Unfiltered Surface Water

We collected water samples at all fixed sampling points and across all sampling periods (except at IID Marsh; table 1). Undisturbed samples of surficial water (with algae and biofilm displaced with a simple mesh strainer) were collected and placed in 500-milliliter (mL) Trace Clean® high-density polyethylene (HDPE) bottles. Samples were placed immediately on ice, acidified (pH=2) with Optima™ nitric acid within 8 hours of collection and refrigerated before analytical chemistry. All water samples were unfiltered and represented concentrations of total recoverable selenium. Although selenium in water alone does not strongly predict bioaccumulation of selenium in upper trophic levels (Hamilton, 2004), estimates of total recoverable selenium in water represented a minimal selenium exposure baseline of biota inhabiting each marsh across all periods sampled.

Chironomidae (Midge larvae) and Corixidae (Waterboatman)

We collected composite samples of Corixidae and Chironomidae at all fixed sampling points where present in sufficient biomass and across all sampling periods (except the IID Marsh; table 1). Corixidae and Chironomidae matrices were sampled extensively in previous studies of selenium risk in Salton Sea wetland food webs (Miles and others, 2009; Saiki and others, 2010; De La Cruz and others, 202216) and are therefore useful for inter- and intra-habitat comparisons. Samples from each taxa and site were collected using a combination of techniques that included a D-ring net swept rapidly in a circular motion though the water column and from surficial sediments <5-centimeter (cm) passed through a 1.0-millimeter (mm) sieve. Invertebrates were placed in clean glass jars filled with water from the respective site for 12–24 hours to allow their guts to purge. Each sample was then sorted and rinsed in deionized water, blotted, and placed in 60-mL HDPE jars until a composite-blotted wet-weight of approximately 0.8 g was achieved. Samples were then frozen at –20 degrees Celsius (°C) before analytical chemistry.

Mosquitofish

We collected composite samples of mosquitofish at all fixed sampling sites where present in sufficient biomass and across all sampling periods (except IID Marsh; table 1). Selenium concentrations in mosquitofish likely represented a pathway of direct selenium exposure to rails given the prevalence of fish in rail diets (McKernan and others, 2016; Eddleman and Conway, 202017). Samples were collected at each point by rapidly sweeping a D-ring net through the water column or visually stalking and netting fish. An average of 12.8 (standard deviation [SD]=5.6) individual fish were collected at each site and sampling period and placed in 500-mL glass jars filled with site water. We then rinsed each fish with deionized water and blotted dry before obtaining individual whole-body mass (to the nearest 0.1 g) and standard length (nose to peduncle, to the nearest 0.1 mm). Individuals were then pooled into a composite sample for each point and sampling period. Mass and standard lengths of fish comprising composite samples averaged 0.19 g (SD=0.26) and 19.1 mm (SD=5.5), respectively. Although variation in selenium concentration and overall body burden related to variation in body size is best controlled through analyses of individual fish (Ackerman and Eagles-Smith, 2010), limited funding for analytical chemistry precluded selenium analysis of individual fish.

Crayfish

We collected crayfish representing three different size classes at fixed sampling sites where present in sufficient biomass and across all sampling periods (except for the IID Marsh; table 1). Unlike other sample matrices, we did not collect crayfish samples for all points and time periods because our objective was to focus on variation among individuals of different weight classes, and crayfish availability and capture success was highly variable. At each point, we set one to three minnow traps baited with chicken legs accessible for consumption. Traps were left out overnight and checked daily. Crayfish were processed and analyzed for selenium primarily on an individual basis within weight classes to evaluate selenium risk to rails that forage on different sizes of crayfish, as opposed to compositing different size classes that could confound measured differences in selenium concentrations among sites (Ackerman and Eagles-Smith, 2010). We rinsed each crayfish with deionized water and blotted dry. Each crayfish was then weighed to the nearest 0.01 g and length measured from rostrum to peduncle to the nearest 0.1 cm. The size classes (based on cut-points in slopes from weight to length relationships) were as follows: small=0.1–5 g, medium=5–15 g, and large>15 g. Crayfish comprising the small size class at a sampling point were composited if their individual mass was too light to allow for selenium analysis. Crayfish were analyzed for selenium on a whole-body basis.

Rail Feather and Blood

We collected samples of rail feathers and whole blood opportunistically from adult individuals captured as part of the rail telemetry part of this study and feathers from adult and juvenile individuals as part of a related rail movement study (Harrity and Conway, 2020). Samples were heavily skewed toward managed marshes because only three birds were captured from unmanaged marshes (all at Morton Bay). Managed marshes included the Hazard and IID Marshes and the Imperial Wildlife Area Wister Unit managed by the California Department of Fish and Wildlife. The Wister Unit borders the northern extent of the FW Marsh. Samples from Hazard and IID were collected from mid-March 2016 through the end of April 2016, whereas samples from Wister were collected during June 2016. Approximately five head feathers, three to five breast feathers, or both head and breast feathers were collected per individual. Feathers were cleaned of any skin or debris and then stored dry in a coin envelope. Collection of blood was dependent on bird condition during handling; blood was only taken if the bird stayed alert during processing. In those cases, blood was subsequently drawn from the femoral artery with heparinized needles and hematocrit tubes and then immediately frozen in chemically cleaned plastic vials.

