ~1 t/d following the completion of dam removal (WY2015-2022; Figure 2B ). Thus, dam removal resulted in three distinct intervals of time with respect to river sediment and turbidity: 1) during dam removal (WY2012-2014), when suspended sediment discharge was high during both winter and summer seasons, 2) immediately following dam removal (WY2015-2016), when annual suspended-sediment discharge was high, but with lower turbidity and sediment discharge during the summer, and 3) several years following dam removal (WY2017-WY2022) when river sediment loads were relatively low in both winter and summer.

Morphologic changes to the seafloor occurred near the Elwha River mouth during dam removal (Gelfenbaum et al., 2015), resulting in vertical sediment deposition exceeding 5 m in some locations ( Figure 3E ). Sediment deposits to the west of the river mouth were composed primarily of mud, and to the east of the river mouth of fine sand ( Figure 3H ). As sediment flux to the nearshore declined after dam removal ended ( Figure 2A ) these new sub-tidal deposits were eroded and re-worked in response to hydrodynamic forces ( Figures 3C, F, I ). Notably, the sand deposit on the east side of the delta eroded, and deposits at all points around the delta coarsened ( Figures 3F, I ). The four sites in the buried site group include those nearest to the river mouth, and at shallower depths. Two sites were in an area of mud deposition to the west of the river mouth, and two sites were in an area of fine sand deposition to the east of the river mouth ( Figures 3E, F ).

The earliest and most pronounced change during dam removal was the decrease of kelp density and species richness at both buried and unburied site groups beginning in 2012 ( Figures 4A, B ). This change is consistent with the timing of an increase in suspended sediment flux from the river following the breach of the Elwha Dam ( Figure 2A ; Rubin et al., 2017), increasing the turbidity of coastal waters (Foley and Warrick, 2017) and reducing the availability of light in the water column (Glover et al., 2019). The trajectory of change following the initial response of kelp to dam removal, however, was different for the buried and unburied site groups. Within the buried site group, density and species richness of kelp did not recover to pre-removal levels by 2022 ( Figures 4A, B ) despite a substantial reduction in suspended sediment flux between 2017 and 2022 ( Figure 2 ).

At the unburied sites, kelp density and richness decreased immediately following dam removal when suspended sediment flux was highest ( Figures 2 , 4A, B ). Unlike the buried sites, the density of kelp at unburied sites started to rebound in 2015 ( Figure 4A ), coinciding with reduced sediment flux from the river, especially during the spring and summer months ( Figure 2B ). The rate of kelp recovery was rapid at unburied sites, largely returning to pre-removal levels by 2015 for richness and 2017 for density, although variability among sites was considerable within the unburied group ( Figures 4A, B ). Density of Desmarestia sp., the other main group of brown algae in the study area besides kelp, total percent cover of algae, and percent cover of red algae, the other main algal group besides brown, showed trends across time generally like kelp at buried and unburied sites ( Figures S1B, C, E, F ).

For invertebrates and fish at sites within the buried group, density and species richness also decreased during dam removal and had not recovered to pre-removal levels by 2022 ( Figures 4C–F ). The decrease in invertebrate and fish density and species richness were most pronounced in 2013 (fish) and 2014 (invertebrates), which coincided with the period of sediment deposition in the buried site group ( Figure 3 ; Gelfenbaum et al., 2015; Rubin et al., 2017). At buried sites, fish density increased briefly in 2017 and 2018, decreased in 2019, and then remained low through 2022 ( Figure 4 ).

Invertebrate and fish density and species richness did not decrease within the unburied site group during dam removal ( Figure 4 ; Rubin et al., 2017). Invertebrate density at unburied sites increased in 2014 and remained high through the end of the time series. Fish density within the unburied site group briefly increased from 2016-2018, but otherwise remained at a similar level throughout the time series. Invertebrate and fish density and species richness were highly variable within site groups between 2011-2022 without any clear trend over time.

