California Clapper Rail (Rallus longirostris obsoletus)
Population monitoring: 2005-2008
Leonard Liu, Julian Wood, Nadav Nur, Diana Stralberg, and Mark Herzog
PRBO Conservation Science
3820 Cypress Drive #11, Petaluma, CA 94954
DRAFT July 29, 2009
California Department of Fish and Game
4001 N. Wilson Way
Stockton, CA 95205
PRBO Conservation Science conducted call count surveys for California Clapper Rail (Rallus longirostrus obsoletus) at 53 sites throughout the San Francisco Bay Estuary from 2005 through 2008. To maximize the spatial coverage of sites, surveys were coordinated with partners conducting call-count surveys (Avocet Research Associates, California Department of Fish and Game, California Coastal Conservancy’s Invasive Spartina Project, and U.S. Fish and Wildlife Service) resulting in a total of 180 sites surveyed. We estimated annual site-specific density estimates using distance sampling and program DISTANCE and detected a negative short-term trend of 20.6% (±3.8%) from 2005 through 2008. There were no significant changes in densities from 2005 to 2006 or from 2006 to 2007. From 2007 to 2008, an Estuary-wide negative change was detected (-46.0%, ±6.75%) which was driven by a dramatic decrease in South San Francisco Bay (-57.4% ±5.0%).
We tested the power to detect a 10-year trend for five different monitoring scenarios and found that detecting an Estuary-wide negative trend of 13.9% or greater over a 10-year period with 80% power is possible under the current monitoring design (90 sites/year with effort allocated equally among sites). Power is improved when higher density sites are visited more often within a season and lower or zero density sites are visited less often. If effort is reduced to 45 sites/year, the minimum negative trend detectable is 19.9% and for 30 sites/year (representing a bay region or group of restoration/treatment sites) the minimum negative trend detectable increases to 25.0% with 80% power. Detecting trends with 80% power at the “marsh complex” level (assumed to be a cluster of 6 marsh sites/year) is not be feasible over a 10-year period.
We also modeled Clapper Rail presence probability based on individual survey-point data. Models were driven primarily by salinity and proportion of wetlands within 1 km. The probability of Clapper Rail presence increased with wetland proportion up to 0.30 then decreased, while the effect of salinity on Clapper Rail presence was positive.
Ongoing monitoring in 2009 and 2010 will help determine if the drop in population from 2007 to 2008 is sustained. Analysis of the effects of invasive Spartina treatment and other potential environmental factors may help identify the causes of these population changes. Using data from all sources (including East Bay Regional Park District and H.T. Harvey and Associates) we estimated a 2005-2008 average population of 1425 ± 22.
Data collected under TE-807078-10 from USFWS and Memorandum of Understanding between California Dept. of Fish and Game and PRBO (3/14/2003-3/14/2005). In addition to our collaborators (Avocet Research Associates, California Department of Fish and Game, California Coastal Conservancy’s Invasive Spartina Project, U.S. Fish and Wildlife Service, East Bay Regional Park District, and H.T. Harvey and Associates) PRBO benefited greatly from the assistance of a number of other individuals, organizations and agencies: CALFED, California Department of Transportation, City of Palo Alto, City of Redwood City, Loring Dales, Daniel Edelstein, Ingrid Hogle, Rick Keck, Marin Audubon Society, Pacific Gas and Electric Company, Ryan Phelan, Port of Sonoma, San Mateo Department of Transportation, Shell Oil Company, and Vallejo Sanitation and Flood Control District. This is PRBO contribution number xxxx.
