Like many wildlife species of management concern, the western snowy plover Charadrius nivosus nivosus is the subject of intensive population monitoring. However, intensive monitoring of reproductive success for this shorebird is time-consuming, financially costly, and potentially disruptive to the birds of interest. These constraints mean that intensive monitoring is not feasible throughout the range of the federally threatened Pacific Coast population. In this study, we used data collected from one intensively monitored subpopulation to assess how reductions in monitoring effort (number of chicks individually marked) would affect the accuracy of estimates of fledging success for western snowy plover chicks. We used monitoring data collected on chicks hatching at 1,845 nests from 2003 to 2012 as a theoretical subpopulation from which to draw random samples for this assessment. As expected, we found that accuracy (as measured by the inverse of percentage difference between sampled and actual fledge rates) increased with increasing percentage of the subpopulation monitored each year. We also found that the day of the week that chicks hatched and were banded had no effect on fledging success. Thus, reducing monitoring effort by banding chicks on specific days of the week is one suitable method for subpopulation sampling that has no embedded biases in the subsequent estimate of fledging success. The results of our analyses provide estimates of the accuracy of different sampling schemes, which should help managers of this threatened shorebird assess appropriate monitoring methods. We recommend use of our methods for others interested in assessing accuracy of sampling schemes for reproductive success of western snowy plover or other birds with similar life-history traits.
Population monitoring is central to effective adaptive management of wildlife populations, especially for recovery of threatened species. Monitoring of immigration, emigration, survival, reproductive success, and other factors limiting recovery is important to inform management decisions and should be planned strategically (Campbell et al. 2002). However, gathering this information can require substantial effort, in terms of both personnel time and funding, and may require trade-offs in terms of study scope, sample size, and data accuracy and precision (Field et al. 2005). Thus, quantitative evaluation of how sampling effort and study design affect accuracy of the results is important for the development of effective and feasible monitoring programs.
The Pacific Coast population of the western snowy plover Charadrius nivosus nivosus is listed as threatened under the US Endangered Species Act (ESA 1973, as amended; USFWS 2007), and like many threatened species, is the subject of considerable monitoring and management. The threatened population of this shorebird nests on sandy beaches and nontidal salt flats along the Pacific Coast from Washington south to central Baja California, Mexico. Nesting occurs from March through early September (Page et al. 2009). Successful reproduction of western snowy plovers requires that the eggs (typically a clutch of three) survive the approximately 30-d incubation period, hatch, and that the precocial chicks survive for a similar interval until fledging (Warriner et al. 1986; Page et al. 2009). Population parameters that provide important information for managers at a regional scale include annual breeding subpopulation size (including estimates for both males and females; we use “subpopulation” here and throughout to mean a regional subset of the entire subspecies population), number of nests, hatching rate of nests, and fledging rate of chicks (USFWS 2007). Monitoring of each parameter presents specific challenges, but monitoring chick fledging success is considerably more difficult than monitoring subpopulation size or nest success because of the movement of precocial chicks (typically cared for by the male parent) away from the nest soon after hatching. Fledging success is most accurately estimated through individual marking of chicks at hatching and monitoring of the brood until the time of fledging. Despite the intensive effort required, fledging success is an important parameter to measure in shorebirds, given that factors affecting hatching and fledging success often differ, and fledging success may have a greater effect on overall reproductive success than hatching success (Neuman et al. 2004; Pauliny et al. 2008; Schulte and Simons 2015).
Western snowy plovers nesting on beaches and coastal salt pannes along the Monterey Bay shoreline in central California have been intensively monitored for >35 y (Warriner et al. 1986; Neuman et al. 2004; Stenzel et al. 2011). As of 2012, this subpopulation of western snowy plovers in Santa Cruz and Monterey Counties accounted for approximately 17% of the Pacific Coast population of the subspecies (USFWS 2012). Through 2012, we monitored fledging success by attempting to locate and determine the fate of all nests, to mark each newly hatched chick with a unique combination of color bands, and to determine the survival of chicks in all broods to 28 d of age. In addition, we individually color-marked a majority of adults as well, either through recruitment of banded chicks to the subpopulation, or through trapping and banding of adults immigrating into the subpopulation (Stenzel et al. 2007, 2011). Management efforts focused on protection of nests from human disturbance and control of predators (Neuman et al. 2004) and resulted in a dramatic local subpopulation increase. After approximately 2012, the subpopulation had grown to the point that it became impractical for us to monitor every nesting attempt and band every hatching chick (499 nests with 390–423 chicks hatching in 2013), indicating the need to develop a sampling approach to estimate fledging success.
