Leveraging Delta Smelt Monitoring for Detecting Juvenile Chinook Salmon in the San Francisco Estuary

Monitoring is an essential component in ecosystem management, and leveraging existing data sources for multiple species of interest can be one effective way to enhance information for management agencies. Here, we analyzed juvenile Chinook Salmon (Oncorhynchus tshawytscha) bycatch data that has been collected by the recently established Enhanced Delta Smelt Monitoring program (EDSM), a survey designed to estimate the abundance and distribution of the San Francisco Estuary’s (estuary) endangered Delta Smelt (Hypomesus transpacificus). Two key aspects of the EDSM program distinguish it from other fish surveys in the estuary: a stratified random sampling design and the spatial scale of its sampling effort. We integrated the EDSM data set with other existing surveys in the estuary, and used an occupancy model to assess differences in the probability of detecting Delta Smelt across gear types. We saw no large-scale differences in size selectivity, and while detection probability varied among gear types, cumulative detection probability for EDSM was comparable to other surveys because of the program’s use of replicate tows. Based on our occupancy model and sampling effort in the estuary during spring of 2017 and 2018, we highlighted under-sampled regions that saw improvements in monitoring coverage from EDSM. Our analysis also revealed that each sampling method has its own benefits and constraints. Although the use of random sites with replicates, as conducted by EDSM, can provide more statistically robust abundance estimates relative to traditional methods, the use of fixed stations and simple methods such as beach seining may provide a more cost-effective way to monitor salmon occurrence in certain regions of the estuary. Leveraging the strengths of each survey’s method can enable stronger inferences on salmon abundance and distribution. Careful consideration of these trade-offs is crucial as the management agencies of the estuary continue to adapt and improve their monitoring programs.


INTRODUCTION
Estuaries are among the most important, heavily affected, and degraded ecosystems on Earth. The majority of the human population lives in coastal areas around estuaries, in part because estuaries are some of the most biologically productive areas in the world (Kennish 2002) and provide valuable ecosystem services (Lotze et al. 2006;Borja et al. 2010). The San Francisco Estuary (estuary) is the largest estuary on the West Coast of North America, and provides important habitat and migratory pathways for over 40 freshwater, estuarine, euryhaline, marine, and anadromous fish species (Moyle 2002). However, human modifications related to flood risk management, water supply to major urban and agricultural areas, as well as urbanization, have resulted in large-scale effects on the landscape, hydrology, and ecology of the estuary (Nichols et al. 1986;Moyle et al. 2010;Castillo et al. 2018). Today, the majority of the estuary's historical wetlands have been drained, tidal rivers have been channelized, and many reservoirs have been constructed on rivers that flow into the estuary. These system alterations have profoundly affected a number of endemic aquatic species and their habitats in the estuary (Stevens and Miller 1983;Nichols et al. 1986;Cloern et al. 2016).
Monitoring is crucial for understanding how species and ecosystems respond to anthropogenic effects and the subsequent management actions to mitigate them. The estuary is one of the most studied and monitored estuaries in the world. Our understanding of this highly complex estuarine ecosystem has been advanced over the years by multiple long-term monitoring programs (Brown and May 2006;Kimmerer et al. 2009; Thomson et al. 2010;Cloern et al. 2017), some of which span over 5 decades and have captured roughly a million fishes since their inception. These monitoring programs can be costly and time intensive. As such, natural resource agencies are often asked to allocate limited funds and maximize the value of each monitoring program (Joseph et al. 2009). One simple way to gain value in monitoring is to leverage data on non-target species to better understand ecosystem changes and inform management actions. A substantial portion of the monitoring efforts in the estuary were designed for a single species. For example, the California Department of Fish and Wildlife's (CDFW) 20-mm and Spring Kodiak Trawl (SKT) surveys target the endangered Delta Smelt (Hypomesus transpacificus) (Dege and Brown 2004;Polansky et al. 2018), while the Delta Juvenile Fish Monitoring Program (DJFMP)'s primary objective is to monitor juvenile Chinook Salmon (Oncorhynchus tshawytscha) rearing and migration through the estuary (IEP et al. 2019a). Despite the focus of these monitoring programs on single species, their data sets can still provide valuable insights on other members of the fish community (Brown and May 2006;Mahardja et al. 2017;Castillo et al. 2018).
One of the most recently established monitoring programs in the estuary is the Enhanced Delta Smelt Monitoring program (EDSM). The EDSM program is a spatially and temporally intensive sampling effort for the endangered and endemic Delta Smelt that was initiated late in 2016 to better assess the abundance and distribution for all life stages of this species (USFWS et al. 2019). Delta Smelt is a highly important species to the estuary as a result of its recent precipitous decline-and effect on California's water management (Moyle et al. 2018). The intensity and breadth of the EDSM program's sampling effort requires a large investment of resources, and the scientific and management value of this monitoring program can be increased by leveraging its bycatch data for other species of concern, such as Chinook Salmon.
The estuary supports fall-, late fall-, winter-, and spring-run Chinook Salmon, named after the timing of the adult upstream migration. Of the four runs of Chinook Salmon, two are listed under the federal Endangered Species Act (NMFS 2009;NMFS 2019): winter-run as endangered and spring-run as threatened. A recent review of the winter-run Chinook Salmon monitoring network in the estuary (Johnson et al. 2017) highlighted key information gaps that preclude accurate assessment of the status and trends of this endangered and endemic run of Chinook Salmon. However, the potential use of information from non-salmon-focused surveys was not considered in Johnson et al.'s (2017) review, likely because monitoring effort at the scale of EDSM did not exist at the time.
The EDSM data set could offer an opportunity to supplement existing monitoring data on Chinook Salmon because it differs fundamentally from most of the estuary's long-running fish surveys. EDSM uses a stratified random sampling design (Stevens and Olsen 2004), whereas other fish monitoring programs in the estuary sample at fixed stations. Additionally, EDSM collects replicate samples at each location to account for imperfect detection (i.e., false zero catch), a relatively uncommon procedure for monitoring programs in the estuary. Here, we aim to explore how the methods used by EDSM can be leveraged to improve inferences drawn by management agencies from the existing salmon monitoring network in the estuary. Our objectives were to: (1) compare the overall capability of EDSM to detect juvenile Chinook Salmon relative to other surveys currently used to monitor the species in the estuary, and (2) assess the value that EDSM adds to the estuary's salmon monitoring network. Note that our investigation is not meant to be a comprehensive analysis of Chinook Salmon capture probability, nor is it a re-evaluation of the overall Central Valley salmon monitoring network. This paper highlights new information that EDSM contributes to Chinook Salmon monitoring. Consequently, our geographic range is also largely limited to the tidal freshwater and brackish portion of the estuary that EDSM monitors.