At the control sites, density and species richness of kelp, invertebrates, and fish, and density for Desmarestia sp. and percent cover for red algae, varied annually between 2011 and 2022, but with no clear trend over time ( Figures 4 , S1A, D ). The decrease in kelp richness at control sites in 2014-2016 is consistent with regional patterns of kelp decline due to a marine heatwave affecting the eastern Pacific (Tolimieri et al., 2023).

The major physical changes in the nearshore during dam removal included increased suspended sediment flux ( Figure 2 ) and sediment deposition ( Figure 3 ). For the buried site group, the dominant mechanism driving community change was burial and an accompanying change in grain size ( Figures 3G–I ). The change in grain size, however, was not the same at all buried sites. At the sites to the west of the river mouth, sand was buried by mud; to the east, gravel was buried by sand ( Figures 3G–I ).

This change in grain size resulted in changes in invertebrate and fish species present within the buried group ( Figure 11 ). For invertebrates, the bivalves Clinocardium nuttallii (cockles) and Panopea generosa (geoduck clams) and the polychaete worms Diopatra ornata and Pista sp. were more abundant at the western buried sites (mud) following dam removal than before ( Figure 11B ). For fish, flatfish and snailfish became more abundant at the western sites when the habitat changed from sand to mud after burial. Poacher decreased after the western sites were buried but were still occasionally present.

For the buried sites on the eastern side of the delta where the habitat changed from gravel to sand, Metacarcinus magister (Dungeness crab) were recorded during our surveys after the sites were buried and had not been observed at those sites before. Sand lance were present during every survey in the new sand habitat at the eastern sites despite not being observed there before deposition ( Figure 11A ). Flatfish (other and starry flounder Platichthys stellatus) also became more abundant following burial at the east sites than before.

The mechanisms driving change in the unburied site group varied depending on taxonomic group. The only dam removal-related physical change at unburied sites was elevated suspended sediment. Kelp community composition was strongly related to FSSL in the river, which was correlated with suspended sediment in the nearshore (Glover et al., 2019), and total kelp density and percent algal cover were strongly negatively correlated with FSSL ( Tables 3 , 4 ; Figure S2A ), suggesting that algal trajectories of change during and after dam removal were primarily driven by high suspended sediment levels during dam removal followed by water column clearing following dam removal ( Figures 2 , 4 – 6 , S1 ). Kelp community composition was also related to Pycnopodia density at unburied sites (p < 0.05; Table 3 ). Kelp community composition ( Table 3 ), kelp density, and algal cover were unrelated to FSSL within the control site group (p > 0.05).

Invertebrates in the unburied site group were affected by changes in suspended sediment and (or) algal cover, and by changes in Pycnopodia density. Invertebrate community composition change within the unburied site group was significantly related to FSSL, algal cover, and the density of Pycnopodia (p < 0.001 for all three terms; Table 3 ). For invertebrates within the control site group, community change was not significantly related to FSSL or algal cover (p > 0.05) but was related to Pycnopodia density (p < 0.05; Table 3 ) even though the ANOSIM test did not detect significant community change across years at control sites ( Table 1 ).

For individual invertebrate species that contributed most to community composition change within the unburied site group ( Table S1 ; vectors in Figure 8C ), the significance of associations with FSSL, algal cover, and the density of Pycnopodia varied widely based on the univariate regressions ( Table 4 ). Except for the seastar Henricia sp., all species that were positively correlated with FSSL ( Figure S2 ) were filter feeders (bivalves and sabellid worms). Most of those species also had a significant negative correlation with algal cover ( Figure S3 ). Comparatively, Cancer productus (crab) density was negatively correlated with FSSL and positively correlated with algal cover. Multiple invertebrate species were negatively correlated with Pycnopodia density, including two species of bivalves, one snail, two sabellid worms, and one urchin ( Figure S4 ). This correlation suggests that the loss of Pycnopodia from our sites between 2013 and 2014 could be driving the increase in abundance of those species ( Figure 8D ), particularly for species that were not correlated with FSSL or algal cover (bivalves, sabellid worm Eudistylia polymorpha, and red urchin Mesocentrotus franciscanus). Mean Pycnopodia density (averaged over site years) was higher at control than at unburied sites, and community composition change was significantly related to the density of Pycnopodia within the control and unburied site groups. However, different species were affected by Pycnopodia within those two site groups. All invertebrate species that were negatively correlated with the density of Pycnopodia within the unburied site group had no relationship with Pycnopodia density within the control site group, possibly because the density of these species was lower in the control than in the unburied site group (< 0.75 times as high).

