Spatio‐temporal metapopulation trends: The coconut crabs of Zanzibar

Abstract Species experience a variety of environmental and anthropogenic conditions across their ranges leading to spatial variation in population dynamics. Understanding population dynamics under different conditions is important but it is challenging to allocate limited effort to spatial and temporal subpopulation monitoring. Using GLMMs, we analyze survey data of a metapopulation of coconut crabs spanning 7 years and 15 sites in and near the Pemba archipelago, Zanzibar, to estimate trends in population size (based on catch per unit effort), weight and sex ratio at the meta‐ and subpopulation level and investigate anthropogenic drivers of these trends. We found that the overall metapopulation has remained stable in terms of size and composition over the survey period, but observed diverging trends in population size and sex ratio at some subpopulations. Formal protection of sites was associated with positive population trends. Of nine sites for which we could estimate site‐specific trends, three showed increasing and two decreasing trends, whereas four sites had stable subpopulations. Although anthropogenic factors affected the average weight, and the incidence of small and large individuals, we found no temporal trends in any weight‐related measures. Furthermore, there were no apparent patterns between weight‐related measures and subpopulation trends. The metapopulation was biased toward males, and exploitation appeared to be associated with declining trends in the proportion of females, likely an artifact of a strong decline in the proportion of females in one of only two exploited sites in the dataset. Educational campaigns implemented in 2020 at six sites were not related to higher population sizes in later surveys. The variable trends in subpopulation sizes and composition highlight the need for spatially replicated monitoring in metapopulations. The analyses further provide a detailed baseline for future subpopulation studies of this vulnerable species in one of its last remaining metapopulations in the Western Indian Ocean.


Supporting Information S1: Additional analyses and results
Table S1              Missing symbols indicate that site was not sampled in a particular survey.

Figure S1 :
Figure S1: Residual plot for best Poisson mixed model (with effort as offset, site random effect, survey random effect nested within site, agriculture as main effect and protection interacting with annual trend) fit to nightly coconut crab capture data from 15 sites in/near Pemba.

Figure S2 :
Figure S2: Residual plot for same model as Figure S1, but using a negative binomial instead of a Poisson distribution (the dispersion parameter was >20, suggesting no overdispersion).

Figure S3 :
Figure S3: Residual plot for same model as in Figure S1, but adding a visit-level random effect to account for extra variability.

Figure S4 :
Figure S4: DHARMa test for zero-inflation for model described under Figure S1.No indication for zero-inflation in the data.

Figure S5 :
Figure S5: Residual plot for Poisson mixed model (with effort as offset, survey random effect nested within site, and fixed site-specific intercepts and trends) fit to nightly coconut crab capture data from 9 data-rich sites in/near Pemba.

Figure S6 :
Figure S6: Residual plot for linear mixed model (with random site-specific intercept, sex, protection and exploitation as fixed main effects, and an annual trend) fit to square-root transformed weights (in kg) of coconut crabs from 13 sites in/near Pemba.

Figure S7 :
FigureS7: Residual plot for mixed logistic regression (with random site-specific intercept, sex and protection as fixed main effects, and an annual trend) fit to binary response indicating whether an individual was small (<=0.33kg) or not for coconut crabs from 13 sites in/near Pemba.

Figure S8 :
FigureS8: Residual plot for mixed logistic regression (with random site-specific intercept, sex and hotel as fixed main effects, and an annual trend) fit to binary response indicating whether an individual was large (>=0.97kg) or not for coconut crabs from 13 sites in/near Pemba.

Figure S9 :
Figure S9: Residual plot for mixed logistic regression (with random site-specific intercept, and exploitation interacting with an annual trend) fit to binary response indicating whether an individual was female or not or not for coconut crabs from 13 sites in/near Pemba.

Figure S10 :
Figure S10: Residual plot for logistic regression (with fixed site-specific intercept and trend) fit to binary response indicating whether an individual was female or not or not for coconut crabs from 9 data-rich sites in/near Pemba.

Figure S11 :
Figure S11: Overall catch per unit effort (crabs captured divided by number of locations surveyed) for student projects on the islands of Chumbe (red) and Misali (blue).Note that CPUE from this plot cannot be compared to CPUE obtained in the present study as detailed effort information was not available for these student projects (data obtained from School for International Training unpublished reports).

Figure S12 :
Figure S12: Proportion of very small coconut crabs (<=0.14kg)caught per survey (yearseason combination) across nine data-rich sites in/near the Pemba archipelago, Zanzibar.Missing symbols indicate that site was not sampled in a particular survey.
: Selection of the best effort measure to use as an offset in a Poisson model of number of coconut crabs caught (CPUE) fit to nightly capture data from 15 sites on/near Pemba collected between 2016 and 2023.

Table S2 :
Selection of visit-level covariates affecting coconut crab capture rates.Moon.bin = binary moon phase (1 = 12-16, i.e., near/full moon, and 0 = all other moon phases), RainDay.sc= precipitation the day of the visit (in mm, scaled), RainWeek.sc= total precipitation in the week preceding the visit (in mm, scaled), Rain.bin = binary

Table S3 :
Estimates of site-specific annual population rates of change for coconut crabs (with standard error SE and 95% Wald confidence interval limits, CI.lower and CI.upper) for 9 datarich sites in/near Pemba, based on annual trend estimates from a Poisson mixed model with fixed site-specific intercept and trend.

Table S4 :
Starting proportion of females in the first survey (P) and linear trend (on the logit scale) over time in nine coconut crab populations on/near Pemba.Trend estimates < -0.1 are considered declining, and > 0.1 increasing.Estimates from a logistic regression with fixed sitespecific intercept and trend.