Why Field Screening Hasn't Worked — And What Had to Change

Field screening has a reputation problem, and it has earned it. Ask a project manager how much weight they put on a field reading and the honest answer is usually "some, but I wouldn't hang a decision on it." That instinct is correct — and it is worth understanding precisely why, because the reasons are specific, documented, and fixable. They are also not a comment on the competence of the people holding the instruments. The tools were designed for detection and triage, and they are being asked to produce decision-grade data. That is a design mismatch, not a skill problem.

Three instrument families dominate field practice. Each fails in its own way, and all three fail for the same underlying reason.

Vapour screening: nothing about the measurement is controlled

The headspace bag method is the most widely used field screening technique in the industry. Soil goes into a plastic bag, the bag is left to warm, and a PID or FID probe is inserted through the wall to read the headspace.

Consider what determines that reading. The mass of soil in the bag. The volume of air above it. The temperature it equilibrated at — and vapour pressure is strongly temperature-dependent, so this one matters enormously. How long it sat. How hard it was worked. Where the probe went in and how quickly it was read. Not one of those is specified, controlled or recorded in normal practice, and every one of them moves the number.

The result is well documented: correlation to laboratory data is typically below 25%, and often below 10%. In a controlled 62-sample comparison, the bag method returned R² = 0.002 against laboratory Total BTEX with a regression slope statistically indistinguishable from zero, and effectively no relationship at all to Total PHC. The same field study measured what the sample temperatures were actually doing: the standardized system held equilibration between 30 °C and 35 °C, while the bag samples ranged across roughly 10 °C to 27 °C over the course of the work.

The detector is not the problem. Both methods in that comparison used the same detector. The difference was process control.

EC meters: measuring everything, speciating nothing

Salinity screening has a different failure mode. A bulk electrical conductivity meter measures the total dissolved ion load of a soil-water extract. It is a genuinely useful instrument for what it is — but the regulatory parameters are chloride and sodium, and EC cannot distinguish either from anything else in solution.

So the reading moves with soil moisture, temperature, mineralogy and background salts as readily as it moves with a produced-water release. In carbonate-rich or naturally saline material the meter reads high in soil that is entirely clean, and in a complex matrix it can read acceptably in soil that exceeds criteria. Both errors happen routinely, and the false negatives are the ones that come back later as failed confirmations and return trips.

Alberta Tier 1, CCME and AER Directive 058 all specify chloride and sodium. Screening against those criteria with an instrument that cannot resolve either is a structural mismatch, not a calibration issue.

Vapour proxies for extractable hydrocarbons: measuring the wrong thing

The third failure is the most conceptual. Extractable petroleum hydrocarbon criteria — F2 and F3, LEPH and HEPH — describe what a laboratory recovers by solvent extraction. A vapour-phase field method measures what partitions into headspace air. Those are different populations of compounds.

The consequence is systematic, not random: light volatiles are over-represented and heavy fractions are systematically underestimated, so a diesel or lubricating-oil impact can screen far cleaner than it is. And because the response varies with temperature, moisture and compound-specific volatility, the bias shifts across a programme rather than sitting at a constant offset that could be corrected for.

The practical effect is that the sites where extractable criteria matter most are the sites where a vapour proxy tells you least. A remediation programme driven by F2 and F3 limits needs a reading that responds to extractability; it is being given one that responds to volatility. Every boundary drawn on that basis is drawn against the wrong variable — and because the error runs conservative in the wrong direction, the soil that screens clean is exactly the soil most likely to fail confirmation.

The root cause is the same in all three cases

Different symptoms, one disease: none of these methods controls the conditions under which the measurement is made, or measures the parameter the regulation is written against. Uncontrolled conditions mean readings are not comparable to each other, and readings that are not comparable cannot be treated as a population. Once you cannot treat them as a population, statistics are unavailable — no distribution, no percentile, no threshold, no stated confidence. You are left with individual numbers and professional intuition.

Precision is what makes a set of readings into a data set. Without it, you do not have a data set. You have a collection of anecdotes.

And underneath that, the sampling paradox

There is a second problem that no amount of instrument improvement alone would solve. Contaminated sites are heterogeneous at a scale finer than conventional programmes sample. EPA guidance documents that matrix heterogeneity influences result variability roughly nineteen times more than the choice of analytical method — which means the dominant source of uncertainty on most sites is not the instrument at all. It is that the sampling density is too sparse to resolve what the soil is actually doing.

Hence the paradox. The more variable a site is, the more data points are needed to characterize it with confidence. But conventional programmes are constrained by per-sample laboratory cost and by the throughput of the field instruments — so they collect the least data at exactly the sites that need the most. The instrument limitation and the density limitation are the same limitation viewed from two directions.

What re-engineering actually required

Fixing this meant addressing each source of variability at the instrument level rather than asking field staff to be more careful. Five things had to change:

• Controlled extraction conditions — standardized sample mass and vessel volume, temperature-held equilibration, fixed timing and agitation, enforced identically on every sample by the hardware and the app rather than left to technique.
• Ion-specific measurement — replacing bulk conductivity with a potentiometric ion-selective electrode array targeted at chloride, the parameter the guideline is actually written against, under a standardized extraction with ionic strength adjustment for matrix independence.
• Laboratory extraction chemistry, in the field — using solvent extraction rather than a vapour proxy for extractable hydrocarbons, so F1, F2 and F3 are resolved separately and each is compared against its own fraction criterion.
• Models trained on paired data — predictive models built on thousands of laboratory-confirmed field readings, translating a field response into a laboratory-equivalent estimate with a confidence attached, and recalibrating against every new result.
• A connected data platform — because high-density screening produces data volumes that spreadsheets and PDFs cannot manage. Real-time aggregation, enforced QA/QC, population and spatial analysis, and reporting that meets data quality objectives.

Those five together are what makes the difference between a reading and a decision. Standardization makes the numbers comparable. Comparability makes the population statistically meaningful. Density makes the population large enough to resolve the site. Correlation grounds the population in laboratory reality. And the platform makes all of it usable while the crew is still standing on the ground.

The One Thing to Remember

Conventional field screening did not fail because the instruments were poorly made or the operators careless. It failed because nothing about the measurement was held constant, and because the sites that most needed dense data were the ones getting the least. Both are engineering problems, and both have now been engineered around.