Following disturbance events, systems can have a range of trajectories. In most cases, systems tend to either return to a condition closely resembling the pre-disturbance state or reset to a different state (Palumbi et al., 2008; Fryirs and Brierley, 2016). In the Elwha, Ritchie et al. (2018) described evidence for both models in the geomorphic system. Similarly, we found evidence for two trajectories of species and community change in the Elwha nearshore depending on the location of a site and whether a site was exposed to the joint stressors of increased turbidity and sediment deposition (buried site group, close to the river mouth) or only increased turbidity (unburied site group, farther away from the river mouth). Within each of these groups, we also found that different taxonomic groups responded differently to burial and increased turbidity, including differences in timing of response.

The geomorphic response in the coastal environment can explain, in part, the response trajectory of the marine community. Within the unburied site group, kelp responded directly to the increased flux of fine suspended sediment as soon as dam removal began (2012; Table 4 ) and started to rebound quickly once turbidity decreased shortly after dam removal was complete ( Figures 4A, B , 5C ). Increased suspended sediment decreased light availability at the unburied site group (Rubin et al., 2017), but may have also resulted in increased scour and temporary deposition, which is known to negatively affect algal recruitment and survival of juvenile sporophytes (Traiger and Konar, 2018; Picard et al., 2022). Kelp density and species richness at the unburied site group responded rapidly ( Figure 4 )—within a year—to the attenuation of FSSL after dam removal was complete. Overall kelp community composition at the unburied site group took longer—closer to ten years—to re-establish a state equivalent to the initial condition ( Figure 5C ). The close similarity between the initial kelp community and the community present at the end of our study, after a major disturbance, is notable. Watson and Estes (2011) followed recovery of kelp after sea otters reoccupied areas and removed red urchins, and found considerable variation in the re-established communities attributable to propagule availability, succession, and demography.

The invertebrate response within the unburied site group was not as universal across species as it was for kelp, nor was it directly tied to the timing of increased turbidity caused by dam removal for most species. Invertebrate density and richness did not immediately change when turbidity increased ( Figures 4C, D ), but overall community composition did change in 2012 ( Figure 7C ). Invertebrate density started to increase in 2013 and remained elevated compared to before dam removal throughout the time series ( Figure 4C ). For some species, increased turbidity and the loss of kelp seemed to allow them to gain a toehold at the unburied sites ( Table 4 ; Figures S2, S3 ), and they were able to persist after kelp recovered and turbidity subsided ( Figures 4C , 8D ). Studies have shown that some filter feeders benefit from increased suspended sediment in the water column, particularly when it is correlated with an uptick in the availability of organic carbon (Dunton et al., 2006; McGovern et al., 2020). Decreases in understory algae can benefit sessile invertebrates by releasing them from competition for space (Arkema et al., 2009), or by increasing filter feeder access to food because the understory inhibits flow rates beneath it and possibly delivery of suspended particulate material as well (Eckman et al., 1989; Eckman and Duggins, 1991). For species whose abundance was positively correlated with algal cover, density tended to increase starting in 2015 ( Figure 8D ) when kelp started to recover.

There was no significant change in fish species composition within the unburied group, suggesting that increased turbidity and loss of kelp had little effect on the fish enumerated in our surveys. The lack of an effect of kelp density on fish abundance is contrary to other studies from kelp forest systems (Bodkin, 1988; Norderhaug et al., 2020). However, many of those studies focused on canopy-forming kelp (e.g., Macrocystis gigantea, Nereocystis luetkeana), while the kelp community in the Elwha nearshore is dominated by prostrate species, such as Alaria marginata, Costaria costata, and Laminaria spp. However, in some cases understory kelp too can be associated with increased fish abundance (Hamilton and Konar, 2007). There were fewer clear long-term invertebrate winners and losers within the unburied site group due to the transitory nature of the dam removal effects at these sites. The invertebrates that were able to sustain an advantage from short-term changes in turbidity and algal cover appear to be the long-term winners following dam removal.

