bumpwise.app/technology
How Bumpwise predicts turbulence
No consumer app had ever fused operational aviation forecasting with live ground truth until now. Bumpwise blends the model air traffic control uses, live pilot reports, real time aircraft telemetry, and a physics based heuristic engine into one honest, per minute intensity timeline.
First of its kind, built from the ground up. Twelve signals. One fused score. What pilots and dispatchers already trust, made clear for passengers.
Core thesis
Most turbulence apps show you a map. Bumpwise pioneered the fusion engine: seven independent data sources sampled along your exact route at the exact minute your aircraft will be there, fused into a single per minute intensity timeline.
Nothing in the app is a vibe. Every number a passenger sees traces back to a physical measurement or a published operational model.
Pioneers in turbulence intelligence
Twelve signals. One fused score.
No consumer app had fused operational aviation forecasting with live ground truth until Bumpwise. We blend the model air traffic control uses, live pilot reports, real time aircraft telemetry, and a physics based engine into one honest, per minute intensity timeline.
Every other turbulence tool mostly renders a government map and asks you to eyeball your route. Bumpwise built the fusion engine airline dispatchers effectively run by hand, automatically, before you board.
7
Independent data sources
3 min
Timeline resolution
NOAA first
When EDR coverage is available
±30 nm
Route corridor coverage
Explore every spoke on the web
Each spoke is an independent measurement the engine pulls, weights, and reconciles for every flight, not a marketing checklist. Hover a node to see how that layer works. Competing apps typically plot one or two of these; Bumpwise is the only consumer product pulling from all twelve in real time.
Built on what pilots and ATC already trust
Where NOAA’s operational EDR forecast is available, Bumpwise defers to it as the primary signal, the same Eddy Dissipation Rate model air traffic control itself uses. Live PIREPs, advisories, and physics fill the rest with capped trust so no single source can overwhelm the truth.
Hover a spoke on the web to explore each signal.
How a forecast is built
From raw atmosphere to a cabin ready timeline, the work a dispatcher would do, automated end to end.
- 01
Ingest seven sources
NOAA GTG, HRRR/GFS, PIREPs, AIRMETs/SIGMETs, convective advisories, FlightLabs schedules, and ADS B telemetry, each contributing a different kind of truth.
- 02
Build the flight corridor
Sample the centerline and ±30 nm edges. Assign each waypoint an ETA from flight progress so forecasts match when you will actually be there.
- 03
Run the physics engines
Surface wind and gust intensity for takeoff and landing. Upper intensity from vertical shear and CAT proxy for cruise, plus horizontal shear between waypoints.
- 04
Fuse with operational truth
Where NOAA EDR is available, the government model leads and the Bumpwise heuristic stays in a supporting role so live PIREPs and advisories can still register. Trust ceilings keep any single layer from dominating.
- 05
Render the per minute timeline
A smooth intensity timeline across the flight, then adjusted for aircraft feel. Average, peak patch, and a plain English ride sentence. Nothing is a vibe.
A per minute intensity timeline
The engine builds an intensity value throughout the flight, then smooths it across a rolling window. Each phase starts from a calibrated baseline. Takeoff and landing run hotter than cruise because ground adjacent air is inherently rougher, then layers in live physics.
Average stays honest. Peak patch never hides.
The average is the mean of the whole timeline, never inflated by spikes. The peak patch is the single bumpiest expected minute, always matching the chart’s high point. Hover a phase to see how that segment is scored.
Corridor, not a city weather guess
Bumpwise never looks up “the weather for a city.” It constructs the actual flight, a time resolved corridor through the atmosphere, and scores each point with the forecast for the moment the aircraft will be there. Hour 3 of a transcontinental flight is scored with hour 3’s forecast, not departure time weather.
The NOAA EDR blend
Where the GTG decoder returns real EDR for a route segment, NOAA’s operational number becomes the primary signal, and the Bumpwise heuristic stays in a supporting role so live PIREPs and advisories can still register.
NOAA operational EDR leads when available
Bumpwise physics and live reports stay secondary so the government model can do its job without disappearing from the forecast.
The app only claims “real NOAA EDR” when this blend actually happened. If coverage fails where it should exist, confidence is docked and the claim is suppressed. We never overstate certainty.
