This project evaluated the differences in habitat suitability of spawning adult Delta Smelt in the upper San Francisco Estuary during March 2006 and March 2008. The purpose of this project is to understand habitat suitability changes during a dry (March 2006) and wet year (March 2008) to help with conservation planning for this species during drought. This was achieved through modeling Delta Smelt presence using water quality data that were collected in the estuary. In the end, habitat suitability was similar in both years in terms of region, but the degree of suitability changed. For both years, habitat suitability (based on the probability of Delta Smelt presence) was highest upstream of Grizzly Bay in Montezuma Slough and Suisun Slough, and lowest on the San Joaquin River. Even so, these results are not conclusive and more work should be done to understand the shifts in habitat suitability under different hydrological conditions.

The Delta Smelt (Hypomesus transpacificus) is only found in the upper San Francisco Estuary in California and its presence is associated with the freshwater-saltwater mixing zone in the estuary, except when spawning (Moyle et al. 1992). During spawning, the Delta Smelt moves to the freshwater portion of the estuary. Starting around 1982, there has been a long-term decline in the abundance of Delta Smelt (Nobriga et al. 2008). The reason for this decline is based on multiple factors, which range from long-term drought during 1987 to 1992 that degraded Delta Smelt habitat to mortality from water diversions in the southern Delta (Moyle et al. 1992). Due to the decline, the United States Fish and Wildlife Service and the California Department of Fish and Wildlife (then Fish and Game) listed the species as threatened under the federal and state Endangered Species Act in 1993. The continued decline of Delta Smelt in the 2000s led to a change in its listing status from threatened to endangered in 2009 under the state Endangered Species Act (Merz et al. 2011). Due to the listing, regulatory actions are in place to protect Delta Smelt in the estuary throughout the year, which includes making reductions in water diversions from the pumping plants of the State Water Project and Central Valley Project (Feyrer et al. 2007).

Figure 1. An overview of the upper San Francisco Estuary, where Delta Smelt inhabit.

Despite water management actions to protect Delta Smelt, the species is still declining, especially in light of the recent drought in California. For example, recent surveys by the California Department of Fish and Wildlife reported only one spawning adult Delta Smelt caught in April 2015. This is a huge drop from the 143 that were caught in April 2012 (Krieger 2015) and stresses the need to understand habitat suitability under wet and dry hydrological conditions. Since Delta Smelt only have a one-year life cycle and have low fecundity, it is highly influenced by changes in the estuary (Moyle et al. 1992). To understand the changes in dry and wet conditions, this project will examine how habitat suitability changes for spawning Delta Smelt in the estuary. This will be done through modeling Delta Smelt presence using water quality data that were collected in the estuary.

The current understanding on habitat suitability for Delta Smelt is heavily based on the long-term fish monitoring surveys that are conducted by the California Department of Fish and Wildlife. These surveys include the 20-mm Tow-Net Survey, the Summer Tow-Net Survey, the Fall Midwater Trawl, and the Spring Kodiak Trawl. Detailed descriptions of these surveys can be found in Merz et al. (2011), Murphy and Hamilton (2013) and Sommer and Mejia (2013). In general, these surveys sample for the different life stages of Delta Smelt at different times of the year. Specifically, the 20-mm Tow-Net Survey samples for larvae and juveniles from around March to July, the Summer Tow Net Survey samples for juveniles during June to August, the Fall Midwater Trawl samples for sub-adults during September to December, and the Spring Kodiak Trawl samples for pre-spawning and spawning Delta Smelt from January to May. These long-term surveys have helped provide an understanding on the spatial and temporal distribution of Delta Smelt (see Merz et al. 2011 for synthesis). However, there are still major uncertainties, which includes the migration strategy for moving out of the freshwater-saltwater mixing zone in the estuary to freshwater for spawning.