Table of Contents
List of Tables
Table 1. Sites surveyed from 2005 to 2008 by Bay with map identification number (for Figure 1), PRBO site code, site area in hectares, and survey type (Appendix 1). 19
Table 2. Five monitoring scenarios used in the power analysis. Sites contain an average of 5.7 listening stations. Scenarios 1, 2, and 3 represent Estuary-wide monitoring scenarios. Scenarios 4 and 5 represent reduced survey efforts. 21
Table 3. DISTANCE vs. Observer-derived estimates of total number of California Clapper Rails, average 2005-08. 22
Table 4. Analysis of inter-annual change and trends in Clapper Rail abundance for San Pablo, South San Francisco, and Central San Francisco Bays, combined and separate. 22
Table 5. Effect of marsh area (natural log-transformed) on Clapper Rail density (natural log-transformed), controlling for annual trend and for Bay Region (San Pablo Bay vs. South San Francisco Bay) including marshes surveyed in 2 or more years where density >0 in 1 or more years, 2005 to 2008. 22
Table 6. Power analysis results from program MONITOR. Power level represents percent chance to detect the stated positive or negative annual and cumulative change. Refer to methods for scenario design details. 23
List of Figures
Figure 1. Map of sites surveyed in San Pablo and Central San Francisco Bays between 2005 and 2008 by PRBO and partners. Sites color-coded by observer-derived density estimate averaged over 2005 to 2008. Site numbers correspond to sites in Table 1 and Appendix 2. 24
Figure 2. Map of sites surveyed in South San Francisco Bay between 2005 and 2008 by PRBO and partners. Sites color-coded by observer-derived density estimate averaged over 2005 to 2008. Site numbers correspond to sites in Table 1 and Appendix 2. 25
Figure 3. Map of sites surveyed in Suisun Bay between 2005 and 2008 by PRBO and partners. Sites color-coded by observer-derived density estimate averaged over 2005 to 2008. Site numbers correspond to sites in Table 1 and Appendix 2. 26
Figure 4. Bay-wide Clapper Rail densities (all sites in San Pablo, Central and South San Francisco Bays), 2005-08. The fitted line assumes a constant percent change per year and is derived from the site by site analysis. The plotted values are shown for illustration and represent pooled mean densities. Error bars represent 1 Standard Error. 27
Figure 5. San Pablo Bay Region Clapper Rail densities, 2005-08. The fitted line assumes a constant percent change per year and is derived from the site by site analysis. The plotted values are shown for illustration and represent pooled mean densities. Error bars represent 1 Standard Error. 27
Figure 6. South San Francisco Bay Region Clapper Rail densities, 2005-08. The fitted line assumes a constant percent change per year and is derived from the site by site analysis. The plotted values are shown for illustration and represent pooled mean densities. Error bars represent 1 Standard Error. 28
Figure 7. Central San Francisco Bay Region Clapper Rail densities, 2005-08. The fitted line assumes a constant percent change per year and is derived from the site by site analysis. The plotted values are shown for illustration and represent pooled mean densities. Error bars represent 1 Standard Error. 28
Figure 8. Predicted California Clapper Rail probability of occurrence based on Maxent model. 29
Figure 9. Maxent-modeled relationship between California Clapper Rail presence and landscape variables (proportion within 1-km radius, except salinity): (a) high-intensity development; (b) low-intensity development proportion; (c) salinity; (d) agriculture; (e) estuarine wetlands; (f) palustrine and estuarine wetlands combined; and (g) elevation. 30
From 2005 to 2008 PRBO Conservation Science (PRBO) and The California Department of Fish and Game (CDFG) in collaboration with Avocet Research Associates (ARA), the California Coastal Conservancy’s San Francisco Estuary Invasive Spartina Project (ISP), U.S. Fish and Wildlife Service (FWS), East Bay Regional Park District (EBRPD), and H.T. Harvey and Associates completed Estuary-wide call count surveys for the California Clapper Rail (Rallus longirostris obsoletus), hereafter Clapper Rail. The primary goal of these surveys was to assess the current population size and trends of the California Clapper Rail. Interannual trends at multiple spatial scales will help identify factors associated with the Clapper Rail’s continued survival.
Clapper Rails have been negatively impacted by a number of historic effects (e.g., loss and degradation of tidal marsh habitat and hunting) and ongoing effects such as pollutants, disturbance, and predation by non-native predators. Habitat alteration, such as the spread and subsequent control of invasive cordgrass (Spartina spp.) may also affect the rail’s Estuary-wide population in a variety of ways. Sea level rise also has the potential to inundate Estuary marshes and drive the population down further. Numerous threats, combined with a small population size characterized by high annual variation, necessitate annual monitoring and critical assessment of threats in order to aid in the recovery of the species.
In this report, we provide an Estuary-wide population estimate, report on short-term trends (2005-2008) and inter-annual changes in abundance at different spatial scales, and analyze the power to detect population trends. We also provide an analysis of spatial patterns of Clapper Rail presence in relation to key landscape and habitat variables as well as recommendations for future monitoring and research.
