Multi-Model Turbulence Analysis
PlaneWX derives Clear Air Turbulence (CAT) potential from the same multi-model soundings used for icing analysis. By computing vertical wind shear and the Richardson number at every altitude level across three independent models, we give you altitude-specific turbulence forecasts with confidence scoring — far beyond what G-AIRMETs and PIREPs alone can provide.
Why Multi-Model Turbulence Matters
Traditional turbulence information comes from two sources: G-AIRMET Tango (large geographic areas with blanket severity ratings) and PIREPs (subjective reports from other pilots). Both have significant limitations for pre-flight planning:
G-AIRMET Tango: Too Broad
- • Cover hundreds of thousands of square miles
- • Single severity for the entire area and altitude range
- • No information about which specific altitudes are roughest
- • Don't indicate the physical mechanism causing turbulence
- • Updated every 3 hours — jet stream patterns shift faster
PIREPs: Subjective & Sparse
- • Severity is subjective — varies by aircraft type and pilot experience
- • "Moderate" in a 737 feels very different than in a Cirrus
- • Sporadic coverage, especially at night and in rural areas
- • Often hours old by the time you see them
- • Absence of PIREPs doesn't mean smooth air
PlaneWX's approach: We compute turbulence potential directly from model wind fields at multiple points along your route. Each model provides wind speed and direction at twenty-four pressure levels (1000–100 hPa) — we calculate the vertical wind shear between each pair of levels and the Richardson number to determine atmospheric stability. Comparing three independent models gives us confidence scoring that no single model can provide.
The Three Models
PlaneWX derives turbulence from the same model soundings used for icing — HRRR, GFS, and ECMWF. Each model resolves the atmosphere differently, and their wind field differences are themselves a signal about forecast uncertainty.
High-Resolution Rapid Refresh
NOAA's highest-resolution hourly model — best for capturing terrain-induced turbulence and mesoscale wind features that coarser models miss.
3 km
Every hour
0–18 hours
Mountain wave, low-level wind shear
Global Forecast System
NOAA's primary global model — good for large-scale wind patterns, jet stream positioning, and upper-level turbulence forecasts.
~13 km
4 times daily
0–16 days
Jet stream CAT, upper-level shear
European Centre for Medium-Range Weather Forecasts
The world's leading global forecast model — provides an independent European perspective on upper-level dynamics and wind patterns.
~9 km
4 times daily
0–10 days
Jet stream dynamics, moisture-driven instability
Why Three Models?
Each model resolves wind fields differently due to different grid resolutions, physics parameterizations, and initial conditions. When all three models agree on turbulence at a given altitude, confidence is HIGH. When they disagree, the atmosphere is uncertain — and that disagreement is itself valuable intelligence for your decision-making.
The Physics: Wind Shear & Richardson Number
PlaneWX derives turbulence from two fundamental atmospheric quantities: vertical wind shear (the engine that creates turbulence) and the Richardson number (the gate that determines whether the atmosphere allows it).
Vertical Wind Shear
Wind shear is the change in wind speed between two altitude levels, expressed in knots per 1,000 feet. When wind speed changes rapidly with altitude, the resulting shear can generate turbulence — especially near jet streams, frontal zones, and temperature inversions.
Example calculation:
18,000 ft: 47 kt from 251°
14,000 ft: 31 kt from 278°
Speed difference: 16 kt over 4,000 ft
Shear = 16 / 4 = 4.0 kt/1000ft
Note: PlaneWX also accounts for directional shear (wind direction changes between levels), which can produce turbulence even when speed shear alone appears modest.
Richardson Number (Ri)
The Richardson number is a dimensionless ratio that measures the balance between thermal stability (which suppresses turbulence) and wind shear(which generates it). It answers: "Is the atmosphere stable enough to resist the shear trying to break it up?"
Ri = (g / θ) × (dθ/dz) / (dU/dz)²
Where g = gravity, θ = potential temperature, dθ/dz = static stability, dU/dz = wind shear
Important Nuance: Ri Is a Gate, Not a Severity Scale
A low Ri tells you the atmosphere can support turbulence — it does NOT tell you how severe that turbulence will be. Ri < 0.25 is common in model data near jet streams and doesn't automatically mean severe turbulence. Think of Ri as a gate: when Ri ≥ 0.5, severity is capped at LIGHT; when Ri < 0.25 and shear ≥ 4 kt/1000ft above 5,000 ft AGL, severity is boosted one level (KHI instability). The KHI boost is suppressed in the boundary layer (≤ 5,000 ft AGL) where low Ri is normal mechanical mixing. How intense the turbulence gets depends on the wind shear magnitude, not on Ri alone.
