Temporal replication and sampling efficiency in high-frequency autonomous and manual eDNA sampling
Environmental DNA (eDNA) sampling is a valuable tool for ecological research, but its reliability as a quantitative approach can be improved by accounting for unexplained spatiotemporal variability and autocorrelation in eDNA concentrations. Sampling designs with temporal replication can separate sampling variation from temporal variation, but obtaining true temporal replicates is difficult when samples are collected sequentially. Autosamplers may facilitate high-frequency eDNA sampling, but their sequential sampling structure raises questions about when closely spaced samples can reasonably be treated as temporal replicates. This 19.5-day study evaluated temporal variation in invasive Round Goby ( Neogobius melanostomus ) eDNA concentration at a single large-river site with short-term hydrologic variability. We used two synchronized autosamplers and a backpack sampler to collect samples as frequently as every 15 min using polyethersulfone (PES) or mixed cellulose ester (MCE) filters. We fit hierarchical temporal models that decomposed sampling and temporal variation, evaluated evidence for residual temporal correlation under alternative priors for the effective range, and used simulation to assess bias when closely spaced sequential samples were treated as simultaneous temporal replicates. We found evidence of residual temporal correlation at short lag scales, but the effective range of temporal correlation was weakly identified and sensitive to prior assumptions. Hydrologic covariates had limited explanatory support once temporal structure was accounted for. Simulations showed that treating sequential samples collected over 15–60 min windows as simultaneous temporal replicates can bias estimates of sampling and temporal variation, especially when temporal correlation is relatively weak. MCE filters produced higher expected recovered copy numbers than PES filters, while evidence for systematic differences among sampler types was weaker. These results emphasize that eDNA temporal replication should be collected as close together in time as possible, and that uncertainty in temporal correlation should be explicitly considered when designing and analyzing high-frequency eDNA surveys.