Environmental DNA (eDNA) is rapidly expanding as a survey tool, while a renaissance of new applications in terrestrial ecosystems has created the need for empirical and theoretical development. Here we provide a framework for informing field deployment of active terrestrial eDNA assays, using an insect pest (Halyomorpha halys) of agriculture as a case study. Our approach consisted of field trials coupled with simulation. We planted known quantities of eDNA (1-16 drops of fecal slurry) on 1-meter sections of tomato vines and sampled for eDNA by conducting ‘spray aggregation’ for 12-120 seconds per section. We used a logistic model to describe how detection probability changed with sampling effort and DNA density. We simulated 1.2 million field surveys with various combinations of effort, abundance, and spatial aggregation assumptions. Each sampling unit received some number of H. halys individuals drawn from a negative binomial distribution, some number of DNA deposits based on a Poisson distribution per individual, and a binomial detection outcome based on the equation-derived detection probability. The number of samples required until first detection was then summarized by effort, abundance, and level of aggregation.
Results/Conclusions
The number of 1-m samples required to achieve at least one detection 95% of the time (‘samples required’) varied from 49 to > 10,000 m, with a median over all scenarios of 203 m. Similar to other rare-species detection studies, we observed a strongly non-linear relationship between abundance and samples required to confirm presence, with a sharp increase noted as mean densities approached 1/1000 m. Effort and spatial aggregation showed approximately linear relationships with samples required and contributed much less to overall variability in confirmed detection. Our approach is easily adapted to eDNA surveys for other terrestrial species (e.g., forest pests, threatened species), with the most important consideration being the choice of an appropriate sampling unit that can be scaled up via simulation. Refinements will result from ongoing efforts to parameterize factors affecting eDNA deposition and degradation rates, detection probability (e.g., capture method, PCR inhibitors), and individual behavior (e.g., habitat preferences). Sensitivity analyses will inform which sources of uncertainty are most likely to affect survey outcomes for a given level of sampling effort. Extending our approach of combining field trials and simulations could ultimately also aid in cost-efficiency analyses, an important step towards operationalizing such emerging technologies.