In considering possible sampling schemes, we had two primary concerns: 1) how a reduction in monitoring effort would affect the accuracy of our annual estimate of fledging success, and 2) whether a systematic sampling scheme that involved skipping monitoring on certain days of the week would introduce a bias into our estimate of fledging success. We considered the possibility of skipping monitoring on any given day of the week, but we were particularly interested in weekend days (Saturday and Sunday), which would be most desirable for monitors. However, recreational activity on nesting beaches on weekend days can lead to reduced chick survival of western snowy plovers (Ruhlen et al. 2003) and related piping plovers C. melodus (DeRose-Wilson et al. 2018). We also considered the possibility that chicks hatching on other days of the week could have reduced survival if the timing of recreational activity (or other stressors) overlapped with sensitive periods of development. Thus, we were interested in assessing whether survival of chicks varied by hatch day.
Here, we use comprehensive monitoring data from a 10-y period (2003–2012) in a resampling procedure to assess the accuracy of a set of hypothetical sampling schemes in estimating western snowy plover fledging success. First, we assessed the accuracy of estimating fledging success as a function of the proportion of nests monitored. Second, we assessed whether there was a bias in estimates of fledging success related to the day of the week chicks hatched and were banded. The methodological framework and results of these analyses should be useful for managers of western snowy plovers and species with similar life histories to help address three questions: 1) What is the accuracy of the estimate of fledge rate if a certain percentage of broods in a subpopulation is monitored in a given year? 2) What percentage of broods needs to be monitored to achieve a desired level of accuracy in the estimate of fledge rate? 3) Will the estimate of fledge rate be biased if a sampling effort is based on particular days of the week?
We used data on western snowy plover fledging success collected at Monterey Bay area coastal beaches and salt pannes. Nesting habitat here primarily occurs on approximately 30 km of contiguous dune-backed ocean beaches. Nesting beaches are managed by California State Parks and U.S. Fish and Wildlife Service and receive moderate human recreational activity. Managers annually erect cable fencing and signage around the most suitable nesting habitat, resulting in at least minimal protection of most nests. Mammalian and avian predators are the primary causes of nest and chick loss (Neuman et al. 2004; Page et al. 2012).
For this study, we used data collected during the 10-y period of 2003–2012. During this period of intensive monitoring, our goal was to find every nest, band every hatched chick with a unique 4-color combination (Figure 1), and monitor every hatched chick to determine fledging status. We considered a chick to have fledged if it was seen alive on or after the age of 28 d. We banded and monitored 6,246 chicks from 2,338 nests. However, we banded some of these chicks ≥1 d after hatching. To avoid any potential bias associated with banding after the day of hatch (e.g., biased fledging estimates related to some chicks dying before they would have been banded), we used a subset of these data including only nests with ≥1 chick banded on the day it hatched. This filtering of the data resulted in 1,845 nests (Table S1, Supplemental Material) that hatched 4,906 chicks (range = 398–573/y), of which 2,053 fledged (range = 116–258/y). There was occasional uncertainty as to the number of eggs hatched and the number of chicks fledged at an individual nest, resulting in a range of values for these parameters for some nests. When this uncertainty occurred, we used the maximum number of eggs hatched and the minimum number of chicks fledged to produce a minimum fledge rate. We calculated fledge rate as the aggregate minimum number of chicks fledged divided by the aggregate maximum number of eggs hatched, for each year. We used this sample of 1,845 broods to represent a theoretical subpopulation that was fully monitored, giving us a “true” theoretical fledge rate (which we refer to as the theoretical fledge rate).
