PFAS — data to prediction
A national PFAS inventory is only useful if something can compute with it.
The same national layers that draw the map are clipped into a runnable watershed model, a process-based engine moves PFAS through soil and stream, and the result is set against independent agency measurements — including where it disagrees.
The holding
Counts are read live from the platform API on page load. They rise as harvests land and are never hardcoded.
Monitoring tells you where PFAS was. It cannot tell you where it goes.
PFAS management has moved into a regime of enforceable numeric limits — a 2024 U.S. drinking-water maximum contaminant level of 4 ng L-1 for PFOA and PFOS, a CERCLA hazardous-substance designation, and state in-stream criteria. That shifts the question from detection toward reach-scale prediction and source attribution. A sample gives a concentration at one point on one day; it cannot say which reach carries mass downstream, how much came from a given source, or what a remedy would change.
NHDPlus HR hydrography, gSSURGO soils, 3DEP terrain, land cover, climate, and observations.
An outlet or gauge becomes a runnable SWAT+ package, with every reach keyed to stable identifiers.
Soil three-phase equilibrium per HRU and layer, mobilisation through runoff and leaching, channel routing with benthic exchange.
Modeled concentrations set against independent agency measurements — including where they disagree.
Questions about the analysis
What does SWATGenX add beyond PFAS monitoring data?
Monitoring tells you a concentration at one point on one day. It cannot tell you which reach carries mass downstream, how much came from a given source, or what a remedy would change. SWATGenX clips the same national layers that draw the map into a runnable SWAT+ watershed model and routes PFAS through soil, runoff, lateral flow, leaching, sediment and the channel network, so the inventory becomes a prediction rather than a picture.
Is the SWAT+ / MODFLOW 6 coupling two-way for PFAS?
Yes. Water is exchanged in both directions, and so is PFAS: leaching from the SWAT+ soil column loads the MODFLOW 6 groundwater transport source term, and PFAS discharged with groundwater returns to the SWAT+ channels, bound live through the MODFLOW 6 BMI/XMI interface. A given study may still prescribe its source from measurements where that fits the question — the Wurtsmith case does, because the plume is a known legacy release — but that is a per-model configuration, not a limit of the engine.
How well does the model reproduce measured PFOS?
On the Rogue River (USGS 04118500), a model of 590 channels and 17,771 hydrologic response units reproduces the observed mainstem PFOS gradient at a log-space RMSE of 0.15 dex along the seven source-bearing mainstem reaches, and 0.74 dex across all 20 gauged reaches and 29 stations. A 120-member Latin-hypercube ensemble encloses five of the seven mainstem observations. This is a first watershed demonstration rather than a national validation, and expert review is appropriate before decision use.
Where does the groundwater model disagree with observations?
At the former Wurtsmith Air Force Base the model reproduces plume extent well and magnitude poorly. Across 22 validation wells with 11 detections, the simulated plume footprint aligns with observations and there are zero deep false positives, but the highest simulated concentration at wells is 1,590 ng/L against a highest observed value of 425 ng/L, an overprediction of about 3.7 times. The model is useful for screening — which wells to sample and where the plume is not — and it is not offered as a compliance number at a single well.
Why should the PFAS engine be trusted?
It is checked against mathematics as well as against data. The soil three-phase equilibrium solver agrees with a 40-digit reference implementation to a worst-case relative error of 1.7e-7 across 29,160 test cases, and the in-stream routine is numerically identical to the established SWAT+ pesticide channel solver, restricted to its linear-partition subset, to single-precision tolerance. The new physics is the soil column rather than the channel mathematics.
Can this be run on another watershed?
Yes. Every layer behind both worked cases is national. Given a USGS station or an outlet, the platform builds the same model for that basin from NHDPlus High Resolution hydrography, national soils, terrain, land cover and climate, without hand-linking any models.
Every layer behind these two cases is national. Give the platform a USGS station or an outlet and it builds the same model for your basin. What it builds is uncalibrated: it ranks and localises where to look and scopes the problem — triage, not design certification. The over-prediction quantified above is why magnitudes need calibration before you rely on them.