A simulation-based evaluation of methods for estimating census population size of terrestrial game species from genetically-identified parent-offspring pairs

Abstract

Estimates of wildlife population size are critical for conservation and management, but accurate estimates are difficult to obtain for many species. Several methods have recently been developed that estimate abundance using kinship relationships observed in genetic samples, particularly parent-offspring pairs. While these methods are similar to traditional Capture-Mark-Recapture, they do not need physical recapture, as individuals are considered recaptured if a sample contains one or more close relatives. This makes methods based on genetically-identified parent-offspring pairs particularly interesting for species for which releasing marked animals back into the population is not desirable or not possible (e.g., harvested fish or game species). However, while these methods have successfully been applied in commercially important fish species, in the absence of life-history data, they are making several assumptions unlikely to be met for harvested terrestrial species. They assume that a sample contains only one generation of parents and one generation of juveniles of the year, while more than two generations can coexist in the hunting bags of long-lived species, or that the sampling probability is the same for each individual, an assumption that is violated when fecundity and/or survival depend on sex or other individual traits. In order to assess the usefulness of kin-based methods to estimate population sizes of terrestrial game species, we simulated population pedigrees of two different species with contrasting demographic strategies (wild boar and red deer), applied four different methods and compared the accuracy and precision of their estimates. We also performed a sensitivity analysis, simulating population pedigrees with varying fecundity characteristics and various levels of harvesting to identify optimal conditions of applicability of each method. We showed that all these methods reached the required levels of accuracy and precision to be effective in wildlife management under simulated circumstances (i.e., for species within a given range of fecundity and for a given range of sampling intensity), while being robust to fecundity variation. Despite the potential usefulness of the methods for terrestrial game species, care is needed as several biases linked to hunting practices still need to be investigated (e.g., when hunting bags are biased toward a particular group of individuals).

Keywords: Fecundity; Genetic markers; Kinship; Mark-recapture; Pedigree; Population monitoring; Wildlife management.

Figure 1. Relative bias (N ˆ/N), mean ± SD) computed for the four kin-based capture-mark-recapture methods.

CKMR and CRE methods estimate adult population size while g-CMR and Moment methods estimate the whole population size and the breeding female population size, respectively. Bias is represented by black circles for the red deer and grey triangles for the wild boar. The long-dashed horizontal line represents the optimal value for an unbiased estimator, estimators above this line are overestimating the true population size while estimators below are underestimating it. Points in the grey area represent estimations within 20% of the true population size.

Figure 2. Relative bias (N ˆ/N), mean ± SD) as a function of the fecundity (ranging from 1 to 12), and the proportion of the population sampled ranging from (0.1 to 0.9) for the (A) CKMR, (B) g-CMR, (C) Moment and (D) CRE methods.

CKMR and CRE methods estimate adult population size while g-CMR and Moment methods estimate the whole population size and the breeding female population size, respectively. The long-dashed horizontal line represents the optimal value for an unbiased estimator, estimators above this line are overestimating the true population size while estimators below are underestimating it. Points in the grey area represent estimations within 20% of the true population size. Results for fecundity > 12 are not shown to increase visibility (original figure can be found in Fig. S2).

Figure 3. Relative bias (N ˆ/N) coefficient of variation (mean ± SD) as a function of the fecundity (ranging from 1 to 12), and the proportion of the population sampled ranging from (0.1 to 0.9) for the (A) CKMR, (B) g-CMR, (C) Moment and (D) CRE methods.

CKMR and CRE methods estimate adult population size while g-CMR and Moment methods estimate the whole population size and the breeding female population size, respectively. The long-dashed horizontal line represents the optimal value of 20% for a precise estimator, estimators above this line are too variable to be useful for wildlife management and conservation. Results for fecundity > 12 are not shown to increase visibility (original figure can be found in Fig. S2).

Figure 4. Relative bias (N ˆ/N), mean ± SD) for a fixed fecundity of 7 with an SD of 0, 2 and 4, and the proportion of the population sampled ranging from (0.1 to 0.9) for the (A) CKMR, (B) g-CMR, (C) Moment and (D) CRE methods.

CKMR and CRE methods estimate adult population size while g-CMR and Moment methods estimate the whole population size and the breeding female population size, respectively. The long-dashed horizontal line represents the optimal value for an unbiased estimator, estimators above this line are overestimating the true population size while estimators below are underestimating it. Points in the grey area represent estimations within 20% of the true population size.

Figure 5. Relative bias (N ˆ/N), mean ± SD) as a function of the fecundity (ranging from 1 to 12), and the number of parent–offspring pairs (POP) sampled for the (A) CKMR, (B) g-CMR, (C) Moment and (D) CRE methods.

CKMR and CRE methods estimate adult population size while g-CMR and Moment methods estimate the whole population size and the breeding female population size, respectively. The long-dashed horizontal line represents the optimal value for an unbiased estimator, estimators above this line are overestimating the true population size while estimators below are underestimating it. Points in the grey area represent estimations within 20% of the true population size. Results for fecundity > 12 are not shown to increase visibility (original figure can be found in Fig. S3).