Does detecting eDNA mean the species is actually present there?

Tournayre et al. (2025) estimated median transport distances of ~18.6 km from UK air quality monitoring stations. Under certain atmospheric conditions — particularly strong winds or storm events — DNA can travel much further. Lynggaard et al. (2024) found that weather conditions influenced which vertebrate taxa were detected, with windy periods producing detection of taxa that were likely not present in the immediate vicinity.

Under certain conditions — low UV exposure, dry conditions, cool temperatures, enclosed spaces — airborne DNA can persist for hours to days after the source organism has departed. Clare et al. (2022) found DNA from zoo animals detectable inside sealed buildings even after the animals had been moved.

For most field monitoring applications, day-to-day persistence is not a major problem — a detection most likely reflects recent presence (within the past 24–48 hours). But for rare taxa in sensitive regulatory contexts, the possibility that a detection reflects a past visit rather than current residency needs to be acknowledged.

An owl that has eaten a field mouse will deposit mouse DNA through its pellets and faeces. A bat that has caught an insect will exhale and shed DNA from its prey. This secondary transfer is generally considered a minor contribution to the total signal, but it cannot be ruled out entirely — particularly for small invertebrates that appear in the diets of many larger species.

Clare et al. (2022) explicitly detected food items from air sampled in zoo enclosures — demonstrating that the food chain is reflected in the airborne DNA pool. They framed this as a feature (revealing ecological interactions), but it is simultaneously a potential complication for species presence inference.

Food products — particularly processed meat, fish, and dairy — are a primary category. A beef burger at a picnic, packaged smoked salmon in a kitchen, or dried pet food in a barn will all shed DNA into the surrounding air. The DNA detected may be genomically identical to the target wild species (or a domestic relative), making bioinformatic discrimination difficult without strict contextual controls.

Manufactured animal-derived materials are a second category: leather, wool, feathers and down, horn, and bone can all release DNA into the air when disturbed, abraded, or worked. A taxidermied specimen in a museum, a wool blanket in a study site shelter, or feathers in a bird ringer's bag represent potential sources of non-living but genetically authentic DNA.

Facility-proximate contamination is a third category: zoo enclosures, wildlife rehabilitation centres, farm buildings, and abattoirs generate very high concentrations of airborne DNA from captive animals. Samples collected near such facilities may detect species that are genuinely absent from the surrounding wild landscape.

The practical implication is that environmental context must always be documented alongside sampling. Detections of taxa that are inconsistent with the landscape type — particularly domestic species or commercially traded organisms — should trigger scrutiny of potential non-natural sources before a conclusion of wild presence is drawn.

The extreme sensitivity of PCR-based methods means that trace amounts of DNA introduced during extraction, amplification or sequencing can produce false positive detections. This is not unique to airborne eDNA — it affects all eDNA methods — but the small total quantities of DNA in many air samples make contamination controls especially important (Alberdi et al. 2018; Goldberg et al. 2016).

All credible airborne eDNA studies include extraction blanks, PCR negative controls, and field blanks (filters handled in the field but not exposed to air). Any species detection that appears in a negative control must be treated as a potential false positive.

The DNA sequence is real and was genuinely present in the sample, but the identification assigned to it by the computational pipeline is incorrect.

Chimeric sequences arise during PCR when an incomplete extension product from one template re-anneals to a different template in the next cycle, producing a hybrid sequence that does not correspond to any real organism. Chimera-detection algorithms (implemented in tools such as UCHIME, USEARCH, and DADA2) remove most chimeras during bioinformatic processing, but no algorithm eliminates them entirely. In complex mixtures with many taxa — as in airborne eDNA from diverse communities — chimera formation rates are elevated relative to simpler samples.

Tag jumping (also called index switching or tag bleed) occurs when multiple samples are multiplexed on the same sequencing run. Adapter-ligated DNA molecules can swap their identifying indices during cluster generation on Illumina flow cells, causing a read from one sample to be misassigned to another. In airborne eDNA studies, where many low-concentration samples are typically run together, tag jumping can produce phantom detections in samples that should be blank. Rodriguez-Martinez et al. (2025) confirmed that T4 DNA polymerase use during library preparation is a major contributor to the process, and that combinatorial tagging protocols are substantially more susceptible than twin-tagging approaches. Estimated contamination rates range from 0.1–10% of reads depending on library preparation method and sequencing chemistry.

Taxonomic misannotation in reference databases propagates errors from the database into identification results. Reference sequences deposited under incorrect species names, chimeric sequences submitted inadvertently, or sequences from misidentified museum voucher specimens are all present in GenBank and BOLD. A sequencing read that is genuinely from organism A may be assigned to organism B if organism B's incorrectly deposited sequence is the best database match. Tournayre et al. (2025) explicitly flagged this risk for some of their unexpected species detections.

Low-complexity sequences and multi-taxon matches represent a fourth class of error: some genomic regions are so conserved that a short barcode fragment matches equally well to many different taxa. In such cases, the pipeline may assign a taxonomic identity at the wrong confidence level, or collapse the read into an incorrect higher-level taxon. This is particularly relevant for bacterial 16S reads and some plant ITS sequences, where short fragments provide insufficient resolution.</span>