Forever Chemicals in the Groundwater
A military base used AFFF firefighting foam for 30 years. PFOS seeped into the aquifer. The municipal well field is 800 meters downgradient. We asked seven questions — starting with when contamination reaches the wells and ending with which remedy the base should choose. The screening estimate and full uncertainty analysis imply very different urgency, while choosing a remedy requires the higher-fidelity model. The contaminant everyone is watching for arrives second.
By Michael Key · ORCID
(UCMR5 Jan 2026 release; ~95% of round complete)
Source: EPA UCMR5 (Jan 2026, 10,299 monitored systems, 1.9M sample results). These are system-level monitoring summaries, not population exposure rates.
Same Site. Same Data. Different Answers.
We asked the same question — when does the plume reach the well field? — at three levels of model fidelity. The answer changed every time. So did the recommended action.
Plume arrival time at the municipal well field (800m downgradient). Of 200 arrival-time realizations, 47% did not reach the well within the 100-year horizon; the percentiles above are conditional on reaching it within that horizon.
A Contamination Crisis in Slow Motion
PFAS are synthetic chemicals with carbon-fluorine bonds — the strongest in organic chemistry — making them nearly impossible to break down. In April 2024, EPA finalized maximum contaminant levels of 4 parts per trillion for PFOS and PFOA. We downloaded the EPA's UCMR5 monitoring dataset — 1.9 million sample results from 10,299 monitored public water systems. In this processed January 2026 snapshot, 59.7% of systems had at least one reported PFAS detection. Among systems where PFOS was detected, 97% exceeded 4 ppt. These are system-level monitoring results, not an estimate of the share of U.S. residents exposed.
Our scenario: a composite military base fire training area, parameterized from published USGS data for Joint Base Cape Cod — one of the best-characterized PFAS sites in the US, with 1,500 hydraulic conductivity measurements and a 1.2 km PFOS plume tracked since the 1970s. DoD has identified 700+ current and former installations with a potential PFAS release. That is an investigation inventory, not evidence that every installation has a Cape Cod-like plume or needs the same remedy.
Data: EPA UCMR5 (1.9M samples), USGS Water Quality Portal (62 monitoring wells at Cape Cod), published aquifer parameters. Full sources listed in the Sources section below.
Three Models. Two Dimensions of Fidelity.
This study uses three distinct models. But model complexity is only one dimension of fidelity. The other — and for this study, the more important one — is how honestly you treat uncertainty in the inputs.
The Models
Domenico Analytical
A closed-form equation for contaminant transport in a uniform aquifer. The standard EPA screening tool. One line of math, one answer, milliseconds.
MODFLOW 6 GWF+GWT
USGS MODFLOW 6 solves groundwater flow on a 200×100 grid with spatially varying hydraulic conductivity. The industry-standard tool for contaminant transport. ~70s/sim.
200-Realization Ensemble
Runs Model B 200 times with different randomly sampled parameters. Instead of one answer, you get a probability distribution.
Deterministic vs. Stochastic
The first two models are deterministic: one set of inputs, one answer. The Monte Carlo is stochastic: it samples hydraulic conductivity (K), sorption coefficient (Kd), and source concentration from realistic distributions — because we don't know these parameters exactly, and pretending we do produces false confidence.
The biggest fidelity gap in this study isn't between models. It's between deterministic and stochastic analysis of the same transport model. MODFLOW 6 didn't change. What changed is whether we pretended we knew the sorption coefficient to two decimal places. That single assumption — known Kd — is what makes the plume look distant when it isn't.
Seven Questions. Seven Deep Dives.
Each question was answered at the fidelity it required. Click any card to see the full analysis, charts, and methodology.
When Does the Plume Reach the Well Field?
Screening says 96 years. MODFLOW 6 says 71–75 years. Among Monte Carlo draws that arrive within the 100-year horizon: P5 = 19 yr, P50 = 49 yr, P95 = 72.8 yr. In 47% of draws, the plume does not arrive within that horizon.
Where Is the Plume Right Now?
