When the first run fails: the refinement ladder

Most regulatory modeling time is not spent on runs that pass — it's spent on runs that don't, yet describe facilities that would genuinely comply. Screening-level choices are deliberately conservative, and the craft of air modeling is largely the discipline of removing that conservatism legitimately: replacing worst-case assumptions with documented realism, one step at a time, with the reviewing authority in the loop. This page is the map; every rung links to the page that covers it fully.

Rung 0 — verify before you refine

A failing run has one boring cause more often than any interesting one: an input error. Before touching methodology, audit what the model actually received — exit temperature in Kelvin (a Celsius value silently kills buoyancy), emission rates in grams per second, area rates per square meter (see The 13 source types), stack parameters from the operating load being modeled, building base elevations on sloping ground (see BPIPPRM), coordinate zones. An hour spent here is never wasted; every later rung assumes the inputs are right.

Rung 1 — the design-value form

Confirm the number that failed is the number the standard defines: the correct rank and averaging (a 98th-percentile daily-maximum form is not the peak hour), annual standards judged on the highest individual year where required, multi-year processing handled as the form demands (see Averaging periods & design-value forms and the rule pages). A demonstration can "fail" on arithmetic that the regulation never asks for.

Rung 2 — receptors and ambient air

Where is the failing receptor? If it sits inside a secured facility boundary, it isn't ambient air and doesn't belong in the analysis (see Ambient air & receptor placement). If it's legitimate, resolve the peak properly — the two-pass network refinement is expressly endorsed — and check the receptor's terrain elevation, which in elevated terrain moves the answer directly (see Terrain).

Rung 3 — background honesty

The background term is often the largest single number in a cumulative design value, and the bluntest. In order of effort: is the monitor actually representative? Is a nearby source being double-counted — modeled explicitly while also inside the monitor's readings (see Nearby sources vs. the monitor)? Would temporal pairing (season, hour) or sector-varying background replace a single worst-case constant with the values the standard's form actually pairs (see Background concentrations)? These refinements are explicitly provided for in the Guideline on Air Quality Models (Appendix W) — they are not tricks.

Rung 4 — source characterization

Is each source the type its physics says it is — a capped vent modeled as POINTCAP rather than an optimistic POINT, a disturbed pile as an AREA with an honest initial dimension (see The 13 source types)? Are operating loads modeled per the load analysis, including the counterintuitive low-load worst cases? Do limited-schedule sources use their permitted hours rather than a smeared average — and never averaged across non-operating hours (see Emission inputs)? Are the building dimensions and GEP determinations built from real geometry (see Building downwash and BPIPPRM)?

Rung 5 — meteorology

The met record is a refinement surface of its own: is the station genuinely representative of the transport between source and peak (see Meteorological data & representativeness)? Does the data situation call for one of AERMET's documented options — the low-wind adjustment, the tower ΔT scheme, the sonic-anemometer threshold (see AERMET's options)? Were the AERSURFACE season, moisture, and sector choices made by the recipe (see AERSURFACE's choices)? A site-specific tower year is the deepest met refinement — preferred by the Guideline itself, at the price of a real monitoring program (see Site-specific towers).

Rung 6 — chemistry and deposition

For NO₂, the tier ladder exists precisely for this moment: escalate from full conversion through the ambient-ratio method to the detailed screens, funded by real data — hourly ozone and a documented in-stack ratio (see NO₂ modeling). For particulates with a measured size distribution, Method 1 deposition and depletion put the coarse mass where physics says it lands — watching for the near-field settling surprise (see Deposition & depletion).

Rung 7 — the design itself

When the model is honest and the number still fails, the remaining choices belong to the project, not the analysis: emission controls, enforceable limits, operating restrictions, or stack height — creditable only up to Good Engineering Practice height (see GEP stack height). And where the preferred model genuinely cannot represent the situation, the alternative-model path exists, with its formal approval process (see The reviewing authority, the protocol & model approval).

The discipline that makes it all defensible

Two habits separate refinement from number-chasing. First, direction of reasoning: every change on this ladder is justified by the data situation — a measured size distribution, a documented in-stack ratio, a monitor that demonstrably carries a neighbor — never by the direction it moves the result. Second, process: changes are recorded in the modeling protocol and agreed with the reviewing authority as they're made, not defended after the fact. A demonstration refined this way arrives at review already answered.

In PlumeSmart

The platform is built around this loop: settings changes re-run the same project rather than rebuilding it, every run's inputs and options are recorded for comparison, and source-contribution and per-receptor views help locate which rung a failing design value is pointing at.