Study System
The estuary's watershed spans about 40% of California, carrying runoff produced in the 163,000-km 2 area bounded by the Cascade and Sierra Nevada mountains (Cloern and Jassby 2012). It is of major socioeconomic importance as a cornerstone of the California water infrastructure, supplying water to a multi-billiondollar national and international agribusiness, and to approximately one-third of California's population (Lund et al. 2008;Lund 2016). The estuary is largely influenced by natural tidal cycles and flows from two main tributaries within the California's Central Valley: the Sacramento River to the north and the San Joaquin River to the south (Figure 1). The Sacramento and San Joaquin rivers converge to form the Sacramento-San Joaquin Delta (Delta). Once a mosaic of river channels, tidal wetlands, floodplains, and riparian forest, the Delta now consists mainly of islands reclaimed for agriculture, separated by a network of leveed channels (Whipple et al. 2012). The tidal freshwater Delta is generally considered the uppermost extent of the estuary (Figure 1). Freshwater exits the Delta, then enters the Suisun Bay region before flowing through Carquinez Strait into San Pablo Bay, and finally passing under the Golden Gate Bridge at the exit of the San Francisco Bay to meet the Pacific Ocean.
Habitat alteration is of great concern for Chinook Salmon, especially at this southern end of their natural range, where water diversions, predation, and temperature increase from climate change pose additional conservation challenges (Yoshiyama et al. 2000;Williams 2006;McLain and Castillo 2009). Historically, salmon populated the entire drainage area of the estuary's watershed (Whipple et al. 2012). Currently, impassable dams reduce available upstream habitat to approximately 5% of the historically available river mileage (Reynolds et al. 1993). Juvenile Chinook Salmon use the estuary for rearing and migration, and are thought to enter the estuary as early as October, with residence time ranging from 41 to 117 days (del Rosario et al. 2013).
Four different runs of Chinook Salmon inhabit California's Central Valley. However, properly identifying these distinct runs during the Chinook Salmon's juvenile life stage has been difficult. The CDFW developed length-at-date criteria in 1989 to assign juvenile Chinook Salmon into the different runs based on timing and size (Fisher 1992); however, the inaccuracy of run assignment for Chinook Salmon based on this length-at-date criteria has been recognized for many years (Hedgecock 2002;Harvey et al. 2014). Yet, length-at-date criteria remains the primary method for identifying spring-and winter-run Chinook Salmon for near-real-time management in the system because of the time and cost currently associated with genetic analysis. Chinook Salmon management in the estuary further requires distinguishing wild-origin fish from hatchery-origin fish, which can also be challenging. Millions of hatchery-reared Chinook Salmon are released each year in the Delta and upstream (Sturrock et al. 2019). Hatcheries contribute substantially to Chinook Salmon populations within the system (Barnett- Johnson et al. 2007;Huber and Carlson 2015;Willmes et al. 2018), but while a considerable number of hatchery fish can be readily identified by the presence of adipose fin clip and coded-wire tag, many are released unmarked. Given our inability to identify the different runs and natal origin of Chinook Salmon with high accuracy, we chose to analyze the species collectively rather than by run-timing or natal origin.

