Peak-dBA noise model — methodology & measurement record
Prepared 2026-10-08; revised 2026-10-19 — added the measurement basis, the 2,889-flight turn-altitude record, the five-flight calibrated cross-check, the receptor characterization, equipment notes, and a plain-language explanation of the law and of common misconceptions. The sensitivity-map figures are unchanged from the 2026-10-10 model.
A measurement record, not assumptions
TrueNoise is a continuous measurement record. Between 1 May and 15 August 2026 the monitor logged roughly 30,000 departure samples across 153 days, grouped into about 3,185 departure passes. From that record:
- 2,889 clean Runway 15R departures were analyzed for turn altitude — the result held steady every month (median 2,325 ft in May, 2,375 in June, 2,292 in July, 2,196 in August), so it reflects a full season, not a snapshot.
- 249 isolated overflights calibrated the loudness-vs-distance law, which then reproduced an independent calibration point (73.3 dBA at 0.55 mi) and the monitor's own measured level (70.5 dBA) without being tuned to either.
- A five-flight blind test against a calibrated sound-level meter put the model within about 0.8 dB of ground truth — slightly better than a commercial noise-tracking device measured in the same window.
What we measure — and why it is not DNL
Single-event peak dBA — what TrueNoise measures — is the loudest instant of one overflight: the spike you actually hear when a plane passes. It is an event, measured directly by the monitor.
DNL (Day-Night average sound Level) is a 24-hour energy average of all noise, quiet hours included, with a 10-dB penalty on night events. It is a bookkeeping figure, not a sound anyone hears; the “65” in federal land-use guidance is a DNL contour.
These are different physical quantities on different scales. A corridor can sit far below DNL 65 as a 24-hour average and still deliver single-event peaks in the low 70s dozens of times a day — which is the Severn situation.
So the figures here must not be compared to, or converted into, a DNL 65 figure, and must not be fed into DNL-based annoyance or reading-score formulas. The model answers one question: how loud is each pass, where.
What goes in (all measured or from public records)
| Input | Source | Description |
|---|---|---|
| Departure tracks | ADS-B track database | 90 clean 15R departures after filtering — the measured climb and west-turn geometry. |
| Turn-altitude record | 153 daily noise logs | ~30,000 departure samples → 2,889 clean 15R overflights with altitude at closest approach. |
| Corridor homes | Anne Arundel County address points | Residential addresses, swept to the 62 dBA peak corridor of either track → 33,206 homes. |
| Noise event logs | IEC 61672 Class 2 monitor | Per-event dBA, slant range, altitude — used to fit and validate the law. |
No modeled manufacturer noise tables (ANP/AEDT) are used; the law is fit directly from the TrueNoise measurements.
The loudness-vs-distance law — in plain terms, then exactly
In plain terms. Sound gets quieter with distance in a regular way, measured directly: every halving of the distance to the aircraft adds about 7 dB, and every doubling subtracts about 7 dB. Anchor that to one measured point — about 73 dBA when the plane's closest approach is 0.55 mi away — and the loudness at any distance follows:
| Closest approach (straight-line) | Modeled peak | Everyday comparison |
|---|---|---|
| 0.25 mi | ~81 dBA | Garbage disposal; diesel truck passing close |
| 0.55 mi | ~73 dBA | The project calibration point (measured 73.3) |
| 1.0 mi | ~66.5 dBA | Busy traffic from the sidewalk |
| 2.0 mi | ~59 dBA | Normal conversation |
| 4.0 mi | ~52 dBA | Light rain; a quiet street |
Exactly. Peak falls with slant range (the 3-D straight-line distance from home to aircraft at closest approach — horizontal offset and altitude combined):
peak_dBA = A + B × log10(slant_range_mi)
A = 66.54 B = -24.37A (66.54) is the intercept — the modeled peak at exactly one mile, where the log term is zero. It is the anchor of the curve, not a threshold. B (−24.37) is the slope — how fast loudness falls with distance (the “~7 dB per halving” above). Slant range is floored at 0.10 mi so the log stays finite directly beneath the path. The law was fit by regressing the single loudest sample of each of 249 overflights against log-distance.
Validation — three independent checks
(1) An untouched calibration point. The fitted law lands at 72.9 dBA at 0.55 mi against the project's measured 73.3 — 0.4 dB, not tuned.
