Convectively Induced Turbulence Due to Thunderstorms

1
UAH
Convectively Induced Turbulence Due to
Thunderstorms A joint project between NCAR/RAL,
UW/CIMSS, and UAH
Bob Sharman (NCAR), John Mecikalski (UAH) and
Wayne Feltz (UW/CIMSS) presented by John K.
Williams and Stan Trier (NCAR)
NASA Applied Sciences Weather Program
Review Boulder, CO November 18, 2008
2
Motivation Turbulence Impacts

• Safety
• Causes many of wx-related accidents 75 Part
  121, 15 Part 135, 23 Part 91 (OFCM Aviation
  weather Programs/ projects 2004 update,
  1995-2002)
• Average 40 fatalities/year (mostly Part 91)
• Economic cost
• Nearly 200 million per year (MCR Federal, 2003)
• Capacity
• Turbulence is ranked as the second leading
  factor likely to affect the NAS (ARTCC/TRACON
  survey of weather factors impacting NAS,
  MOSAICATM/AvMET report)

C-V
TB
3
Convectively-induced turbulence (CIT)
CIT
Source P. Lester, Turbulence A new
perspective for pilots, Jeppesen, 1994
4
Convective and convectively-induced turbulence
(CIT)

• Responsible for 60 or more of all turbulence
  encounters
• Not directly addressed by current GTG
• FAA thunderstorm avoidance guidelines inadequate
  for balancing safety and capacity
• Turbulence is not always well-correlated with
  precipitation intensity

5
Scales of Aircraft Turbulence
cm
10s km
Aircraft responds to scales from few m - km
Aircraft-scale turbulence inferred from
lower-res data
satelliteresolution
Operational NWP modelresolution
Mesoscale modelresolution
RUC / WRF
6
Project goals and approach

• Goal Develop CIT nowcast/forecast system for
  aviation
• Overall approach
• Understand CIT source mechanisms through case
  studies, climatology, and fine-scale model
  simulations
• Develop nowcast/forecast diagnostic
  (Diagnose-CIT, DCIT) based on artificial-intellige
  nce-based combinations of data from
• Satellite and other observations
• NWP (RUC/WRF) derived diagnostics
• Tune and verify against in situ EDR turbulence
  data
• Mainly from UAL 757 aircraft
• Position accuracy lt 15 km, time accuracy lt 1 min
  much better than PIREPS
• Implement into FAA AWRP GTG-N (experimental
  11/09, operational 2/11) ? NextGen 4D data cube
• External collaborators Todd Lane, Univ. of
  Melbourne, Rob Fovell, UCLA

7
Connection to NextGen 4D data cube
EN-2430 Weather Forecasts - Consolidated
Turbulence - Level 1. Near-term predictive
models and current weather observations are fused
to provide a consolidated turbulence forecast
that is available to users over a network-enabled
infrastructure. This capability will include
North America from 10,000 feet to FL450, 0-18
hours, updated hourly, and will forecast clear
air and mountain wave turbulence.
HOL
NWP Model (s)
NOAA
PIREPS
Indices
GTG
FAA AWRP Turb RT
Insitu data
QC
FAA AWRP other RT
DCIT
GTGN
NSSL 3-D DBZ
NASA AvWx
Mosaic
NTDA
Primary funding source color code
Feature extractor
CoSPA
Satellite data
24X7 Processing center
Red ? Single Authoritative Data Source (SAS)
8
Example of CIT Results from Lane et al. study
(JAS 2003)
9
GeneralizationsT. P. Lane and R. D. Sharman
Some influences of background flow conditions on
the generation of turbulence due to gravity wave
breaking above deep convection. J. Appl.
Meteor. Clim., 47, 27772796.
Baseline simulation
Reduced stability above cloud top
Stronger shear near cloud top
10
Major work areas

• Develop better understanding of CIT mechanisms
  (NCAR)
• Turbulence climatology
• Case studies
• Numerical simulations
• Develop satellite-based inferences of CIT (CIMSS,
  UAH)
• Over-shooting cloud tops
• Transverse bands
• Wave identifier
• Develop and test real-time blender (data
  fusion) of relevant available observations /
  features and NWP model output (NCAR)
• Implement and test resulting DCIT (NCAR)
• Coordinate with FAA AWRP DCIT/GTG-N work

11
Schedule

• Now on year 3 of 3-year grant
• Work accomplished
• Two detailed simulation case studies (June 17 and
  August 5, 2005)
• Lane above-cloud turbulence case generalized
• Satellite-based overshooting cloud top algorithm
  (CIMSS,UAH) ready for integration and test in the
  DCIT module
• Library of CIT events on CIMSS website
  http//www.ssec.wisc.edu/alenz/EDRevents.html.
• Preliminary DCIT statistical studies using random
  forests
• 3 referreed journal papers, 6 conference papers
• 3 CIT mini-workshops, including external
  attendees
• To do
• Finish case studies
• Integrate and test final DCIT data fusion
  algorithm
• Handoff DCIT algorithm to GTGN development team
  (will ultimately populate NextGen 3D data cube)