Homogeneous model says “drill on the centerline.” Heterogeneous model shows the plume bifurcated — centerline wells miss it.
P&T vs. PRB: Which Remedy Wins?
P&T contains the plume in all 50 sampled MC realizations under the encoded priors (the well never exceeds detection in those draws; deterministic peak ~1.1 ppt, ~3.5× below the 4 ppt MCL). NPV $31.9M flat vs PRB $80.9–99.1M.
Is Better Site Characterization Worth It?
Value-of-information (VOI) on Kd ≈ $0M. Sharper sorption tightens the median by 6 years but doesn’t change the build, the schedule, or the cost. The honest answer: skip the campaign.
What If the MCL Changes?
Tightening 10→0.5 ppt roughly 2.2x the plume area (89,038→197,775 m²) and quadruples the exceedance fraction (13.3→53.3%). Site NPV stays flat at ~$93.2M — tighter regs make MORE sites expensive, not this one. (Note: this $93M is the 100-year continuous-containment cost-of-compliance scenario; it is NOT comparable to the Q3 $32M figure, which is the early-intervention P&T capex-dominated NPV.)
Where Does Fidelity Change the Decision?
Screening suffices for go/no-go. Monte Carlo tests the remedy across the encoded uncertainty: P&T contained the plume in all 50 sampled realizations here, while sharper measurements did not change the decision.
Scope: Three models (Domenico analytical, MODFLOW 6 GWF+GWT, Monte Carlo at 30–200 realizations per question), ~1,200 lines of Python, published USGS aquifer parameters, 1.9M EPA monitoring records. Limitations documented honestly: no unsaturated zone transport, no density-driven flow, no multi-species precursor reactions. Each question got the model it needed — nothing more.
Finding the Right Model for the Decision
Each question above was answered at the fidelity it required. But scattered across seven questions, it's easy to miss the pattern: which dimensions of reality actually change the answer? Here it is in one place — four decisions evaluated at three model fidelities, showing where the answer converges and where it breaks.
| Decision | Screening (A) | MODFLOW 6 (B) | Monte Carlo (C) |
|---|---|---|---|
| Is there a problem? | Yes — ~96 yr | Yes, 71–75 yr with heterogeneity | Among within-horizon arrivals, P5 / P50 / P95 = 19 / 49 / 72.8 yr; 47% did not arrive within 100 yr |
| Which remedy? | Can't evaluate | P&T plausible | P&T has lower NPV in 100% of 50 sampled remedy draws; NPV $31.9M flat |
| How much will it cost? | Can't estimate | ~$30M capex (no horizon view) | $31.9M flat NPV; PRB $80.9–99.1M |
| Which species arrives first? | Can't distinguish | PFOA at 20yr, PFOS at 25yr | Distribution per species |
The deterministic models say P&T is plausible. In a 50-draw remedy comparison, P&T has lower modeled cost in every sampled realization under the encoded priors. Separately, 47% of the 200 arrival-time realizations did not reach the well within the 100-year horizon. The gap isn’t whether the conventional remedy is right; it’s whether you can test that choice across a stated distribution. This finite ensemble does not cover every physical condition or omitted process. Real aquifers have unknown K and Kd. A planning process that ignores both isn’t wrong by accident — it is incomplete on the questions that matter (does this work across the sampled uncertainty? is sharper data worth paying for?).
Methodology note: Our composite scenario uses published USGS data from Joint Base Cape Cod — one of the most extensively studied PFAS sites in the US, with 1,500 hydraulic conductivity measurements and a 1.2 km PFOS plume tracked since the 1970s. Monte Carlo samples K from log-normal(mean=10, σlog=0.5), Kd from log-normal(mean=1.5, σlog=0.8) anchored to Anderson et al. (2019) AFFF-site Kd statistics (95% range ≈ 0.30–7.43 L/kg, with a 0.1 L/kg floor), and source concentration from normal(100, 25). Realization counts vary by question (Q1 arrival uses 200; Q3–Q5 use 50; Q6 uses 30 per MCL level). Note: the sensitivity table below reports the full Anderson 2019 literature range for Kd (0.5–20 L/kg); the MC log-normal centers within the site-relevant subset of that range.