Secretive Marsh-Bird Surveys

Call counts were done for rails by using protocols based on the North American Marsh Bird Monitoring Protocol (NAMBMP; Conway, 2011) along 14 current or historically occupied marshes along the southeastern edge of the Salton Sea and southern Imperial Valley. This survey method uses a series of listening stations (hereafter, transect or marsh) located at least 200 m apart and along the perimeter of accessible rail habitats. Survey period for all stations extended through the crepuscular periods and included no more than a half hour of twilight periods (before sunrise or after sunset) or an hour and a half of daylit periods (after sunrise or before sunset). Each transect was surveyed three times between March 8 and April 17, 2016, with alternating a.m. and p.m. survey periods. Two transects were surveyed a fourth time, which also was included in the analysis. Each station was surveyed for 10 minutes and included 5 minutes of passive call detection followed immediately by playback of a sequence of recorded secretive marsh bird calls. Each 1-minute-long sequence consisted of three calls played 15 seconds (second=0, 15, and 30) apart, followed by 30 seconds of silence. The species selected for call playback can be tailored to the local bird community, but we elected to use the same sequence used for NAMBMP during operational survey development in San Francisco Bay because we were only interested in providing detection information for Yuma Ridgway’s rail. The species present in our recordings were, in order of playback, black rail, Yuma Ridgway’s rail, sora, Virginia rail, and American bittern. Upon detection of a call, observers recorded the approximate (±20 m) location on a map of recent satellite imagery and distance and direction calculated to the call, along with the time of detection. Subsequent detections of the same individual were determined by visual identification when possible or “dead-reckoning” of multiple calls from the same location. Subsequent calls were recorded within survey and when the same individual was heard or seen at subsequently visited stations in a transect, but only the initial detection was analyzed in occupancy and abundance modeling.

Rail Population Modeling

Rail call count data were aggregated to represent the total number of Yuma Ridgway’s rail detections per survey session per station and analyzed using an imperfect-detection abundance model with repeated measures (Royle, 2004). This model allowed us to estimate relationships between detection of rail vocalization and temporal and environmental variables. An important clarification relevant to territorial secretive marshbirds is that our detection probabilities are not for individuals but rather occurrence and detection of a vocalization by a rail. Survey site occupancy is static between survey rounds and not necessarily by specific individuals but by a specific number of individuals. Given the despotic nature of territorial rails before breeding, we feel this assumption is valid and that any population turn-over between survey rounds represents the exchange of individuals and not the addition or removal of individuals. Additionally, we assume that calling rates are equal across individuals after accounting for environmental or seasonal factors. The resulting detection probabilities were used to estimate approximate abundance and relationships of environmental covariates, such as habitat composition and spatial arrangement, with abundance. Four factors, including some known to influence detection probability of Ridgway’s rail in San Francisco Bay (Liu and others, 2012), were assessed: period of survey (a.m. versus p.m.), Julian date, time of survey relative to dawn and dusk, and temperature. Temporal covariates were determined based on Global Positioning System (GPS) location of survey station and local sunrise and sunset timing. Temperature was recorded hourly at the closest METeorological Aerodrome Report (METAR) station (Station: KIPL, Imperial, CA) to all survey locations. Models assessing factors that affect abundance of calling Yuma Ridgway’s rail included site-level habitat characteristics derived from aerial or satellite orthoimagery.

Two methods were used to classify habitat type into two sets of habitat metrics used in subsequent analysis of rail abundance. For the first set of metrics, proportion of “Potential” habitat was delineated from orthophotos based on current and past surveyed habitats and consisted of all contiguous non-agricultural vegetation and wet areas. For the second set, potential habitats were further analyzed by performing a maximum likelihood classification of contemporaneous USGS Landsat imagery to separate vegetated from barren and open-water landcovers, resulting in a vegetated “Classified” habitat. The proportions of potential and classified habitats were measured at three spatial scales from each station: 300, 1,000, and 5,000 m. Large contiguous habitats are suspected to have greater abundance of rails in San Francisco Bay. Although use of one candidate variable, perimeter:area ratio, would have allowed investigation of the relationship between size of habitat patches and rail abundance, we excluded this variable from our analysis because it was apparent that new levees and berms had been constructed between the time of available orthoimagery and data collection.

Habitats were classified based on management type and broad composition classes. Managed marshes involved some degree of water management and anthropogenic impoundment, usually with active vegetation management (for example, mowing/burning). Unmanaged marshes contained suitable vegetation, but any impoundments were naturally formed and included no vegetation or direct water management. Habitat type was considered “dominant” if at least 80 percent of the available habitat within 300 m of a survey station was of a single management class. This classification scheme resulted in the habitat classes: “managed marsh dominant,” “unmanaged marsh dominant,” “managed: unmanaged marsh edge,” “managed: non-habitat edge,” and “unmanaged: non-habitat edge.”

Quadratic relationships were included in candidate models that estimated the effect of classified habitat type on abundance and for temperature, date, and time of day effects on detection probability. Model uncertainty was assessed using the Akaike information criterion (AIC) with ranking of competing models within the suite of candidates identified by a delta-AIC (increase in AIC scores relative to the best performing model). Models with delta-AIC scores <2 are considered nearly equivalently effective at explaining dependent variables; models with delta-AIC scores <6 are presented. We performed a priori analysis comparing model performance of the similar two groups of habitat variables “potential” versus “classified” habitat. Models incorporating the potential habitat metrics outperformed equivalent models with classified habitat metrics (in other words, at same 300-, 1,000-, and 5,000-m scale) by 3 to 6 AIC units. Therefore, we omitted classified habitat metrics from our final suite of candidate models to reduce overall model selection uncertainty. Dependent variables from the best performing model according to AIC model selection are presented as back-transformed values of variables analyzed on the logit (detection probability) or log scale (abundance). In total, we evaluated 13 formulations of detection probability and 25 formulations of rail abundance, resulting in 325 candidate models plus a null model to evaluate relative model fit.

