Sea‐level rise, habitat loss, and potential extirpation of a salt marsh specialist bird in urbanized landscapes

Abstract Sea‐level rise (SLR) impacts on intertidal habitat depend on coastal topology, accretion, and constraints from surrounding development. Such habitat changes might affect species like Belding's savannah sparrows (Passerculus sandwichensis beldingi; BSSP), which live in high‐elevation salt marsh in the Southern California Bight. To predict how BSSP habitat might change under various SLR scenarios, we first constructed a suitability model by matching bird observations with elevation. We then mapped current BSSP breeding and foraging habitat at six estuarine sites by applying the elevation‐suitability model to digital elevation models. To estimate changes in digital elevation models under different SLR scenarios, we used a site‐specific, one‐dimensional elevation model (wetland accretion rate model of ecosystem resilience). We then applied our elevation‐suitability model to the projected digital elevation models. The resulting maps suggest that suitable breeding and foraging habitat could decline as increased inundation converts middle‐ and high‐elevation suitable habitat to mudflat and subtidal zones. As a result, the highest SLR scenario predicted that no suitable breeding or foraging habitat would remain at any site by 2100 and 2110. Removing development constraints to facilitate landward migration of high salt marsh, or redistributing dredge spoils to replace submerged habitat, might create future high salt marsh habitat, thereby reducing extirpation risk for BSSP in southern California.


| INTRODUC TI ON
Salt marshes shift their distributions in response to sea-level rise (SLR) through vertical accretion, landward inundation, and retreat to formerly dryland sites (Donnelly & Bertness, 2001).
Anticipated future rapid rates of SLR could obviate the benefits of accretion, and coastal development will, in many cases, prevent inland retreat (Roman, 2017). Along the Pacific Coast, recent modeling efforts have predicted a complete loss of coastal salt marshes in California (Thorne et al., 2018). This is of particular concern in areas such as southern California (USA), where small, "urban" salt marshes are hotspots and refugia for sensitive endemic species (Zedler, 1982), including the state endangered Belding's savannah sparrow (Passerculus sandwichensis beldingi; BSSP). Here, we examine how BSSP habitat may respond to SLR.
Estuarine sparrows have small home ranges and narrow elevation niches. Any benefits of sediment contributions could be obscured, reducing niche availability, and potentially geographic ranges, for those species dependent upon estuarine margins. For instance, salt marsh sparrows (Ammodramus caudacutus) on the USA east coast declined by 9% annually, from 1998 to 2012, primarily due to reductions in habitat availability. Such losses could be compounded by SLR, with studies predicting extirpation by 2035 (Field et al., 2017) or 2050 (Correll et al., 2016). Furthermore, salt marsh habitat for seaside sparrow (Ammodramus maritimus) populations in Georgia, USA, could contract by 81% by 2100 (Hunter, Nibbelink, & Cooper, 2016).
Similarly, in the San Francisco Bay Estuary (SFBE), the tidal marsh song sparrow (Melospiza melodia) could be vulnerable as highelevation habitat becomes low elevation habitat (Veloz et al., 2013). However, Kirwan, Temmerman, Skeehan, Guntenspergen, and Fagherazzi (2016) suggest that past models based on constant accretion rates do not incorporate the self-adaptive capacity of salt marshes, thereby over-estimating habitat loss.
Future projections of habitat response to SLR that account for the dominant processes dictating tidal marsh elevation (Morris, Sundareshwar, Nietch, Kjerfve, & Cahoon, 2002;Schile et al., 2014;Swanson et al., 2014;Thorne et al., 2018), species distribution modeling, and population size projections should provide more robust projections of suitable habitat. Similar to Atlantic Coast and SFBE salt marsh sparrows, BSSP is a nonmigratory salt marsh specialist with a narrow geographic range from Goleta Slough, California, USA, to Bahia San Quintin, Baja California, Mexico. BSSP breeding depends on middle-to highelevation marsh habitats dominated by pickleweed (Sarcocornia pacifica) and salt grass (Distichlis spicata) (Bradley, 1973(Bradley, , 1994Grinnell & Miller, 1944;Powell, 1993Powell, , 2006. Inundation limits BSSP's lower elevation niche, whereas territorial song sparrows (M. melodia) can displace BSSP from the upland transition zone (Zembal, Hoffman, & Patton, 2015). Furthermore, studies of BSSP song dialects suggest that individuals do not often disperse between sites (Bradley, 1994;Burnell, 1996), subjecting them to local extirpation. In 1988, BSSP occupied only 2,150 ha of salt marsh vegetation, salt flats, and small tidal channels among 27-30 sites on the Southern California Bight (SCB) (Zembal, Kramer, Bransfield, & Gilbert, 1988), varying in area from <1 ha to 620 ha (Powell, 2006). The 2015 regional population estimate found an increase of 11.3% from counts in 2010 (Zembal et al., 2015), perhaps due to greater nesting success and survival in a warmer and drier period. How SLR affects this apparent recovery trajectory could affect plans for delisting.
Salt marsh habitats lie within the intertidal zone and rely on a balance between accretion and erosion, as well as uplift and subsidence to maintain elevations with SLR. Salt marshes can trap mineral sediment and accumulate organic matter to maintain their position with rising seas (Kirwan et al., 2016), and they might migrate inland as upland habitats recede (Raabe & Stumpf, 2016). However, coastal development in the SCB acts as a backstop to transgression and likely reduces sediment available for accretion (Callaway & Zedler, 2004;Thorne et al., 2018). Nonetheless, diverse land uses within each salt marsh catchment lead to variable accretion rates within and across salt marshes in the SCB (Callaway, Borgnis, Turner, & Milan, 2012;Day et al., 1999;Thorne et al., 2018). The uncertainty created by the range in accretion rates led us to create vulnerability scenarios for individual salt marshes using site-specific data, as well as build upon recent Pacific Coast research that predicts salt marsh eradication within 100 years under high rates of SLR (Thorne et al., 2018).
To project future salt marsh elevations at sites that are important to BSSP, we calibrated the wetland accretion rate model of ecosystem resilience (WARMER). WARMER is a one-dimensional soil cohort model that projects future salt marsh elevation based Removing development constraints to facilitate landward migration of high salt marsh, or redistributing dredge spoils to replace submerged habitat, might create future high salt marsh habitat, thereby reducing extirpation risk for BSSP in southern California.