(2) End-to-end, against the monitor's own exposure. With the measured climb in place, the full model reproduces the receptor's level: 70.6 dBA modeled vs 70.5 measured (median of 249 logged overflights there). The climb and geometry were not tuned to this — it validates the climb curve and the law jointly, and it is what exposed and fixed an earlier contaminated climb.
(3) A live, blind test against a calibrated meter — next.
Live cross-check against a calibrated meter
On 15 October 2026, five consecutive 15R departures over a home ~130 ft from the monitor were measured three ways at once: the model (from the ADS-B slant range, computed before the logs were opened), a commercial airnoise.io device, and the calibrated IEC 61672 Class 2 meter (ground truth).
| # | Aircraft | Slant (mi) | Model | airnoise | Calibrated | Model−cal |
|---|---|---|---|---|---|---|
| 1 | 737-800 | 0.58 | 72.3 | 71.3 | 72.8 | −0.5 |
| 2 | A320 | 0.85 | 68.2 | 66.5 | 67.4 | +0.8 |
| 3 | 737-800 | 0.83 | 68.5 | 67.0 | 67.3 | +1.2 |
| 4 | 737-800 | 0.66 | 71.0 | 68.9 | 70.2 | +0.8 |
| 5 | A300F | 0.97 | 66.8 | 66.7 | 66.6 | +0.2 |
Against the meter the model ran +0.50 dB mean, 0.76 dB RMS, 1.2 dB worst case; the commercial device −0.78 dB mean, 0.98 dB RMS. Both landed well inside the ±7 dB screening band, the model slightly ahead. Caveat: n = 5, one 10-minute window, one weather state, no near-overhead pass — a confirmation, not a second calibration.
How high the aircraft actually are
The most-contested point in community discussion is how high aircraft are when they turn. It has a measured answer on a large sample, and the answer is that there is no single turn altitude — the turn is a process, not a point.
Across 2,889 clean 15R departures, altitude at closest approach has a median of 2,325 ft (middle half 2,141–2,617; only 7.8% below 2,000 ft). Closest approach falls a median 3.46 nm out — the turn-completion zone.
Putting the geometry together: departures begin the turn around 1,800 ft (near 2 nm), complete it around 2,300 ft (2.5–3 nm out, over the community), and keep climbing west. They are lowest over the close-in neighborhoods and schools nearest the runway and higher by the western end. Quoting any one height for “the turn” is what fuels the confusion.
| Month (2026) | Median completion altitude |
|---|---|
| May | 2,325 ft |
| June | 2,375 ft |
| July | 2,292 ft |
| August | 2,196 ft |
This also sets the stakes for proposals that turn aircraft lower: on the measured falloff, dropping the turn from ~2,300 ft to ~1,000 ft adds on the order of 6 dB to the peak directly beneath it.
Altitude convention. These figures use the logging convention in place through 2026-10-19. From 2026-10-20 the measurement app references altitude to mean sea level (GNSS geoid / barometric-QNH correction), which reads about 100 ft higher for the same aircraft; figures recomputed on new-convention data will sit correspondingly higher, and old and new logs must not be pooled. The new datum (WGS-84 ellipsoid) is pending confirmation against a live reference-track measurement.
The decisions that define the model
- Metric = single-event peak dBA. DNL and LAeq,16h are deliberately excluded.
- Both procedures as idealized single lines. WARYN-BRGHT is notional, so it has no measured spread; for a fair comparison today's 15R is also its median centerline, not its cloud.
- Today's line = the ~79% west-curve majority, built along-track and extended to the published BOBYJ leg.
- WARYN-BRGHT line shares the runway segment (so near-in differences are ~0), then runs WARYN → BRGHT.
- Altitude = the measured 15R climb, draped by along-track distance — the piece independently confirmed by the 2,889-flight record and the receptor check.
- Peak = the law at the 3-D minimum slant range between each home and the line.
How precise the numbers are
Error band = ±7 dB (95%). The residual standard deviation of the 249-point fit is 3.7 dB, reflecting real variation across aircraft type, weight and weather. Differences smaller than this band are not individually distinguishable per home. The defensible reading is the spatial pattern of where exposure shifts, not any single home's exact value — except the receptor's own level, which is anchored to measurement. (Modeling aircraft type explicitly tightens the band to about ±6.3 dB; that refinement is in progress and is not used in the map figures.)
