Analysis of Methods to Estimate Abundance of River Cooters, Pseudemys concinna: An Example from the Santa Fe River, Florida
A multitude of different statistical models are commonly used to monitor trends in wildlife populations. Most are used to estimate abundance or survival (or both), and these estimates are then examined over time to infer trends in a population. The choice of which model to use is influenced by the key research question of interest and the types of data available. The accuracy and precision of any estimate from a population model are determined by whether the data meet the model assumptions. We assessed the performance of both closed and open capture–recapture models for determining trends in abundance and survival of River Cooters, Pseudemys concinna, in the Santa Fe River, Florida from 2009–2019. We fit three closed models to estimate abundance, one open model to estimate survival, and two robust design models to estimate both abundance and survival. We then used simulation to generate three datasets that represented different sampling designs, including one that mimics our field data, to assess model performance and compare tradeoffs in sampling design. We recommend using the robust design framework when possible as this design and model estimation returned accurate and precise estimates of abundance and survival. This model estimated survival ranging from 0.69–0.95 and capture probability from 0.21–0.25. This design requires consistent sampling of at least three events per year during a closed period, repeated over at least five years, to estimate survival between years. In situations where samples could not be repeated across years, closed population models are likely the most reliable framework in terms of model precision and accuracy. Overall, sampling designs that allow for repeated sampling and align the biology of the study species and the assumptions of the statistical model are likely the most informative approaches for sampling River Cooters and similar species.
(A) Map of Florida showing the location of Alachua, Columbia, and Gilchrist Counties and the Santa Fe River in relation to the rest of the map. (B) Map of the Santa Fe River showing the start and end points of the stretches of river sampled. (C) Map of the Santa Fe River showing the portions of the river sampled.
(A) Abundance estimates and 95% credible/confidence intervals for sequential Bayesian model and Chapman estimate for 2010, 2011, and 2019 when there were multiple capture events per year. B1 is the first sequential Bayesian estimate of the year; B2 is the second estimate of the year. C1 is the first Chapman estimate of the year; C2 is the second estimate of the year. (B) Abundance estimates and 95% confidence intervals for open and closed models that explicitly account for capture probability; for all models, only Rum Island data were used unless otherwise stated. CJS = Cormack-Jolly-Seber-derived estimates, Huggins = Huggins model in years when capture probability could be estimated by the model (2010, 2011, 2019), Huggins H and Huggins L = years when capture probability had to be assumed, Huggins H is the estimates when p = 0.26, Huggins L is the estimates when p = 0.17, Robust Design = Robust design framework, RD All Sites = Robust design framework using all sites (Ginnie Springs, Poe Springs, Rum Island) when available.
Survival rate estimates and 95% confidence intervals for the two models that we were able to calculate survival estimates, the CJS and robust design, using only Rum Island data.
River discharge plots for the Santa Fe River Fort White location (USGS 02322500) from 2007–2020. (A) Average discharge in cubic feet per second with the average for the period of record (1986–2020) represented by the blue dashed line. (B) Variance in discharge for each year with the blue dashed line representing the average variance over the period of record. (C) Coefficient of variation for each year, with the average for the period of record represented by a blue dashed line. (D) Boxplots of river discharge by month for the period of record (1986–2020). The red points represent the monthly average for the period of record. The blue points represent the average river discharge for the month Tropical Storm Debby hit and the month after in 2012. The yellow points represent the average river discharge for the month Hurricane Irma hit and the month after in 2017.
(A) Yearly abundance estimates using simulated capture histories. Design 1—sampling design with three sampling occasions per year for all 11 years. Design 2—follows the ideal sampling design but accounts for the years in the field data when sampling could not occur due to environmental conditions. Design 3—follows the opportunistic sampling design of the field data. Design 4—follows the opportunistic sampling design of the field data, with 2016 survival estimate replaced with null model estimate. True—the true number of turtles in the population. (B) Yearly survival estimates using simulated capture histories. Design 1—sampling design with three sampling occasions per year for all 11 years. Design 2–follows the ideal sampling design but takes into account the years in the field data when sampling could not occur due to environmental conditions. Design 3—follows the opportunistic sampling design of the field data. Design 4—follows the opportunistic sampling design of the field data, with 2016 estimate replaced with null model estimate. True—the true survival.
Contributor Notes
Associate Editor: M. J. Lannoo.