How much water should I sample for fish eDNA? How many replicates do I need? What’s the best way to take a sample so I can make good comparisons?
These are the most common questions we get about eDNA sampling design at Jonah.
The answers to these questions are no surprise: it depends on your objective.
If your goal is qualitative bioassessment — characterising species composition at a site — you’re aiming for a representative snapshot of the assemblage. A single sample may be sufficient to capture composition qualitatively.
But if your goal involves real-world comparisons of whole assemblages where detection varies over space or time — you need to think differently.
Comparing richness between rivers with different turbidity? Testing the effects of a restoration that shifts biomass? Monitoring a rare species whose abundance changes over time? You will never collect a “perfect” sample that stands alone.
Even with a fixed water volume, individual eDNA samples are not equivalent ecological units.
Why?
•Turbidity may prevent filtering the same volume
•Waters vary in inhibitors or DNA degradation rates
•Hydrology affects transport and mixing
•A flush of DNA from one species can mask others
Two “1-liter” samples are unlikely to represent the same biological window into a community.
That makes single-sample comparisons fragile.
Instead of treating filters as directly comparable, it’s more defensible to treat them as replicate draws from a stochastic detection process.
Inference should rest on the sampling framework — not the success of any one sample.
This is where replication and species accumulation curves become powerful.
Instead of asking:
“Did I filter enough water?”
Ask:
“Have we sampled enough to begin to saturate detection?”
Accumulation approaches allow you to:
•Estimate asymptotic richness
•Standardize effort across sites
•Reduce sensitivity to shifting detection probability
Base comparisons on replicated structure rather than single samples
Which leads to the next question:
How many samples are needed to reliably estimate asymptotic richness?
We don’t really know. Our science hasn’t filled this gap yet. It’s going to be more than one, for sure.
Most importantly, we need a mental shift — from how do I compare individual samples? to how do I compare sampling collections?