Enhanced Delta Smelt Monitoring Program
The EDSM program is a year-round weekly sampling program conducted by the US Fish and Wildlife Service (USFWS) that provides: (1) fine-scale temporal resolution of Delta Smelt abundance and distribution, (2) early warning of potential adult and juvenile Delta Smelt entrainment into the Delta's water pumps (Smith et al. 2019), and (3)  • Phase 2 samples post-larvae and small juveniles using larval tow nets from approximately April through June.
• Phase 3 samples juveniles and sub-adults using Kodiak trawls from approximately July through November.
The initiation and duration of phases can be dynamic, depending on contemporary environmental conditions, catches, and management needs, but remained constant throughout the first 2 years of EDSM implementation, and will likely stay static for years in the future. The Kodiak net is connected to the boats using a 2.3-m rope bridle attached to a 30.5-m tow rope, which is attached to the metal bar on each side of the net. Starting in 2018, all Kodiak tows were standardized to 10 minutes in length under normal conditions. (Before this, the duration of tows ranged between 2.5 and 10 minutes.) All fish ≥ 25 mm fork length (FL) are identified to species or run and then measured to the nearest 1 mm FL. If more than 50 individuals of a juvenile Chinook Salmon run are captured within a single haul, a random sub-sample of 50 individual fish is measured for FL, and the rest of the captured fish are counted but not measured.
The sampling region of EDSM is dynamic because it varies with Delta Smelt life stage and expected distribution. In general, the study area is defined as estuary waters that Delta Smelt occupy ( Figure 2). The study area is divided into spatially defined, temporally dynamic strata. During Phase 1 of 2016−2017, four spatial strata corresponded to perceived risk of Delta Smelt entrainment into the South Delta water export facilities (see USFWS et al. 2019, Figure A1). As the program evolved, strata were modified to better reflect geographic boundaries or historical Delta Smelt distribution. Within each stratum, sampling locations are selected each week using a generalized random-tessellation stratified (GRTS) design (Stevens and Olsen 2004). The GRTS sampling procedure yields random samples that are spatially well distributed across a stratum. Field crews sample 3 to 5 days per week for a total of 24 to 37 sites per week (2 to 15 sites per stratum). To account for false zeroes, at least two replicate tows are generally conducted at each site. From the beginning of the survey in December of 2016 through Phase 3 of 2017 (July 2017-November 2017), if no Delta Smelt were caught at a site after the second replicate tow, up to five total tows were completed at sites within strata of (presumed) high Delta Smelt density, and up to eight total tows were completed at sites within strata of (presumed) low Delta Smelt density ( Figure A1). In Phase 1 of 2017-2018, the maximum number of tows per site in low-density strata was reduced from eight to six. The EDSM program applies a "stopping rule" to the number of tows conducted at each site to reduce the sampling take's potential effect on the Delta Smelt population. Sampling with replicate tows at a site stops after at least one Delta Smelt is observed and at least two full tows are completed, unless 25 or more Delta Smelt are captured in the first tow. Generally, if 3 to 24 Delta Smelt are captured in the first tow, the duration of the second tow is reduced. If approximately 25 or more Delta Smelt are captured in the first tow, no replicate or additional tows are conducted.

Delta Juvenile Fish Monitoring Program
The USFWS DJFMP has used a combination of surface trawls and beach seines to evaluate the relative abundance and distribution of juvenile fishes in the estuary since 1976 (IEP et al. 2019a). Since 2000, three fixed trawl sites and 58 beach seine sites have been sampled weekly or every 2 weeks within the estuary and the lower Sacramento and San Joaquin rivers. Beach seines are used to assess the spatial distribution of juvenile Chinook Salmon in and upstream of the Delta by targeting the shallow (≤ 1.2 m depth) near-shore habitats where small juvenile Chinook Salmon can typically be found. Beach seine sites are sampled with a single haul using a 15.2-m-by-1.3-m beach seine net with 3-mm mesh. Beach seines are deployed along the shoreline by two crew members within unobstructed habitats (e.g., boat ramps, mud banks, sandy beaches) starting from the downstream portion of each site to limit disturbance (e.g., displacement of sediment into the site).
DJFMP trawls are used to examine the relative abundance of juvenile Chinook Salmon migrating in and out of the Delta: Sacramento and Mossdale trawl sites for entry points into the Delta at the Sacramento River and San Joaquin River, respectively, and Chipps Island trawl site for the exit point of the Delta at the confluence between the Sacramento and San Joaquin rivers ( Figure 2

