faa turbulence pdt turbulence forecasting graphical turbulence guidance (gtg) –nwp-model-based...

Overview of FAA Nowcast and D-CIT Overview of FAA Nowcast and D-CIT Goals and ApproachGoals and Approach

John WilliamsJohn Williams

CIT Mini-Workshop 2CIT Mini-Workshop 2

October 12, 2006October 12, 2006

FAA Turbulence PDT

Turbulence Forecasting• Graphical Turbulence

Guidance (GTG)– NWP-model-based

turbulence forecasts with dynamically-tuned “ensemble of experts”

• GTG Nowcast (GTGN)

En-route Turbulence• Automated in situ EDR

reports from commercial aircraft– United g-load method– New Southwest/Delta

method

Turbulence Remote Sensing• NEXRAD turbulence detection

algorithm (NTDA)– Nationwide radar-based

real-time detection• Other remote sensors

– Satellite data– Profilers– TDWR radar

• CIT avoidance guidelines– Evaluate current guidelines– Recommend alternatives– Develop “Diagnose-CIT”

algorithms

GTG-Nowcast (GTG-N) Motivation

• The Graphical Turbulence Guidance (GTG) product is an analysis/forecast product driven by NWP grids, PIREPs and (in next release) in situ EDR reports.

• Detection capabilities include PIREPs, in situ EDR reports, NEXRAD Turbulence Detection Algorithm (NTDA) grids, and satellite-based detection.

• A turbulence nowcast product, GTG-N, will fit between these timescales, integrating turbulence measurements, diagnostics, and (eventually) short-term convective wx. forecasts

NTDA grids

Cockpit display or alert ADDS: Dispatch, ATC, etc.

Graphic courtesy of virtualskies.arc.nasa.gov

GTG grids

CIT diagnostics

Satellite data

Conv. Wx nowcasts

In-situ reports, PIREPs, MDCRS winds and temperatures

GTG-N

GTG-N Concept

In situ Turbulence Measurement and Reporting System

Goal: Augment/replace subjective PIREPs with objective and precise turbulence measurements.

Features:

• Atmospheric turbulence metric: eddy dissipation rate (EDR).

• EDR can be scaled into aircraft turbulence response metric (RMS-g).

• Adopted as ICAO Standard

Increase in Spatial/temporal Coverage: UAL In situ Reports Compared to PIREPs

1.3 million EDR reports/month from 100 or so aircraft - compared to 55,000 pireps from all aircraft. Imagine with 400 SWA aircraft added!

737

757

737 +757

Radar Detection of In-cloud Turbulence

• NEXRAD Turbulence Detection Algorithm (NTDA) performs data quality control and produces in-cloud EDR on a polar grid

• To be installed on all NEXRADs, ingested and mosaicked at NCEP using NSSL mosaic algorithm

• Grids available to GTG-N and other users

NTDA Demonstration/EXPERIMENTAL TURBULENCE FI UAL███/AN N███UA UPLINK-- 05 Sep 2006 21:38:13Z FL 300 orient. 83 deg'+'=waypoint, '*'=route, 'X'=aircraft at 38.3N, 80.6W' '=no_data, 'o'=smooth, 'l'=light, 'M'=mod, 'S'=severe-----------------------(52 to IAD)------------------------ | * | *MM | *MM | *MM | l lll M *MMM | lollo *lMl | oolo *l | oo * | * |080 * M |llllllll ll *MM |lllllllllllllll l *lll |lllllllllollllllll *MMl l |MMllllllllooollllll *MMl l |MMMl lllllllllollll *MM l |MMMM llll llol ll *MM | MMMM ll *MMS | MMM *MSS | MM *MSS +PUTTZ +MSS | * SSSSSSMSS M | l S*SSSSSSSSSMSS M | lllS*SSSMSSSSSMSS l | lSMS*SMMMMSSSSMSS l | SSMM*MMMMMMMMMSS M | MM*SMMMMMMMMlllllMSS M | M*SSMMMMMMMMllllllMMM M | *MMMMMMMMMMMMMllllMMM | * MMlllMMMlMllllllMlll |040 * lllllllllllllllllMMllll | * llllllllllllllllS | * llllllllllMMS | * lllllllSSS | * lllllSSS | * lllMMS | * l llMM | * MM M | MM*MM oMSS | SM*MMMMM | MM *MM lMM | MMMSS S * l llM | MMMSSM * lllllll ol |l MMMM * lllllo | MMMMM * llllll | M * lllll | SSS * llllll | SSS M l * llll | SSS Mlllll *llM | MM lMlllll * l-------|---------valid-|--------X--------|-----------------90 +90|Left 40 2135Z (18 from 3819N/8058W) Right 40

Java web-based turbulence display

Sample uplink message Flight information

Legend

Route

Severe turbulence

Aircraft position

Waypoint

Vertical cross-section

Moderate turbulence

Severe turbulence

In situ turbulence

reports

Moderate turbulence

CIT Diagnostics (D-CIT) for GTG-N

Possible approach to CIT in GTG

GTG-N (0-30 min?)

