86,000 events · interactive analytics
The accident record, sliced.
Every NTSB-investigated accident and incident on file — — events from —, plotted, ranked, cross-tabbed. Click any bar, year, or cell to drop into a filtered map. No editorializing — just the agency's own counts, surfaced.
Loading 86 k events…
01 · Events per year
Yearly volume, severity-stacked.
Line height = events that year. Color band shows severity mix (fatal / serious / minor / none). Click any year to open the map filtered to it.
02 · Top aircraft makes
By total events.
Top 25 manufacturers by NTSB event count. Bar segments show severity mix. Click any bar to filter the map by that make.
03 · Top models
By model.
Pick a make to drill into its top 25 models. Without a make selected, this is the global top 25 across all manufacturers.
04 · By US state
Where the events happen.
Top 20 US states + territories by event count. Fatal-event count called out separately. Click any bar to deep-link the map to that state.
05 · Light × Severity
Time of day, by outcome.
Cell intensity scales with count. Fatality rate per cell shown as a percentage of total events in that light condition.
06 · Seasonality
By calendar month.
Total events per month across the entire dataset. Summer GA peak vs. winter trough is visible without commentary.
07 · Severity mix
Outcome mix overall.
The full record by highest-injury classification per event. Fatal and Serious make the news; None dominates the volume.
07b · Numbers
| Severity | Events | % of total | Fatalities |
|---|
08 · Time of day
When the day breaks.
Every event with a recorded time, bucketed into 30-minute slices of the local clock and laid around a 24-hour radial. — events plotted. Notice the late-afternoon ridge — and the 5 a.m. notch where almost nothing is flying.
09 · Relative to civil twilight
The twilight risk window.
For every event with a date, time, and coordinates we compute the actual sun position — civil twilight at that exact lat/lon on that exact day — and bin the event by minutes-from-sunset. Nobody else surfaces this because nobody else does the per-event astronomy. Watch where the fatal mass sits.
Sun-position math: Spencer 1971 / NOAA solar position algorithm — exact within a few seconds.
11 · Hour × weekday
When the week breathes.
— events with both a known time and a parseable date, plotted as a 7 day × 24 hour heatmap. The Sunday-afternoon recreation peak. The Tuesday-morning training peak. Saturday-night silence.
12 · Causal findings
What the agency calls the cause.
33,427 causal-factor findings (post-2008 events with the formal NTSB taxonomy). Bars show events by severity per top-level category; the bullet list shows the top sub-categories within each. Personnel decisions outweigh aircraft failure roughly 1:1 — the post-2008 data is honest about pilot causation.
Personnel issues
14,984 findings · 44.8%
- Task performance 8,804
- Action/decision 4,440
- Psychological 1,360
- Physical 178
- Experience/knowledge 161
- Miscellaneous 41
Aircraft
13,710 findings · 41.0%
- Aircraft oper/perf/capability 8,263
- Aircraft systems 1,887
- Aircraft power plant 1,498
- Fluids/misc hardware 1,242
- Aircraft structures 343
- Aircraft propeller/rotor 305
Environmental
2,984 findings · 8.9%
- Conditions/weather/phenomena 1,570
- Physical environment 1,350
- Operating environment 50
- Task environment 14
Not determined
1,650 findings · 4.9%
- Not determined 1,650
Organizational
99 findings · 0.3%
- Development 40
- Management 32
- Support/oversight/monitoring 27
Source: NTSB CAROL findings table · post-2008 events with taxonomy-coded findings · cause_factor = C
10 · Density altitude × fatality rate
Hot & high, in numbers.
For every event where altimeter, OAT, and field elevation were recorded, we recompute density altitude using the standard ISA-deviation formula. — events binned by 1,000-ft DA bands. Bar height = events in that band; orange line = fatality rate within that band. The hot-and-high curve isn't anecdote.
Method.
Source
NTSB CAROL avdata MDB (1982 → current) + the Pre-1982 narratives MDB. Coordinates, severity, light, weather, make, model, state are extracted as published by the agency.
Severity
Highest-injury per event — Fatal > Serious > Minor > None > Unknown. The agency's own classification; we don't re-bucket.
Click-through
Each click navigates to /atlas/maps/accidents/ with URL parameters (?year=…&make=…&sev=…&light=…) so the filtered view is shareable and bookmarkable.
Caveats
Counts are events, not normalized rates — they don't divide by fleet hours, type-population, or exposure. A make with more events isn't necessarily less safe; it might just be more common. The data is what the agency publishes; the inferences are yours.