For example, Sommer et al. (2011), which looked at the mean position of Delta Smelt from the Fall Midwater Trawl, found that Delta Smelt moved upstream to the eastern portion of the Sacramento-San Joaquin Delta for spawning. Delta Smelt that are in the eastern portion of Delta are more susceptible to entrainment into the pumping plants of the State Water Project and Central Valley Project. Looking at similar data, Murphy and Hamilton (2013) did not find migration to be unidirectional. Instead, Delta Smelt that rear in Suisun Bay would migrate in the northern direction to Montezuma Slough and Suisun Marsh to spawn. In addition, Delta Smelt that rear in the lower Sacramento River could move in many directions for spawning, which includes further upstream of the Sacramento River, to the eastern region of the San Joaquin River, and western Montezuma Slough. The differences between Sommer et al. (2011) and Murphy and Hamilton (2013) show how the understanding on Delta Smelt is shaped by the surveys conducted by the California Department of Fish and Wildlife. However, the utility of data from these surveys will likely decrease in future years since Delta Smelt catch has declined to low levels during the recent drought (Krieger 2015). This would make it harder for these surveys to detect Delta Smelt even if the fish are present. Moreover, Delta Smelt could move into areas that are not sampled by the surveys for spawning (Murphy and Hamilton 2013).

As an alternative to using survey data, habitat suitability based on water quality collected could provide insight on the spatial and temporal distribution of the Delta Smelt. Many studies have made use of generalized addictive modeling to predict habitat suitability for Delta Smelt based on water quality in the upper San Francisco Estuary (see Feyrer et al. 2007, Nobriga et al. 2008, and Sommer and Mejia 2013). A generalized addictive model is useful in depicting the relationships of nonlinear relationships among dependent and independent variables.

As an example, Feyrer et al. (2007) was able to develop a generalized addictive model between the probability of pre-spawning adult Delta Smelt presence using Secchi depth (a surrogate for turbidity), electrical conductivity (a surrogate for salinity), and water temperature during the fall, using data collected from the Fall Midwater Trawl. The model accounted for 25.7% of the deviance for delta smelt presence. Generally, Feyrer et al. (2007) found that there was a long-term decline in habitat suitability over a 36-year period, with the largest declines in the southeastern and western regions of the estuary.

Similarly, Nobriga et al. (2008) was able to develop a generalized addictive model for the probability of Delta smelt presence for juveniles and sub-adults during the summer, using the same variables as Feyrer et al. (2007). Instead of using data from the Fall Midwater Trawl, Nobriga et al. (2008) used data from the Summer Tow-Net Survey. Based on this study, habitat suitability during the summer was highest in the Confluence of the Sacramento and San Joaquin rivers.

Lastly, Sommer and Mejia (2013) developed a similar generalized additive model for larvae and juveniles using data from the 20-mm Tow-Net survey. However, Sommer and Mejia (2013) also included copepod densities, an important food source for Delta Smelt, into the model. Overall, these studies showed that Delta Smelt usually reside in areas that are less than 6 psu in salinity, water temperatures less than 25 degrees Celsius, and in fairly turbid environments. In fact, Delta Smelt are usually only present during the fall when Secchi depth is less than 1 meter. Turbid environments help Delta Smelt avoid predators and help with feeding (Sommer and Mejia 2013). As a result, generalized additive models are a useful tool for evaluating habitat suitability.

A generalized additive model for predicting delta smelt habitat suitability has been published for larvae/juveniles (based on 20-mm Townet Survey; see Sommer and Mejia 2013), juveniles/sub-adults (based on Summer Tow-Net Survey; see Nobriga et al. 2008), and pre-spawning adults (based on Fall Midwater Trawl; see Feyrer et al. 2007). Unfortunately, no model has been published in literature for spawning adult Delta Smelt, which is the life stage of interest for this project. As a result, there was a need to create a generalized addictive model between Delta Smelt presence from the Spring Kodiak Trawl and selected water quality parameters. Therefore, Delta Smelt absence/presence and water quality data were acquired online at the California Department of Fish and Wildlifes’s website (http://www.dfg.ca.gov/delta/projects.asp?ProjectID=SKT). The data used for the model was between January and May from 2002 to 2015 and the stations sampled is found in Figure 2.