The development of abundance estimates tied to environmental variables that influence estimates allow for extrapolation to regional estimates encompassing areas where surveys were not performed. Relationships between predictor variables and rail abundance from the best performing imperfect detection model were used to extrapolate regional estimates of rail abundance. Confounding the relationship between calls heard and abundance is that the operational NAMBMP resulted in non-independence among survey stations. Survey stations in the NAMBMP are placed 200 m apart, but rail detection extends well beyond 200 m. This non-independence between stations means that extrapolating rail abundance using a census with the same protocol as NAMBMP (200-m spacing) would effectively overrepresent parts of the landscape where adjacent survey stations are. Therefore, we provide an estimate of the total number of birds expected to be counted from a survey effort consisting of independently placed survey stations (400 m apart) that saturate available habitats in the study area. Statistical modelling was performed using the ‘pcount’ function in R package “unmarked” (https://cran.r-project.org/web/packages/unmarked/index.html; Fiske and Chandler, 2011). Habitat classification using maximum likelihood supervised classification and extrapolation of abundance estimate were performed using ArcGIS—ArcMap software (Environmental Systems Research Institute, 2011).

Rail Ultra High Frequency Telemetry

There were 16 rails captured by hand, with a long-handled net, mist net, or with the use of modified drop-door traps (Zembal and Massey, 1983; Conway and others, 1993; Albertson, 1995; Bui and others, 2015). All birds were measured, banded, radio-marked, and released in situ with minimal handling time. Capture occurred in early morning or evening to minimize heat stress. Transmitters were Ecotone© Sterna model GPS that transmitted data through ultra high frequency (UHF) radio frequencies to small portable base stations. Transmitters could be set to record locations from 1-minute to 4-hour intervals and would suspend data collection when battery levels dropped below a threshold and would be restarted upon solar recharging by top-mounted solar cells. Transmitters were attached using backpack harnesses previously used on Ridgway’s rails (Albertson, 1995; Overton and others, 2014) and all marking occurred between April 3 and April 27. Whole blood was collected (up to 1 mL) using brachial, medial metatarsal, or jugular venipuncture on 14 individuals using 24- to 28-gauge heparinized needles. To establish a tissue comparison, we collected two to three breast and head feathers collected at the time of capture to assess selenium concentrations.

All transmitters were initially deployed with a 15-minute GPS collection interval that could be maintained under full, direct sunlight (for example, while testing transmitters) but was too taxing for continuous data collection when deployed on a rail due to shading by vegetation. Transmitters that could be reconnected with a base station were adjusted to a 4-hour cycle for the remainder of the study. However, many transmitters could not reconnect and did not sufficiently recharge to enable continued tracking following the initial period of deployment. As a result, we were not able to determine the fate of most individuals due to battery failures. Geographic (World Geodetic System 1984) location data were projected to North American Datum of 1983 Universal Transverse Mercator zone 11 before analysis to facilitate estimation of home range and movement patterns in metric units.

Location data were analyzed across each individual to determine patterns of space use as indicated by spatial distribution. Point pattern analyses were used to quantify the amount of space typically used by an individual, which has relevance to territorial patterns, potential reproductive patterns (nesting), and potentially for habitat quality. Available data were analyzed across all individuals to determine the duration at which sufficient relocations had been obtained to calculate stable estimates of home range size. Once a suitable temporal range was identified, estimates of home range size across this duration were calculated with a daily incrementing start date. The time necessary for home range size to equilibrate to within 20 percent of “final” home range size for an individual was used to gauge the minimum sample size needed to develop robust home range estimates. Smaller (approximately 1.5 ha) home ranges tended to take longer to reach stability than median (approximately 2.5 ha) or larger (>4 ha) home ranges, but 75 percent of individuals reached stable home range size estimates around 5 days. Therefore, we decided to calculate home range size in weekly increments through the available period of data. Given substantial heterogeneity of vegetation and landscape features in the study area that could influence the shape and sizes of rail home ranges, weekly home range size was calculated using a kernel utilization distribution and local convex hull methods. Both kernel and local convex hull are accepted methods for analyzing space use of wild animal populations. Although kernel methods have been suggested to perform better than local convex hull generally, for species that exhibit sharp boundaries in availability of habitat or demonstrate spatial gaps due to territorial aggression among neighbors, local convex hull may more effectively limit unused area in home ranges that otherwise result in elevated bias in home range size (Lichti and Swihart, 2011). Kernels were based on least square cross-validation to estimate a smoothing parameter/bandwidth (Worton, 1989). Local convex hulls were based on the nearest neighbor method (LoCoH-k) to develop hulls across eight neighbors (Getz and Wilmers, 2004). Utilization distributions were calculated using the 95 percent-isopleth for each method, with all analysis employing the R package “adehabitatHR” (Calenge, 2006), and movement patterns were calculated using the ‘dist’ function within the core R stats package (R Development Core Team, 2017).