Before Bumpwise
Static forecast map. One data source.
No sense of time of day, aircraft type, or how far into the flight a cell of weather actually is. The passenger does the fusing, in their head, badly.
The Bumpwise standard
Seven sources fused per minute, per mile, per aircraft.
The engine does the work a dispatcher would do, automatically, before you’ve boarded. Same physics aviation meteorology already trusts; the pioneering part is the fusion, honesty rules, and route and time exact sampling nobody else in the category had built.
Engineering deep dive
Everything that goes into one number
Twelve signals. One fused score. Each item below is an independent measurement the engine pulls, weights, and reconciles for every flight, not a marketing checklist.
Signal 01
NOAA DAFS GTG (Eddy Dissipation Rate)
The official EDR forecast, the international standard unit of turbulence, and the same Graphical Turbulence Guidance product air traffic control and airline dispatchers use. We decode raw GRIB2 grids and sample them along your route at cruise altitude. When it is available, Bumpwise defers to it as the primary signal.
Signal 02
Vertical Wind Shear
The classic clear air turbulence signature: altitude weighted wind change between FL300, FL340, and FL390, plus jet stream and peak shear bonuses. This is the physics pilots train to recognize in the cruise layer.
Signal 03
CAPE and Atmospheric Instability
Vertical motion and lifted index instability feed a CAT proxy. Convection is deliberately conservative: CAPE alone never moves the ride number. Realized precipitation is required before storms can raise intensity.
Signal 04
Surface Wind and Gusts
Departure and arrival wind and gust intensity drive takeoff and landing bumps. Gust differential is weighted more heavily than steady wind. That is what passengers actually feel on final approach.
Signal 05
Pilot Reports
Aviation’s only live ground truth turbulence signal. Reports within 100 nm of route and 5,000 ft of cruise from the last 6 hours, weighted by recency. A PIREP from 30 minutes ago counts far more than one from 5 hours ago. Trust is capped so reports can never overwhelm the physics.
Signal 06
NWS AIRMET Advisories
Official National Weather Service turbulence advisory polygons. Applied to cruise at a deliberately low weight. These polygons can cover half the country, and most of the air inside them is still smooth.
Signal 07
SIGMET Advisories
The more severe NWS advisory tier for significant turbulence. Still blended conservatively so a broad advisory never masquerades as a route specific forecast.
Signal 08
Convective SIGMETs
Official thunderstorm advisories applied only where your route actually intersects the storm polygon, not painted across the whole map. Mid flight, only the most serious cells register meaningfully, because crews vector around storms at altitude.
Signal 09
ADS B Aircraft Telemetry
The aircraft’s own real time radio beacon for position, altitude, and groundspeed, used for movement truth. Status flips to In Air within 1 to 2 minutes of wheels up and Landed within minutes of touchdown, far faster than conventional status feeds.
Signal 10
Flight Schedule and Status Data
Schedules, live status, and delay estimates from FlightLabs so each waypoint is scored against when your aircraft will actually be there, including revised ETAs when the flight slips.
Signal 11
Aircraft Type Cabin Feel
Same air, different cabin. After the atmospheric blend is complete, aircraft type is applied last so the score reflects what you will feel in that seat, from wide body steadiness to regional jet and turboprop sensitivity.
Signal 12
Route Corridor and Time Sampling
Not city weather. A sampled corridor through the atmosphere. Centerline every 25 to 50 miles (capped at 30 points), plus ±30 nm perpendicular points so a storm just off the airway still registers. Each waypoint gets its own ETA so hour 3 is scored with hour 3’s forecast.
Two independent detectors, blended honestly
Surface physics scores takeoff and landing. Upper physics scores cruise. Convection is written to match how crews actually operate.
Surface wind intensity
Drives takeoff and landing bumps
- Stronger winds contribute more near the ground, with deliberate ceilings so one reading cannot runaway.
- Gusts weigh more than steady wind because that differential is what passengers feel.
- Gust differential is what a passenger actually feels on final approach.
Upper intensity
The cruise signal, max of two detectors
- Vertical wind shear between FL300, FL340, and FL390 catches classic clear air turbulence.
- A CAT proxy blends vertical motion and lifted index instability.
- Horizontal shear between adjacent waypoints catches fronts and jet stream exits at lighter weight.