Severity Classification
PlaneWX classifies turbulence severity using a multi-step process. First, wind shear magnitude determines the baseline severity. Then, the Richardson number acts as amodifier — capping severity in stable conditions and boosting it when dynamic instability is detected.
Step 1: Shear-Based Severity
Wind shear magnitude is evaluated first using GA-calibrated thresholds (see Step 2 below). The Richardson number then modifies this baseline.
Step 2: GA-Calibrated Shear Thresholds
Wind shear magnitude determines the baseline severity:
< 6 kt/1000ft
SMOOTH
No significant turbulence detected by models. This does not guarantee smooth air — localized turbulence from convection, terrain, or wake is always possible and model resolution cannot capture all sources.
≥ 6 kt/1000ft
LIGHT
Slight, erratic changes in altitude and/or attitude. GA pilots report light chop at energy levels well below what transport-category aircraft pilots notice.
≥ 12 kt/1000ft
MODERATE
Changes in altitude and/or attitude occur. The low-moderate threshold where GA structural stress begins. Occupants feel definite strains against seat belts.
≥ 20 kt/1000ft
SEVERE
Large, abrupt changes in altitude and/or attitude. Model-derived shear ≥ 20 across a wide layer is almost always a severe event for light GA aircraft (3,000–6,000 lb).
GA-Calibrated Thresholds
PlaneWX thresholds are specifically calibrated for General Aviation aircraft (3,000–6,000 lb). GA aircraft experience turbulence at much lower energy states than transport-category aircraft — a “Light” report from a Boeing 737 corresponds roughly to “Moderate” for a Cessna 172 or DA40. Additionally, model data at 2,000–6,000 ft vertical resolution smooths peak shear by 40–60%, so a measured 8 kt/1000ft across a bulk layer often masks a 15+ kt/1000ft shear in a 500 ft sub-layer. The 6/12/20 kt thresholds account for both GA airframe sensitivity and model smoothing.
Stability Cap & KHI Boost
When Ri ≥ 0.5 (stable or transitionally stable), severity is capped at LIGHT — never overridden to SMOOTH. Bulk Richardson numbers from NWP model layers (2,000–6,000 ft thick) can miss thin shear layers near fronts, low-level jets, and mountain waves where turbulence co-exists with a stable bulk layer (Sharman et al. 2006). Capping at LIGHT instead of clearing to SMOOTH prevents dangerous false clears.
Conversely, when Ri drops below 0.25 (dynamic instability / Kelvin-Helmholtz onset) above 5,000 ft AGL and shear is at least 4 kt/1000ft, severity is boosted by one level. This captures turbulence from KHI billows in environments near the tropopause or convective boundaries — a source of false negatives in shear-only models.
Boundary layer suppression: The KHI boost is suppressed below 5,000 ft AGL. Low Richardson numbers in the atmospheric boundary layer are routine mechanical mixing from surface friction and gusty winds — not the free-atmosphere dynamic instability that KHI detection targets. This prevents strong surface gusts (e.g., 40 kt) from triggering false SEVERE classifications in the lowest flight levels during approach and departure.
Severity band downgrading: Separately from the KHI boost suppression, any SEVERE turbulence band that lies entirely below 5,000 ft AGL is automatically downgraded to MODERATE and flagged as low-level wind shear. Strong surface gusts during approach and departure can produce high model-derived shear in the lowest layers, but this is a surface phenomenon — not the kind of en-route severe turbulence that warrants a SEVERE classification in your briefing. The 5,000 ft threshold is computed from airport elevation when available, so it works correctly at high-elevation airports like Denver (5,431 ft MSL → cap at ~10,431 ft MSL).