Unfiltered Surface Water

Selenium was detected in all but one water sample. Mean concentrations (plus or minus 2 SE) are reported in table 2, and marsh sampling distributions of selenium concentration are illustrated in figure 3. Mean concentrations ranged from a low of 0.4 μg/L in managed IID Marshes during April to 2.0 μg/L in the unmanaged FW Marsh during June. Distributions of selenium concentrations of unmanaged marshes largely overlapped each other, but unmanaged marsh concentrations were higher than those from managed marshes (fig. 3). At unmanaged marshes, concentrations of samples varied substantially within marsh location and season combinations. Median concentration differed relatively little among seasons at FW Marsh, whereas the concentration median and variability of among samples declined from February to June at Morton Bay (fig. 3). At managed marshes, distributions of selenium from Hazard Marshes were distinct and higher than those at IID that most closely resembled background concentrations, and seasonal trends were not apparent at either marsh (fig. 3). Average concentrations from HZ9A were higher across all seasons (particularly in June) in comparison to concentrations averaged across all Hazard Marshes (table 2).

A suggested toxicity threshold for total recoverable selenium in water is 2.0 μg/L, whereas concentrations greater than 1–2 μg/L are considered elevated above background concentrations (U.S. Department of the Interior, 1998; Hamilton, 2004). Across the study, concentrations of selenium in 29 and 33 percent of water samples from unmanaged marshes at FW Marsh and Morton Bay, in comparison to 3 percent and 0 percent of water samples from managed marshes at Hazard and IID exceeded 2.0 μg/L, respectively (fig. 4). If we assume that our measures of total recoverable selenium reflect a minimum concentration of 1.5 μg/L total dissolved selenium over a 30-day period and more than one sample collected at a fixed point over a 3-year period exceeds that concentration according to the new U.S. Environmental Protection Agency (2016) criteria for lentic systems, then chronic risk to waterborne selenium exposure is roughly 30 percent higher for aquatic life in unmanaged compared to managed marshes. Selenium concentrations consistently exceeded 2.0 μg/L across all sampling periods at FW-3 and during two (February and April) of three sampling periods at MB-1, MB-7, and MB-10 (all points near or at termini of drainage canals; see fig. 2, sampling locations; fig. 4, selenium concentrations). Finally, only two samples across the study (FW-4 during April, FW-5 during June) exceeded a previously used threshold of 5.0 μg/L for protection of aquatic life (U.S. Environmental Protection Agency, 1987).

Chironomidae (Midge larvae) and Corixidae (Waterboatman)

Selenium was detected in all Chironomidae and Corixidae samples. Mean concentrations (plus or minus 2 SE) are reported in table 2 and sampling distributions are illustrated in figure 5. For Chironomidae, mean concentrations ranged from a low of 1.45 μg/g (dry weight) in managed IID Marshes during June to >6.0 μg/g in unmanaged FW Marsh and the managed HZ9A Marsh in Hazard during February (table 2). For Corixidae, mean concentrations ranged from a low of 1.10 μg/g in managed IID Marshes during June to 3.13 μg/g in managed Marsh HZ9A during April (table 2). Distributions across all sampling periods and marsh complexes indicated higher and largely non-overlapping concentrations of selenium in Chironomidae compared to Corixidae (fig. 5). For Chironomidae, variation in concentrations steadily tracked sampling periods, whereby concentrations at unmanaged FW Marsh and managed Hazard and IID Marshes decreased steadily across time, with relatively higher concentrations during February and April compared to much lower concentrations during June. In contrast, concentrations in Chironomidae from unmanaged Morton Bay followed an opposite seasonal pattern that increased from February to June. For Corixidae, seasonal trends were only evident at Morton Bay that followed a similar February to June increase in concentrations as seen in Chironomids from the same marsh (fig. 5). At managed marshes, distributions of selenium in both taxa from Hazard were distinct from, and higher than, those at IID. Also, distributions of selenium in both taxa from Hazard more closely resembled those at both unmanaged marshes.

Suggested dietary toxicity thresholds for selenium in invertebrate avian prey range between 3.0 and 4.0 μg/g dry weight (Hamilton, 2004). For Chironomidae across the study, concentrations of selenium in 73 percent and 85 percent of samples from unmanaged marshes at FW Marsh and Morton Bay and 52 percent and 0 percent of samples from managed marshes at Hazard and IID, respectively exceeded a more protective 3.0 μg/g threshold (see fig. 2, sample locations; fig. 6, selenium concentrations). The highest concentration of selenium across all samples was detected in Chironomidae at HZ6-1 (19.9 μg/g) during February. However, abundance of Chironomidae was too low to effectively sample HZ6-1, a marsh inlet, during April and June. Chironomidae samples from FW-7 and MB-3 harbored selenium concentrations that consistently exceeded 6.0 μg/g across all three sampling periods, whereas samples from MB-7 and MB-5 exceeded 6.0 μg/g across two (April and June) of three sampling periods (see fig. 2, sample locations; fig. 6, selenium concentrations). At the managed Hazard Marsh complex, percentages of Chironomidae samples from HZ6, HZ3, and HZ9 with selenium concentrations exceeding 3.0 μg/g decreased markedly from 71 percent during February and April to 0 percent during June (fig. 6). In contrast, selenium concentrations in 33 percent and 50 percent of Chironomidae samples across the study from HZ10 exceeded 6.0 μg/g and 3.0 μg/g, respectively. For Corixidae, selenium concentrations exceeded 3.0 μg/g in 13 percent and 11 percent of samples from unmanaged FW Marsh and Morton Bay and 16 percent and 0 percent of samples from managed Hazard and IID Marshes, respectively (fig. 7). No sampled concentrations of selenium in Corixidae exceeded 4.0 μg/g. “Hot spots” with Corixidae concentrations exceeding 3.0 μg/g among unmanaged marshes included FW-7 during April and June, MB-7 during April, and MB-3 and MB-8 during June. Among managed marshes, similar hot spots occurred at all sampling points in HZ9A and HZ10-1 during April and at HZ10-2 during February. These overall patterns of higher selenium concentrations in Chironomidae compared to Corixidae, as well as relatively high percentages of Chironomidae samples from both unmanaged marshes and from Hazard that exceeded dietary thresholds are similar to patterns reported by Miles and others (2009) and De La Cruz and others (2022) in the same marsh complexes from 2006 through 2010.