The response of kelp, invertebrates, and fish within the buried site group was consistent with previous studies in other marine ecosystems following large sediment deposition events where communities tend to shift to an alternate state (Miller et al., 2002; Lohrer et al., 2006; Connell et al., 2008; Watanabe et al., 2016). The species response at the buried site group highlighted winners and losers due to dam removal. Species that were dependent on the gravel substrate at the eastern buried sites were lost locally and had not recovered by 2022. For the sand-dependent species, the area of new habitat created by deposition of relatively fine sediment associated with dam removal resulted in density increases. Within the buried site group, kelp responded initially to the increased turbidity in 2012 and recovery was precluded due to burial and a change in habitat suitability when the cobble substrate at the eastern buried sites was buried by sand ( Figure 3I ), which reduced or eliminated substrate suitable for holdfast attachment. For invertebrates, the effects were more species specific, although the overall trajectory was decreased density and species richness ( Figure 4 ). The initial increase in turbidity resulted in an increase in filter feeding sabellid worms, followed by a steep decline once the sites were buried in 2014 ( Figure 8A ). For other species, burial provided new habitat that was rapidly colonized by species that rely on finer grained habitat than cobble. For fish, increased turbidity at the start of dam removal in 2012 did not result in drastic changes to species density and richness or overall community composition, but burial and related grain size changes in 2013 and 2014 resulted in decreased richness and a significant change in species composition ( Figures 4 , 9 ). For some species, particularly the ecologically important forage fish Pacific sand lance, the change in habitat at the river mouth was a boon for their population ( Figure 11 ). Frick et al. (2022) also saw an increase in sand lance at their Elwha sites after dam removal, but the increase was not significant. For all three taxonomic groups at the burial sites, despite their different initial responses to turbidity and burial, they all reached a relatively stable level of density and richness, and consistent community composition by 2016, following the peak of sediment deposition at the buried sites ( Figures 3 – 5 , 7 , 9 ). The evolution of seafloor deposits formed during dam removal is on-going ( Figure 3 ), with important implications for the species found within the buried site group. As the deposits of mud and sand formed during dam removal are eroded by tides and waves and the substrate coarsens ( Figure 3I ), it is possible that the composition of the benthic community associated with those new habitats will continue to evolve until geomorphic processes stabilize.

Within the control site group, change was unrelated to the effects of dam removal. Kelp community composition did significantly change over the course of our study, but the trajectory of change was highly variable over time ( Figure 5D ) and changes in density and species richness did not correspond with the timing of changes within the buried and unburied site groups ( Figure 4 ). There were regional changes that occurred at our sites (see Other drivers of change below) that likely had a stronger effect on the variability of kelp, invertebrates, and fish within the control site group.

While dam removal was the dominant driver of change for kelp and some invertebrate and fish species from 2012-2015 for the unburied and buried site groups, dam removal was not the only disturbance playing out in the Pacific Northwest during our study period. As mentioned previously, SSWS decimated populations of Pycnopodia in the Strait of Juan de Fuca between our 2013 and 2014 surveys (Montecino-Latorre et al., 2016; Harvell et al., 2023) and populations had not recovered by 2022 ( Figure 8C ). Correlation analysis showed that the changes in the density of Pycnopodia was significantly negatively correlated with densities of six invertebrate species in the unburied site group, including (from strongest effect to weakest) Mesocentrotus franciscanus, Saxidomus gigantea, Schizobranchia sp., Tresus capax, Eudistylia polymorpha, and Nucella sp. ( Table 4 ; Figure S4 ). Pycnopodia is a generalist in most subtidal communities (Duggins, 1983), so the loss of Pycnopodia may have affected other invertebrate species besides those we included in our analysis. We limited our analysis to those species that contributed to the overall change in invertebrate community composition within the unburied site group during our study. To the north of our study sites, but still within the Salish Sea, Schultz et al. (Schultz et al., 2016) found evidence of a trophic cascade in the subtidal communities in Howe Sound following the loss of Pycnopodia, whereby the population of green urchins (Strongylocentrotus droebachiensis) increased, and kelp cover declined. Although invertebrate community composition did not change significantly across years within the control site group ( Table 1 ; Figure 7A ), community composition at control sites was significantly related to the density of Pycnopodia ( Table 3 ). This mixed result provides some support for a more extensive regional effect of the loss of this important sea star predator.