What the model finds
Corridor (33,206 homes), single-event peak dBA — counts by band (<55 / 55–60 / 60–65 / 65–70 / ≥70):
| Scenario | <55 | 55–60 | 60–65 | 65–70 | ≥70 |
|---|---|---|---|---|---|
| Today's 15R | 879 | 3,492 | 10,019 | 14,512 | 4,304 |
| WARYN-BRGHT | 4,501 | 5,317 | 11,627 | 7,210 | 4,551 |
By per-home change: quieter by ≥1.5 dB 16,772, within ±1.5 dB 7,256, louder by ≥1.5 dB 9,178; median change −1.75 dB — net relief by roughly two-to-one, at the cost of a real increase on the southeast side.
Schools (peak dBA, ±7 dB), most-exposed-today first:
| School | Today | WARYN-BRGHT | Δ |
|---|---|---|---|
| Quarterfield Elem. | 74.5 | 70.9 | −3.6 |
| Glen Burnie Park Elem. | 70.2 | 71.3 | +1.2 |
| Rippling Woods Elem. | 64.3 | 68.8 | +4.5 |
| Severn Elem. | 62.1 | 58.3 | −3.7 |
| Southgate Elem. | 59.9 | 63.1 | +3.2 |
The most-exposed school today, Quarterfield, is among those helped most; two schools (Rippling Woods, Southgate) see real increases, stated plainly. The robust, cross-option fact is about altitude over children: WARYN-BRGHT is the only option studied that never sends a pass below ~2,000 ft over any school, while a low turn drops a ~1,800-ft overflight onto the schools nearest the runway.
Common misconceptions
“TrueNoise's numbers are above the FAA's 65 limit.”
No — these are single-event peaks (the loudest instant of one flyover); 65 is a DNL 24-hour-average land-use contour. Different quantities, different scales. Our numbers cannot be compared to, or converted into, a DNL 65 figure.
“What altitude do they turn at?” — there isn't one number.
On 2,889 measured flights, the turn begins around 1,800 ft and completes around 2,300 ft over the community, climbing west. Aircraft are lowest over the close-in neighborhoods and schools. Quoting one height — high or low — is what fuels the confusion.
“The monitor's readings are the worst case.”
The opposite. The fixed monitor sits ~3.4 nm out, in the turn-completion zone, where its measured median peak is 70.5 dBA. Homes closer to the turn, where aircraft are lower, hear louder passes. The published receptor figures are a conservative, low-end anchor, not a maximum.
“Spreading the flights out will fix it.”
Spreading moves noise; it does not lower it. Loudness is set by altitude, not by how tightly the tracks bunch. Fanning a low turn across more streets spreads a louder-per-event overflight onto more homes — and if the turn is forced lower to do it, every home under the wider band is louder. Only raising the turn makes each pass quieter for everyone beneath it.
Equipment and its limits
Measurements come from an IEC 61672 Class 2 monitor and a co-located airnoise.io device. Class 2 is the right grade here for two reasons. First, the analysis is comparative — today's procedure against a proposed one at the same homes — so any fixed instrument offset applies to both and cancels in the difference. Second, the instrument has been checked against a calibrated meter and found within ~0.8 dB, well inside the ±7 dB screening band. This is a measured community screening, not a certified Class 1 network and not an FAA AEDT model; where a regulatory-grade number is required, these figures show where to look, not the final certified value. The measurement app’s full signal chain — A-weighting, SPL computation, calibration, psychoacoustic analysis, and ADS-B correlation — is documented in the Aircraft Noise Tracker technical brief.
Scope limits and honest caveats
- Metric: peak dBA only — not DNL, not LAeq,16h. Do not convert to either.
- The west-curve share is load-bearing. The whole west-relief / southeast-increase split rests on ~79% of today's departures curving west; that share is measured but not yet ground-truthed against a full operations record.
- The single line is weakest where flights fan out — between the turn and BOBYJ the measured tracks diverge.
- The climb is a median; individual departures run higher or lower (part of the ±7 dB band).
- Idealized single lines understate how many receptors real dispersed operations affect; precise dispersion is for FAA AEDT modeling.
- WARYN-BRGHT is unflown — projection, not observation.