Understanding “SMOOTH” Values in the Table
When you look at the turbulence table in the Models tab, you’ll often see many levels showing SMOOTH across all models. This means no significant turbulence signal was detected by the models — not a guarantee of smooth air. Here’s why:
Low Shear (< 6 kt/1000ft)
Wind shear below the GA detection threshold at model resolution. You might see Ri = 0.41 with shear = 4.5 showing SMOOTH — the atmosphere is marginally unstable, but the shear isn’t strong enough for models to predict perceptible turbulence. Localized turbulence from convection, terrain, or wake is always possible.
Stable Layers Show LIGHT, Not SMOOTH
When Ri ≥ 0.5 (stable atmosphere) and shear is above the light threshold, severity is capped at LIGHT rather than cleared to SMOOTH. This conservative approach prevents dangerous false clears — bulk Ri from NWP model layers (2,000–6,000 ft thick) can miss thin shear layers near fronts, low-level jets, and mountain waves.
What Highlighted Rows Mean
Levels with turbulence (LIGHT, MODERATE, SEVERE) are highlighted with an amber background. These are the altitudes where wind shear is strong enough to generate bumps. A table with mostly SMOOTH and a few highlighted rows tells you where the models detect turbulence — but remember that SMOOTH means “no significant modeled signal,” not “guaranteed smooth air.”
Reading the Color-Coded Ri Column
The Ri column in the turbulence table uses color coding to help you quickly assess atmospheric stability:
Red
Ri < 0.25
Unstable
Amber
Ri 0.25 – 0.5
Transitional
Green
Ri ≥ 0.5
Stable (cap at LIGHT)
Multi-Model Consensus
Each model independently computes wind fields, which means each has its own shear and Ri values. PlaneWX combines them into a consensus assessment using a conservative-max approach.
Conservative Max Consensus
For safety-critical aviation, PlaneWX uses a conservative max consensus: the highest severity predicted by any model is used, unless that model is an outlier by two or more categories (in which case it is tempered down by one level).
- Safety first: If the high-resolution HRRR detects severe turbulence that GFS and ECMWF smooth away, a median-based approach would report “Light” — a dangerous false clear that could lead a pilot into a KHI zone or rotor.
- Outlier protection: If one model says MODERATE and two say SMOOTH (a 2-category gap), the consensus is tempered to LIGHT rather than accepting the outlier value at face value.
- Per-model columns: You can still see each model's individual assessment in the table. If models disagree, the per-model columns make that visible so you can factor it into your decision.
HIGH
Unanimous
All models agree on turbulence severity at this altitude. The forecast is reliable.
MODERATE
Majority
Most models agree. One model sees different conditions. The majority view is likely correct, but consider the outlier model's perspective.
LOW
Split
Models disagree significantly. This happens near jet stream boundaries where small position differences produce big shear differences. Plan conservatively.
Route Sampling
PlaneWX samples 3 to 7 points along your route (depending on route length) and queries all three models at each point. This captures how turbulence conditions change along your path — critical for long flights that cross jet stream boundaries or frontal zones.
At Each Sample Point, We Compute:
- Wind speed and direction at 24 pressure levels (1000–100 hPa) from each model
- Vertical wind shear between each pair of adjacent levels
- Richardson number using temperature gradient and shear
- Per-model severity assessment at each level
- Multi-model consensus severity and agreement
- Turbulence layer boundaries (base and top altitudes)
What We Detect (and What We Don't)
PlaneWX's model-derived analysis is designed to detect Clear Air Turbulence (CAT) — turbulence caused by wind shear in clear air, typically near jet streams, frontal zones, and temperature inversions. This is the most common type of turbulence at cruise altitudes.
What Our Model Analysis Detects
- • Clear Air Turbulence (jet stream CAT)
- • Upper-level wind shear zones
- • Frontal zone turbulence
- • Temperature inversion shear layers
- • Tropopause-level turbulence
- • Mountain wave turbulence (cross-barrier flow detection)
Covered by PIREPs & G-AIRMETs
- • Convective turbulence (thunderstorms)
- • Mechanical turbulence (terrain/buildings)
- • Wake turbulence
- • Low-level wind shear (LLWS)
PlaneWX includes PIREP-reported turbulence and G-AIRMET Tango advisories alongside the model analysis to provide complete coverage of all turbulence types.
Mountain Wave Detection
Mountain wave turbulence is a leading cause of GA accidents, and standard vertical wind shear models can miss it entirely if the shear is low but terrain-forced gravity waves are present. PlaneWX includes a cross-barrier flow check that detects conditions conducive to mountain wave turbulence independently of the shear/Ri analysis.