To assess the effect of monitoring effort on accuracy of the sampled annual fledge rate estimates, we used a resampling scheme to simulate different levels of monitoring effort (percent of broods monitored). We wrote a custom resampling routine in Program R (R Core Team 2017; version 3.4.3) to randomly resample the data set of all nests (N = 1,845; range = 152–215/y) and associated number of eggs hatched and chicks fledged. We calculated fledge rate based on random samples of 10–90%, at 10% intervals, of nests monitored. At each level of effort we 1) selected a random sample (without replacement) of that percentage of nests for a given year, 2) calculated the fledge rate based on that sample, 3) calculated the percentage difference between the sampled fledge rate estimate and the theoretical fledge rate for that year based on the full data set, 4) repeated the first three steps for each of the 10 y of data, and 5) repeated all of the above steps 1,000 times with different random samples chosen. We then calculated error estimate bounds (EEB), by year, calculated as the 95th percentile (950th ranked value) of the 1,000 percentage difference values for a given year. Thus, EEB was an estimate of the maximum typical (0.95 of the time) error (measured as percentage difference) for a given year. Finally, combining data from each of the 10 y, we calculated mean EEB for each level of effort, with 95% confidence limits (CL; ±1.96 standard deviations). In addition to these analyses conducted at intervals of 10% increases in nests monitored, we also calculated mean EEB for two specific levels of sampling: 85.7% and 71.4% (corresponding to skipping 1 and 2 d out of 7 d of monitoring, respectively).
We used generalized linear mixed models to assess whether estimates of fledging success were likely to be biased by the day of the week a chick hatched and was banded. We coded the response variable, fledging success, as either 1 (fledged) or 0 (not fledged) and associated it with the binomial family of error distribution using a logit-link function. We modeled fledging success for each banded chick, and we excluded chicks banded after hatch day (day zero). This resulted in 4,478 banded chicks from 1,845 nests (Table S2, Supplemental Material). We modeled dependence within broods (chicks from the same nest) and variation among years as random effects. The main effect, day of the week, was categorical and stratified to contrast various day of the week combinations (see below). We fit generalized linear mixed models with the “lme4” library and using maximum likelihood (Bates et al. 2015) in Program R (R Core Team 2017).
We formulated a set of candidate models to assess whether systematic bias in the estimation of fledging success was likely to occur under a simple set of reduced banding-effort scenarios. Specifically, we used scenarios in which no banding occurred on either any 1 d of the week or on any 2 consecutive d of the week, which correspond to a 14.3% and a 28.6% reduction in effort, respectively. The candidate set of models (Table 1) included 2 reference models and 14 reduced-effort scenarios. One of the reference models estimates a single, uniform fledging success for all days of the week together (Model 1), and the other reference model (Model 16) estimates fledging success for each day of the week separately. The 14 reduced-effort scenarios included 2 subgroups of 7 models that contrast either 1 d of the week against the remaining 6 d (Models 2–8) or 2 consecutive d of the week against the remaining 5 d (Models 9–15). The 14 reduced-effort scenarios were not nested and differed in the way that days of the week were partitioned. As a consequence, this subset of models was not amenable to model averaging.
We modeled fledging success on the logit scale, so model coefficients were inverse-logit-transformed to generate the corresponding fledging probabilities. We used parametric bootstrapping to generate 95% confidence intervals for these fledging probabilities. We ranked models within the candidate set using Akaike Information Criterion (AIC) differences (ΔAIC = AIC of each model − lowest AIC in model set). We used the criteria outlined by Burnham and Anderson (2002) to characterize empirical support for models within the candidate set as follows: models within 2 ΔAIC of the model with the lowest AIC score have substantial empirical support, models with ΔAIC of 4.0–7.0 have considerably less empirical support, and models with ΔAIC >10.0 have no empirical support.
Accuracy of the fledge rate estimate increased with percentage of the theoretical subpopulation sampled (Figure 2). This analysis shows that if, for example, we wanted to estimate fledging rate in this subpopulation with a desired EEB of 10% (meaning that the fledge rate estimates would be within 10% of the actual fledge rate 95% of the time), we would need to monitor approximately 80% of broods (i.e., hatched nests) to achieve this accuracy with 95% CL. Conversely, if we knew we had monitored, for example, a third (33%) of broods, we can see that the error in our estimate of fledge rate was most likely (95% of the time) within approximately 17% (95% CL = ∼11–25%) difference from the actual fledge rate. With a theoretical sampling scheme corresponding to skipping 1 d out of every 7 d (85.7% effort), EEB was 5.5% (95% CL = 3.3–7.8%), and with a scheme corresponding to skipping 2 d out of every 7 d (71.4% effort), EEB was 8.5% (95% CL = 4.7–12.3%).
