Detecting population declines via monitoring the effective number of breeders (Nb)

Estimating the effective population size and effective number of breeders per year (N-b) can facilitate early detection of population declines. We used computer simulations to quantify bias and precision of the one-sample LDNe estimator of N-b in age-structured populations using a range of published species life history types, sample sizes, and DNA markers. N-b estimates were biased by similar to 5%-10% when using SNPs or microsatellites in species ranging from fishes to mosquitoes, frogs, and seaweed. The bias (high or low) was similar for different life history types within a species suggesting that life history variation in populations will not influence N-b estimation. Precision was higher for 100 SNPs (H approximate to 0.30) than for 15 microsatellites (H approximate to 0.70). Confidence intervals (CIs) were occasionally too narrow, and biased high when N-b was small (N-b < 50); however, the magnitude of bias would unlikely influence management decisions. The CIs (from LDNe) were sufficiently narrow to achieve high statistical power (>= 0.80) to reject the null hypothesis that N-b = 50 when the true N-b = 30 and when sampling 50 individuals and 200 SNPs. Similarly, CIs were sufficiently narrow to reject N-b = 500 when the true N-b = 400 and when sampling 200 individuals and 5,000 loci. Finally, we present a linear regression method that provides high power to detect a decline in N-b when sampling at least five consecutive cohorts. This study provides guidelines and tools to simulate and estimate N-b for age structured populations (https://github.com/popgengui/agestrucnb/), which should help biologists develop sensitive monitoring programmes for early detection of changes in N-b and population declines.

Automated summary

Estimating the effective population size and effective number of breeders per year using computer simulations showed a bias of 5%-10% when using SNPs or microsatellites across different species. However, precision was higher for SNPs than for microsatellites, and confidence intervals were narrow enough to achieve high statistical power in rejecting null hypotheses for different true values of N-b and sample sizes.