How It Works
- PlaneWX fetches a detailed terrain elevation profile along your route using a dedicated elevation API.
- Ridge orientation is estimated dynamically from the actual terrain gradient, with confidence predicates that assess gradient magnitude and bearing coherence across neighboring points. Low-confidence bearings trigger conservative fallbacks.
- Model sounding winds are interpolated at ridge level and decomposed into perpendicular-to-ridge and parallel-to-ridge components.
- Relief-scaled thresholds: Wind speed thresholds are dynamically adjusted based on terrain relief. Lower-relief terrain (Appalachians, coastal ranges) uses lower thresholds to avoid under-detection, while major barriers (Rockies, Sierra) retain higher thresholds. The formula is: base = 18 + (relief_ft / 1,000) × 4 kt, with SEVERE = base + 10 kt, MODERATE = base + 5 kt, LIGHT = base.
- Advisory tier: 15+ kt perpendicular flow over terrain with ≥2,000 ft relief generates a marginal mountain wave advisory in your briefing. This carries no WX Score penalty but alerts you to conditions where light to moderate mountain wave turbulence is possible near ridgelines.
- Sub-2,000 ft terrain promotion: Lower-relief terrain like the Appalachians, Ozarks, and coastal ranges (<2,000 ft crest-to-valley relief) can still produce significant turbulence when flow is blocked. When the Froude number indicates blocked flow (Fr < 0.60) and cross-ridge winds are ≥25 kt, severity is promoted by one level even over modest ridges. This prevents under-detection of real hazards in lower-relief terrain.
- Low-confidence ridge fallback: When terrain gradient is weak or ridge orientation cannot be determined with confidence, PlaneWX uses 70% of the total wind speed as a conservative estimate of the perpendicular component (the RMS value of |sin(θ)| over all possible ridge orientations). This avoids worst-case 100% assumptions that would over-alert, while still flagging genuine risk.
- Results feed into pre-flight checks, WX Score deductions, and the LLM briefing prompt as terrain-specific hazards.
Data availability: Mountain wave and rotor detection require model wind data at ridge level. If model soundings are unavailable (e.g., fetch timeout), PlaneWX will prominently note that wind-terrain interaction analysis was not performed and only terrain clearance was assessed. The briefing will explicitly warn when mountain wave analysis is unavailable over mountainous routes.
Mountain wave turbulence can occur in perfectly clear skies with no G-AIRMET coverage. Rotors and lee-side downdrafts near ridgelines are particularly hazardous for light GA aircraft with limited climb performance. When PlaneWX flags mountain wave potential, treat it seriously.
Shear Guard Protection
NWP models report winds on thick pressure layers (2,000–7,500 ft apart) that can smooth out real-world shear. The bulk Richardson number (Ri) — which PlaneWX normally uses to cap severity at LIGHT when Ri ≥ 0.5 — can therefore mask significant shear signals. The Shear Guard prevents this false-quiet failure mode.
Altitude-Banded Thresholds
When resolved shear exceeds the guard threshold for that altitude band, the Ri cap is bypassed and the shear-based severity stands:
Low Level
1000–750 hPa
≥ 18 kt/1000ft
Mid Level
749–500 hPa
≥ 14 kt/1000ft
Upper Level
< 500 hPa
≥ 12 kt/1000ft
Lower-altitude layers are relatively thinner and resolve shear more faithfully, so a higher threshold is appropriate. Upper layers are much thicker (~2,500–7,500 ft between pressure levels), heavily smoothing shear — a lower threshold protects weaker signals that likely represent stronger real-world shear.
These thresholds were validated against 1,034 GA-specific PIREPs and calibrated to keep false alarms within the FAA-accepted ~10% tolerance. The methodology was independently reviewed and stress-tested before deployment. Read the full methodology review
Mountain Rotor Detection
Rotors are violently turbulent circulations that form on the lee side of mountain ridges when airflow is partially or fully blocked. They are among the most dangerous phenomena for GA aircraft, capable of producing updrafts and downdrafts exceeding aircraft performance. PlaneWX uses Froude number analysis to detect rotor risk.
