The four study sites are located on the Kitsap Peninsula in Washington State (Figure 1). The four sites selected included: The first site is the pre-site, Harper Creek Estuary (47˚30'00.9"N 122˚30'58.4"W) in Port Orchard, Washington. This pre-restoration site within this study contained two working culverts (60 cm in diameter). Soil directly adjacent to this site is composed of Tacoma silt loam and Harstine gravelly ashy sandy loam [20]. The second site, Carpenter Creek Estuary (47˚47'42.471"N 122˚30'26.7114"W), located in Kingston, Washington, was restored in 2013 after removal of a 3 m × 3 m box culvert on South Kingston Road and installation of a 27.4 m bridge that spans the entire channel width of Carpenter Creek and can withstand natural tidal inundation from Appletree Cove. The majority of soil adjacent to this site is classified as an alluvial beach deposit with minor components of Poulsbo gravelly sandy loam [20]. The third site sampled was the nine-year-old site at the Beaver Creek Estuary (47˚34'12.1938"N 122˚33'7.2468"W), located in Manchester, Washington, at the head of Clam Bay and was restored in 2007. Restoration of Beaver Creek with the removal of the lower culvert and installation of a 6.7 m bottomless arch bridge at the head of Clam Bay were completed in 2007. Soil in this region is composed of Kapowsin gravelly ashy loam [20]. The forth site was the 12-year-old site located at the Dogfish Creek Estuary (47˚44'48.3144"N 122˚39'7.3368"W), located in Poulsbo, Washington at the head of Liberty Bay. This site was restored in 2004 with the removal of a 1.5 m culvert, which was replaced with a 26.8 m bridge. Soil composition adjacent to Dogfish Creek includes urban land-Alderwood complex and Kitsap silt loam [20].

Transects were placed parallel to the water along the lower bank edge where the first perennial vegetation was located nearest the edge of water, as described in the PacFish InFish Biological Opinion Monitoring Program [21]. Three, 50-m

transects were place above the headwater of the restoration location on the freshwater influence side and three 50-m transects were placed below the tailwater on the marine influence side. The initial transect began at the bridge and subsequent transects followed the remaining estuarine vegetation, along both sides of the stream channel. Line-point intercept data was recorded along each meter of each transect wherein the species the nearest stem, leaf, or plant base intercepted was recorded using a four-letter code based on the first two letters of the genus and species. A total of 48 transects were completed in this study (4 sites × 2 visits × 6 transects). Along each transect, three 1 m × 1 m quadrats were randomly placed (4 sites × 6 transects × 3 quadrats). Within each quadrat, soil was collected and vegetation height (cm) from the center and each corner, along with visual percent cover of each species present was recorded. A 25 cm × 25 cm sample of biomass, including all plant matter one cm above the soil, was collected from a fourth randomly chosen quadrat (4 sites × 6 biomass samples). Hitchcock and Cronquist [22] and Pojar and MacKinnon [23] were used to identify plant species.

Soil was dried at 45˚C for one week, sieved (screen 1 cm × 1 cm) to remove rocks greater than 1 cm in diameter. From each sieved soil sample (n = 72), approximately 60 g was taken from each sample to make a homogenized composite sample representing each transect, totaling 24 composite samples. From each composite sample, a 30 g sub-sample was ground into a fine powder using a SPEX Mixer Mill in the Geology Department at WWU, which uses a hardened stainless-steel mortar and two stainless steel ball bearings. The ground composite sub-samples were used to test for P and CEC at Spectrum Analytical (Washington Court House, OH). Soil organic matter from the composited soil samples was determined using an optimized weight loss-on-ignition (WLOI) methodology specific to estuarine sediment [24]. Soil was dried overnight at 110˚C to remove any adsorbed water. Approximately two to three grams of soil was added to a ceramic crucible and the combined weight of the sample and crucible was recorded. The samples were then placed in a muffle furnace at 550˚C for four hours. Crucibles were removed and cooled for 10 minutes before being weighed. Carbon content was determined by percent WLOI and was calculated by determining the difference of final soil weight after samples were heated for four hours, from the initial soil weight.

Dried, homogenized soil and vegetation samples were analyzed for total carbon and nitrogen using a Thermo Electron NC Soil Analyzer Flash EA 1112 Series (Thermo Electron Corporation, Milan, Italy). A mass of approximately 100 mg was placed in tin capsules and compressed to remove any air prior to carbon-to-nitrogen analysis. A calibration curve for nitrogen was established using an analytical standard of atropine, which contained 48.4 g∙kg⁻¹ nitrogen. The calibration curve for nitrogen had an R² value of 0.99 and the mass of atropine ranged from 0.05 mg to 0.96 mg. The calibration curve for carbon also had an R² value of 0.99 and the mass of atropine ranged from 0.77 mg to 13.67 mg.