Spring Kodiak Trawl
The SKT survey was established by the CDFW in 2002 to monitor the distribution and relative abundance of spawning Delta Smelt in the estuary (Souza 2002;Polansky et al. 2018). The core SKT survey samples from January to May, with a single tow each month at 40 fixed stations that cover the range of adult Delta Smelt ( Figure 2). The SKT survey is conducted with a Kodiak trawl net (almost identical to that used by EDSM) for 5 or 10 minutes at near-idle speed. Although Delta Smelt is the target species for the SKT, this survey has caught a substantial number of Chinook Salmon over the years (Castillo et al. 2018), and the similarity of its gear to the EDSM program's is useful for comparison between randomized and fixed stations.

Knights Landing Rotary Screw Trap
The CDFW established the Knights Landing rotary screw trap sampling site on the upper Sacramento River in 1995 to provide an early warning of juvenile salmonids emigrating into the Delta and to trigger water operation modifications ( Figure 2

Size Distribution Comparison
To explore differences between sampling programs, we constructed a series of bean plots that illustrate size-frequency distributions of Chinook Salmon catch over time. Bean plots provide a convenient way to characterize the distribution shape of continuous data, which is given by a kernel density estimate computed within the "beanplot" function and library in R (Kampstra 2008; R Core Team 2018). We limited data to those that overlap with the randomtessellation EDSM data used in this study (i.e., December 2016-March 2017, December 2017-March 2018. For this size distribution comparison, we further limited the data by excluding catch without length measurements, and catch of known hatchery-origin Chinook Salmon that were identified by a clipped adipose fin. Sample locations from all monitoring programs were spatially joined in ArcGIS (version 10.6.1) by geographic proximity to EDSM region ( Figure 2) to ensure comparisons were made with the same or nearby EDSM region. We also summarized these data to quantify the differences in catch and size distributions of juvenile Chinook Salmon in the estuary (see Table 2).

Occupancy Model
To assess large-scale relative differences in detection probability of juvenile Chinook Salmon between surveys, we used an occupancy-model framework (MacKenzie et al. 2002). An occupancy model uses replicate samples conducted within each site in a set of sites to simultaneously estimate occupancy (the probability that a randomly selected site in the study area is occupied) and detection (the probability of detection at a site conditional on occupancy) for a species of interest. A site's detection history consists of a series of 1s (indicating detection) and 0s (indicating non-detection) that reflect the outcomes of the replicate samples (e.g., 011 for non-detection, detection, detection). Detection histories can be used to construct a likelihood for estimating occupancy and detection probabilities. It is generally assumed that the occupancy status (occupied or not occupied) of a site remains constant throughout the period during which replicate samples are collected; this is known as the closure assumption (MacKenzie et al. 2002).
We defined our study area as the San Francisco Estuary (Figure 1), our study time-frame as December through March, and our species of interest as juvenile Chinook Salmon. We divided the estuary into 39 geographic sub-regions (edited slightly from EDSM sub-region cut-offs to be more applicable for Chinook Salmon; see Figure 2), defined a site as a unique combination of sub-region and date, and treated samples collected by EDSM, DJFMP, SKT, and YBFMP in a given sub-region-date as replicates. We modeled occupancy probability, Ψ, and detection probability, p, in terms of three categorical variables: region, month, and gear: Region reflects a coarse spatial partitioning of the estuary, with each of the 39 sub-regions falling into only one of the 11 regions (Figure 2), and Month reflects the month (December, January, February, or March) during which a sample was collected. We included Region and Month to account for spatiotemporal variability in abundance that can affect occupancy and detection, but our primary interest was in gear, which we used to assess relative differences in detection probability of juvenile Chinook Salmon between surveys. We divided gear into six categories: EDSM Kodiak trawl, DJFMP Sacramento Kodiak trawl, DJFMP Chipps Island midwater trawl, DJFMP Mossdale Kodiak trawl, CDFW spring Kodiak trawl, and beach seine (DJFMP and YBFMP combined). We kept the various Kodiak trawls separate to implicitly account for design differences between surveys (e.g., differences in tow durations, fixed vs. random site-selection methods). We did not include common-habitat predictor variables such as water-quality parameters because they are beyond the scope of our study.
For further clarification on how we structured the data, suppose two EDSM samples and one beach seine sample were collected in a given subregion on a given date (i.e., a given site). Then the detection history for this site would be a vector of length three, for example (0,1,1), and the gear covariate vector would be (EDSM, EDSM, seine). The region vector would consist of the region value (corresponding to the given sub-region) repeated three times. Similarly, the month vector would consist of the month value (corresponding to the given date) repeated three times.
We fit separate models for the December 2016-March 2017 time-period and the December 2017-March 2018 time-period. The former corresponds to water year 2017, which was a record wet year with fairly high juvenile Chinook Salmon abundance; the latter corresponds to water year 2018, which had below-average precipitation and modest juvenile Chinook Salmon numbers. (We note that the water year in California begins in October and ends in September). The 2 years provide good contrast, and running separate models allowed us to account for inherent differences between these 2 years while avoiding a great reduction in degrees of freedom through interaction terms. We fit both models using the "unmarked" package (Fiske and Chandler 2011) in R (R Core Team 2018).