Figure 2. Data used for the project in the upper San Francisco Estuary

To develop a model for habitat suitability of spawning adult Delta Smelt, this project acquired the programming script that was used in Sommer and Mejia (2013) for the 20-mm Tow-Net Survey and modified it for the Spring Kodiak Trawl data. The generalized additive model for spawning adult Delta Smelt was done in R using the GAM function from the MGCV package (R Development Corm Team 2011). A logit-link function was used to describe the relationship between absence/presence (i.e., binomial) data and the water quality parameters in the generalized addictive model. In the end, the model developed used the same water quality parameters (i.e., Secchi depth, electrical conductivity, water temperature), as Feyrer et al. (2007) and Nobriga et al. (2008). Like in those past studies, all three parameters were significant predictors of delta smelt presence. However, the deviance for the model developed for spawning adult Delta Smelt during winter and spring was only 18.7%. This was lower than the deviance of 25.7% for adult Delta Smelt in the fall by Feyrer et al. (2007) and the deviance of 29% for juvenile/sub-adult Delta Smelt during the summer by Nobriga et al. (2008). This shows that the ability to predict suitable habitat for Delta Smelt is not as strong during the winter/spring. Since the model developed is a generalized addictive model and uses smoothers, it cannot be easily displayed in equation format. As a result, the model itself is not described in this report.

After developing the model for spawning adult Delta Smelt, the model was applied to the continuous water quality data in R. The water quality data were collected by the California Department of Water Resources during two cruises on a research vessel in the upper San Francisco Estuary from March 20 to March 24, 2006, and March 12 to March 28, 2008 (see Figure 2 for route). These two cruises, selected for the analysis, were based on two different hydrological conditions. Based on the water year type classification, 2006 was a wet year, while 2008 was a dry year. Moreover, March was selected for comparison purposes since the presence of spawning adult Delta Smelt start to peak in March (Murphy and Hamilton 2013).

In the end, the model produced habitat suitability for spawning Delta Smelt in the estuary during March 2006 and March 2008 (see Figure 3 and Figure 4). During the continuous water quality data collection process, each collection point was georeferenced with a latitude and longitude in the field on a vessel attached GPS, so it could be displayed in ArcMap 10.3.1. As a result, data were displayed on an ESRI base map, using California Teale Albers projection, to help make comparisons during a dry and wet year.

To help compare differences, a spatial join was conducted in ArcMap on the points collected in March 2006 and March 2008 to combine points into one dataset. This helps provide consistency in the observations since the water quality transects between years were not identical. Afterwards, the habitat suitability for a given point from 2006 was subtracted from 2008 to document quantitative differences (Figure 5). This was done using the field calculator in ArcMap.

In March 2006 (a wet water year), habitat suitability (based on the probability of Delta Smelt presence) is highest upstream of Grizzly Bay in Montezuma Slough and Suisun Slough, and lowest on the San Joaquin River (Figure 3). These trends appear to hold true during a drier year in March 2008 (Figure 4); however, the habitat suitability decreases in some areas and increases in other areas (Figure 5). For example, habitat suitability appeared to be higher in Suisun Bay and on the San Joaquin River in March 2008 than in March 2006. On the other hand, habitat suitability appeared to be higher on the Sacramento River and upstream of Grizzly Bay in March 2006 than March 2008. Despite these results, the differences may not be statistically significant and this project did not test for significant differences.

Although differences exist between years, habitat suitability was not particularly high in the eastern portion of the estuary on the San Joaquin River during these two months. This does not support the eastern migration strategy for spawning Delta Smelt that is discussed in Sommer et al. (2011). However, this finding is not necessarily conclusive, as the project only analyzed one month in two different hydrological years.

Lastly, there appears to be higher quality habitat in the reach between Suisun Bay and the San Pablo Bay in March 2006 than March 2008. Even so, Delta Smelt do not establish in San Pablo Bay, but could be washed out to the region under high flows.

Figure 3. Habitat Suitability of Spawning Adult Delta Smelt during March 2006, a wet year.

Figure 4. Habitat Suitability of Spawning Adult Delta Smelt during March 2008, a dry year.

Figure 5. Differences between Habitat Suitability of Spawning Adult Smelt in 2006 and 2008.

In the end, this project was able to depict differences in habitat quality during March in a wet (i.e., 2006) and a dry year (i.e., 2008) year. However, results from this project are hardly conclusive and does not explain the reason for the change between years. For example, results are not conclusive since the study only examined one month of an entire water year. There is likely to be variations between months. Future analyses should include other months in the water year.

