Environmental DNA (eDNA) sampling can provide accurate, cost-effective, landscape-level data on species distributions. Previous studies have compared the sensitivity of eDNA sampling to traditional sampling methods for single species, but similar comparative studies on multi-species eDNA metabarcoding are rare. Using hierarchical species occupancy-detection models, we examined whether key choices associated with eDNA metabarcoding (primer selection, low-abundance read filtering, and the number of positive water samples used to classify a species as present at a site) affect the sensitivity of metabarcoding, relative to backpack electrofishing for fish in freshwater streams. Under all scenarios (teleostei and vertebrate primers; 0%, 0.1% and 1% read filtering thresholds; 1 or 2 positive samples required to classify species as present), we found that eDNA metabarcoding is, on average, more sensitive than electrofishing. Combining vertebrate and teleostei markers resulted in higher detection probabilities relative to the use of either marker in isolation. Increasing the threshold used to filter low abundance reads decreased species detection probabilities but did not change our overall finding that eDNA metabarcoding was more sensitive than electrofishing. Using a threshold of two positive water samples (out of 5) to classify a species as present typically had negligible effects on detection probabilities compared to using one positive water sample. Our findings demonstrate that eDNA metabarcoding is generally more sensitive than electrofishing for conducting fish surveys in freshwater streams, and that this outcome is not sensitive to methodological decisions associated with metabarcoding.

Raw FASTQ files generated from a MiSeq run, prior to any processing.

Sample information (metadata) for both the FISH and VERT datasets have been included - these are needed to perform demultiplexing of the raw FASTQ files.