Short Answer: Yes — and this is one of the most exciting and underexploited aspects of airborne eDNA. By combining DNA detections with atmospheric back-trajectory modelling (tools already used routinely by meteorologists to track pollutants, volcanic ash and pollen), it is possible to estimate where detected organisms likely came from. Sullivan et al. (2025) demonstrated this for a 34-year archive, tracing vertebrate DNA signals back to probable source regions using HYSPLIT modelling. This is not yet standard practice in eDNA studies, but the tools exist and the potential is transformative for spatial biodiversity assessment.
1. The source attribution problem
A detection of hedgehog DNA in an air sample tells us that hedgehog DNA was in the air. But where did the hedgehog that shed that DNA actually live? Was it in the field immediately next to the sampler, or did the DNA arrive on the wind from a woodland 5 km away? Did the DNA come from a single animal, or from multiple individuals across a landscape?
This is the source attribution problem, and it is fundamental to using airborne eDNA for spatial biodiversity mapping. Without knowing the source region, a detection is of limited value for conservation management: you cannot protect a population you cannot locate.
Fortunately, atmospheric science has developed powerful tools for exactly this purpose, developed not for eDNA but for tracking pollutants, radioactive material, pollen and volcanic ash.
2. Atmospheric back-trajectory modelling
2.1 The HYSPLIT model
The Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model, developed by NOAA's Air Resources Laboratory (Stein et al. 2015), is the most widely used atmospheric trajectory model in the world. It computes the path a parcel of air has followed before arriving at a given location and time, using archived or forecast meteorological data.
HYSPLIT back-trajectories have been used for decades to:
- Source atmospheric pollutants (e.g. PM2.5 episodes to specific industrial emitters)
- Track long-range pollen transport from specific source regions (Skjøth et al. 2007; Sofiev et al. 2006)
- Trace radioactive material from nuclear accidents (Chernobyl, Fukushima)
- Track volcanic ash clouds for aviation safety
The application to airborne eDNA source attribution was first demonstrated systematically by Sullivan et al. (2025), who computed adjoint (receptor-based) dispersion models using HYSPLIT to estimate the geographic origin density of 5, 22, and 60 µm particles captured at their Swedish radionuclide station.
2.2 What Sullivan et al. (2025) found
The Sullivan et al. (2025) catchment analysis produced the following results:
2.3 The FLEXPART and SILAM models
Other atmospheric dispersion models used in aerobiology and environmental research include FLEXPART (a Lagrangian particle dispersion model: Stohl et al. 2005; Pisso et al. 2019) and SILAM (System for Integrated modelling of Atmospheric coMposition, developed at the Finnish Meteorological Institute: Sofiev et al. 2015). SILAM has been used extensively for pollen forecast modelling across Europe and could be adapted for eDNA catchment analysis. The existence of multiple validated models increases confidence in the approach and allows cross-validation.
3. Lessons from pollen aerobiology: 70 years of source attribution experience
The airborne eDNA field is not starting from scratch in source attribution. Aerobiology has accumulated 70 years of experience tracking pollen from known source regions to receptor stations, validating trajectory models against observations, and quantifying source-receptor relationships.
Key lessons from aerobiology directly applicable to airborne eDNA:
4. The geographic source estimation approach
Lennartz et al. (2021) demonstrated a complementary approach to source attribution that does not require atmospheric modelling. They used the distribution of plant DNA in airborne dust samples from different geographic regions to estimate the origin of the dust — essentially using plant community composition as a geographic fingerprint. This has potential forensic applications (geographic origin estimation from dust samples) and ecological applications (characterising regional vegetation composition from deposited dust).
This approach is analogous to pollen analysis in archaeology and palynology, where the assemblage of pollen types in a sample is used to reconstruct the vegetation of the source region. For airborne eDNA, the diversity and composition of detected organisms could serve as a biogeographic fingerprint.
5. Practical catchment estimation: the Tournayre approach
Tournayre et al. (2025) used a simpler but practically useful approach to catchment estimation: comparing eDNA detections with iNaturalist citizen science records at different radii around each UK air quality monitoring station. By finding the radius at which eDNA and iNaturalist detection rates converged, they estimated a median catchment radius of approximately 18.6 km.
This approach does not require atmospheric modelling and can be applied retrospectively to any dataset where co-located citizen science records exist. Its limitations are that it provides a landscape average rather than a directional, meteorologically-informed catchment, and that it depends on the quality and completeness of citizen science data in the surrounding area.
6. Current limitations and research needs
HYSPLIT has important limitations when applied to settling particles. The model was designed for passive scalar tracers — gases and fine particles with negligible settling velocity. For particles larger than ~20 µm (which includes most pollen grains and larger biological particles), gravitational settling during transport is non-trivial and HYSPLIT's standard trajectory calculation will underestimate deposition along the path, tending to overestimate how far such particles travel. The adjoint dispersion approach used by Sullivan et al. (2025) partially addresses this by computing receptor-oriented source contributions rather than simple forward trajectories, and by running separate models for different particle sizes (5, 22, and 60 µm). However, researchers working with eDNA from large-particle taxa should be aware that HYSPLIT source attribution is most reliable for particles below ~20 µm. FLEXPART is a validated alternative that handles particle settling and deposition more explicitly in some configurations and may be preferable for heavier particles.