Mosquitofish (Gambusia spp.)

Selenium was detected in all mosquitofish (Gambusia spp.) samples. Mean concentrations (±2 SE) are reported in table 2 and sampling distributions are illustrated in figure 8. Mean concentrations ranged from a low of 2.51 μg/g in managed IID Marshes during June to a high of 5.42 μg/g in Hazard Marsh HZ9A also during June (table 2). Distributions of selenium concentrations across all marsh complexes showed relatively low variation among sampling periods. Among marshes, distributions indicated generally higher selenium concentrations in mosquitofish from unmanaged marshes and the managed Hazard Marshes (where all median concentrations exceeded 3.0 μg/g) compared to the IID Marshes (fig. 8).

To assess relative dietary risk to rails from consuming mosquitofish, we used the upper suggested dietary threshold of 4.0 μg/g (dry weight) for taxa consuming invertebrates in aquatic food webs (U.S. Department of the Interior, 1998; Hamilton, 2004) as a surrogate for consuming fish. Across the study, concentrations of selenium in 67 percent and 56 percent of samples from unmanaged marshes at FW Marsh and Morton Bay and 43 percent and 0 percent of samples from managed marshes at Hazard and IID, respectively exceeded 4.0 μg/g (fig. 9). No sampled concentrations in mosquitofish exceeded the more recent chronic value of 8.5 μg/g (dry weight) for selenium in whole fish samples (U.S. Environmental Protection Agency, 2016). However, during June, mosquitofish from FW-7 harbored 7.9 μg/g of selenium, which approached the 8.5 μg/g chronic value. Among unmanaged marshes, relative hot spots where selenium exceeded 4.0 μg/g across all sampling periods included FW-3, FW-5, FW-7, MB-3, MB-4, MB-7, and MB-8. Among managed marshes, HZ10-1, HZ10-2, and all points within HZ9A exceeded 4.0 μg/g across all sampling periods (see fig. 2, sample locations; fig. 9, selenium concentrations).

Crayfish

Selenium was detected in all crayfish samples. Mean concentrations (±2 SE) are reported in table 2, and sampling distributions for selenium concentrations and body burden are illustrated in figures 10 and 11, respectively. Mean concentrations of selenium in all size classes of crayfish were generally lowest (1.02–2.13 μg/g) for managed IID Marshes across sampling periods, although concentrations also were relatively low for medium-sized crayfish in Hazard Marshes across periods (1.96–2.13 μg/g). Concentrations in small and large crayfish were highest (small=4.84 and large=2.51 μg/g) during June in the unmanaged Morton Bay Marshes; concentrations in medium-sized crayfish were highest (3.47 μg/g) in FW Marsh during April. No crayfish of any size-class were captured from HZ9A during any sampling event (table 2), and sample sizes varied substantially across marshes and sampling periods (table 1).

Distributions of selenium concentrations across all marsh complexes showed high variation among sampling periods (fig. 10). Among small crayfish, distributions from FW and Hazard Marshes indicated relatively higher selenium concentrations during February and June and lower concentrations during April. Median concentrations of both unmanaged marsh complexes exceeded 3.0 μg/g across all periods, whereas in managed marshes, concentrations did not exceed 3.0 μg/g in any period. For medium-sized crayfish, distributions from FW Marsh indicated relatively higher selenium concentrations during February and lower concentrations during April and June, whereas distributions from Morton Bay Marshes followed an opposite seasonal pattern. No strong seasonal pattern was evident for distributions at any managed marsh. Among large crayfish, steadily decreasing distributions over sampling periods from Hazard Marshes was the only seasonal pattern evident. Among marshes, distributions indicated generally higher selenium concentrations in crayfish from all size-classes and sampling periods from unmanaged compared to managed marshes. Among size-classes, concentrations were generally higher in small crayfish and lower in large crayfish (fig. 10).

When selenium concentrations are converted to body burdens of total selenium to account for size-related variation in dietary exposure to higher trophic levels, distributions were highly variable among marshes and sampling periods (fig. 11). However, distributions of body burden selenium for large crayfish were generally at least 10 times higher compared to those from small crayfish. As within concentrations, distributions of body burden selenium were higher at unmanaged than managed marshes across most sampling periods and size class, although burdens were notably low for small size-class crayfish from FW Marsh during April. Across sampling periods and marshes, selenium body burdens followed patterns opposite of those observed for concentration, where maximum burdens for small crayfish approached 4 micrograms (μg) of selenium, medium crayfish approached 15 μg, and large crayfish approached 30 μg.