12
RUC-Based Climatology and WRF Convection-Permittin
g Simulations of Convection-Induced Turbulence
(CIT)

• Stan Trier
• work with Bob Sharman
• collaborator Rob Fovell, UCLA

13
05 August 2005 0230 UTC incident
WRF Simulations
FL370
FL390
In situ EDR Key 0.05 grey, 0.15 purple,
0.25 blue, 0.35 green, 0.45 yellow, 0.55
orange, 0.65 red, 0.75 pink
From Prof. Rob Fovell (UCLA)
w (m/s)

• Outflow from WRF-simulated storm has regions of
• low Ri 1
• Storm-Induced gravity waves cause further
  decrease

14
16 17 June 2005 Case Study
0905 UTC 16 June
43 N
40 N
37 N
0745 UTC 17 June
40 N
37 N
15
Four-Year (2003-2006) June-August Climatology of
25 of Heaviest Rain Cases over 100 W to 90 W
Longitude (e.g., W Kansas- W Illinois)
Time (hour UTC)
North Latitude (degrees)

• RUC analyzed 200-mb zonal wind enhancement
  (left) several hundred km north and several hours
• after heaviest rainfall with nocturnal
  convection
• Leads to enhanced vertical shear near flight
  levels (right) in similar location persisting for
  several hours
• beyond sunrise and enhanced threat of
  turbulence from lowering of Richardson Number

16
(No Transcript)
17
12-h Numerical Simulations Using WRF ARW Version
2.2

• Control simulation (with convection) compared to
  Adiabatic simulation (no latent heat
  release) to assess MCSs role in generating
  turbulence

Single Large Domain

• Domain Characteristics
• - 600 x 500 horizontal points (D3km)
• - 65 vertical levels (model top at 10 mb)
• - 10-km deep diffusive damping layer
• Physical Parameterizations
• - MYJ PBL scheme
• - Thompson microphysics
• - Noah LSM
• - RRTM longwave, Dudhia shortwave
• - Simple KM 2-D Smagorinsky diffusion
• Initial and Boundary Conditions
• - Hourly RUC Hybrid Analysis (up to 70 mb)
• - GFS Global Analysis (50 to 10 mb)

18
Observations
Model Control Simulation
0430 UTC
0730 UTC

• Control simulation is reasonable representation
  of the observed MCS
• Observed turbulence and simulated TKE several
  hundred km north of active thunderstorms in MCS

19
12-h Animation with 12.5-km Control Adiabatic
Winds, 11.75-km Control TKEand Control
Reflectivity
20
Vertical Wind Profiles from Different Sides of
MCS Anvil Edge
Flight levels where turbulence observed

• Strong vertical shear at flight levels due
  almost entirely to MCS outflow on north side
  (top)
• Vertical shear at flight levels on south side
  (bottom) weaker because easterly outflow winds
• and shear are opposed by their westerly
  environmental (adiabatic) counterparts

21
Temporal Variability Near North Anvil Edge Where
Turbulence is Observed

• Strong vertical shear (top) contributes low
• Ri lt 1 background (bottom) in anvil outflow

• Low-frequency (T 3 h) variations in moist
• static stability (top) governs further
  decreases
• of Ri to lt 0.25 supporting protracted large
• TKE episodes (bottom)

22
Climatology and Simulations Summary

• RUC-based climatology and convection-resolving
  WRF simulations used to
• examine CIT occurring hundreds of km from
  storms (i.e. large radar reflectivity)
• Simulated TKE at flight levels occurs when
  strong vertical shear and weak
• static stability in MCS upper-level outflow
  results in Ri lt 0.25
• TKE favored near north edge of MCS upper-level
  outflow where vertical shear
• is strongest and most unopposed by background
  vertical shear
• Operational RUC analyses may be useful in
  diagnosing turbulence potential
• for cases of large-scale MCS-induced
  upper-level circulations

Reference Trier, S. B., and R. D. Sharman 2008
Convection-permitting simulations of the
environment supporting
widespread turbulence within the upper-level
outflow of an MCS. Monthly Weather Review,
conditionally accepted.
(Copies available)
23
Statistical Analysis and Diagnose-CIT (DCIT)
Data Fusion System Development

• John Williams

24
Convectively-Induced Turbulence Signatures in
MODIS, AVHRR, and GOES Multi-spectral Imagery
Rapid Convective Growth
Gravity Waves and Overshooting Tops
Banded Cirrus Outflow
FL 40000 ft T205 K Satellite Temp233 K
Rapid Anvil Expansion