The California Clapper Rail is one of three subspecies of Clapper Rail recognized by the American Ornithologist’s Union (AOU 1957) and is listed as both a state and federally endangered species. It occurs entirely within the San Francisco Bay Estuary and is dependent on tidal marsh habitat which has decreased over 80% from its historical extent (Goals Project xxxx). Historically, the California Clapper Rail is thought to have been abundant in the Estuary, as “thousands” were reported to have been killed in a single day in 1859 for consumption in San Francisco and Sierra goldfields (Wilbur and Tomlinson 1976). Market hunting was arrested in 1913 (Wilbur and Tomlinson 1976) and California Clapper Rails began re-colonizing marshes in the first half of the 20th century (Grinnell and Miller 1944). The total California Clapper Rail population in the estuary was first estimated in the 1970s at 4,500-6,000 birds (Collins et al. 1994). Based on surveys from the mid-80s, the total population was placed at 1,200 to 1,500 individuals. In 1988 the population estimate dropped to 700 individuals and in 1990-91 the estimate dropped further to 300-500 (Albertson and Evens 2000). In the mid- to late-90s the population appeared to increase to an estimated 1,040 to 1,264 individuals (Albertson and Evens 2000). Predation by red foxes is blamed for the precipitous population decline in the late 1980s, and their ongoing control since then has been credited with the population’s rebound.
Assessing the population status of Clapper Rails is made difficult by the Clapper Rail’s secretive behavior and inconsistent and variable vocalizations. In addition, summarizing decades of surveys and assessing long-term trends is also difficult due to the spatial and temporal variation in survey effort and variation in methods used to collect and store data.
Field Surveys- Surveys by different partners were conducted for different reasons. Consequently, slightly different protocols were used (Appendix 1) but we believe these differences did not seriously affect our results. All of the surveys upon which this report is based were conducted between 19 December and 26 May from 2005 through 2008. All marshes (“sites”) were surveyed 1 to 4 times per year (Table 1) by experienced, permitted biologists. Listening stations were primarily located at marsh edges, levees bordering and within marshes, boardwalks, boat-accessible channels within the marsh, and in the case of 11 marshes in San Pablo Bay, within the marsh itself. Stations were placed 70 to 400 m apart. Sites were located throughout San Pablo Bay, South San Francisco Bay, Central San Francisco Bay and Suisun Bay (Figure 1-3).
PRBO surveyed 53 sites using a call-count method (Type A-PRBO), with 10 minutes per listening station (Appendix 1). San Pablo Bay National Wildlife Refuge surveyed 5 sites using the same method. All Clapper Rails (as well as other rail species, including California Black Rail [Laterallus jamaicencis coturniculus], Virginia Rail [Rallus limicola], and Sora [Porzana carolina]) detected from a listening station were recorded with the time, direction and distance from the listening station. The actual number of rails detected was recorded, or if the detection was not heard clearly because of confounding circumstances (e.g., distance from observer or environmental conditions) a range of number of rails (e.g., 1 to 2, 2 to 4) was recorded. If no Clapper Rails were detected within 200 m of a listening station after 2 passive surveys, playback (up to 1 minute) of Clapper Rail vocalizations was used to stimulate a response on a third survey. Playback surveys consisted of 5 minutes of passive listening (with no Clapper Rails detected), then 1 minute of playback followed by 4 minutes of passive listening. Clapper Rails detected between listening stations and before or after the 10-minute listening period were also recorded, but not used in estimating densities based on distance sampling.
At a total of 99 sites, both ARA and ISP conducted Type A-ISP surveys, which differ from the Type A-PRBO surveys in that every detection, regardless of the actual number of rails detected, was recorded as a range of number of rails (e.g., a single “kek” was recorded as 1-2 Clapper Rails, a “clatter duet” was recorded as 2-4 birds). At 7 sites, surveyors from ARA also employed a stationary method (Type B), remaining at a listening station for 30 to 120 minutes. ISP conducted presence-absence surveys (Type C) at 65 sites judged to have low potential for Clapper Rail (McBroom 2007). Playback of Clapper Rail vocalizations was performed during the first survey (and up to 2 subsequent surveys) if no Clapper Rail was detected in the first 5 minutes of the survey. Surveys were discontinued upon Clapper Rail detection. In Suisun Bay, DFG also performed Type C surveys, consisting of 10 minutes of passive listening at each station then up to 1 minute of playback of Clapper Rail calls followed by 1 minute of listening (Estrella 2007). FWS used a similar method (Type D) at 11 narrow strip marshes with medium to high Clapper Rail densities, conducting 1 to 3 surveys with vocalization playbacks at each listening station. Sites were not resurveyed in the same season if a Clapper Rail was detected. FWS summarized the results of their surveys for use in this report. Surveys of Types A-D were conducted at a total of 178 sites, with more than one type of survey used at some sites in different years (Table 1 and Appendix 2).