Occupancy Model Interpretation
Our analysis represents the novel application of an occupancy-model framework to data that were not collected as part of a dedicated occupancy study. Because of this, further discussion of model interpretation is warranted. With traditional occupancy models, individual locations are surveyed multiple times over the course of the study, and the occupancy state (i.e., occupied or unoccupied) at a given sampling location does not change over the study (see Kéry 2010). Using this survey design, the interpretation of occupancy is the proportion of locations sampled in which at least one fish was present. This differs from our survey design and interpretation of occupancy. Based on how we defined a site (i.e., a sub-region sampled on a given date), occupancy represents the proportion of sub-region-date combinations during which at least one fish was present in the sub-region on that date. Thus, occupancy for a given region and month is then the proportion of sub-region-date combinations (within the given region-month combination) that are occupied. Note that this is different from-and more abstract than-the interpretation of occupancy as the proportion of days during which a given subregion was occupied.
Each tow or seine haul constitutes a sub-sample of a sub-region. It can be argued that even temporal replicates (such as those conducted by the EDSM Kodiak Trawl, Sacramento Kodiak Trawl, Chipps Island Midwater Trawl, and Mossdale Kodiak Trawl) equate to spatial subsamples, since water in the estuary is constantly moving, and a gear cannot realistically sample the exact same "patch" of water multiple times.
This introduces the concept of local occupancy, i.e., occupancy at the sample level conditional on occupancy at the sub-region level (Kendall and White 2009; Guillera-Arroita 2011). In our model, what we refer to as detection probability (Equation 2) is therefore an effective detection probability equal to the product of the probability of local occupancy and the probability of detection at the site level. Among other variables, local occupancy probability is a function of the proportion of sub-region water volume sampled, with the probability of occupancy increasing from 0 to 1 as the proportion sampled increases from 0 to 1. From this perspective, local occupancy can change from sample to sample. Here, however, we are attempting to capture large-scale relative changes in both local occupancy and sample-level detection through the Gear variable.
Defining site at the sub-region-date level allowed us to have replicate samples while minimizing variability in occupancy and detection. For modeling, however, we used region and month covariates to keep the number of parameters relatively low while still accounting for changes in abundance that can affect detection, regardless of which gear is used. The model would have a different interpretation if all samples from a given region or month were treated as replicates, and in that case, occupancy and detection estimates would be higher.

Cumulative Detection Probability
In addition to the single-sample detection probability estimates the model provided, we investigated the ability of a given gear to detect at least one Chinook Salmon across "replicate" samples in a given month and region. We calculated gear-specific cumulative detection probability γ r,m,g as γ r,m,g = 1 -(1 -pˇr ,m,g ) n where pˇr ,m,g is the model-estimated detection probability for gear g in month m and region r, and n = 1,…,10 is a hypothetical number of replicate samples.
We summarized the benefits of EDSM to Chinook Salmon monitoring efforts through a synthesis of our understanding of salmon biology, the use of existing surveys, and our modeling results. However, to provide a quantitative assessment of such benefits, we calculated the probability of detecting Chinook Salmon at least once in a given month and sub-region, conditional on the species' presence and on a particular level of sampling effort across gear types. We used 39 sub-regions (Figure 2), each of which falls into a single region, to examine detection differences during the study period on a finer geographic scale. We calculated the across-gear cumulative detection probability Γ s,m for sub-region s and month m as where γ s,m,g is the gear-specific cumulative detection probability for gear g, month m, and sub-region s, and the product is across the set of all gears G s,m that were used to sample sub-region s and month m. Here, we calculated γ s,m,g as γ s,m,g = 1 -(1 -p^r ,m,g ) n s,m,g where n s,m,g is the actual number of "replicate" samples taken by the gear, and p^r ,m,g corresponds to the region that contains sub-region s. We calculated Γ s,m under two scenarios-one with EDSM samples excluded and one with EDSM samples included-and subsequently calculated the increase in detection probability that resulted from the inclusion of EDSM samples.  Table 2). Out of this total, we observed the highest catch counts in January (n = 9,261) followed by February (n = 5,705),