NTDA

D-CIT (detection and NWP env. data)

Clear air: NWP and satellite-based diagnostics with in situ, MDCRS nudging and dynamic tuning

Forecast (30 min – 6 hours?)

D-CIT (NCWD conv. wx. forecast and NWP data)

Clear air: NWP and satellite-based diagnostics with dynamic tuning

“CIT Avoidance Guidelines” task goals

• Evaluate the FAA’s current thunderstorm avoidance guidelines for their effectiveness in mitigating CIT encounters

• Propose alternative guidelines if appropriate

• Develop CIT diagnostics module (“Diagnose-CIT”) for GTGN

Current avoidance guidelines• Don’t attempt to fly under a thunderstorm even if you

can see through to the other side. Turbulence and wind shear under the storm could be disastrous.

• Do avoid by at least 20 miles any thunderstorm identified as severe or giving an intense radar echo. This is especially true under the anvil of a large cumulonimbus.

• Do clear the top of a known or suspected severe thunderstorm by at least 1,000 feet altitude for each 10 knots of wind speed at the cloud top.

• Do circumnavigate the entire area if the area has 6/10 thunderstorm coverage.

• Do regard as extremely hazardous any thunderstorm with tops 35,000 feet or higher whether the top is visually sighted or determined by radar.

Data sources• Turbulence “truth” data

– In situ EDR reports from commercial aircraft– TAMDAR?– PIREPs (but may be too imprecise)– Field program data (NASA, IHOP, BAMEX, etc.)– NTSB accident cases, FDR data– NTDA EDR data?

• Thunderstorm feature data– Radar data (VIL, echo tops, NSSL 3-D reflectivity, NTDA)– Lightning data (NLDN)– GOES satellite data, IR-derived cloud tops– Conv. Wx. nowcast data (NCWD, NCWF)– Aircraft data (winds, temperature, EDR)

• Environment data– NWP model data and derived diagnostics

Approaches to Diagnose-CIT

• Detailed case studies

• Data mining based on comparisons of thunderstorm features and environment data with “truth” data (data mining = discovery of patterns or relationships through the automated analysis of data)– Improve understanding of processes involved – Build/tune automated diagnosis algorithm

• Fine-scale numerical modeling of interesting cases

• Feedback between these approaches

Statistical comparisons and Statistical comparisons and initial data mining resultsinitial data mining results

John WilliamsJohn Williams

CIT Mini-Workshop 2CIT Mini-Workshop 2

October 12, 2006October 12, 2006

Initial results: evaluating FAA thunderstorm avoidance auidelines

• For all aircraft turbulence reports over 11 months in summers of 2004 and 2005, computed– horizontal proximity to thunderstorms, indicated by

NCWD VIL values (e.g., >= 3.5 kg m-2, moderate rain)

– vertical proximity to thunderstorm tops, indicated by radar echo tops

• Stratified data by proximity values• Determined frequency of different levels of turbulence

in each proximity range and compared to average frequency to get relative “risk”

Results: Horizontal proximity

Turbulence categories:

Results: Vertical proximity

Turbulence categories:

“Dartboard” comparisons

• Motivations– Investigate turbulence severity dependence on

whether convection is upwind or downwind– Investigate turbulence dependence on intensity

and size of convective activity• Approach

– For each aircraft measurement, orient “dart board” based on aircraft wind direction (5-10, 10-20, 20-40, 40-80, 80-160 nmi rings, 60 wedges)

– For each in situ measurement, compute • distance to convection within each wedge

(convection given by NCWD VIL)• coverage by convection within wedge “rings”

Wedge orientation

Wedge 0

Wedge 1

Wedge 2

Wedge 5

Wedge 4

Wedge 3

Wind vector

“Dartboard”

51020 40 80 160 nmi

Wedge distance to convection (alt > 20kft)

Peak EDR = 0.05 (null)

Peak EDR >= 0.35 (MoG)

Red: VIL = 0.9 kg m-2

Orange: VIL = 5

Green: VIL = 10

Lt. Blue: VIL = 15

Blue: VIL = 30

Distribution of conditional distance to VIL pixels above different intensity thresholds in the 6 dartboard sectors.

Numbers on plot represent mean distances for the different thresholds.

Wedge distance to convection (alt > 20kft)

Avg. EDR = 0.05 (null)

Avg. EDR >= 0.35 (MoG)

Red: VIL = 0.9 kg m-2

Orange: VIL = 5

Green: VIL = 10

Lt. Blue: VIL = 15

Blue: VIL = 30

Coverage: 5-10 nmi (alt > 20kft)

Peak EDR = 0.05 (null)

Peak EDR >= 0.35 (MoG)

Red: VIL = 0.9 kg m-2

Orange: VIL = 5

Green: VIL = 10

Lt. Blue: VIL = 15

Blue: VIL = 30

Distribution of conditional coverage fractions for different intensities (VIL values) in the 6 dartboard sectors.