K E Y W O R D S
Belding's savannah sparrow, California, conservation, dynamic salt marsh accretion model, sea-level rise, species distribution model on (a) the dynamic relationship between organic matter accumulation and elevation, (b) the nonlinear relationship between inorganic matter accumulation and elevation, and (c) temporally variable SLR (Swanson et al., 2014;Thorne et al., 2018). Unlike the regional and temporally constant accretion rate used to calibrate wetland models for other sites in other studies (Kirwan et al., 2016), we used insitu historical sediment accumulation rates to dynamically project organic and mineral accretion for each salt marsh.

| MATERIAL S AND ME THODS
To assess the vulnerability of BSSP habitat to SLR, we (a) collected baseline habitat information, (b) estimated salt marsh area and elevation gains or losses with SLR using a dynamic one-dimensional elevation model, (c) determined current habitat suitability, and (d) projected habitat suitability under three plausible SLR scenarios. Fine-scale, site-specific data were leveraged to answer these research questions at 5-m horizontal resolution across six study sites in the SCB.

| Study sites
We modeled habitat suitability for the six fully tidal salt marshes  (Figure 1) where BSSP breed now and enough data were available to parameterize WARMER (see Elgin, 2012.

| Salt marsh topography
For evaluating salt marsh vulnerabilities, salt marsh elevations were defined relative to the local tide datum. Swanson et al. (2014) define z* as a unit free "elevation relative to the tidal range of the site," which is calculated as: By definition, z = the absolute elevation relative to North American Vertical Datum 1988 (NAVD88). Because z* = 0 when z = mean sea level (MSL), and z* = 1.0 when z = MHHW for all sites, we were able to compare vulnerabilities across sites.