Size Distribution Comparison
Despite substantial differences in total catch across monitoring programs, the size frequency distributions of captured Chinook Salmon were relatively consistent ( Figure 3; Table 2; Figure A2). The contrast between EDSM and Chipps Island Trawl was a notable exception. The Chipps Island Trawl captured larger salmon on average between December and March ( Figure 3; Table 2). It may also be worth noting that size distribution differences between EDSM trawl and other surveys appear to differ more in the month of March (Figures 3 and A2). Variation in fish size was greatest in December (mean coefficient of variation [CV] = 0.412) and March (mean CV = 0.251) across monitoring programs ( Figure 3; Table 2). We observed the least amount of variation in fish size in February (mean CV = 0.152; Figure 3; Table 2).

Occupancy Model
The data set used to fit the model for water year 2017 consisted of 698 sites (sub-region-date combinations), with the number of replicate samples per site ranging from 1 to 32. Seventynine percent of sites had 1, 2, 5, 8, or 10 replicate samples (Table A1). The data set used to fit the water year 2018 model consisted of 903 sites, with replicate sample sizes that ranged from 1 to 25. Eighty percent of sites had 1, 2, 5, 6, or 10 replicate samples (Table A1).
Overall occupancy probabilities were higher in water year 2017 than in water year 2018 (  Figure A3). Detection probability for the beach seine was consistently higher than for any other gear (with the exception of Mossdale Kodiak Trawl); detection probability for EDSM was  consistently lowest ( Figure A4). SKT detection probability was similar to that of the Sacramento Kodiak trawl and the Chipps Island midwater trawl, except in water year 2018, when detection at Chipps Island was higher than SKT. In a given water year, temporal detection patterns were similar for all gears. For example, in water year 2017, detection increased from December to February, and decreased in March.

Cumulative Detection Probability
Although EDSM had the lowest single-sample detection probability, as few as two or three replicate EDSM samples resulted in a cumulative detection probability similar to a single-sample detection probability by SKT in both years, Sacramento Kodiak trawl in both years, and Chipps midwater trawl in 2017. Because EDSM conducts between 2 and 10 tows per site, typically with multiple sites per sub-region, the cumulative detection probabilities for EDSM were generally comparable to the single-sample detection probability of the other gears (Figure 4). The primary gains in cumulative detection probability from the addition of EDSM occurred in the lower estuary (i.e., San Pablo Bay/Napa River, Suisun Bay, and Suisun Marsh), the lower San Joaquin River, the Sacramento River, and Cache Slough-Liberty Island, with the most dramatic increases occurring in March of each year (Figures 5 and  A5).