• Rapid new storm growth, overshooting tops,
  convective gravity waves, banded cirrus outflow,
  and rapid horizontal expansion of anvil cloud
  were the satellite signatures found to be
  associated with turbulence in this study

25
1 km AVHRR
Objective Day/Night Overshooting Top Detection
Using Current and Future GOES

• IR window imagery is used to identify
  overshooting tops occurring during both day and
  night
• A study of 450 enhanced-V producing overshooting
  top cases (Brunner et al. (WAF, 2007)) shows that
  tops are
• isolated clusters of pixels (lt 12 km2 area)
  colder than 215 K and the GFS tropopause temp
• significantly colder (gt 6.5 K) than the
  surrounding anvil cloud
• Higher ABI spatial resolution leads to better
  observation of cold BT minima and improved
  overshooting detection

Overshooting Detection Using 4 km GOES-12
Overshooting Detection Using 2 km GOES-R ABI
26
Overshooting Top-EDR Turbulence Relationships

• Flight very close (lt 10 km) to overshooting tops
  produces a significant increase in the frequency
  and intensity of turbulence compared to uniformly
  cold cloud tops

• Failure pixels indicate the presence of a
  large area (gt 40 km in diameter) of lt 215 K
  cloud tops.
• - Larger, mature thunderstorms induce a higher
  frequency of turbulence encounters at greater
  distances from the cold BT region

27
NLDN vs. EDR Relationships (UAH)

• Recent efforts
• When lightning strike is present, check EDR
  reports within 0.1 degrees and 10 minutes of
  detected strike.
• Examine turbulence distribution as a function of
  distance, altitude and lightning polarity
• Future work
• Will expand this study over a one year period to
  have a large dataset of cases.
• Will incorporate cloud top temperature into the
  analysis for identification of large, mature
  convective storms.

28
DCIT Data Fusion Goals

• Combine data to provide a 3-D gridded assessment
  of CIT
• near-storm environment and clear-air turbulence
  diagnostics derived from RUC
• lightning and satellite-derived features and
  turbulence signatures developed by CIMSS, UAH
• storm features from radar in-cloud turbulence and
  3-D reflectivity (from case studies and
  simulations)
• United Airlines in situ EDR reports used to tune
  an empirical model
• Generate a deterministic (eddy dissipation rate,
  EDR) grid and also probability estimates of
  light-or-greater (LoG) moderate-or-greater (MoG)
  and severe-or-greater (SoG) turbulence
• DCIT grids will be used in a rapid-update
  nowcast version of GTG, called GTGN

29
Preliminary DCIT development

• Used a statistical analysis technique (random
  forests) to select diagnostic variables and
  create a preliminary empirical mapping to
  turbulence
• initial prototype DCIT running in real-time at
  NCAR/RAL since July 2008
• Additional features (e.g., latest CIMSS
  overshooting tops product) are being evaluated
  for future inclusion
• field values and transformations (local disc
  statistics, feature distances, directional
  distances) have been compiled into a MySQL
  database of over 1500 candidate fields and
  predictors

30
Random Forest (RF)

• A non-linear statistical analysis (a.k.a. machine
  learning, data mining) technique
• Produces a collection of decision trees using a
  training set of predictor variables (fields and
  features) and associated truth (in situ EDR)
• each decision trees forecast logic is based on a
  random subset of data and predictor variables,
  making it independent from others
• during training, random forests produce estimates
  of predictor importance
• importance of each predictor evaluated in context
  of others
• does not presuppose a form of the empirical model

continued
31
RF, continued

• A trained random forest provides an empirical
  model that can be applied to new data
• the trees function as an ensemble of experts,
  voting on the predicted classification for each
  new data point
• the classification is the consensus winner, or
  the vote distribution may be used to derive a
  probability or uncertainty estimate

Data pt.
Data pt.
Data pt.
Data pt.
Data pt.
Tree 2
Tree 3
Tree 100
Tree 1
Tree 4

Vote 0
Vote 0
Vote 0
Vote 1
Vote 1
gt 40 votes for 0, 60 votes for 1 consensus
category 1
32
MySQL EDR / Feature Database

• Over 1500 candidate fields and features,
  including
• Satellite GOES visible and IR, 10.7 micron
  cooling rates, CuMask, cloud type, winds, channel
  differences, overshooting tops, CINowcast and
  transformations (statistics from surrounding
  discs max, min, mean, std, coverage)
• Radar echo tops (10, 18, 30 dBZ), NTDA EDR tops,
  composite dBZ, VIL, 3-D NSSL dBZ, 3-D NTDA EDR,
  selected disc statistics, vertical, and
  direction-dependent horizontal distances
• Lightning NLDN density and distance
• Model RUC fields (inc. CAPE, surface conditions,
  and 3-D fields), and derived turbulence
  diagnostics, and model/satellite temperature
  differences