In addition, the water quality transect used for this project to model habitat suitability does not encompass all of the regions that Delta Smelt could inhabit, as seen in Figure 2. Future water quality sampling should include other regions, where Delta Smelt could be present. Results may change based on what is sampled. Nonetheless, modeling habitat suitability from water quality data will be a useful tool in the future, as the ability to detect Delta Smelt from field sampling may decrease due the low fish abundance.

Lastly, the habitat modeling used for the project did not consider all of the biotic and abiotic factors that could influence the probability of delta smelt presence. The model only included Secchi depth, turbidity, and water temperature, as these were the parameters that were available for modeling. However, biotic factors like predation and food availability could influence habitat suitability. Similarly, abiotic factors like dissolved oxygen could also influence habitat suitability (Feyrer et al. 2007).

In summary, there were differences in habitat suitability for spawning adult Delta Smelt during March between dry and wet years. Future analyses should include other months of the year in order to fully understand the differences in habitat suitability during years with dry and wet hydrological conditions. This will be particularly important in light of recent drought in California. As a result, there needs to be a stronger understanding on how to conserve this species during dry periods, especially since the long-term decline since 1982 is associated with the drought from the 1987 to 1992 period. If not, the Delta Smelt may go extinct in the upper estuary.

Overall, the biggest challenge for this project was understanding the data collection and data storage process for the Spring Kodiak Trawl. Luckily, there was a vast amount of information published about the data collection process of the Spring Kodiak Trawl (e.g., Merz et al. 2011). In particular, I am thankful for past and present staff at the California Department of Fish and Wildlife for making this data publically available on a website:
http://www.dfg.ca.gov/delta/projects.asp?ProjectID=SKT.

On a similar note, I was lucky to also be able to use water quality transect data from the California Department of Water Resources, which is currently not publically available. I was able to acquire the data from Scott Waller, a former colleague, at the California Department of Water Resources.

In addition, one of the most challenging parts of this project was creating a generalized addictive model to predict habitat suitability for spawning Delta Smelt, using the R Statistical Software. I was very lucky to be able to get a hold of the programming script for the habitat modeling done with the 20-mm Tow-Net Survey from Francine Mejia, a co-author of Sommer and Mejia (2013). Although challenging, I was able to modify the script to develop a generalized addictive model for spawning adult Delta Smelt, using the Spring Kodiak Trawl.

Feyrer, F., M.L. Nobgria, and T.R. Sommer. 2007. Multidecadal trends for three declining fish species: habitat patterns and mechanisms in the San Francisco Estuary, California USA. Can. J. Fish Aquat. Sci 64: 723-724.

Krieger, L. M. "California drought: Delta smelt survey finds a single fish, heightening debate over water supply". San Jose Mercury News

Merz, J.E., S. Hamilton, P.S. Bergman, and B. Cavallo. 2011. Spatial perspective for Delta Smelt: a summary of contemporary survey data. California Fish and Game 97(4): 164-189.

Moyle, P.B., B. Herbold, D.E. Stevens, and Lee W. Miller. 1992. Life history and status of delta smelt in the Sacramento-San Joaquin Estuary, California. Transactions of the American Fisheries Society 121:67-77.

Murphy, D.D., and S.A. Hamilton. 2013. Eastward Migration or Marshward Dispersal: Exercising Survey Data to Elicit an Understanding of Seasonal Movement of Delta Smelt. San Francisco Estuary and Watershed Science 11(3).

Nobgria, M.L., T.R. Sommer, F. Feyrer, and K. Flemming. 2008. Long-term Trends in Summertime Habitat Suitability for Delta Smelt. San Francisco Estuary and Watershed Science 6(1).

R Core Team. 2011. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.

Sommer, T., F. Mejia, M.L. Nobgria, F.Feyrer, and L. Grimaldo. 2011. The spawning migration of Delta Smelt in the upper San Francisco Estuary. San Francisco Estuary and Watershed Science 9(2).

Sommer, T., and F. Mejia. 2013. A Place to Call Home: A synthesis of Delta Smelt Habitat in upper San Francisco Estuary. San Francisco Estuary and Watershed Science 11(2).