DISCUSSION
Effective management in a dynamic estuarine system can be challenging, given the number of species in decline, limited resources, various interacting environmental drivers that continually change the system, and imperfect information to guide management and conservation actions.
Monitoring is a crucial component of ecosystem management, and leveraging existing data sources for multiple species can be one effective way to enhance information when management decisions are made. Here, we explored juvenile Chinook Salmon bycatch data collected by the recently established EDSM program (USFWS et al. 2019). Our examination of juvenile Chinook Salmon size-frequency distribution indicates that, in general, fish surveys in the estuary  capture similar sizes of juvenile Chinook Salmon from December to March (Figure 3). However, around Chipps Island, Kodiak trawls (as used by EDSM and SKT) appear to be under-sampling larger-sized salmon, while the DJFMP midwater trawl seems to be under-sampling smaller-sized salmon. This is in contrast to a previous study showing that Kodiak trawls catch larger salmon than midwater trawls on the Sacramento River (McLain 1998). The DJFMP midwater trawl used at Chipps Island has a larger net opening and mesh size compared to the DJFMP midwater trawl used on the Sacramento River (Table 1), which may explain this discrepancy (IEP et al. 2019a). If absolute abundance estimation for the various salmon runs in the estuary is the goal (Perry et al. 2016), then a relative gear efficiency assessment that uses existing data (Walker et al. 2017) or an additional side-by-side gear comparison for juvenile salmon may be warranted to better understand the fish size bias associated with net and mesh dimensions (Mitchell et al. 2019).
The spatiotemporal patterns in occurrence and detectability we observed in our occupancy model aligned with our understanding of Central Valley salmon life history. We expect to see detection probability increase with salmon density (i.e., number of salmon available to be caught). As such, both occupancy and detection probability estimates were at their highest in February and March-the months in which we would expect a higher catch of salmon catch within our December−March study period (Yoshiyama et al. 1998;Sturrock et al. 2015). Occupancy and detection probability estimates also tend to be higher in regions that are within the migratory pathway of salmon (e.g., Sacramento River, Suisun Bay), whereas backwater areas, such as the Sacramento Deep Water Shipping Channel, had low cumulative detection probability estimates despite the amount of sampling that occurred ( Figure 2). The wet water year of 2017 (December 2016-March 2017) saw considerably higher occupancy and detection probabilities (Table 3), consistent with previous studies that demonstrated a positive relationship between outflow and salmon occurrence in the estuary (Kjelson et al. 1982;Brandes and McLain 2001;Munsch et al. 2021). One potential reason for the higher occurrence of juvenile salmon in the estuary during high flow years ( Figure A3) is floodplain inundation such as that observed in Yolo Bypass during 2017, which can increase habitat and create additional migratory pathways (Sommer et al. 2001). In the drier water year of 2018 (December 2017-March 2018), moderate to high occupancy probability estimates were primarily restricted to the Sacramento River and downstream (Table 3), likely reflecting the well documented low survivorship of juvenile salmon along the San Joaquin River and the interior Delta (Newman and Brandes 2010;Perry et al. 2010;Buchanan et al. 2013;. We found considerable differences in detection probability among gear types, and these differences remained consistent between the 2 years. In general, given a single sampling event, beach seines had the highest probability of detecting juvenile Chinook Salmon in a region if the species were present, followed by the fixed station trawl, and then by the EDSM randomstation trawl. Multiple factors likely led to this result. Beach seines occur in shallow-water, nearshore habitat, whereas the trawls take place in open water. Juvenile Chinook Salmon may rear in higher density in nearshore habitat than in open water (Kjelson et al. 1982). The fixed station trawls that DJFMP conducts ( Figure 2) are set in the migratory path of juvenile Chinook Salmon by design; therefore, we can expect these stations to have a higher detection probability than randomly chosen sites. It is less clear why fixed sites for a Delta Smelt monitoring program such as the SKT would have higher detection probability for juvenile salmon than those selected at random. However, fixed stations are typically determined based on their higher fish catch, and may comprise higher-quality habitat for fish in general (McClelland and Sass 2012). Random sites as sampled by EDSM are meant to provide a snapshot of the estuary and may inadvertently survey microhabitats not used by juvenile salmon in a particular region.
It is also important to consider the assumptions of our model and how they may affect our results. We defined a site as any sub-region with at least a single sample on a given day. Based on this definition, a sub-region can become occupied or unoccupied with little consideration for temporal correlation aside from the month variable. However, the occupancy of a sub-region-date is naturally correlated with the occupancy of the same sub-region on the previous date and subsequent date. In particular, if Chinook Salmon are present on the adjoining dates, it is more likely they will have been present on the intervening date. This lack of independence in occupancy between sub-region-date combinations can lead to biased parameter estimates, overdispersion in the model, or both. A potential solution would be to incorporate extinction and colonization probabilities into the model (e.g., MacKenzie et al. 2003) to account for local exodus and re-occupancy of sub-regions, but this was beyond the scope of our objectives.
Our model also assumed that temporal and spatial replicates are exchangeable. EDSM and DJFMP trawls had temporal replicates; the SKT and DJFMP beach seines had only spatial replicates (because of the lack of temporal replicates in their original survey designs). Spatial replicates may induce bias in occupancy estimates, depending on sampling design (e.g., with or without replacement), the system, and species dynamics (Kendall and White 2009;Guillera-Arroita 2011;Charbonnel et al. 2014).
The discrepancy in replicate types may also have contributed to some of the differences in the gear detection probabilities we observed.
Having spatial replicates may lead to higher detection probability estimates merely because samples from multiple locations within the same sub-region and day would likely have more independence (i.e., lower correlation with one another) than multiple samples taken from a single location within the same sub-region and day. Teasing apart the different factors that affect detection probability is outside the scope of our study. However, we expect that the relatively low detection probability of EDSM is partly a result of differences in the number of spatial and temporal replicates.
Despite the relatively low detection probability of the EDSM trawl, we found substantial improvements in our juvenile Chinook Salmon monitoring coverage for water years 2017 and 2018 (Figures 5 and A5). This is largely a result of the wide geographical scope of EDSM and the frequency at which it conducts its sampling. At a single location, EDSM would typically conduct anywhere between two and ten replicate tows, which would increase the program's cumulative detection probability to levels comparable with other surveys (Figure 4). This was done 4 days per week throughout December to March in our study period across a large portion of the estuary, resulting in improved cumulative detection probability for juvenile salmon in regions that were generally under-sampled by other surveys (provided that salmon are present in detectable numbers). This added information was notable downstream of the Delta (Figures 5 and A5), where EDSM has observed fish that were winterrun and spring-run length-at-date sizes ( Figure 6).
Having information in these key regions of the salmon migratory pathway can help the species' life history variability be better understood (Sturrock et al. 2015;Goertler et al. 2018;) as well as how species interact with environmental drivers such as water year type ( Figure 6). Moreover, fixed stations are not likely to represent the estuary as a whole (IEP SAG 2013; Peterson and Barajas 2018) and may bias abundance estimation (McClelland and Sass 2012; Kiraly et al. 2014;Li et al. 2015). Incorporating random station data from EDSM can potentially aid the estimation of absolute abundance for juvenile Chinook Salmon through the proper calibration of fixed station random effects. Data from EDSM can also be used to better account for imperfect detection (i.e., observation error), because the program's replicate tows are conducted within a fairly short time-frame and should not violate the closure assumption excessively (Peterson and Barajas 2018). However, aspects of the EDSM data could limit its use for Chinook Salmon monitoring under the current state. The juvenile Chinook Salmon out-migration window in the estuary extends into the early summer months, and the larval fish gear that EDSM uses in these months does not efficiently capture juvenile salmon. The EDSM program also currently uses length-at-date criteria instead of genetic analysis, which may lead to erroneous assignments for the various Central Valley Chinook Salmon runs (Hedgecock 2002;Harvey et al. 2014).