As with mosquitofish, we used the upper suggested dietary threshold of 4.0 μg /g (dw) for taxa consuming invertebrates in aquatic food webs (U.S. Department of the Interior, 1998; Hamilton, 2004) as a surrogate for consuming fish. Patterns indicated higher risk on a concentration basis for small versus large crayfish. Among small crayfish across the study, concentrations of selenium in 38 percent and 67 percent of samples from unmanaged marshes at FW Marsh and Morton Bay compared to 17 percent and 0 percent of samples from managed marshes at Hazard and IID, respectively exceeded 4.0 μg/g (fig. 12). Among medium-sized crayfish across the study, concentrations of selenium in 14 percent and 9 percent of samples from unmanaged marshes at FW Marsh and Morton Bay and 11 percent and 0 percent of samples from managed marshes at Hazard and IID, respectively exceeded 4.0 μg/g (fig. 13). Among large size-class crayfish across the study, concentrations of selenium in 0 percent and 7 percent of samples from unmanaged marshes at FW Marsh and Morton Bay compared to 4 percent and 0 percent of samples from managed marshes at Hazard and IID, respectively exceeded 4.0 μg/g (fig. 14). Sampling at fixed points was too variable to assess hot spots exceeding 4.0 μg/g across sampling periods and size-classes.

Rail Blood and Feathers

Selenium was detected in all rail tissue matrices. Mean concentrations (±2 SE) and sample sizes from opportunistic collections from rails captured as part of the movement component of the study are shown in table 3, and sampling distributions for selenium concentrations in feathers and blood are illustrated on figures 15 and 16, respectively. Additionally, spatially explicit selenium concentrations for feathers and whole blood are illustrated in figure 17. All blood samples (n=12) were collected in April. Feather samples were collected in April for rails at Hazard and IID Marshes (except for one in June); April (n =3) and June (n=2) for rails at Morton Bay; and June for rails at Wister. No rails were captured and sampled from HZ9A.

Low sample size precluded strong inference to populations, but general trends were apparent. Distributions of selenium concentrations in head and body feathers were generally higher in rails from Morton Bay compared to those from all managed marshes (fig. 15). Head feather selenium concentrations were generally higher than concentration in body feathers among marshes (fig. 15) and head and body feather concentrations were strongly correlated with each other among paired samples (r² 0.92, n=14). Among limited blood samples, the lone sample from Morton Bay had a higher selenium concentration (1.51 μg/g, ww) than samples from managed marshes, although, one sample from Hazard Marshes had similar concentration (1.37 μg/g, ww; fig. 16).

Avian species display a wide range of sensitivity to selenium (Burger and others, 2015), and tolerance to selenium generally increases with salt-tolerance (Hamilton, 2004; Presser and Luoma, 2010). U.S. Department of the Interior (1998, p. 167) suggested feather and blood selenium concentrations below 4.0 μg/g (dw) and 0.4 μg/g (ww), respectively, reflect background concentrations, and McKernan and others (2016) suggested a concentration of 5.0 μg/g (dw) in rails that could warrant concern. Among unmanaged marshes, 100 percent of head feathers and 40 percent of body feathers sampled exceeded 5.0 μg/g (fig. 15). Among managed marshes, 50 percent and 0 percent of head and body feather samples, respectively, from Hazard Marsh, and 0 percent of all head and body feathers from IID, exceeded 5.0 μg/g. Only one (8 percent) of the body feather samples from Wister exceeded 5.0 μg/g. All blood samples, except one at IID, exceeded background concentrations of 0.4 μg/g (fig. 16).

Time Series of Selenium in Unmanaged Marshes with Lowering Salton Sea Elevation

Distributions of total recoverable selenium in unfiltered water, collected from five fixed-sampling points (FW-2, FW-3, MB-1, MB-2, MB-3), indicated an overall drop in concentrations from 2006 to 2010 when the Salton Sea was at higher elevation and the FW Marsh was less vegetated, compared to conditions in 2016 (fig. 18). In particular, during 2016, distributions from FW-2, MB-2, and MB-3 were lower and completely non-overlapping from those in 2006–10. Median selenium concentrations for 4 out of 5 points were at or below the 2.0 μg/L threshold during 2016. Selenium concentrations in Chironomidae also appeared to drop over time in some fixed sampling points (fig. 19). During 2016, distributions of selenium from FW-2 and MB-1 were lower and did not overlap those from 2008 to 2010, whereas distributions from FW-3, MB-2, and MB-3 overlapped among time periods. Greater proportions of pre-2016 selenium concentrations in Chironomidae exceeded dietary thresholds of 3.0–4.0 μg/g compared to those from 2016. We note that conditions were too saline in Morton Bay to support Chironomidae during 2006–08. Distributions of selenium in Corixidae among time periods were more variable, and non-overlapping declines during 2016 were most evident at FW-3, MB-1, and MB-2 (fig. 20). Like Chironomidae, selenium concentrations for Corixidae before 2016 (especially 2008–10) exceeded 3.0–4.0 μg/g in higher proportion than during 2016 (fig. 20). These overall patterns among water and invertebrate selenium distributions correlated with the state-transition that occurred following 2010, when Morton Bay became isolated completely from the Salton Sea and was colonized by cattails and the FW Marsh became more densely vegetated. However, this pattern may be confounded, to some extent, by differences in selenium analytical chemistry, and results should not be broadly extrapolated.