Froude Number (Fr)
Fr = Uperp / (N × h), where all quantities are in SI units:
- Uperp — perpendicular wind component across the ridge in m/s (converted internally from knots; 1 kt = 0.514 m/s)
- N — Brunt-Väisälä frequency (atmospheric stability, s-1), computed from potential temperature lapse between valley and crest
- h — effective ridge height (meters), max-min elevation within ±2.7 NM window, requiring ≥500 ft relief
SEVERE Rotor
Fr < 0.60 and Uperp ≥ 25 kt (12.9 m/s)
Blocked flow — hydraulic jump
WX Score: NO-GO
MODERATE Rotor
Fr 0.60–0.75 and Uperp ≥ 20 kt (10.3 m/s)
Partial blocking — trapped lee waves
WX Score: −30 points
Rotor turbulence occurs below ridge crest level and is invisible to conventional turbulence forecasts. When PlaneWX flags rotor risk, avoid flying on the lee side of the ridge below crest altitude. Consider routing upwind of the ridge or crossing well above ridge crest level. Note: Froude-number rotor detection requires terrain relief ≥ 500 ft and a high-confidence ridge orientation (gradient ≥ 100 ft/NM with coherent neighboring bearings). When ridge geometry is uncertain, the system skips rotor analysis and falls back to the cross-barrier wind mountain wave heuristics described above, using a conservative wind estimate.
Scorer Parameter Gate
Not all blocked-flow conditions produce rotors. Rotor turbulence requires trapped lee waves, which only form when the atmospheric profile supports wave trapping. PlaneWX computes the Scorer parameter (l²) from model soundings between ridge crest and ~4 km above crest to verify this condition:
l² = (N² / U²) − (1/U)(d²U/dz²)
- When min(l²) ≥ 0.25 between crest and crest+4 km, wave trapping is supported and the rotor alert stands.
- When min(l²) < 0.25, waves propagate vertically rather than trap — rotor formation is suppressed and the alert is removed. This prevents false alarms in common winter jet-stream scenarios where strong upper-level winds prevent wave trapping.
- When sounding data is insufficient to compute the Scorer parameter, the system falls through to the Froude-number analysis alone (conservative).
Complex Terrain Modifier
Simple 2D Froude analysis can over-predict rotor severity in complex 3D terrain where multiple ridge orientations exist (sinuous valleys, overlapping ridges). PlaneWX computes the circular variance of ridge-crest orientations within the analysis window. When variance exceeds 0.35 (indicating highly complex terrain), the Froude number is inflated by 25% to account for 3D flow-splitting effects that weaken rotors. Deeply blocked flows (Fr < 0.35) are exempt from this adjustment because 3D splitting can concentrate downslope flow in confined valleys.
Methodology Review
PlaneWX’s mountain wave and rotor detection algorithms are validated against published mountain meteorology literature (FAA AC 00-57, Durran 1990, Vosper 2004, Reinecke & Durran 2008) and continuously verified through automated regression testing against known mountain routes including Rocky Mountain, Sierra Nevada, Appalachian, and Colorado Front Range crossings.
PIREP GA Normalization
A “Moderate” turbulence report from a Boeing 747 (~875,000 lb) is a vastly different experience than “Moderate” in a Cessna 172 (2,300 lb). PlaneWX uses a three-tier weight classification to normalize PIREP turbulence intensity for GA aircraft sensitivity.
Weight Categories
Light
< 15,500 lbs
C172, DA40, SR22
No adjustment
Medium
15,500 – 300,000 lbs
B737, A320, CRJ, E-jets
Half-step bump
Heavy
> 300,000 lbs
B747, B777, A380, A330
Full-category bump
Normalization Examples
- Heavy aircraft (B747, B777, A330, A380): “Light” becomes “Moderate” and “Moderate” becomes “Severe” in your briefing.
- Medium aircraft (B737, A320, CRJ, E-jets, biz jets): “Light” becomes “Light-Moderate” and “Moderate” becomes “Moderate-Severe.”
- “Smooth” and “Negative” reports are not adjusted — smooth air is smooth regardless of aircraft weight.
- Reports from Light GA aircraft are not adjusted — they are already calibrated for your experience.
- Unknown aircraft types are flagged with a low-confidence note in the briefing. The intensity is displayed as-reported (not normalized), and the briefing explicitly notes that the aircraft type could not be identified. This helps you weigh the report appropriately.
