PFAS Study → Question 7

Where Does Fidelity Change the Decision?

Not every question needs the most complex model. The right model depends on the decision you're making. This is the ADM thesis demonstrated.

Background

What Is the Domenico Equation?

The Domenico equation (1987) is the standard EPA screening tool for contaminant transport in groundwater. It's a closed-form analytical solution that calculates how a contaminant plume spreads through a uniform aquifer over time, accounting for advection (bulk flow), dispersion (spreading), and sorption (attachment to soil).

Its strength is transparency: one equation, a handful of parameters, an answer in milliseconds. Every hydrogeologist knows it. Every EPA screening assessment uses it. It tells you the order of magnitude and the direction of travel.

Its weakness is what it assumes: everything is homogeneous. The aquifer has uniform conductivity. The flow field is perfectly one-dimensional. There are no preferential flow paths, no heterogeneous sorption zones, no geological surprises. In a real aquifer, none of these things are true — and the gap between the Domenico answer and reality is exactly where fidelity decisions live.

The Comparison

Five Decisions, Three Model Fidelities

The same site, the same data, the same question — but the answer depends on which model you ask. Green cells are sufficient for the decision. Gold cells give a partial answer. Rose cells can't participate.

Decision	Screening (A)	MODFLOW 6 (B)	Monte Carlo (C)
Is there a problem?	Yes — ~96 yr	Yes — faster with heterogeneity	5% chance within 5 yr
Where to drill?	Centerline only	Off-center lobes	Probability-weighted zones
Which remedy?	Can't evaluate	P&T appears cheaper	PRB wins risk-adjusted
How much will it cost?	Can't estimate	~$80M (ignores tail)	$80M median, $120M+ at P95
Where to invest next?	No guidance	No guidance	$500K characterization saves $15M+
Which species arrives first?	Can't distinguish	PFOA at 20yr, PFOS at 25yr	Distribution per species
What depth to screen wells?	N/A (2D only)	3D: peak at 25-35m depth	3D probability by layer

Finding

For “is there a problem?” — screening suffices. For remedy selection — screening can't participate, deterministic gives the wrong answer, and only Monte Carlo gets it right.

Model Roles

What Each Model Is Good For

No model is universally best or worst. Each occupies a fidelity niche where it's the right tool for the decision at hand.

Model A — Screening

Domenico Analytical

Good for: initial screening, go/no-go decisions, order-of-magnitude timelines.

Not for: remedy selection, cost estimation, well placement.

One equation. One answer. Milliseconds. When you need to know IF — not HOW MUCH — this is the right model.

Model B — 2D/3D Transport

MODFLOW 6 GWF+GWT

Good for: plume mapping, monitoring well design, understanding preferential flow, multi-species comparison (PFOS/PFOA/PFHxS), 3D vertical structure, monitoring well depth optimization.

Not for: cost estimation under uncertainty, risk-adjusted decisions.

Spatially resolved. Physically grounded. When you need to know WHERE — this is the right model.

Model C — Monte Carlo

200-Realization Ensemble

Good for: remedy selection, cost estimation, value-of-information, regulatory planning.

Required for: any decision involving tail risk or uncertainty.

200 possible futures. When you need to know HOW MUCH and HOW CERTAIN — this is the right model.

The right model depends on the decision. Screening tells you IF. Deterministic tells you WHERE. Monte Carlo tells you HOW MUCH and HOW CERTAIN. Using screening for remedy selection is like using a napkin for an engineering drawing.

Synthesis

What the Seven Answers Tell Us

Across the full investigation, a clear pattern emerges: the questions that matter most for decisions are the ones that require the highest fidelity.

Conclusion	Evidence	Model Required
There is a problem	Plume moving; 5% chance within 5 yr	A (confirmed by C)
Standard remedy riskier than it looks	P&T P95 = 120+ years	C
Better option exists	PRB wins risk-adjusted NPV	C
Most valuable investment is information	$500K saves $15M+	C
Regulatory risk dominates	4→2 ppt triples cost	C
Right model depends on the decision	All three answer different questions	All
PFOA is the early warning	Arrives 5 years before PFOS at every heterogeneous realization	B (multi-species)

The ADM Thesis

The right model isn't the most accurate model — it's the one at the right fidelity for the decision it needs to support. Screening is perfect for go/no-go. Deterministic is perfect for spatial planning. Monte Carlo is required the moment money or risk enters the equation. Matching fidelity to the decision is the discipline. That's Analysis Driven Modeling.

Three models, ~1,200 lines of Python, USGS MODFLOW 6, published parameters, 1.9M EPA records.

← Previous Q6: Regulatory Study Back to Study →