% values represent mean coverage.

First bin is 1/100 of actual value.

Coverage: 10-20 nmi (alt > 20kft)

Peak EDR = 0.05 (null)

Peak EDR >= 0.35 (MoG)

Red: VIL = 0.9 kg m-2

Orange: VIL = 5

Green: VIL = 10

Lt. Blue: VIL = 15

Blue: VIL = 30

Coverage: 20-40 nmi (alt > 20kft)

Peak EDR = 0.05 (null)

Peak EDR >= 0.35 (MoG)

Red: VIL = 0.9 kg m-2

Orange: VIL = 5

Green: VIL = 10

Lt. Blue: VIL = 15

Blue: VIL = 30

Coverage: 40-80 nmi (alt > 20kft)

Peak EDR = 0.05 (null)

Peak EDR >= 0.35 (MoG)

Red: VIL = 0.9 kg m-2

Orange: VIL = 5

Green: VIL = 10

Lt. Blue: VIL = 15

Blue: VIL = 30

Coverage: 80-160 nmi (alt > 20kft)

Peak EDR = 0.05 (null)

Peak EDR >= 0.35 (MoG)

Red: VIL = 0.9 kg m-2

Orange: VIL = 5

Green: VIL = 10

Lt. Blue: VIL = 15

Blue: VIL = 30

Conclusion: There is a correlation between convection proximity and intensity and the likelihood of turbulence.

Question: How do we exploit this relationship to create a CIT diagnostic?

Learning a Predictive Algorithm: Random Forests

• Basic idea– “grow” multiple decision trees to predict turbulence

based on “dartboard” values, each using a random subset of data (“bagging”) and random splitting variables

– trees function as “ensembles of experts”– trees “vote” to determine consensus categorization;

they also create a “probability distribution” over classes

Vote: 4 Vote: 2 Vote: 4 Vote: 4 Vote: 1

=> consensus vote: 4 (“confidence” 3/5)

Initial results (ROC curves)

• Discriminate light or greater (in situ 1/3 > 0.1 m2/3 s-1) vs. null turbulence (random and biased summertime data training samples, with and without wind magnitude)

Initial results (ROC curves)

• Discriminate moderate or greater (in situ 1/3 > 0.3 m2/3 s-1) vs. less than moderate turbulence

Current efforts• Interpret data mining results to understand importance

of different variables• Build a fuzzy logic algorithm to duplicate RF

performance• Develop an “augmented in situ database” with quality-

controlled EDR values, RUC model, and variables derived from RUC model data

• Perform “dart board” comparisons that incorporate cloud top information for each region, use RUC model winds for orientation, and incorporate augmented in situ data

• Add temporally-lagged data to the analysis• Incorporate thunderstorm “objects” and features• Use results from case studies and simulations to

modify/refine approach

Synergies with NASA work

• Interest in using satellite detection and feature-based diagnostics of CIT

• Other data sources (e.g., NTDA) available for comparison

• “Augmented in situ database” supports case studies, verification, tuning

• Infrastructure for processing and displaying data in real-time

Challenges• Large number of data sources • Analysis/tracking/extrapolation/feature identification from multiple

4-D data fields required– “Curse of dimensionality”

• In situ EDR data not ideal “truth” • Many different sources and conditions, and their interactions,

cause observed turbulence• Aircraft data are not representative samples of the atmosphere• Need additional data sources

– “echo tops” data are only to nearest 5,000 ft– object information (size or severity) of nearby storm– history and track of the storm?– environmental conditions that may affect propagation of

turbulence, breaking of “gravity waves”• Need physical basis for narrowing down fields/features to

include in analysis

Available RUC and derived fieldsConvective Parameters

• CAPE

• CIN

• Showalter Index

• Totals Indices

• Lifted Index

• Precipitable Water

• SWEAT (Severe Wx Threat Index)

• K-Index

• Bulk Richardson Number

• Richardson Number

• Lapse Rate

• DTF3 (Diagnostic TKE Formulations)

• Vertical Shear

• Horizontal Shear

• 1/Stability

• EDR (Structure Function derived Eddy Dissipation Rates)

• SIGW (Structure Function derived Sigma Vertical Velocity)

• Divergence

• Vorticity

• Dutton

• NCSUI (NC State U. Index)

•Colson-Panofsky

• Ellrod1

• NCSUI (N.C. State U. Index)

• Saturated Richardson Number

• Frontogenesis Function

• LAZ (Laikhman-Alter-Zalik)

• NGM1 and NGM2

• ABSIA

• UBF (Unbalanced Flow)

• NVA (Negative Voriticity Advection)

• Tropopause Height

• Wind Speed

Turbulence Indices