Estimates of fledging success were not influenced by the day of the week that chicks hatched. The candidate set of models returned a narrow and uniform set of fledging probabilities that overlapped broadly in their 95% confidence intervals (Table 1). The reference models demonstrate that the overall fledging probability of a chick banded on the day it hatched was 0.41 (95% CI = 0.34–0.49, Model 1) and that fledging probability was independent of day of the week (range of estimated fledging probabilities = 0.39–0.45, Model 16). Similarly, fledging probability was independent of day of the week for all 7 of the reduced-effort scenarios that contrast 1 d of the week against the remaining 6 d (Models 2–8) and for all 7 scenarios that contrast 2 consecutive d of the week against the remaining 5 d (Models 9–15). The data set was large relative to the number of modeled parameters, and after accounting for Year and Nest, all of the proposed 4-parameter models in the candidate set returned a narrow and particularly uniform set of fledging probabilities that were not differentiable from a single estimate of fledging probability for all days combined.
There was substantial empirical support for 15 of the 16 models in the candidate set. The reference models, which included the minimum and maximum number of stratification levels for the main effect, included one of the two top-ranked models (Model 1) and the bottom-ranked model (Model 16). After accounting for Year and Nest, all reduced-effort scenarios (Models 2–18) were characterized by ΔAIC <2.0. For the reduced-effort scenario that contrasted Friday against the remaining 6 d (Model 7), ΔAIC = 0.0, but this model was less parsimonious than the top-ranked model (Model 1). As a consequence, there was no indication that estimates of fledging success would be biased by implementing any of the reduced-effort scenarios considered here.
We found that accuracy of estimates of western snowy plover fledging success declined with declining monitoring effort and that reducing monitoring effort by omitting banding of chicks on weekends or other particular days of the week should not bias estimates of fledging success. Combined, these findings will allow us to both design optimal sampling schemes in the future and to assess accuracy of monitoring results from years after 2012, when monitoring of all broods was no longer feasible. It should be noted that our analyses used a subset of our overall monitoring data from 2003 to 2012, using only broods with ≥1 chicks banded the day they hatched. Given potential biases associated with chicks banded several days after hatching, we have since established a protocol of banding chicks only within 1 d of hatching, and if other chicks or juveniles are banded later for purposes other than monitoring fledging success, those birds will not be considered for estimates of annual fledging success.
In order to monitor a given percentage of hatching nests, it will still be necessary to monitor all or most nests to determine whether they hatch or not. In addition, monitoring hatching success and causes of nest loss provides managers with important information on predator impacts that can be applied toward adaptive management. However, once it has been determined that a nest has hatched, substantial effort can be saved by intensively monitoring a reduced set of chicks. Color-banding and monitoring only a percentage of, but not all, hatching chicks, will not only save effort on the part of the researchers, but will reduce human disturbance to the study subjects and will reduce the risk of band-related injuries (Amat 1999).
Assuming that all western snowy plover nests are monitored to determine hatching success, and monitoring effort is reduced by banding a subsample of hatching chicks, a scheme would still need to be developed to achieve the desired sample. For example, if an accuracy of 10% (including CL) is desired, researchers may thus attempt to monitor 80%, or 4 out of 5, of broods (Figure 2). However, it would not be efficient to simply skip banding and monitoring of every fifth brood. For example, some nests pre-identified to be skipped may be relatively easy to monitor in conjunction with other nests hatching that day, whereas some nests pre-identified for monitoring may be difficult to monitor if responding to that hatching nest was the only reason for researchers to be in the field that day. Reducing effort by omitting entire days from banding would be more efficient, allowing researchers to accomplish other tasks (or take time-off) on those days. Based in part on previous studies (Ruhlen et al. 2003; DeRose-Wilson et al. 2018), we were concerned that increased human use of nesting beaches on weekend (Saturday and Sunday) days might negatively affect survival of recently hatched chicks, and thus might bias estimates of fledging success if chicks hatching on certain days were systematically dropped from the monitoring scheme. However, we found that day of the week that chicks hatched did not affect fledging success. Thus, designing a sampling scheme based on any days of the week, including skipping weekend days, should provide similar accuracy at any given level of sampling. In our theoretical subpopulation, our data indicated that if no chicks hatching on weekends were monitored (a sampling effort of 71.4%), accuracy of the fledging success estimate would be 8.5% different from the actual rate, with a positive CL of 3.8%, for a total error typically within 12.3%. Any sampling scheme based on a reduction of days in the field, or based on proportion of nests monitored, should include appropriate proportional spatial allocation of sampling to avoid geographical biases, as well as proportional temporal allocation throughout the breeding season. Stochastic events such as synchronous depredation of nests could potentially result in uneven sample sizes of hatching nests available to monitor on different days of the week.