For some species, density increases following the loss of Pycnopodia may have resulted from behavior changes rather than reduced predation mortality. Urchins sometimes come out of hiding and switch from passive to active foraging (Smith et al., 2021) and are preyed upon by and show an escape response to Pycnopodia (Duggins, 1983), however, susceptibility of red urchins (Mesocentrotus franciscanus) to Pycnopodia predation is greatly reduced when they reach full size (Duggins, 1983). Although we did not take size measurements, the red urchins we censused appeared to be full sized. They were usually aggregated on boulders or more spread out on gravel-cobble substrate, suggesting a change in foraging behavior rather than a release from predation.

The other regional event that coincided with the tail end of increased suspended sediment associated with the dam removal was the largest marine heatwave on record from winter 2013-2015 (Di Lorenzo and Mantua, 2016). Given the regional nature of this event, we would expect to be able to see the effects, if any, of the marine heatwave at our control sites. For kelp, there was significant change in community composition across years that was not temporally aligned with change at Elwha sites (little change from 2011-2013 at control sites compared to great change at Elwha sites; Figure 5 ). The waters along the outer coast of Washington, including the western entrance to the Strait of Juan de Fuca, experienced anomalously warm temperatures in 2013 and 2014 (Tolimieri et al., 2023), and canopy kelp extent was anomalously low on the outer coast and in the Strait of Juan de Fuca in 2014 (Pfister et al., 2018). Kelp density at our control sites decreased from 2013 to 2016 with some recovery in 2017 ( Figure 4 ), and Nereocystis leutkeana and Laminaria setchellii were lost in 2014 but reappeared by 2016-2017 ( Figure 6C ), accounting for the large change in kelp community composition between 2013 and 2017 ( Figure 5D ). These results suggest some correspondence between kelp response at our control sites and regional trends in ocean temperatures and canopy kelp abundance. We would expect that regional oceanographic conditions would affect kelp at Elwha sites as well, but we were unable to separate regional from dam removal effects based on our sampling design because of the overlap of the two stressors.

The correlations we observed between community composition and the loss of sea stars ( Tables 3 , 4 ; Figure S4 ), along with the potential connection with climate variability (Tolimieri et al., 2023) cloud, to a degree, our ability to characterize the long-term dam removal response in the marine ecosystem, as well as to predict the ongoing trajectory of change. Invertebrate community composition within the unburied site group, for example, was indirectly related to dam removal (correlation to algal cover), but directly affected by SSWS. The relaxed predation pressure likely drove increased densities of bivalves, urchins, and some worms, suggesting that both dam removal and SSWS contributed to the response trajectory. It is possible that dam removal sparked changes in the marine community that were then further shaped by additional stressors in the system, some of which we could directly test and others that we could not.

While dam removal is accelerating in the United States (O’Connor et al., 2015), the number of dam removals motivated by, or that considered, marine impacts or restoration in their management is small (Ralston et al., 2021; Cancel Villamil and Locke, 2022). The removal of the two dams on the Elwha was notable for its influence on multiple aspects of the marine system (Gelfenbaum et al., 2015; Rubin et al., 2017; Shaffer et al., 2017; Glover et al., 2019; Warrick et al., 2019). It is likely that a variety of factors contributed to those effects, including the characteristics of the impounded sediment (volume, erodibility, and grain size distribution), length and steepness of the river system, and distribution of dams within the watershed (i.e., tributary versus mainstem), which have been well characterized, particularly for dam removals in the western United States (Foley et al., 2017a; Major et al., 2017). The characteristics of marine waters into which the watershed drains—waves, currents, and bathymetry—influenced sediment dynamics in the nearshore. While data on the marine influence of dam removals besides Elwha is generally lacking, sediment dynamics in the nearshore from other sediment disturbances (e.g., land use change, landslides) have been studied (Crain et al., 2009; Erftemeijer et al., 2012; Booth, 2020; Rangel-Buitrago et al., 2023). Below we highlight processes that contributed to the geomorphic changes and ultimately ecological responses and how they may be used to consider effects of sediment disturbances, including dam removal, in other locations.

