PFAS in private well water — state and aquifer breakdown
Predicted probability of detecting any PFAS compound at typical private-well drinking-water depths, across the continental United States.
In 2024, USGS published the first national-scale model of PFAS occurrence in U.S. groundwater at the depths actually used for drinking water (Tokranov et al., Science, October 2024). The model uses an XGBoost classifier trained on 1,238 wells with measured PFAS data across 21 compounds. Inputs include depth-to-water, soil clay content, septic density, population density, urban land use, groundwater recharge, and distance to known PFAS-emitting facilities. Output is a probability surface at 1 km resolution: at every 1 km × 1 km cell in the contiguous United States, the predicted probability that at least one PFAS compound would be detected if you drilled a private well to typical residential depth.
The probabilities below are aggregated from that surface over two geographies: U.S. states (Census TIGER 2024 boundaries) and the 63 USGS Principal Aquifers (the named groundwater bodies that supply most U.S. well water). "Mean" is the average probability across the geography; "% area > 50%" is the share of that geography where the model predicts greater than 50% chance of any detection — the share of land where a freshly drilled domestic well is more likely than not to have detectable PFAS.
Top 10 states by predicted PFAS probability (private-well depth)
| Rank | State | Mean probability | % area > 25% | % area > 50% | % area > 75% |
|---|---|---|---|---|---|
| 1 | District of Columbia | 0.60 | 89.7% | 66.2% | 30.9% |
| 2 | Connecticut | 0.56 | 79.0% | 57.3% | 37.1% |
| 3 | Rhode Island | 0.55 | 81.4% | 55.5% | 29.8% |
| 4 | New Jersey | 0.54 | 82.6% | 52.7% | 28.4% |
| 5 | Delaware | 0.44 | 83.9% | 34.2% | 7.9% |
| 6 | Massachusetts | 0.43 | 65.0% | 38.4% | 18.1% |
| 7 | Florida | 0.32 | 47.2% | 20.8% | 9.5% |
| 8 | Ohio | 0.31 | 45.4% | 18.3% | 7.2% |
| 9 | North Carolina | 0.30 | 46.7% | 19.8% | 5.7% |
| 10 | Michigan | 0.29 | 44.9% | 16.1% | 5.5% |
The Northeast Corridor — DC, Connecticut, Rhode Island, New Jersey, Delaware, Massachusetts — dominates the high end. Old industrial activity, dense population, military bases, and shallow groundwater all push these states toward higher predicted detection rates. New Jersey's reputation as the most-studied PFAS state in the country is borne out in this national model.
Full state ranking
| Rank | State | Mean | % > 25% | % > 50% | % > 75% |
|---|---|---|---|---|---|
| 1 | District of Columbia | 0.60 | 89.7% | 66.2% | 30.9% |
| 2 | Connecticut | 0.56 | 79.0% | 57.3% | 37.1% |
| 3 | Rhode Island | 0.55 | 81.4% | 55.5% | 29.8% |
| 4 | New Jersey | 0.54 | 82.6% | 52.7% | 28.4% |
| 5 | Delaware | 0.44 | 83.9% | 34.2% | 7.9% |
| 6 | Massachusetts | 0.43 | 65.0% | 38.4% | 18.1% |
| 7 | Florida | 0.32 | 47.2% | 20.8% | 9.5% |
| 8 | Ohio | 0.31 | 45.4% | 18.3% | 7.2% |
| 9 | North Carolina | 0.30 | 46.7% | 19.8% | 5.7% |