Rail Population Modeling

Among the 14 transects surveyed, call count surveys for rails resulted in 269 individual rail detections (85 stations; table 4), which included a total of 245 individual surveys across stations. Zero rails were detected at 3 (21 percent) transects, the greatest total number of detections along a transect was 48, and the greatest number of detections in a single round of surveys was 25 calls. The greatest number of calls detected at a station was six calls (in a single survey).

On average, across all observations, estimated call detection probability at a survey station was 19.6 percent and mean number of rail calls detectable around each station was 5.1 calls among stations where detection was confirmed (approximately 1 call per station among all stations). Our candidate suite of models was effective at identifying factors associated with variation in rail call detection probability and estimated rail abundance. Our top model performed 43.8 delta-AIC units better than the null model; that did not account for variation in detection probability or rail abundance (table 5). The formulation of the detection probability was the primary source of model uncertainty of the imperfect detection modelling. Among the 326 models, the top 7 differed only in the factors identified as being most closely associated with difference in detection probability (table 5). For each of these models, which included all models up to a delta-AIC of 4, abundance of rails was estimated to vary non-linearly with the quantity of habitat adjacent to the survey station (within 300 m; fig. 21) and increased as the total habitat within 5 kilometers (km) of the survey station increased (fig. 22). These models suggest that Yuma Ridgway’s rail abundance, like Ridgway’s rail abundance in San Francisco Bay (Liu and others, 2012), is influenced by landscape level habitat availability and that smaller and isolated habitats have fewer rails. Detection probability was nearly identically described (delta-AIC <2) in four models that included temperature at the time of the survey (table 5). Thus, model selection indicated that although temperature and time of day are strongly correlated variables, call occurrence was more precisely estimated using temperature at the time of the survey than the time of the survey more generally. The top model included only a temperature effect on detection probability and estimated just over a 1.8 percent decrease in detection probability for every degree increase in temperature during the survey. The three other competing models each included temperature along with additional influence of either specific habitat type at the survey station (delta-AIC=1.15), curvilinear (delta-AIC=1.4), or linear (delta-AIC=1.999) relationship with date of the survey. The relationship between detection probability and habitat was not very strong. Detection probabilities were highest at stations near the edges of managed or unmanaged marshes (about 50 percent of the land around a station was either habitat type), whereas detection probability was significantly lower where managed and unmanaged marshland met (100 percent habitat cover). Detection probabilities within marsh fragments (<25-percent habitat around a point) or at contiguous habitats, either managed or unmanaged, were not significantly different than any other habitat class.

Classification of remotely sensed imagery using unsupervised maximum likelihood classification techniques did not result in a stronger relationship between estimated rail abundance and habitat metrics. The top 21 models (delta-AIC <10.2) included only potential habitat as identified from past orthophoto interpretation, which could include large areas of open water and barren ground that do not provide high quality rail habitat on their own. Lack of support for the predicted relationship between call abundance and classified habitat, based on more refined maps, could be due to the relatively narrow range of classified habitats at our observation points but is likely influenced by the observed greater abundance (and potentially detection probability) along habitat edges. As a result, models including potential habitat estimated lower abundance per hectare of habitat than models including classified habitat but had a more consistent relationship across all observed data. In addition, the top performing classified habitat maps tended not to include the amount of classified habitat at the 300-m scale, and this could result from imprecision in estimates due to the relatively large, 30-m pixel size present in Landsat data.

We produced population size estimates by extrapolating the habitat relationships of the best performing imperfect detection model to characteristics of the landscape with survey stations situated to perform a complete census of available habitat (fig. 23). Major limitations of this approach include the inability to assign fine-scale habitat relationships such as vegetation composition or plant cover to imperfect detection models. In addition, the model is only able to provide estimates of the number of uniquely calling individuals detected and it remains ambiguous how this value relates to the total number of individuals within the population. Despite its limitations, our approach allowed a complete survey without double-counting rails. Such a complete survey of the southern Salton Sea and Imperial Valley result in an expected detection (that is, accounting for imperfect detection) of 1,752 rails. This estimate for rail population size in 2016 could be low compared to most years because other research indicates counts of Yuma Ridgway’s rail populations at the Salton Sea likely were lowest in 2016 for the 13-year period of 2006–18 (fig. s11 in Harrity and others, 2020).

Although we did not evaluate relationships of finer-scale vegetation cover characteristics as indicated earlier, landscape characteristics used to define potential habitat (fig. 23) and supported by model selection (table 5) are heuristically related to other remotely sensed vegetation characteristics important in other models predicting relative rail abundance (Harrity and others, 2020). Similarities and differences between the two models could help determine how the models may be complementary for identifying suitable rail habitat in the future. Also, estimation of rail population sizes based on respective models would allow comparison and replication of their predictions based on differing assumptions and methods. In turn, managers might better be able to assess risk to the Yuma Ridgway’s rail in the Salton Sea by using a range of predicted population estimates.