The Elwha River delivered 20 Mt of sediment to the nearshore during and immediately following dam removal, resulting in greater than 5 m of deposition near the river mouth ( Figure 3E ). The watershed steepness, proximity of dams to the coast, and lack of additional impoundments all contributed to the rapid flux of sediment to the nearshore and subsequent reduction following the completion of dam removal (Ritchie et al., 2018). The tidal currents in the Strait of Juan de Fuca are a dominant driver of sediment transport at the Elwha and likely contributed to a reduced amount of fine sediment retention (Warrick et al., 2008; Miller et al., 2011) compared to other systems with weaker tidal forcing. Of the 20 Mt of sediment delivered to the nearshore, approximately 60% was fine silt and clay (fine suspended sediment), which was predominantly dispersed away from the Elwha River delta by the strong tidal currents (Gelfenbaum et al., 2015). This sediment was transported predominantly to the east of the river mouth, affecting the spatial distribution of water column turbidity (Gelfenbaum et al., 2015; Foley and Warrick, 2017; Glover et al., 2019) and resulting impacts to nearshore communities (Rubin et al., 2017).

During the dam removals, the large flux of coarse sediment to the nearshore overwhelmed the littoral transport processes resulting in widespread accumulation of mostly sand-sized sediment ( Figures 3E, H ) that deposited on top of formerly eroding lag deposits of cobbles and boulders (Warrick et al., 2008). The two dams were operated as run of river dams, allowing for the passage of a substantial portion of fine sediment to reach the nearshore while the dams were in place, while retaining most of the sand and coarse sediment in the reservoirs. With the removal of the dams, it is likely that sand delivery to the nearshore will be greater than when the dams were in place. While the large influx of sediment initially resulted in deposition near the river mouth, those deposits are actively being reworked by tidal currents and waves, resulting in net erosion on the subaqueous delta (Warrick et al., 2019) and a coarsening of grain size ( Figures 3C, F, I ). The ongoing changes at the delta and the magnitude of deposition and erosion will likely affect future habitat suitability for the species in the new deposits.

Based on our observations of geomorphic and ecological response after the Elwha dam removals, the major factors to consider when evaluating the potential effects from dam removals or large sediment fluxes from watersheds to the marine environment (e.g., landslides, wildfire debris flows) include the type, magnitude, and duration of the disturbance, character of the sediment, and coastal sediment transport processes that result in dispersal and sorting of sediment in the nearshore. In the case of the Elwha, ecological change was driven by two types of effects, sediment deposition of coarser grain sediment and increased turbidity from finer grain sediment. The duration of the disturbance depended on the type of effect. Sediment deposition had longer lasting effects on the ecosystem than increased fine suspended sediment, in part due to the more variable effects of suspended sediment in space and time. In addition, species likely had different sensitivities to stressor types depending on their life history, morphological structure, mobility, and habitat needs. The magnitude of effect on the Elwha ecosystem was likely a function of both the overall sediment flux from the reservoirs to the coast as well as the hydrodynamics of the Elwha nearshore. Coastal areas without strong tidal currents would likely be most affected by burial due to a lack of transport capacity in the system, particularly if the amount of sediment being delivered is greater than background conditions. While the exact trajectories of community response to a disturbance are difficult to predict, assessing these conditions can be used to inform the development of conceptual models of potential outcomes that can inform monitoring and modeling efforts before and after dam removal.