What Drives the Answer?
Across the 200-realization arrival-time ensemble, two parameters control 81% of the variance. Everything else is noise for this decision.
| Parameter | Range | Variance Share | Decision Impact |
|---|---|---|---|
| K (hydraulic conductivity) | 1–100 m/d | ~51% | Determines whether plume arrives in years or decades |
| Kd (sorption) | 0.5–20 L/kg | ~30% | Controls PFOS vs. PFOA differential arrival |
| Gradient (i) | 0.002–0.008 | ~18% | Secondary; scales with K |
| Porosity / Dispersivity | 0.2–0.4 / 5–50 m | ~2% | Affects plume width, not arrival |
K and Kd together drive 81% of the arrival-time variance. That’s where a characterization budget would have the most leverage — but whether to spend it depends on what decision is on the table. For this site, with containment already the lower-cost remedy in all 50 sampled realizations under these priors, sharper Kd doesn’t change the build (see Q5: VOI ≈ $0M). For a site weighing monitored natural attenuation (MNA) vs. active remediation, the same measurement could be worth millions. Sensitivity tells you where the leverage is; VOI tells you whether to pull it.
Sensitivity from Q1 tornado diagram (arrival time) and Q5 variance decomposition (cleanup time). Q1 ensemble: 200 realizations, USGS-sourced parameter ranges.
Model Validation
Before projecting remediation scenarios, we ran our MODFLOW 6 model against real monitoring data from Joint Base Cape Cod — one of the most extensively studied PFAS sites in the country. 62 USGS monitoring well measurements, same model, same algorithm.
| Model Configuration | Predicted Plume Front | vs. Observed (2,700 m) | Explanation |
|---|---|---|---|
| Cape Cod params (K=95, Kd=0.4) | 2,990 m | +11% | Site-specific data matches observed plume |
| Generic literature params (K=10, Kd=1.5) | 780 m | −71% | 3.5x ratio of observed to predicted (equivalent to 71% under-prediction relative to observed) without site data |
| Full 3D model (10 layers, 200K cells) | 2,910 m | +8% | Vertical structure matches qualitatively |
The honest reading: With published USGS aquifer parameters (K=95 m/d, n=0.39, back-calculated Kd=0.4 L/kg), our model predicts the plume front within 11% of what USGS actually measured. Generic literature parameters miss by 3.5x. This is why site characterization matters — and why the conditional arrival distribution among realizations that reached the well within 100 years (P5 / P50 / P95 = 19 / 49 / 72.8 years) captures variation the deterministic base case misses. The 47% that did not reach within the horizon remain outside those percentiles. The 3D model confirms vertical structure: PFOS peaks at 25–35m depth, consistent with recharge pushing the plume downward.
Observed data: 49 PFOS detections (1.3–610 ng/L) at 62 monitoring wells, USGS Water Quality Portal (2019–2020 sampling). Predicted: MODFLOW 6, 200×100 grid, 55-year simulation.
What We Modeled and What We Didn't
Transport modeling uses USGS MODFLOW 6 (v6.6.3). Aquifer properties from Joint Base Cape Cod USGS studies (K=60–110 m/d, n=0.39, αL=0.96m from 1,500 borehole flowmeter tests). National context from EPA UCMR5 (1.9M samples). Every parameter sourced.
| Parameter | Base Value | Range | Source |
|---|---|---|---|
| Hydraulic conductivity (K) | 10 m/d | 1–100 m/d | USGS, Gelhar (1992) |
| Effective porosity (n) | 0.30 | 0.20–0.40 | Freeze & Cherry (1979) |
| PFOS Kd | 1.5 L/kg | 0.5–20 L/kg | Anderson et al. (2019) |
| Source concentration | 100 ppb | 50–200 ppb | DoD fire training area (FTA) data |
| Decay rate | ≈0 | — | “Forever chemicals” |
What we didn't model: Unsaturated zone transport, density-driven flow, multi-species precursor reactions, air-water interface sorption, co-contaminant interactions. Each would increase fidelity — and each would increase computation 10–100x. The open question is: which of these actually changes the remedy selection decision? That’s the follow-up study.