We are careful to ascribe the earlier estimate (1,752 rails) to the “number of detected rails” because survey methodology relies on a bird to vocalize in order to be “present” for the survey. The call frequency of Yuma Ridgway’s rail is unknown and likely varies according to breeding stage, age, and sex. If rails call frequently and most of the population is “available” to be heard within a survey period, then our extrapolation is likely close to a true population estimate. However, if rails call infrequently and only a small fraction of the population is detected during a survey, then our abundance estimates for “calling” rails could be much lower than actual population size. At the time of this study, there is no clear mechanism available to relate detected calls, even with extrapolation to account for detection probability, to the number of individuals in the population. Estimates of calling rate or means to assign individual identity to calls heard are two approaches that could provide insight to this relationship. For example, if the average time between calls for an individual rail is substantially less than the duration of a survey at a station then our estimated number of detected calls should be much closer to the true population size than if a rail only calls once per hour. Surveying captive birds with known population size or marking individuals with sound sensing transmitters are two approaches that can be leveraged to estimate a relationship between calls heard and actual population size.

Rail Intra- and Inter-Marsh Movements

Over 5,000 locations were obtained from 15 radio-marked Yuma Ridgway’s rail covering a period from early April through mid-November 2016 (table 6). A sixteenth individual was radio-marked but the transmitter failed and never provided any locations. Due to an error in programing the transmitter duty cycle, each transmitter was initially deployed to collect locations once every 15 minutes. This schedule was too taxing on the lithium polymer batteries and transmitters were remotely reprogrammed to collect data every 4 hours; however, several transmitters suffered premature battery failure due to this early setting (table 6). Only four birds disappeared while transmitters still retained enough charge for data transmission, indicating that they moved from the marshes and could not be relocated due to the limited transmission range of the UHF transmitters (a few hundred m; table 7). The transmitter for a fifth individual was recovered with a broken harness consistent with a depredation, but no carcass or feathers were found at the site, suggesting it could have broken due to wear or damage.

Weekly home range size, estimated using kernel utilization distributions, ranged from 0.27 to 29.67 hectares (ha), and the average weekly kernel utilization distribution was 3.33 ha (median: 2.5 ha; fig. 24). Home range size for Yuma rails was larger than the estimated size of home ranges for Ridgway’s rail in San Francisco Bay (Rohmer, 2010; U.S. Geological Survey, unpub. data, 2015). Due to abrupt edges in habitat patches in this landscape as well as territorial behavior between rails, local convex hull methods are likely to provide a more accurate measure of space use than kernels. The weekly LoCoH-k size ranged between 0.14 and 11.07 ha, and the mean weekly LoCoH-k utilization distribution was 1.48 ha (median=1.33 ha; fig. 25). Individual variation in mean weekly home range size was between 1.0 and 4.97 ha for kernel estimates and between 0.65 and 2.93 ha for LoCoH-k estimates. There was no relationship between duration of tracking and average home range size for either method, but the LoCoH method consistently resulted in space use estimates that were about half the size of those estimated using kernel methods. The largest space use estimates were from a single individual who relocated its territory while the wetland unit that it was using, and the surrounding units, were being drained for vegetation management purposes (indicated by black dot in fig. 25). This same individual showed periods of substantial increase in space use during the early fall as well as when it moved back into the previously used area following re-flooding of the wetland units.

Kernel estimates of space use showed substantial overlap among individuals total home range (95-percent contours; figs. 26–28). Likewise, core areas (50-percent isopleths) in the Imperial Irrigation District managed marshes showed almost complete overlap among three pairs of individuals, suggesting we had radio-marked mated pairs (fig. 29). The amount of core area overlap among adjacent mated pairs in Salton Sea was approximately the same as observed for pair-bonded Ridgway’s rails in San Francisco Bay. For adjacent unmated rails, by contrast, core area overlap was less in the Salton Sea than in San Francisco Bay (Overton, 2013). One explanation for regional differences in core area overlap is that the tidal environment of San Francisco Bay is much more dynamic on short time scales than within marsh environments of the Salton Sea and requires rails within the bay to move throughout the marsh in response to tidal inundation. Because tidal inundation does not affect rail movement in the Salton Sea, home territoriality and space use can be compressed, resulting in less overlap in core areas and greater population density than within tidal marshes. The area used by rails, in response to changes in habitat, also suggests that tidal marshes in San Francisco Bay that have the most habitat heterogeneity at very local (1 ha or less) scales should require the least amount of movement in response to tidal inundation, which in turn could facilitate greater population density. This pattern of movement related to habitat appears to exist within San Francisco Bay and is consistent with habitat association models developed for rails in that tidal environment (Liu and others, 2012; U.S. Geological Survey, unpub. data, 2015).

Distance between animal relocations can be a useful descriptor of behavioral choices and ecological processes affecting individuals. The maximum distance, or total displacement, between relocations of a single individual ranged between 192 m and 1.7 km, with a mean of 737 m. However, GPS location error is extremely non-normally distributed with nearly exponential error distances in extreme cases and it is impossible to distinguish between temporary extra-territorial movements and most GPS errors. A more robust measure of spatial spread is the distance range for 95 percent of all relocations for an individual. The 95-percent Maximum Displacement Distance (MDD) ranged from 141 to 305 m (mean=228 m), excluding a single outlier (MDD=677 m) with a relatively long tracking history and several territory shifts. Tracking duration ranged from 3 to 126 days (223 days for the outlying individual) and neither total displacement nor 95-percent MDD was affected by tracking duration. These findings indicate that extra-territorial movements are infrequent, but individuals can, and most likely do (despite the occurrence of GPS location error), make substantial movements that range across multiple adjacent territories.