Optimising sampling and analysis protocols in environmental DNA studies

Abstract

Ecological surveys risk incurring false negative and false positive detections of the target species. With indirect survey methods, such as environmental DNA, such error can occur at two stages: sample collection and laboratory analysis. Here we analyse a large qPCR based eDNA data set using two occupancy models, one of which accounts for false positive error by Griffin et al. (J R Stat Soc Ser C Appl Stat 69: 377-392, 2020), and a second that assumes no false positive error by Stratton et al. (Methods Ecol Evol 11: 1113-1120, 2020). Additionally, we apply the Griffin et al. (2020) model to simulated data to determine optimal levels of replication at both sampling stages. The Stratton et al. (2020) model, which assumes no false positive results, consistently overestimated both overall and individual site occupancy compared to both the Griffin et al. (2020) model and to previous estimates of pond occupancy for the target species. The inclusion of replication at both stages of eDNA analysis (sample collection and in the laboratory) reduces both bias and credible interval width in estimates of both occupancy and detectability. Even the collection of > 1 sample from a site can improve parameter estimates more than having a high number of replicates only within the laboratory analysis.

Figure 1

A schematic representation of the Griffin model demonstrating true and false results at both stages. ψ: species presence; ${\theta }_{11}$: Stage 1 true positive; ${\theta }_{01}$: Stage 1 false negative; ${\theta }_{00}$: stage 1 true negative; ${\theta }_{10}$: stage 1 false positive; $p_{11}$:Stage 2 true positive; $p_{01}$: Stage 2 false negative; $p_{00}$: Stage 2 true negative; $p_{10}$: Stage 2 false positive.

Figure 2

(a) Mean bias in occupancy (ψ) estimate with varying numbers of sites (S), samples collected (M) and qPCR replication (K). (b) Mean 95% PCI width for occupancy (ψ) plotted on a logarithmic scale, with varying numbers of sites (S), samples collected (M) and qPCR replication (K).

Figure 3

(a) Mean Bias in Stage 1 false positives (${\theta }_{10}$) estimate with varying numbers of sites (S), samples collected (M) and qPCR replication (K). (b) Mean 95% PCI width for Stage 1 false positives (${\theta }_{10}$) plotted on a logarithmic scale, with varying numbers of sites (S), samples collected (M) and qPCR replication (K). (c) Mean Bias in Stage 1 true positives (${\theta }_{11}$) estimate with varying numbers of sites (S), samples collected (M) and qPCR replication (K). (d) Mean 95% PCI width for Stage 1 true positives (${\theta }_{11}$) plotted on a logarithmic scale, with varying numbers of sites (S), samples collected (M) and qPCR replication (K).

Figure 4

(a) Mean Bias in Stage 1 false positives ($p_{10}$) estimate with varying numbers of sites (S), samples collected (M) and qPCR replication (K). (b) Mean 95% PCI width for Stage 1 false positives ($p_{10}$) plotted on a logarithmic scale, with varying numbers of sites (S), samples collected (M) and qPCR replication (K). (c) Mean Bias in Stage 1 true positives ($p_{11}$) estimate with varying numbers of sites (S), samples collected (M) and qPCR replication (K). (d) Mean 95% PCI width for Stage 1 true positives ($p_{11}$) plotted on a logarithmic scale, with varying numbers of sites (S), samples collected (M) and qPCR replication (K).

Figure 5

Boxplot of site posterior mean occupancy estimates for the Stratton et al. model (Left) and Griffin et al. model (Right). The median value for all sites, with inter-quartile range, 1.5 inter-quartile range inner fence and suspected outliers presented.

Figure 6

Paired site posterior mean occupancy probabilities for the Stratton et al. model (x-axis) and Griffin et al. model (y-axis), the black line represents the line through the origin.

Figure 7

The posterior conditional probabilities of species absence (1 − ψ (x)), given x amplifying qPCR replicates, calculated using the Griffin et al. model.