Conservative convection rules
CAPE alone never moves the ride number
- Huge CAPE with no rain is just a summer afternoon, so precipitation has to be realized.
- Near departure and arrival, storms matter more because crews have less room to route around cells.
- Mid flight, only the most serious cells register meaningfully, because crews vector around storms at altitude.
Ground truth, weighted with restraint
Live human and advisory layers are blended in, but capped, so a handful of reports or a broad advisory polygon can never overwhelm the physics underneath.
| Layer | Applies to | Role | Rationale |
|---|---|---|---|
| PIREPs | Cruise, climb, descent | Capped ground truth | Recency weighted reports from nearby aircraft. Climb and descent use less trust than cruise so reports never overwhelm the physics. |
| AIRMETs | Cruise | Light advisory weight | Advisory polygons can cover half the country. Most of the air inside them is still smooth. |
| SIGMETs | Cruise | Stronger advisory weight | More severe advisory tier, still blended conservatively. |
| Convective SIGMETs | Route intersecting waypoints only | Only where routes cross | Only applied where the route position actually falls inside the storm polygon. |
Same air, different cabin experience
Aircraft type is applied last, after the atmospheric blend is complete, so it never contaminates the underlying physics.
| Airframe | Examples | Cabin feel |
|---|---|---|
| Wide body | 777 / 787 / A350 | Steadiest cabin ride |
| Narrow body | 737 / A320 | Standard reference ride |
| Regional jet | CRJ / ERJ | More sensitive to bumps |
| Turboprop | Propeller regionals | Most sensitive cabin feel |
The two headline numbers
Average
The honest mean of the whole timeline. Never inflated by spikes. A mostly calm flight with one bumpy patch stays calm overall.
Peak patch
The timeline’s exact maximum. The single bumpiest expected minute, and it always matches the chart’s highest point.
Ride quality sentence
Computed separately and spike aware. It counts distinct excursions and sustained minutes above thresholds, so a calm average can never hide a real bumpy stretch, and a single blip can never scare anyone unnecessarily.
A rigor check no user ever sees
Bumpwise does not surface a confidence percentage to passengers. Instead, an internal checklist governs what the app is allowed to claim, not what it displays.
Strong coverage
The app is permitted to badge a segment as backed by real NOAA EDR data.
Partial coverage
The full timeline still renders. The NOAA badge is withheld until coverage improves.
Thin coverage
The engine leans harder on PIREPs and advisories rather than presenting thin data as settled fact.
An EDR expected but missing condition subtracts confidence outright. The gate exists to keep the engine honest with itself. What the user sees is always the same: a clear timeline and a plain English ride description.
Always fresh
Forecasts recompute on a cadence that tightens as departure approaches. Delay feeds stamp revised estimates. ADS B flips status to In Air within 1 to 2 minutes of wheels up.
| Window before departure | Recompute interval |
|---|---|
| Beyond 48 hours | Every 6 hours |
| Inside 48 hours | Every 2 hours |
| Inside 6 hours | Every 30 minutes |
| Inside 12 hours, screen open | Silently, every 10 minutes |
Never fully down
Resilience is designed in. The forecast requires at least one real turbulence source, but individual feeds are wrapped so one outage cannot break the product.
- ✓Winds use a circuit breaker so a flaky feed cannot freeze the search.
- ✓EDR has a short ceiling, then the heuristic keeps the timeline alive.
- ✓PIREPs and advisories degrade to empty rather than inventing coverage.
- ✓Delay and ADS B wrappers cannot fail a whole search.
- ✓Only when every turbulence source is down does the user see a friendly retry card, never a broken screen.
Tuned against industry benchmarks. Validated in the air.
None of the constants in this engine are guesses. They were tuned against industry reference routes such as JFK to LAX jet stream crossings and MIA to JFK summer storm corridors, then validated against real flights.
In flight validation
The model correctly placed the bumpy final 30 minutes of an actual MIA to ATL flight flown during development.
Storm logic rebuilt
Rebuilt after a live test against the sky out the window. That is the origin of the realized precipitation only rule.
The physics is standard aviation meteorology. The engineering is the fusion, the honesty rules, and the route and time exact sampling nobody else in the category had built before. That is what pioneering looks like in this space: not a new gauge, a new discipline.