What's Actually Happening
The analysis above answers “what if.” This section answers “what is.” Real contamination has been measured. Real regulations have been finalized. Real remediation is underway. Here's how reality maps onto our model.
What's Coming: The Scale of the Problem
We downloaded the EPA's UCMR5 monitoring dataset — 1.9 million sample results from 10,299 public water systems. The numbers are stark.
| Metric | Value | Context |
|---|---|---|
| Monitored systems with at least one reported PFAS detection | 59.7% | 6,148 of 10,299 systems |
| PFOS detections exceeding 4 ppt MCL | 97% | Nearly every detection is an exceedance |
| Median detected PFOS concentration | 6.8 ppt | Nearly twice the legal limit |
| Systems currently exceeding MCL | 1,693 | Each must remediate or find alternative supply |
| DoD installations identified with a potential PFAS release | 700+ | Federal screening inventory; not 700+ confirmed Cape Cod-like plumes or remedies |
The Regulatory Landscape
| Event | Date | Impact |
|---|---|---|
| EPA finalizes 4 ppt MCL for PFOS/PFOA | Apr 2024 | Potential-release sites proceed through investigation and, where warranted, cleanup |
| EPA proposes a PFOA/PFOS exemption option and a separate rescission rule for four other PFAS provisions | May 2026 | Both actions remain proposed; the 2024 final rule still supplies the current enforceable text |
| Several states propose 2 ppt or lower | Ongoing | Our Q6 shows site NPV is flat but exceedance footprint scales 4× |
| UCMR5 monitoring ~95% complete | Jan 2026 | 59.7% of monitored systems in this snapshot have at least one reported PFAS detection |
Specific Sites in the Pipeline
| Site | Status | Estimated Cost | Key Challenge |
|---|---|---|---|
| Joint Base Cape Cod, MA | Active remediation since 2015 | $100M+ (ongoing) | 1.2 km plume in sand/gravel; 6,200-acre zone |
| Pease AFB, NH | Remedial investigation/feasibility study (RI/FS) underway, report mid-2026 | Estimate pending final RI/FS | Municipal water supply contaminated |
| Luke AFB, AZ | Preliminary assessment complete | Estimate pending final RI/FS | Arid climate reduces recharge but concentrates plume |
| Broader DoD potential-release inventory | 718 current and former installations identified with a potential PFAS release as of June 2024 | More than $9.3B in estimated future investigation and cleanup costs beginning in FY2025 | GAO’s bounded estimate, not a lifecycle ceiling; inclusion does not establish a plume, remedy, or site cost |
The portfolio number and the site model answer different questions. GAO reported more than $9.3 billion in estimated future PFAS investigation and cleanup costs beginning in fiscal year 2025; it did not present that figure as a full lifecycle ceiling. The $32M containment NPV belongs only to the composite Cape Cod-parameterized scenario. DoD’s 718-site count is a potential-release inventory whose sites differ in plume evidence, soils, wells, investigation stage, and whether remediation will be required. The uncertainty is therefore both which sites ultimately require action and what that action entails (see Q6).
The Remediation Market
The PFAS remediation market is expanding. Clean Harbors, Arcadis, Tetra Tech, Geosyntec — the companies doing this work face a fundamental challenge: deterministic models do not quantify containment across the encoded uncertainty, and they can’t tell a client when a $500K characterization spend would be wasted. The Monte Carlo work isn’t about pricing the tail bigger — at this site the cost is flat. It tests both the recommended remedy and the decision to stop spending on data across the sampled priors; it does not prove performance under every unmodeled condition.
What We'd Recommend
Seven questions, three model fidelities, hundreds of Monte Carlo runs across the questions — and now grounded in what's actually being measured, regulated, and remediated.