| Wetland Accretion Rate Model of Ecosystem Resilience
WARMER is a dynamic, one-dimensional elevation model that incorporates the self-adaptive capacity of salt marshes to respond to SLR based on site-specific inundation, sedimentation, climate, and vegetation characteristics (Callaway, Nyman, & DeLaune, 1996;Swanson et al., 2014). Based on a cesium-137 analysis of a soil core from Carpinteria via Elgin (2012), existing soil core parameters from Sweetwater via Thorne et al. (2014) and soil core parameters from the remaining sites via Thorne et al. (2016), we used the full WARMER model, and all associated inorganic sediment and organic matter functions (Morris et al., 2002;Swanson et al., 2014), to pro- of Fish and Wildlife (Zembal et al., 2015). In summary, biologists walked transects and sightings within 100 feet of the transects were recorded on field maps and later georeferenced over an aerial photograph. Breeding habitat was indicated by several categories of BSSP behaviors such as singing (s) or scolding, perching together of mates, including feeding of young, nest building, and aerial chases between two territory holders. Posted males or foraging birds were not included in the breeding habitat or breeding period model inputs. All areas surveyed were accessible on foot.
The Carpinteria transect surveys occurred on two consecutive days monthly from January 2012 to March 2013, spanning high and low tides (Lafferty, Stewart, & Hechinger, 2017). A survey occurred on a single day at Mugu Lagoon in February 2018 using a similar transect survey strategy. The walking transect was designed to sample the breeding and foraging habitat exhaustively.
All observed BSSP were recorded to the nearest 10 m on hard copy maps, and points were later digitized in ArcGIS. BSSP, especially nonsinging males, are secretive and difficult to detect at distance (Powell, 2006). Distance sampling was performed post hoc, so every digitized sparrow location was assigned a distance to the transect. In this study, counts declined by half when 30 m from an observer so that we had high confidence in detections within 10 m of a transect.
The compiled data at Border Field State Park followed the 5-year statewide breeding survey protocol (Zembal et al., 2015), where oc-

| Current distribution
Suitable breeding habitat occurred in the middle, high, and upland transition zones of the saltmarsh, while foraging habitat was more extensive and variable across sites. This was defined by Maxent modeling, but quantiles explain the elevation differences. For example, the 90% quantile of z* (a dimensionless elevation value; Swanson et al., 2014) for breeding habitat, breeding period, and foraging habitat were 3.9, 2.9, and 2.4, respectively ( Table 2). The response to elevation was better than the random species distribution model (Figure 2), and results of the Maxent modeling indicate that 99% of the Carpinteria salt marsh was suitable foraging habitat and 91% (59 ha) was suitable breeding habitat (Figures 3 and 4). Approximately 67% (93 ha Figures 3 and 4).
Thus, within and across fully tidal salt marshes, there was a mosaic of suitable breeding and foraging habitat. However, these relative percentages changed with projected SLR.

| Low rates of SLR (0.44 m/100 year)
The distribution of habitat was predicted to shift extensively under a low rate of SLR. However, Carpinteria was an exception.
Under this scenario, 99% of the salt marsh was predicted to be suitable foraging habitat by 2110, and Carpinteria was expected to have no net loss of suitable breeding habitat (Figures 3 and 4).
Mugu also was projected to gain suitable foraging habitat under this scenario. However, Mugu was predicted to lose breeding habitat so that only 47% (65 ha) of the salt marsh could be suitable by 2110 (Figures 3 and 4). At Seal Beach, foraging habitat was expected to be reduced to 5% (9 ha), and breeding habitat was expected to be eliminated by 2110 (Figures 3 and 4). Newport foraging habitat was predicted to be reduced to less than 36% (54 ha (Figures 3 and 4). At Mugu, suitable foraging habitat was predicted to be reduced to 16% (22 ha) of the total area under this scenario, and less than 1 ha of the salt marsh was predicted to be suitable breeding habitat by 2110 (Figures 3 and 4). At Seal Beach, foraging and breeding habitat were expected to be reduced to less than 1 ha by 2110 and 2070 (Figures 3 and 4). Further south, Newport foraging habitat was expected to be reduced to 4% (6 ha) of the initial salt marsh area, while breeding habitat could be reduced to 1% (2 ha) of the initial salt marsh area (Figures 3 and 4)

| High rates of SLR (1.66 m/100 year)
This scenario showed complete submergence of foraging and breeding habitat at all six sites by 2110. Under this scenario, 5% (3 ha) of Carpinteria was predicted to be suitable foraging and suitable breeding habitat by 2100, and no BSSP habitat was expected to be left by 2110. At Mugu, suitable foraging habitat was expected to be reduced to 1% (2 ha) of the total area by 2090 under this scenario, and less than 1 ha of the salt marsh was predicted to be suitable breeding habitat by 2100 (Figures 3 and 4). At Seal Beach, foraging and breeding habitat were eliminated by 2070 and 2060 (Figures 3 and 4). Only 2% (3 ha) of Newport foraging habitat was expected to be left by 2090, while breeding habitat was predicted to be reduced to 4% (7 ha) of the initial salt marsh area by 2070 (Figures 3 and 4). At Sweetwater, less than 1 ha of foraging and breeding habitat was expected to be left by 2090 and 2080. At Tijuana, where less than 1 ha was predicted to be foraging habitat by 2100, and less than 1 ha was expected to be breeding habitat by 2080. All suitable habitats were expected to be submerged in Tijuana by 2110. Thus, BSSP is expected to have no suitable habitat left under a high SLR scenario by 2110.