A multiple analysis of variance (MANOVA) was used followed by a two-way fixed-effects analysis of variance (ANOVA) to compare percent carbon, soil organic matter, macro and micronutrients, species richness, species abundance using the Shannon-Wiener (H'), and number of invasive species occurrences at each site. The fixed effects were site (pre, 3-Yr, 9-Yr, and 12-Yr) and location (above and below) with an interaction term. Levene’s test and Shapiro-Wilk’s was used to assess how well data fit the assumption of homogeneity of variances and normally distributed populations, respectively. A log transformation was applied, when needed, prior to the ANOVA. A post-hoc pairwise comparison was used for sites and locations with significant ANOVA results (P-value < 0.05) using the “holm” P-adjustment method. When data could not be transformed to meet the assumptions of ANOVA, a Kruskal-Wallis non-parametric test was used.

The Chao and bootstrap population estimations were used to assess sampling efficiency. Principal components analysis (PCA) was used to determine which combination of species explained the most variance in the plant communities [25]. Because several species comprised nearly 20 percent of the entire vegetation population across all sites, a row-centered and scaled PCA was used to create homoscedasticity in the dataset, and thus scale species with large counts. Using the prcomp function in R, based on a singular value decomposition of the data matrix, identification of which components accounted for the most variation and separation in each principal component was determined. Analysis of similarity (ANOSIM) was used to calculate dissimilarity by using a ranked dissimilarity matrix. Based on these assumptions, a two-way ANOSIM was calculated between site and location, with 1000 permutations. All data was analyzed by R Version 3.3.0; [26].

Cation exchange capacity (CEC) differed per site with higher levels associated with the oldest restoration sites and the and pre-site and lowest levels recorded in sediments associated with the 3-year-old sites, below the bridge (P = 0.002). The newest restoration site (3-Yr) below the point of culvert restoration was significantly lower in soil carbon when compared across the other sites. Pair-wise post-hoc tests indicate similarities between the pre-restoration site and the oldest post-restoration site (12-Yr) with intermediate values observed from the 9-Yr plots. There were also significant differences in percent soil organic matter between sites (P = 0.003). Similar with soil carbon, the newest site (3-Yr) below the location of culvert restoration was significantly lower in soil organic matter when compared to the other sites (Table 1). There was a significant interaction between site and location (above and below) when concentrations of N and P were compared among the sites driven by a significant N and P nutrient decreases below the bridge at the 3rd year, newest site (P < 0.05; Table 1).

No differences existed when plant height and aboveground biomass was compared among sites and locations. A significant difference between site was noted

Superscript letters indicate homogenous subsets, as determined by Tukey’s HSD.

for plant species diversity by site (F_(3,16) = 23.58, P < 0.001; Table 1). The pre-restoration site was the only site with significant differences in plant species diversity between above vs. below location (P < 0.005). Additionally, the Shannon-Wiener (H’) index of species diversity was significantly highest at the oldest post-restoration site (1.98 ± 0.04) than any other site. A trend toward a more diverse plant community assemblage can be seen over time. A similar trend was noted for plant species richness between site and location (Table 1).

Summer vegetation surveys documented a total of 64 plant species (Table 2). The Chao estimated 73 (±9) species followed by the bootstrap method, which estimated a population size of 66 (±3). The most abundant native plant species throughout all the sites included: pickleweed (Salicornia virginica; 19.2%), orache (Atriplex patula; 9.5%), gumweed (Grindelia squarrosa; 8.0%), saltgrass (Distichlis spicata; 6.4%), meadow grass (Hordeum brachyantherum; 4.9%), and dune grass (Elymus mollis; 4.2%). ANOSIM indicated significant dissimilarities between location (above and below) at all sites (All P < 0.05; Table 3).

The principal components ordination was used to illustrate the vegetation community sampled during this study. Principal components I-II accounted for 32.3% of the total variance and illustrates salinity and vertical gradients based on species variable loadings (χ² = 23.4; df = 6; P < 0.0001; Figure 2). Principal component I describes the species distribution by salt-tolerance, whereas salt tolerant species (illustrated by green ellipse) cluster in the upper left of the ordination and differentiate into freshwater marsh species (illustrated by blue ellipse)

*Significant at P < 0.05; **Significant at P < 0.01

into the right quadrate of the graph. Principal component II likely describes a vertical gradient, wherein older, woody native and non-native upland species (orange ellipse) differentiated from herbaceous marsh perennials and annuals (Figure 2).

Invasive species were relatively uncommon at each site and marginal evidence of a difference was observed in the interaction of site and location of the number of invasive occurrences in point-intercept data (F_(3,40) = 2.52, P = 0.07; Table 2). Nine invasive plant species were documented: Himalayan blackberry(Rubus armeniacus; 1.3%), Scotchbroom (Cytisus scoparius; 0.5%), reed canary grass(Phalaris arundinacea; 0.5%), hairy cat’s ear (Hypochaeris radicata; 0.1%), Canada thistle (Cirsium arvense; 0.1%), ox-eye daisy (Leucanthemum vulgare; <0.1%), European beachgrass (Ammophila arenaria; <0.01%), common tansy (Tanacetum vulgare; < 0.1%), field bindweed (Convolvulus arvensis; <0.1%), and Queen Anne’s lace (Daucus carota; <0.1%).