For Remediation Contractors
Don’t lead with PRB premium pricing here — P&T wins on NPV by ~$50M ($31.9M vs $80.9–99.1M) and contains the plume in all 50 sampled realizations under the encoded priors. The contract risk isn’t the cleanup time; it’s scope creep if the MCL tightens to 0.5 ppt and the exceedance fraction quadruples (Q6). Price the wells, document the basis, and build a re-opener for regulatory shifts.
For Site Owners
Resist the urge to fund a $500K Kd campaign reflexively. At this site the VOI is essentially zero — sharper sorption data doesn’t change the build, the schedule, or the bill. The right next investment is operations: get the wells on, document the capture zone, and put characterization money toward sites where it actually flips a decision.
For Monitoring Programs
PFOA travels faster than PFOS (lower Kd). At sites with mixed AFFF contamination, PFOA arrives at the well field 5 years earlier. If your monitoring plan only tests for PFOS, you're missing the leading edge. Test for the full PFAS suite — PFOA and PFHxS are your early warning system.
Compliance Deadlines
The 2024 final rule requires compliance by April 2029. On May 18, 2026, EPA proposed an option for eligible systems to request two additional years, to 2031, while keeping the 4 ppt PFOA and PFOS limits. EPA separately proposed rescinding the PFHxS, PFNA, HFPO-DA, and Hazard Index provisions. Neither proposal is final as of July 25, 2026, so the final rule’s 2029 deadline remains the current baseline. In the processed January 2026 UCMR5 snapshot, 1,693 monitored systems exceeded at least one final-rule MCL. States with stricter standards — New Jersey (58% exceedance), Massachusetts (40%), North Carolina (39%) — face the earliest and most expensive compliance requirements.
Model results and source-data snapshot: March 2026. Regulatory text reviewed: July 25, 2026. EPA regulations, state MCL proposals, and UCMR5 monitoring data are evolving.
Explore the Data
See for yourself — adjust the parameters and watch the plume move.
3D Plume Replay
Rotate, zoom, and scrub through 55 years of PFAS migration. Watch PFOA race ahead of PFOS through 10 aquifer layers.
Remediation Decision Tool
Adjust sorption, conductivity, and regulatory assumptions. Watch the winning remedy — and its cost — change in real time.
Sources
Regulatory & Monitoring Data
EPA (2024). PFAS National Primary Drinking Water Regulation, 89 FR 32532.
EPA (2025). Announced intent to retain PFOA/PFOS limits and pursue rulemaking.
EPA (2026). Proposed PFOA/PFOS Compliance Extension Rule
and Proposed PFAS Rescission Rule.
EPA UCMR5 Occurrence Data (Jan 2026, 10,299 systems).
EWG PFAS Contamination Map (9,728 sites).
Transport & Sorption
Domenico (1987). Analytical transport model. J. Hydrology 91:49–58.
Anderson et al. (2019). PFAS Kd values. J. Contaminant Hydrology 220:59–65.
Brusseau (2018). Air-water interface sorption. Sci. Total Environ. 613–614:176–185.
Gelhar et al. (1992). Field-scale dispersivity. Water Resources Res. 28(7):1955–1974.
Hydrogeology
Bear (1972). Dynamics of Fluids in Porous Media.
Freeze & Cherry (1979). Groundwater.
Walter et al. (2018). USGS SIR 2018-5139. Cape Cod MODFLOW model.
LeBlanc et al. (1991). Cape Cod tracer test. Water Resources Res. 27(5):895–910.
Remediation Costs
EPA (2021). PFAS Treatment Technologies. EPA/600/R-21/164.
ITRC (2023). PFAS Technical and Regulatory Guidance.
GAO (2025), Persistent Chemicals:
718 current and former installations identified with a potential PFAS release as of June 2024;
nearly all had completed initial assessment and site inspection.
Validation Data
USGS via Water Quality Portal.
62 monitoring well measurements at Joint Base Cape Cod (2019–2020).
49 PFOS detections, 1.3–610 ng/L, plume front ~2,700 m after 55 years.
Modeling Tools
Langevin et al. (2024). MODFLOW 6. Groundwater.
USGS MODFLOW 6 v6.6.3 with GWT transport model.