| 10 | Michigan | 0.29 | 44.9% | 16.1% | 5.5% |
| 11 | Kentucky | 0.28 | 45.2% | 16.1% | 3.7% |
| 12 | New York | 0.28 | 40.2% | 16.1% | 6.9% |
| 13 | Pennsylvania | 0.28 | 42.5% | 14.9% | 4.5% |
| 14 | Tennessee | 0.27 | 39.3% | 15.7% | 4.2% |
| 15 | Maryland | 0.26 | 36.8% | 16.3% | 4.9% |
| 16 | South Carolina | 0.26 | 36.4% | 16.9% | 6.4% |
| 17 | Indiana | 0.25 | 34.3% | 11.5% | 3.7% |
| 18 | West Virginia | 0.25 | 36.5% | 11.1% | 3.6% |
| 19 | North Dakota | 0.25 | 38.1% | 15.4% | 3.0% |
| 20 | New Hampshire | 0.24 | 35.9% | 10.4% | 1.7% |
| 21 | Alabama | 0.24 | 35.3% | 13.5% | 4.3% |
| 22 | Virginia | 0.22 | 29.1% | 12.0% | 3.2% |
| 23 | Vermont | 0.21 | 29.2% | 4.1% | 0.8% |
| 24 | Georgia | 0.21 | 27.6% | 11.8% | 4.1% |
| 25 | Minnesota | 0.21 | 27.1% | 11.2% | 3.3% |
| 26 | Wisconsin | 0.20 | 25.5% | 8.4% | 1.5% |
| 27 | Louisiana | 0.20 | 28.7% | 6.3% | 0.8% |
| 28 | Oklahoma | 0.20 | 23.2% | 6.4% | 2.3% |
| 29 | Kansas | 0.18 | 21.6% | 4.3% | 1.4% |
| 30 | Maine | 0.18 | 19.4% | 5.7% | 1.6% |
| 31 | Nebraska | 0.18 | 18.4% | 4.5% | 0.9% |
| 32 | Illinois | 0.18 | 19.7% | 7.1% | 3.0% |
| 33 | Texas | 0.17 | 20.3% | 6.5% | 1.5% |
| 34 | South Dakota | 0.17 | 23.3% | 3.0% | 0.4% |
| 35 | Arkansas | 0.16 | 19.7% | 4.3% | 0.9% |
| 36 | California | 0.16 | 14.6% | 6.1% | 3.0% |
| 37 | Washington | 0.15 | 15.9% | 6.6% | 3.5% |
| 38 | Missouri | 0.15 | 17.8% | 3.8% | 0.9% |
| 39 | Arizona | 0.14 | 11.2% | 3.1% | 1.1% |
| 40 | Colorado | 0.14 | 12.3% | 2.9% | 0.9% |
| 41 | Mississippi | 0.13 | 13.0% | 2.2% | 0.2% |
| 42 | Montana | 0.13 | 11.8% | 2.1% | 0.2% |
| 43 | Nevada | 0.13 | 6.5% | 1.3% | 0.5% |
| 44 | New Mexico | 0.11 | 7.6% | 2.0% | 0.6% |
| 45 | Wyoming | 0.11 | 6.4% | 0.6% | 0.1% |
| 46 | Oregon | 0.11 | 8.6% | 2.5% | 0.7% |
| 47 | Iowa | 0.11 | 7.6% | 1.9% | 0.7% |
| 48 | Utah | 0.10 | 4.6% | 0.9% | 0.4% |
| 49 | Idaho | 0.10 | 6.5% | 1.2% | 0.3% |
Alaska, Hawaii, Puerto Rico, and the territories are outside the model's continental U.S. coverage and are not ranked.
Top 10 USGS Principal Aquifers by predicted PFAS probability
| Rank | Aquifer | Mean | % > 25% | % > 50% | % > 75% |
|---|---|---|---|---|---|
| 1 | Early Mesozoic basin aquifers | 0.44 | 61.1% | 40.7% | 21.7% |
| 2 | Puget Sound aquifer system | 0.42 | 55.7% | 40.7% | 27.4% |
| 3 | Biscayne aquifer | 0.41 | 46.8% | 30.3% | 24.6% |
| 4 | Piedmont and Blue Ridge carbonate-rock aquifers | 0.41 | 62.3% | 33.8% | 13.4% |
| 5 | Ordovician aquifers | 0.40 | 68.8% | 31.0% | 8.9% |
| 6 | Willamette Lowland basin-fill aquifers | 0.37 | 60.6% | 31.1% | 13.4% |
| 7 | Valley and Ridge carbonate-rock aquifers | 0.37 | 60.5% | 29.4% | 8.4% |
| 8 | California Coastal Basin aquifers | 0.37 | 46.5% | 33.6% | 20.3% |
| 9 | New York sandstone aquifers | 0.36 | 63.9% | 21.9% | 6.0% |
| 10 | New York and New England carbonate-rock aquifers | 0.33 | 51.6% | 21.2% | 7.3% |
Linked names go to our hub page for that aquifer; unlinked aquifers are listed under the USGS name we haven't yet built a dedicated hub for.