H.T. Harvey and Associates provided Clapper Rail densities for six sites that they surveyed (Type H) without using a playback method (H.T. Harvey and Associates 2007). EBRPD also provided summarized data for 5 sites from 2005 that were surveyed using a combination of Type A and B methods (EBRPD unpubl. data).
Winter high tide surveys (Type E) were conducted by FWS and EBRPD at several South San Francisco Bay marshes; data from 2 of these marshes were included in this report (Table 1). During a very high tide, an airboat would traverse a marsh and refugia were examined for Clapper Rails, which were then counted. Airboat survey data were not used in the trend analysis.
Population Estimates- Estuary-wide Clapper Rail population estimates were developed in two different ways. In one method, the program DISTANCE (Buckland et al. 2001) was used to estimate Clapper Rail densities at all marshes with call count data. Distance sampling helps to overcome the problem of a decline in detection probability of a bird with increasing distance from an observer (Thomas et al. 2002). Data from all visits to a listening station were used to create detection curves from which density for each site in each year was calculated. Only detections with a recorded distance or distance range were used. Detection distance ranges were averaged, and when detections were associated with a possible range of number of birds detected (e.g., 1 to 2 Clapper Rails), the lower estimate was used. Density estimates were not applied to sites where call count data were lacking or unavailable.
The other method of estimating population size involved combining all unique detections recorded during a survey of a marsh, regardless of survey methodology and including birds detected outside the 10-minute survey periods at listening stations. This method does not account for the decline in detection probability of a bird with increasing distance from an observer. However, this method was used to incorporate data from a broader range of survey methods. This method may also be more suitable for comparing to population estimates with historic surveys. If a range of individuals was recorded, the lower estimate was always used in order to be consistent with previous efforts to estimate population size and because methods varied for recording the upper range of individuals detected. At sites where marsh coverage at listening stations was greater than 75% (marsh coverage being the area within a 200-m radius around each listening station), observers’ survey results were used directly. At each site where coverage was less than 75%, detections within 200 m of all listening stations for that site were used to calculate a density which was extrapolated over the entire site area. When multiple surveys of a site were conducted in one season, only the survey with the highest count of Clapper Rails for that site was used to calculate the population estimate.
To produce two complete and comparable population estimates (Observer-derived and DISTANCE-derived), winter high tide survey results and “summarized survey results” (count data summarized over the entire marsh site) were added to both estimates for remaining areas. Areas not covered by any surveys were not included in the total population estimate.
Trend Analysis- To analyze the between-year change in Clapper Rail density we used the DISTANCE-derived mean density estimates for each site in each year. We analyzed natural log-transformed mean density; the linear trend in log density thus obtained implies a constant percent change in density over time (Nur et al. 1999). To account for differences in Clapper Rail density among sites, we controlled for the variation in density among sites as a “random effect.” We then performed a mixed-effect linear regression using Stata 10.1 (StataCorp 2009) to fit a common slope for all sites (Estuary-wide trend) or for each region (San Pablo, Central and South San Francisco Bays). Suisun Bay was excluded because of the very low detection rate of California Clapper Rails in that region. The coefficients obtained were back-transformed into percent change per year. We calculated the standard errors of the back-transformed results as the geometric mean of two values: the estimate of the back-transformed “upper” S.E. (= e(Y + 1 s.e.) – eY ) and the estimate of the back-transformed “lower” S.E. (= eY - e(Y + 1 s.e.)), where Y = the estimate from analysis on the log-transformed values. (Note that the back-transformed S.E.s are asymmetric). We used this approach for analyzing trend over the entire time period, 2005 to 2008, and for examining year-to-year changes.
Power Analysis- We performed a power analysis using the DISTANCE-derived density estimates and assessed the statistical power of various monitoring scenarios to detect trends over time. In general, as sample size increases, the ability or power to detect a population trend improves (i.e., a smaller trend can be detected). The analysis presented here seeks to evaluate the statistical power of the current study design as well as to provide guidance for future monitoring programs.