CONCLUSIONS
For all fish monitoring programs, there will inevitably be trade-offs in temporal and spatial scales of measurement as a result of limited resources and, at times, multiple objectives (Radinger et al. 2019). To adjust the estuary's salmon monitoring network, recommendations have been made, such as the addition of new gears, collection of fish condition information, or transition into randomized stations (IEP SAG 2013; Johnson et al. 2017). Stratified random sampling design offers many advantages, and is

Figure 6
Size and life stages of Chinook Salmon caught in EDSM Kodiak trawls from Suisun Bay, Suisun Marsh, and San Pablo Bay regions (as seen in Figure 2), demonstrating contrast in numbers and sizes of Chinook Salmon observed within this area between the wet water year of 2017 and the moderately dry year of 2018. Only data from measured fish are shown. Colors denote different life stages of Chinook Salmon as classified by field crew, and "n/p" indicates that it was not recorded (generally for adipose fin-clipped, hatchery fish).
generally preferable for estimating the abundance of a species given unlimited resources (IEP SAG 2013; Kiraly et al. 2014;Peterson and Barajas 2018). However, results from our model indicate that certain methods (i.e., fixed station beach seines) are more cost-effective at detecting juvenile Chinook Salmon if the species is present in some areas (because it generally involves less staff and gear). For fixed station surveys, such as the DJFMP beach seine, modifying protocol to include some form of random station selection could provide similar benefits (e.g., high detection probability, cost-effectiveness) while allowing for better abundance estimation. Plans are currently being developed to implement a stratified random sampling design for the DJFMP beach seine survey in accordance with these recommendations (IEP SAG 2013), and it would be prudent to assess how detection probabilities change once this new design is implemented. However, for some aspects of juvenile Chinook Salmon management that focus on their occurrence at certain regions, such as the Delta Cross Channel gate operations (NMFS 2009; NMFS 2019), having higher detection probability at specific regions may be more desirable than a proper estimation of abundance.
This study serves as a first step in leveraging Delta Smelt monitoring data collected by EDSM to better understand juvenile Chinook Salmon monitoring in the estuary. Our results indicate that using EDSM data along with the traditional salmon surveys can improve our monitoring of under-sampled regions of the estuary, and increase the spatial resolution of surveys within each region of the estuary. With data collected under a stratified random design, we can also better infer the true proportion of the estuary that is occupied by salmon at a given time period. Lastly, we demonstrated that trade-offs exist between various sampling designs undertaken by the fish monitoring programs we analyzed. By leveraging the strengths from each program, we can make stronger inferences about juvenile Chinook Salmon abundance and distribution patterns. Each survey design (e.g., fixed station vs. random station) offers advantages that are tied to specific monitoring goals. Careful consideration of these trade-offs and the overall monitoring objectives is crucial as management agencies of the estuary continue to adapt and improve their monitoring programs.