Full aquifer ranking
| Rank | Aquifer | Mean | % > 25% | % > 50% | % > 75% |
|---|---|---|---|---|---|
| 1 | Early Mesozoic basin aquifers | 0.44 | 61.1% | 40.7% | 21.7% |
| 2 | Puget Sound aquifer system | 0.42 | 55.7% | 40.7% | 27.4% |
| 3 | Biscayne aquifer | 0.41 | 46.8% | 30.3% | 24.6% |
| 4 | Piedmont and Blue Ridge carbonate-rock aquifers | 0.41 | 62.3% | 33.8% | 13.4% |
| 5 | Ordovician aquifers | 0.40 | 68.8% | 31.0% | 8.9% |
| 6 | Willamette Lowland basin-fill aquifers | 0.37 | 60.6% | 31.1% | 13.4% |
| 7 | Valley and Ridge carbonate-rock aquifers | 0.37 | 60.5% | 29.4% | 8.4% |
| 8 | California Coastal Basin aquifers | 0.37 | 46.5% | 33.6% | 20.3% |
| 9 | New York sandstone aquifers | 0.36 | 63.9% | 21.9% | 6.0% |
| 10 | New York and New England carbonate-rock aquifers | 0.33 | 51.6% | 21.2% | 7.3% |
| 11 | Northern Atlantic Coastal Plain aquifer system | 0.31 | 46.5% | 21.9% | 9.1% |
| 12 | Castle Hayne aquifer | 0.30 | 46.3% | 20.0% | 5.6% |
| 13 | Marshall aquifer | 0.30 | 45.1% | 16.0% | 6.9% |
| 14 | Central Oklahoma aquifer | 0.30 | 40.8% | 18.6% | 11.2% |
| 15 | Piedmont and Blue Ridge crystalline-rock aquifers | 0.30 | 43.8% | 21.1% | 7.3% |
| 16 | Valley and Ridge aquifers | 0.28 | 41.7% | 17.0% | 5.2% |
| 17 | Floridan aquifer system | 0.28 | 41.7% | 14.1% | 5.4% |
| 18 | Pennsylvanian aquifers | 0.27 | 41.5% | 13.7% | 4.3% |
| 19 | Silurian-Devonian aquifers | 0.24 | 32.4% | 12.8% | 5.0% |
| 20 | Mississippian aquifers | 0.23 | 32.6% | 12.4% | 3.8% |
| 21 | Jacobsville aquifer | 0.23 | 36.1% | 8.3% | 1.2% |
| 22 | Surficial aquifer system | 0.22 | 29.9% | 11.8% | 4.3% |
| 23 | Blaine aquifer | 0.21 | 27.4% | 10.4% | 1.3% |
| 24 | Coastal lowlands aquifer system | 0.21 | 28.4% | 8.5% | 2.1% |
| 25 | Southeastern Coastal Plain aquifer system | 0.20 | 24.8% | 7.7% | 2.2% |
| 26 | Central Valley aquifer system | 0.19 | 22.0% | 8.0% | 3.8% |
| 27 | Denver Basin aquifer system | 0.19 | 23.1% | 10.1% | 5.0% |
| 28 | Seymour aquifer | 0.19 | 23.1% | 9.7% | 1.6% |
| 29 | Mississippi River Valley alluvial aquifer | 0.19 | 27.0% | 5.7% | 0.5% |
| 30 | Cambrian-Ordovician aquifer system | 0.19 | 22.8% | 8.4% | 2.1% |
| 31 | Upper Cretaceous aquifers | 0.19 | 24.3% | 9.7% | 2.5% |
| 32 | Snake River Plain basin-fill aquifers | 0.18 | 21.0% | 7.8% | 3.7% |
| 33 | Pecos River Basin alluvial aquifer | 0.18 | 25.0% | 6.2% | 0.9% |
| 34 | Columbia Plateau basin-fill aquifers | 0.17 | 16.3% | 4.7% | 2.0% |
| 35 | Ozark Plateaus aquifer system | 0.16 | 19.9% | 3.8% | 0.9% |
| 36 | Rio Grande aquifer system | 0.16 | 14.8% | 4.7% | 2.1% |
| 37 | Rush Springs aquifer | 0.16 | 13.0% | 1.5% | 0.4% |
| 38 | Southern Nevada volcanic-rock aquifers | 0.16 | 7.8% | 4.9% | 0.1% |
| 39 | High Plains aquifer | 0.15 | 14.2% | 3.4% | 0.8% |
| 40 | Northern Rocky Mountains Intermontane Basins aquifer system | 0.15 | 15.4% | 3.9% | 0.9% |
| 41 | Lower Tertiary aquifers | 0.15 | 16.9% | 5.2% | 0.8% |
| 42 | Basin and Range basin-fill aquifers | 0.14 | 10.0% | 3.0% | 1.2% |
| 43 | Mississippi embayment aquifer system | 0.14 | 12.8% | 3.7% | 0.9% |
| 44 | Upper Tertiary aquifers | 0.13 | 9.7% | 0.6% | 0.1% |
| 45 | Ada-Vamoosa aquifer | 0.13 | 8.2% | 1.3% | 0.4% |
| 46 | Pacific Northwest basin-fill aquifers | 0.13 | 11.3% | 2.9% | 0.8% |
| 47 | Lower Cretaceous aquifers | 0.13 | 10.7% | 2.3% | 0.7% |