| Spatial patterns of habitat loss within salt marshes
Across the SLR scenarios, upland transition zone habitats would become suitable for BSSP with increasing inundation depth and frequency, while low elevation foraging areas were forecasted to be submerged ( Figure 4). SCB salt marshes face habitat migration restrictions, such as highways, roads, and adjacent development (Figure 4), so landward movement was assumed to be minimal. Habitat loss would occur first in low areas (e.g., open water, bayward edge, and tidal creeks; Regardless of the progression, all currently suitable habitat areas were predicted to be inundated by SLR through the next century.

| D ISCUSS I ON
Our models suggest that under all projected SLR scenarios, and without adaptation by BSSP or accommodation by humans, near complete loss of BSSP habitat is likely throughout the SCB under high SLR scenarios. Carpinteria, currently the smallest study site, could support the last remaining BSSP population within fully tidal basins due to its relatively high-elevation marsh.
Our results are consistent with projected declines in other mid to high salt marsh species. Seaside sparrow habitat in Georgia is expected to decline between 2025 and 2050 (Hunter et al., 2016).
Under high SLR scenarios, two high-elevation salt marsh birds, the Common Yellowthroat (Geothlypis trichas) and Marsh Wren (Cistohorus palustris), will likely become extirpated from the SFBE salt marsh within a century (Veloz et al., 2013). Small mammals, such as the salt marsh harvest mouse (Reithrodontomys raviventris), could be extirpated from areas currently dominated by pickleweed as sealevels rise and that habitat disappears (Shellhammer, 1989 (Powell & Collier, 1998). Furthermore, in the following year, 45.5% of banded male BSSPs in that same site occupied the same territory that they occupied when they were originally banded, highlighting their site fidelity. Reduced dispersal will make restoration more difficult if local populations are extirpated.
BSSP extirpation could occur before all habitats are submerged.
For example, salt marshes smaller than 10 ha have been shown not to support BSSP breeding populations (Powell & Collier, 1998;Zembal et al., 1988). As habitat shrinks in area due to increasing inundation, it may also decline in quality, which might lead to breeding failure be-  (Zembal et al., 2015). Common raven (Corvus corax) and American crow (Corvus brachyrynchus) are known nest predators of several threatened and endangered species in California (Liebezeit & George, 2002), and these impacts could also increase if BSSP habitat concentrates near uplands.
Future marsh elevation and associated habitat change depend on the extent that sediment supply will make up for SLR. Large storm events in the SCB have been known to rapidly increase elevations in mudflats and low marsh zones. For example, in Tijuana, high sedimentation rates during storms have led to an increase in elevation, and low to high marsh zone habitat conversion (Ward, Callaway, & Zedler, 2003). The same is true of Mugu, where low elevation areas have been repeatedly filled with sediment during storm episodes (Onuf, 1987 and has altered connectivity to freshwater sources through concrete channelization (Sadro et al., 2007). Because sediment availability is dependent on infrequent storm events that are difficult to predict (Warrick & Farnsworth, 2009) Our analyses suggest that the recent increase in BSSP counts in the SCB (Zembal et al., 2015) will likely reverse in the near future.
Even before Pacific Coast salt marshes are completely submerged in 2110 (Thorne et al., 2018), our modeling predicts that there will be no suitable habitat for BSSPs under a high SLR scenario. Although habitat suitability could temporarily increase in two of the six salt marshes we studied under low SLR scenarios, local extirpations may occur. These losses could possibly be ameliorated with management intervention, restoration, and increasing transgression upland refugia habitat.

ACK N OWLED G M ENTS
The authors would like to thank the volunteers, graduate students, and staff at U.S. Geological Survey, University of California,

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interest to declare.

AUTH O R S CO NTR I B UTI O N
JR conceived the study. JR and KB did the species distribution modeling, modeling of occurrences, acquired/interpreted the sediment core date data, and drafted the work. KL, RH, TS, MH, RP, and JC conceived, designed and conducted the bird surveys, habitat delineation, and revised the work critically for important intellectual content. KT and KB conducted the WARMER modeling, and revised the work critically for important intellectual content. JT, KT, RA, and GM conceived/designed the geomorphic modeling study, acquired the data for warmer modeling, and revised the work critically for important intellectual content.

DATA ACCE SS I B I LIT Y
All data needed to evaluate the conclusions in the paper are present in the paper. Additional data related to this paper may be requested from the authors. All raw data are archived at the U.S. Geological Survey