We estimated the magnitude of trend that could be detected, by using the program MONITOR (Gibbs 1995). All trend values refer to annual trends, unless otherwise specified. MONITOR calculates the power associated with a trend by simulating a hypothetical dataset repeatedly (in this case, 1,000 iterations), based on a set of input values and allowing for stochastic variation in the data. For each iteration, the program calculates whether a statistically significant trend was detected given the simulated data. The proportion of trials resulting in a statistically significant result is the measure of statistical power (i.e., probability of detecting a significant trend, given such a trend exists in the data).
The program MONITOR calculates a trend for each sampling unit (using the mean and variability of the simulated data as described above) and then calculates an average trend across all sampling units. MONITOR allows for correlation of trends across units. We assumed that trends across listening stations displayed an intermediate degree of covariation; we reasoned that it is unrealistic to assume that trends for each station are identical across the study area. Thus, MONITOR picks a station-specific trend for each station with a specified mean value but with some variability around that trend specific to a station (i.e., some stations will demonstrate stronger trends than others, but the underlying mean trend is specified).
We investigated our ability to detect population trends under five monitoring scenarios over a projected period of 10 years (Table 2). We varied the level of “effort” (the number of sites surveyed or number of surveys per year or per season at those sites), and we varied the distribution or allocation of those sites among 6 categories of Clapper Rail densities (zero to very low= 0 to 0.030 birds/ha, medium low= 0.031 to 0.128 birds/ha, medium= 0.129 to 0.225 birds/ha, medium high= 0.226 to 0.458 birds/ha, high= above 0.459 birds/ha. For the purposes of this analysis, each site contained the average number of listening stations (5.7). Scenario 1 represents our current effort and monitoring design (90 sites equally distributed among sites with a range of Clapper Rail densities and each surveyed three times per year). Scenario 2 represents a reallocation of our current effort. This scenario examines the improvement in the ability to monitor trends with more effort (more surveys) at high density marshes and less effort (30 very low density sites surveyed on a 3-year cycle of 10 sites per year) at low-density marshes. Scenario 3 represents a 50% reduction in effort (number of sites) but retains an equal distribution of sites varying in Clapper Rail density. Scenario 4 represents a 66% reduction in number of sites to 30 sites, and also retains equal allocation of effort among sites in relation to rail density. This scenario examines the ability to detect trends in one bay region (e.g., South San Francisco Bay) or in one habitat (or treatment) type. Scenario 5 represents a 93% reduction in the number of sites to just 6 sites and examines the ability to detect trends at a complex level (e.g., within a National Wildlife Refuge or Ecological Reserve composed of multiple marshes).
Landscape Analysis- We used presence-only data from each listening station to develop a predictive model based on several environmental variables: point-level salinity and elevation; and the proportion of several land cover types (estuarine wetlands, all wetlands, high-intensity development, low-intensity development, and agriculture) within a 1-km radius area. 2001 land cover data were obtained from the NOAA Coastal Change Analysis Program (C-CAP; http://www.csc.noaa.gov/crs/lca/pacificcoast.html). Land cover grids (30 m) were used to develop continuous moving-window representations (proportion with a 1-km radius) of each land cover type of interest using ArcGIS Spatial Analyst. A 30-m summer salinity grid was generated by interpolating point data from three water quality monitoring data sources: the Integrated Regional Wetlands Monitoring Program (http://irwm.org/), the Department of Water Resources (http://www.iep.ca.gov/suisun/dataReports/index.html), and the San Francisco Estuary Institute (http://www.sfei.org/). We used a simple inverse distance-weighted algorithm (power = 2) within ArcGIS 9.2 to generate the interpolated salinity surfaces. For elevation we used the National Elevation Dataset from USGS (http://ned.usgs.gov/). We used a machine learning algorithm called MAXENT (Phillips et al. 2006) to predict Clapper Rail distributions based on observed presence locations. MAXENT is based on the principle of maximum entropy, and uses information about a known set of species occurrence points, compared with environmental background data, to develop parsimonious models of species occurrence. We allowed linear, quadratic, product, and hinge feature, and built the model on a random 75% of the presence data, reserving 25% for testing. We used an area-under-the-curve (AUC) statistic (Fielding and Bell 1997) to evaluate model predictive power.