| 48 | Edwards-Trinity aquifer system | 0.13 | 12.0% | 2.7% | 0.6% |
| 49 | Arbuckle-Simpson aquifer | 0.12 | 3.8% | 0.1% | 0.0% |
| 50 | Texas coastal uplands aquifer system | 0.12 | 10.4% | 2.2% | 0.5% |
| 51 | Roswell Basin aquifer system | 0.12 | 11.3% | 2.3% | 0.9% |
| 52 | Colorado Plateaus aquifers | 0.11 | 6.0% | 0.7% | 0.1% |
| 53 | Paleozoic aquifers | 0.10 | 5.1% | 0.4% | 0.1% |
| 54 | Basin and Range carbonate-rock aquifers | 0.10 | 2.1% | 0.3% | 0.0% |
| 55 | Columbia Plateau basaltic-rock aquifers | 0.10 | 5.1% | 0.9% | 0.2% |
| 56 | Pacific Northwest basaltic-rock aquifers | 0.09 | 4.5% | 0.7% | 0.1% |
| 57 | Snake River Plain basaltic-rock aquifers | 0.09 | 3.6% | 0.6% | 0.2% |
| 58 | Upper carbonate aquifer | 0.07 | 2.4% | 0.6% | 0.2% |
Hawaiian, Puerto Rican, and Virgin Islands aquifers (Kingshill, Hawaiian Volcanic-rock, Hawaiian Sedimentary deposit, Puerto Rico north and south coast) have zero coverage in this CONUS-only model and are omitted.
What "probability" means here, and what it doesn't
The model output is a statistical prediction of detection, not a measured concentration. A cell with predicted probability of 0.6 means the model — having seen the characteristics of that cell's geology, hydrology, and surroundings — would expect about 60% of comparable wells in the training data to have shown any detectable PFAS at any level (across 21 compounds). It does not mean any specific well will be contaminated. It does not say at what concentration.
The "detection" threshold in the training data is roughly 4 ng/L (4 parts per trillion) for most compounds — the minimum the analytical methods used could distinguish from zero. For comparison, EPA's enforceable drinking-water limits set in April 2024 are 4 ng/L for PFOA and PFOS and 10 ng/L for PFHxS, PFNA, and HFPO-DA. So a "detection" in this model maps roughly to "above the new federal MCL" for PFOA/PFOS specifically. But the model lumps all 21 compounds into one binary; some detections will be of compounds with higher action levels.
Two important caveats apply to interpretation at any one location:
- Resolution. The model is at 1 km. Local conditions can differ dramatically within a 1 km cell — a well 200 m from a fire-training facility will look nothing like a well 800 m on the other side.
- Depth. "Domestic" depth is parameterized as the typical depth of private drinking-water wells in each region. Your well's actual depth may be shallower (more vulnerable) or deeper (more protected). PFAS, being persistent and mobile, can move down through groundwater systems over decades.
For your specific situation, no model is a substitute for an actual measurement. Affordable certified PFAS testing for private wells is now available (see our testing guide). If your area shows up high in the table above and you've never tested, that's the strongest argument for testing.
Training data: actual measurements from the USGS-sampled wells
The model above is a prediction surface. The underlying training set is itself worth surfacing directly. USGS sampled 1,238 wells for the published model; 383 of those are domestic (private) wells across 32 states and 23 aquifers — the subset that maps cleanly to the private-well question. The measured detection rates in that subset are independent of the model and can be checked against the source dataset directly.