Back-trajectory uncertainty grows rapidly with time. Trajectories computed forward or backward beyond 48–72 hours have positional uncertainties of hundreds of kilometres, driven by accumulated meteorological analysis errors. Sullivan et al. (2025) used 120-hour back-trajectories in some analyses, which is at the outer edge of reliability for individual event attribution. Results at this time horizon should be interpreted as identifying probable source regions in a statistical sense, not precise source points. For individual detection events, back-trajectory analysis beyond 72 hours should be treated as indicative rather than definitive.
Transport models currently do not incorporate DNA degradation. HYSPLIT, FLEXPART, and related atmospheric dispersion models simulate particle movement — where particles physically travel, how they dilute, and where they deposit. They do not incorporate DNA decay rates. A particle that the model places within the source catchment may have arrived with its DNA intact and detectable, or may have arrived too degraded to amplify. The modelled catchment is therefore a physical source envelope — the region from which particles could have arrived — not a detectable source envelope, which would be a subset of that region constrained by how long DNA remains amplifiable in transit under the actual atmospheric conditions (UV exposure, temperature, humidity) along the trajectory. For macrobial airborne eDNA, this gap has not yet been quantified. The practical implication: back-trajectory outputs should be read as upper bounds on source area, not precise maps of where detectable organisms were present. Integrating degradation kinetics into dispersion models is a priority research gap identified by the field (Tulloch et al. 2025).
Source attribution modelling is not yet standard practice. Only Sullivan et al. (2025) has published systematic source attribution modelling for macrobial airborne eDNA. The tools exist and are validated, but their integration into routine eDNA monitoring pipelines requires collaboration between ecologists and atmospheric scientists — a cross-disciplinary bridge not yet commonly built.
Particle size distribution of eDNA-carrying particles is incompletely known. HYSPLIT modelling requires knowing the particle size of the carrier particles. For pollen, this is well-established. For vertebrate skin fragments or faecal particles, the relevant size distribution has not been systematically measured for airborne eDNA applications.
Reference biodiversity data is needed for validation. Validating source attribution requires independent biodiversity data (e.g. from camera traps, acoustic monitoring, or structured field surveys) at potential source locations. Building these validation datasets is resource-intensive.
7. The spatial mapping opportunity
The combination of atmospheric source attribution modelling with airborne eDNA detection opens a genuinely novel possibility: constructing spatially explicit biodiversity maps from fixed or moving sampling stations, without any field surveys in the source area.
Imagine a network of thousands of air sampling stations across Europe, each producing weekly samples. For each week, back-trajectory analysis assigns each detection a probability distribution of source locations. Summed across many weeks and many stations, this produces a landscape-scale biodiversity map with spatial resolution determined by the density of monitoring stations and the precision of the atmospheric model. Over time source allocation will become better and a fascinating new field of bioscience will emerge.
This is not yet operational, but all the scientific components to build it exist. Sullivan et al. (2025) have demonstrated the approach at a single fixed station over three decades. The next step is multi-station, multi-year validation.
References
- Sullivan AR et al. (2025). Airborne eDNA captures three decades of ecosystem biodiversity. Nature Communications 16:11281. https://doi.org/10.1038/s41467-025-67676-7
- Stein AF et al. (2015). NOAA’s HYSPLIT atmospheric transport and dispersion modeling system. BAMS 96:2059–2077. https://doi.org/10.1175/BAMS-D-14-00110.1
- Sofiev M et al. (2006). Towards numerical forecasting of long-range air transport of birch pollen. International Journal of Biometeorology 50:392–402. https://doi.org/10.1007/s00484-006-0027-x
- Tournayre O et al. (2025). First national survey of terrestrial biodiversity using airborne eDNA. Scientific Reports. https://doi.org/10.1038/s41598-025-03650-z
- Skjøth CA et al. (2007). Long-range transport of birch pollen from Poland and Germany causes pre-season concentrations in Denmark. Clinical & Experimental Allergy 37:1204–1212. https://doi.org/10.1111/j.1365-2222.2007.02771.x
- Lennartz C et al. (2021). Geographic source estimation using airborne plant environmental DNA in dust. Scientific Reports 11:16238. https://doi.org/10.1038/s41598-021-95702-3
- Stohl A et al. (2005). Technical note: The Lagrangian particle dispersion model FLEXPART version 6.2. ACP 5:2461–2474. https://doi.org/10.5194/acp-5-2461-2005
- Pisso I et al. (2019). The Lagrangian particle dispersion model FLEXPART version 10.4. Geoscientific Model Development 12:4955–4997. https://doi.org/10.5194/gmd-12-4955-2019
- de Weger LA et al. (2016). The long distance transport of airborne Ambrosia pollen to the UK and the Netherlands from Central and south Europe. International Journal of Biometeorology 60:1829–1839. https://doi.org/10.1007/s00484-016-1170-7
- Sofiev M et al. (2015). MACC regional multi-model ensemble simulations of birch pollen dispersion in Europe. Atmospheric Chemistry and Physics 15:8115–8130. https://doi.org/10.5194/acp-15-8115-2015