Methodology

The Conceptual Site Model: Applying ASTM E1689 on Korean Sites

Second in the methodology series. Every investigation carries a theory of the site — where contamination came from, where it went, who it could reach. The only question is whether that theory is written down, tested, and updated, or carried unstated in someone's head. ASTM E1689 is the discipline of writing it down; here is how it works on Korean ground.

What the standard actually asks for

E1689's demand is structural honesty about what you believe and why:

• Sources, stated. Each suspected release — the tank, the trench, the fill event — identified with its contaminants of concern, its mechanism, and its era. "The site is contaminated" is not a source statement; "the 1980s degreasing operation likely released chlorinated solvents through the floor trench" is.
• Pathways, physical. How contamination moves: through what soils, above or below what water table, along which utility corridors — with the site-specific geology and hydrogeology doing the work, not adjectives.
• Receptors, named. Who or what the pathways could reach: site workers, neighbors, wells, surface water. The receptor list is what later connects the model to risk logic — the subject of the E2081 piece in this series.
• Uncertainty, admitted. The model flags what is assumed versus known — and those flags are precisely where the next samples belong. A CSM with no admitted uncertainty is either a finished remediation or a fiction.
• Revision, built in. Every data round either confirms the model or breaks it, and a broken model is progress — it means the site just taught you something the next round can use, the iteration engine of E1903 design.

Where it meets the Korean system

• Korean ground breaks imported assumptions. Much of urban and coastal Korea sits on engineered fill — the "native geology" on the regional map may start meters down, with decades of placed material above it. A CSM that skips the fill question misdraws every pathway; on such sites the fill horizon is often the first source and the first pathway at once, the pattern behind several of our investigation triggers.
• Density changes the receptor map. Korean industrial sites abut housing, commerce, and other plants at distances that would be remarkable elsewhere. Off-site sources send plumes in; off-site receptors sit close enough to matter fast. A fence-line-bounded model is structurally wrong here more often than not.
• The monsoon is a model parameter. Seasonal water-table swings change gradients, mobilize smear zones, and move concentrations — a single-season model quietly assumes the site away from half its year, the same seasonality problem we flagged for groundwater programs.
• The model is the bilingual layer. Korean statutory instruments test numbers against tables; ASTM-frame reviewers reason in source–pathway–receptor. The CSM is where the two meet: it explains to the international reader why a Korean exceedance matters (or doesn't), and organizes for the Korean file why the sampling went where it went. In dual-audience work, the CSM section is the report's load-bearing wall.

The mistakes that repeat

1. Geology by regional map. The model imports a textbook stratigraphy the first boring contradicts — and then nobody updates the model, so every later depth decision inherits the error.
2. Pathways without flow data. Groundwater arrows drawn from topography or wishful thinking. Until gradient is measured, pathway statements are placeholders and should be flagged as such.
3. The utility-corridor blind spot. Backfilled trenches move contamination sideways in ways no natural-geology model predicts — on old Korean industrial parcels, the pipe map is a pathway map.
4. Receptors stopping at the fence. The neighboring apartment block, the downgradient well, the stream two parcels away — omitting them doesn't make them go away; it makes the model useless for the risk conversation that follows.
5. The unfalsifiable model. Written vaguely enough that no data could contradict it. If you cannot name a result that would break the model, it isn't doing any work — the same no-decision-rule failure that afflicts sampling design.

Last reviewed: 31 August 2026 · Dime Works