National: 383 private wells sampled. 63 (16%) had detectable PFAS across the 21 analyzed compounds. 27 (7%) had detectable PFOA; 34 (9%) had detectable PFOS. 7 wells exceeded EPA's 4 ng/L PFOA MCL; 12 exceeded the PFOS MCL.
By aquifer (private-well subset, aquifers with at least 3 sampled wells)
| Aquifer | Wells | Any PFAS | %Det | PFOA det | PFOS det | Above 4 ng/L MCL |
|---|---|---|---|---|---|---|
| Central Valley | 185 | 29 | 16% | 13 | 17 | 7 |
| Other aquifers (Sierra Nevada) | 49 | 13 | 27% | 7 | 8 | 9 |
| Glacial aquifer system | 24 | 10 | 42% | 2 | 1 | 0 |
| Basin and Range basin-fill aquifers | 36 | 3 | 8% | 0 | 2 | 0 |
| California Coastal Basin | 14 | 3 | 21% | 3 | 3 | 2 |
| Coastal Lowlands aquifer system | 22 | 2 | 9% | 1 | 2 | 1 |
| High Plains aquifer | 24 | 2 | 8% | 1 | 0 | 0 |
| Mississippi Embayment–Texas Coastal Uplands | 6 | 0 | 0% | 0 | 0 | 0 |
| Southeastern Coastal Plain aquifer system | 17 | 0 | 0% | 0 | 0 | 0 |
The Glacial Aquifer System covers the upper Midwest and Northeast — Ohio through Michigan, Wisconsin, Minnesota, much of New York and New England. Its 42% detection rate in the training subset is the highest of any aquifer with substantial private-well sampling, and notably higher than the Central Valley (16%) where most of the training-set wells are concentrated. The Glacial sample is small (n=24), so the confidence interval is wide; the model-output rankings above use the same training data plus geographic predictors to extrapolate spatially.
By state (private-well subset, states with at least 2 wells, ordered by detections)
| State | Wells | Any PFAS | %Det | PFOA det | PFOS det |
|---|---|---|---|---|---|
| California | 286 | 49 | 17% | 23 | 31 |
| Ohio | 23 | 9 | 39% | 1 | 1 |
| Louisiana | 22 | 2 | 9% | 1 | 2 |
| Kansas | 9 | 1 | 11% | 1 | 0 |
| Texas | 10 | 1 | 10% | 0 | 0 |
| Colorado | 2 | 0 | 0% | 0 | 0 |
| Oklahoma | 3 | 0 | 0% | 0 | 0 |
| Mississippi | 9 | 0 | 0% | 0 | 0 |
| Alabama | 13 | 0 | 0% | 0 | 0 |
| South Carolina | 2 | 0 | 0% | 0 | 0 |
California dominates the count because that's where USGS focused most of the public-well sampling that the training data inherits — 286 of the 383 private wells (75%) are in California. The other states have small samples, so the per-state rates have wide error bars. The Ohio result (9/23 detect, 39%) is the strongest non-California signal in the published private-well subset; that detection rate is consistent with the Glacial Aquifer System rate Ohio sits inside.
Source: USGS Predictions of PFAS Occurrence in Groundwater dataset, model archive (training subset). The wells CSV and the aggregation code that produced these tables are at ~/PrivateWellWater.org/usgs-pfas-data/ in the source repo — ask at hello@privatewellwater.org for the working copy.
For journalists, researchers, and data users
The underlying raster is publicly available at the USGS ScienceBase catalog: national_pfas_gw model archive (DOI: 10.5066/P93RXTKJ). The peer-reviewed paper is in Science: Tokranov et al., 2024, "Predictions of Groundwater PFAS Occurrence at Drinking Water Supply Depths in the United States" (doi:10.1126/science.adp6638).
Our aggregations of that surface — per-state and per-aquifer — are available as JSON for direct use:
/data/per_state_pfas.json— 56 states/territories, mean and threshold percentages/data/per_aquifer_pfas.json— 63 USGS Principal Aquifers, mean and threshold percentages, with our hub slug if mapped
The aggregation code (Python, ~80 lines using rasterio and rasterstats) is reproducible from the USGS source raster and the Census TIGER 2024 state polygons. We re-ran the zonal statistics on a single laptop in under 15 seconds. The full methodology is what you'd expect: reproject the WGS84 state polygons to USGS Albers (EPSG:5070), call rasterstats.zonal_stats with all_touched=True, summarize mean and threshold-area percentages per polygon. We are happy to share the scripts if useful; ask at hello@privatewellwater.org.