Correlation with PIREPs
Your turbulence briefing includes both model analysis and recent PIREPs. PlaneWX correlates the two to give you a complete picture:
Models and PIREPs Agree
When model-predicted turbulence at an altitude matches PIREP-reported turbulence nearby, confidence in the forecast increases. Your briefing will note this correlation.
PIREPs Exceed Model Predictions
Pilots reporting worse turbulence than models predict is common — models can underpredict due to coarse resolution. Your briefing will flag this discrepancy and prioritize the PIREPs as ground truth.
Models Show Turbulence, No PIREPs
Absence of PIREPs doesn't mean smooth air — it may mean no one has filed a report. Model-predicted turbulence without PIREP confirmation should still be taken seriously, especially for future flights where PIREPs don't exist yet.
How Turbulence Affects Your WX Score
Turbulence deductions use your personal minimums — the same soft/hard limit system used for icing, winds, and visibility.
Soft / Hard Limit System
Comfort (Soft Limit)
The turbulence severity you're comfortable with. Exceeding this puts you in the caution zone.
Example: Soft = Light. If moderate turbulence is forecast, you'll see a caution deduction.
Limit (Hard Limit)
The maximum turbulence you'll accept. Exceeding this is unfavorable.
Example: Hard = Moderate. Severe turbulence forecast = unfavorable.
Reading the Turbulence Table
The turbulence table in the Models tab shows all the raw data behind the analysis. Here's what each column means:
| Column | What It Shows |
|---|---|
| Alt (ft) | Altitude of the upper level in each adjacent pair |
| Wind | Consensus wind speed at this altitude (average across models) |
| Dir | Consensus wind direction (degrees true) |
| Shear | Vertical wind shear in kt/1000ft between this level and the one below. Color-coded: red ≥ 9, amber ≥ 6, yellow ≥ 3 |
| Ri | Richardson number. Color-coded: red < 0.25 (unstable, KHI boost above 5,000 ft AGL), amber 0.25–0.5 (transitional), green ≥ 0.5 (stable, capped at LIGHT) |
| HRRR / GFS / ECMWF | Per-model severity assessment (SMO = smooth, LIG = light, MOD = moderate, SEV = severe) |
| Severity | Multi-model consensus severity (conservative max with outlier tempering) |
| Agreement | How well models agree: unanimous, majority, or split |
Limitations & Important Notes
Coarse Vertical Resolution
Weather models provide data at 24 pressure levels (1000–100 hPa), with layer thickness ranging from ~700 ft at low altitudes to ~7,500 ft at upper levels. Sharp shear layers that exist between these levels can be smoothed out in the data. Real turbulence can be more localized than what the model resolves. PIREPs from pilots at specific altitudes remain valuable for filling these gaps.
CAT Focus Only
The shear/Ri model analysis primarily detects Clear Air Turbulence (wind shear-driven). Mountain-wave risk is evaluated separately via the terrain-based cross-barrier flow detection described above. Convective turbulence (thunderstorms) and mechanical turbulence are covered by PIREPs, G-AIRMETs, and SIGMETs — all of which PlaneWX includes separately in your briefing.
Forecast, Not Observation
Model-derived turbulence is a forecast of where conditions will support turbulence, not a real-time measurement. Always cross-reference with current PIREPs for immediate flights. For flights hours or days out, the model analysis provides crucial advance intelligence that PIREPs cannot.
Not a Substitute for Official Briefing
PlaneWX's multi-model turbulence analysis supplements — but does not replace — official weather products. Always cross-reference with current G-AIRMET Tango, SIGMETs, and PIREPs. The pilot in command is always responsible for the final go/no-go decision per 14 CFR 91.3.
About This Documentation
PlaneWX's multi-model turbulence analysis is derived from the same HRRR, GFS, and ECMWF soundings used for icing analysis, accessed via the Open-Meteo API. The shear and Richardson number calculations are performed server-side in real-time for each briefing. As operational turbulence products (like GTG) become available via API, they can be integrated to further improve accuracy.
Research
- Iterative AI Validation of Turbulence Detection — 8 debates, 3 days, shear guard & rotor detection
- Adversarial AI Validation of Mountain Wave Detection — Scorer parameter, relief-scaled thresholds, 6 improvements