Choosing how to reduce sampling effort is difficult and depends entirely on the objectives of the monitoring program (Reynolds et al. 2016). In many cases, concern regarding accuracy of monitoring data for conservation planning is related to power to detect annual trends or differences among management treatments (Steidl et al. 1997; Field et al. 2005). However, understanding the accuracy of estimates for key life-history parameters within a given year may be of interest as well. Long-term trend data are valuable, but data within a single year can provide important information on effects of management that year, and this information is critical for effective adaptive management. For our study, in which western snowy plover fledging success is an important yet time-consuming variable to monitor, understanding the accuracy of this estimate under different levels of sampling effort will help inform allocation of effort.
Our project is based only on one area, with a fairly large subpopulation, and results from our study should not be considered as directly applicable to other western snowy plover subpopulations. Many factors can affect fledging success of western snowy plovers, including mammalian predators, avian predators, and human disturbance (Ruhlen et al. 2003; Neuman et al. 2004; USFWS 2007), and spatial and temporal variation in these factors may potentially affect the accuracy of fledging success estimates. Nevertheless, our results provide a helpful guide for others interested in assessing the influence of sampling effort on the accuracy of estimates in reproductive success of western snowy plovers, and our analytic approach may be similarly employed by researchers working with other wildlife species.
Many intensive long-term ecological studies have amassed large detailed data sets that may allow for retrospective analyses similar to ours. These exploratory analyses can provide important additional information on how to strategically apply effort to monitor and manage wildlife populations more efficiently, potentially reducing both the cost of monitoring and research-associated impacts on wildlife. Despite some similar resampling efforts (e.g., Lahoz-Montfort et al. 2014), resampling of existing data sets appears to be an underutilized method for determining appropriate future sampling effort. We encourage others to consider similar methods to assess accuracy of monitoring efforts for western snowy plovers and other species.
Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.
Table S1. Data file for 1,845 western snowy plover Charadrius nivosus nivosus nests that hatched on Monterey Bay, California, beaches between 2003 and 2012 and were used for analyses in this study. Data include the maximum number of eggs hatching at each nest and minimum number of chicks fledged.
Found at DOI: https://doi.org/10.3996/102019-JFWM-088.S1 (59 KB XLSX).
Table S2. Data file for 4,478 western snowy plover Charadrius nivosus nivosus chicks banded the day they hatched on Monterey Bay, California, beaches between 2003 and 2012 and used for an analysis in this study. Data include the day of the week (DOW) the chick hatched and whether it eventually fledged (1 = fledged, 0 = not fledged).
Found at DOI: https://doi.org/10.3996/102019-JFWM-088.S2 (429 KB XLS).
Reference S1. Page GW, Neuman KK, Warriner JC, Warriner JS, Eyster C, Erbes J, Dixon D, Palkovic A. 2012. Nesting of the snowy plover in the Monterey Bay area, California in 2012. Petaluma, California: PRBO Conservation Science, Unpublished Report.
Reference S2. [USFWS] U.S. Fish and Wildlife Service. 2007. Western snowy plover (Charadrius alexandrinus niovosus) Pacific coast population recovery plan. Sacramento, California: U.S. Fish and Wildlife Service.
Reference S3. [USFWS] U.S. Fish and Wildlife Service. 2012. 2012 Summer window survey for snowy plovers on U.S. Pacific Coast with 2005–2011 results for comparison. Arcata, California: U.S. Fish and Wildlife Service, Unpublished Report.
We thank Gary Page, Jane Warriner, and John Warriner for field and logistical support, and many field observers, particularly Carleton Eyster, Jenny Erbes, and Dave Dixon, for their substantial contribution to the data collection effort. Matt Reiter, the Associate Editor, and two anonymous reviewers provided useful input on this manuscript. This is contribution No. 2292 of Point Blue Conservation Science.
Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.
Citation: Henkel LA, Neuman KK, Stein RW, Stenzel LE. 2020. Assessing accuracy of sampling schemes to estimate western snowy plover reproductive success. Journal of Fish and Wildlife Management 11(1):210–216; e1944-687X. https://doi.org/10.3996/102019-JFWM-088
The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the U.S. Fish and Wildlife Service.