3 Results

3.1.1 Capture and spatial recapture rates

A total of 7,210 fecal samples were collected and 6,865 were successfully genotyped (average 95.1% genotyping success), resulting in the identification of 1,755 unique individuals from the seven populations (Table 1). Only four allelic dropout amplification errors occurred (error rate <0.001%). We obtained adequate number of captures, number of unique individuals, number of recaptures, and number of spatial recaptures for all seven populations (Table 1), with the lowest spatial recapture rates being in the smaller populations of Little Smoky and Slave Lake. We had similar recaptures and spatial recaptures rates for females and males (Table S2.1, Table S2.3).

3.1.2 Empirical model performance

Density estimation for both sexes combined and for females had good precision (CV <30%; Table 2, Table S2.1), but modelling males separately led to density estimates for some populations having poor precision (Table S2.2). The average detection probability was low (g0 < 0.06; Table 2) for all populations except the first sampling occasion for Slave Lake (g0t1 = 0.66, g0t2 = 0.036, g0t3 = 0.44). \(\sigma\) differed among populations, ranging from 1,226 m in Slave Lake to 3,363 m in Cold Lake (Table 2).

3.1.3 Assumptions of homogeneous distribution

Results of simulations showed that clustering of caribou detections did not impact the precision or relative bias of the density estimates (Appendix 2). Median density estimates remained similar and slightly above the starting density for all levels of clustering density (\(\mu\)) for the three simulated populations. The simulated Cold Lake population estimates retained the highest precision and were relatively unbiased, despite clustering, which corresponds with the precision found for the empirical model (Table 2). The simulated Little Smoky and Slave Lake population density estimates had lower precision than Cold Lake when caribou were clustered, but median density estimates were not affected by clustering, and density estimates from both populations remained unbiased (Appendix 2). Using a threshold value for precision of CV <30%, Little Smoky and Slave Lake had inadequate median levels of precision at all levels of \(\mu\). These populations had similar (Little Smoky \(\sigma\) = 1600 m) or smaller (Slave Lake\(\sigma\) = 1200 m) \(\sigma\) values compared to the chosen detector spacing of 1500 m (see Appendix 4). The detector spacing of 1500 m for the empirical studies for these populations was too wide relative to\(\sigma\), with very few spatial recaptures of individuals (36 in Little Smoky, 38 in Slave Lake over three occasions), as the detector spacing was larger than \(\sigma\).

3.1.4 Precision and relative bias of reduced sampling designs

In total, 36 different subsampling scenarios were run for each population, for a total of 252 models. Precision and relative bias were positively correlated for all sexes (both sexes r = 0.557,p < 0.0001, female r = 0.597, p< 0.0001, male r = 0.634, p < 0.0001), with decreasing precision (increased CV) and increasing relative bias (divergence from the estimate from the full dataset) with increased transect spacing and reduced number of occasions (Figs 3-4). Several scenarios failed to converge for Little Smoky and Slave Lake at 6 km and 9 km due to low numbers of individuals and no recaptures, resulting in 227 completed models. The Little Smoky and Slave Lake ranges are two of the geographically smallest ranges (Table 1; Fig. 2), and samples in these areas were clustered geographically (Fig. 2). The detection function scaling parameter (\(\sigma\)) for the empirical data for Little Smoky and Cold Lake were smaller than the detector spacing of 1500 m and reducing the number of transects increased the detector spacing even further, leading to the detector spacing being significantly larger than the \(\sigma\) estimates for these populations.
Precision of the subsampling scenarios were influenced by the number of unique individuals, number of recaptures, and number of spatial recaptures (Fig. 5). Precision was negatively correlated with the number of individuals, with precision decreasing with fewer captured individuals (Table S2.5, Fig. 5); all models that failed to run had no recaptures of individuals. The larger ranges of Cold Lake, ESAR, WSAR and Red Earth had more unique individuals than the smaller ranges of Little Smoky, Nipisi and Slave Lake (Fig. 5). When determining the influence of the number of individuals on model precision, all models with three occasions had adequate precision (<30% CV) for both sexes in the larger populations. The number of unique individuals had a greater influence in the smaller ranges, leading to inadequate precision in Little Smoky, Nipisi and Slave Lake (Fig. 5), with no significant correlation between precision and the number of unique individuals in Slave Lake (both sexes) and Little Smoky males (Table S2.5). Precision was negatively correlated with the number of recaptures (Table S2.6) and spatial recaptures (Table S2.7), with lower precision in the smaller populations compared to the larger populations. All models with three occasions for the larger populations fell below the 30% CV threshold for all sex models (Fig. 5). Even when decreasing the number of occasions to two, the larger ranges still performed well with adequate precision, as these subsets still provided an adequate number of recaptures of individuals for the models to run and precision was significantly correlated to the number of recaptures (Table S2.6, Fig. 5). The smaller ranges did not perform as well when the data was reduced to two occasions; several models only retained one recapture of an individual, which resulted in a CV of nearly 100% (Fig. 5) and the number of recaptures or spatial recaptures was not significantly correlated with precision (Slave Lake both sexes, Little Smoky males, Slave Lake males; Table S2.6-Table S2.7).
While there was a strong relationship between precision and the number of individuals and recaptures, this was not the case for relative bias (Tables S2.5-S2.7; Fig. 5). Except for Nipisi (all sexes) and Red Earth females, the number of captures, number of unique individuals, recaptures or spatial recaptures was not significantly correlated with relative bias (Tables S2.5-S2.7). Removing the third session resulted in more bias compared to removing the first and second sessions (Fig. 6).