33
Example conditional histograms

• Good predictors

EDR lt 0.2 m2/3 s-1
EDR gt 0.2 m2/3 s-1
34
Example conditional histograms

• Good predictors

EDR lt 0.2 m2/3 s-1
EDR gt 0.2 m2/3 s-1
35
Example RF imp., 107TB gt 273 K
(Rank, Importance, Field)
1 2.27 RUC_derived_Frontogenesis_func 2
1.62 RUC_derived_Structure_function 3 1.61
RUC_derived_Ellrod_1 4 0.76 RUC_derived_NGM
5 0.72 Altitude 6 0.65 RUC_Pressure 7
0.65 RUC_derived_Saturated_richards 8 0.63
RUC_derived_Totals_indices 9 0.63
RUC_derived_Diagnostic_turbule 10 0.62
Longitude 11 0.60 RUC_derived_Vertical_shear
12 0.59 RUC_Geopotential_height 13 0.57
RUC_Humidity_mixing_ratio 14 0.55
RUC_derived_Unbalanced_flow_al 15 0.53
RUC_derived_NGM2 16 0.51 RUC_Virtual_potential
_temperat 17 0.49 RUC_Mean_sea_level_pressure
18 0.46 RUC_derived_Showalter_index 19
0.45 RUC_derived_Severe_weather_thr 20 0.44
RUC_Precipitable_water 21 0.43
RUC_derived_Lapse_rate 22 0.40
RUC_derived_Richardson_number 23 0.39
RUC_derived_Absolute_xxx_advec 24 0.39
RUC_Turbulent_kinetic_energy 25 0.38
RUC_derived_Structure_function 26 0.30
RUC_derived_Lifted_index 27 0.29 Latitude 28
0.28 RUC_derived_K_index
29 0.28 f107TB_40Npts 30 0.27
RUC_derived_Precipitable_Water 31 0.27
RUC_Easterly_wind 32 0.27 f107TB_40StdDev 33
0.27 f107TB_20Mean 34 0.26 f107TB_20Min
35 0.25 RUC_Potential_temperature 36 0.25
f107TB_40Min 37 0.25 f107TB_20StdDev 38
0.25 f107TB_10Min 39 0.24 RUC_derived_Vortici
ty 40 0.24 RUC_derived_Horizontal_shear 41
0.24 f107TB_20Max 42 0.24 f107TB_10Mean 43
0.23 RUC_Northerly_wind 44 0.23
f107TB_20Npts 45 0.23 f107TB_40Mean 46
0.22 f107TB_40Max 47 0.20 RUC_derived_Inverse
_stability 48 0.20 f107TB_10StdDev 49 0.19
VIL_30_0 50 0.19 RUC_Soil_temperature 51
0.19 f107TB_10Max 52 0.18 RUC_derived_Channel
_4_infrared 53 0.18 f107TB 54 0.17
RUC_Convective_available_poten 55 0.17
RUC_Convective_inhibition 56 0.16
f107TB_10Npts
57 0.15 RUC_Pressure_vertical_velocity 58
0.14 VIL_15_0 59 0.10 RUC_derived_Divergence
60 0.08 RUC_derived_Bulk_richardson_nu 61
0.06 RUC_derived_Colson_panofsky 62 0.04
RUC_derived_Laikhman_alter_zal 63 0.03
RUC_Ice_mixing_ratio 64 0.02
RUC_Convective_precipitation 65 0.01 Top 66
0.01 f107TB_15mindiff 67 0.01
RUC_Ice_particle_number_concen 68 0.01
OvershootMask_80fract 69 0.01
OvershootMask_160fract 70 0.01
f107TB_15mindiff_20Npts 71 0.00
f107TB_15mindiff_20Max 72 0.00
f107TB_15mindiff_20Mean 73 0.00
OvershootMask_20fract 74 0.00
f107TB_15mindiff_10Npts 75 0.00
OvershootMask_40fract 76 0.00
f107TB_15mindiff_10Mean 77 0.00
f107TB_15mindiff_20Min 78 0.00
f107TB_15mindiff_10StdDev 79 0.00
f107TB_15mindiff_10Min 80 0.00
f107TB_15mindiff_10Max
36
Example RF imp., 107TB lt 273 K
(Rank, Importance, Field)
1 2.29 f107TB_40Min 2 2.16
f107TB_20Min 3 1.97 f107TB_10Min 4 1.63
Altitude 5 1.54 RUC_derived_Frontogenesis_fun
c 6 1.53 RUC_Virtual_potential_temperat 7
1.38 RUC_Geopotential_height 8 1.29
RUC_derived_Structure_function 9 1.27
RUC_Pressure 10 1.27 f107TB_10Mean 11 1.23
f107TB_20Mean 12 1.20 RUC_derived_Unbalanced_
flow_al 13 1.15 f107TB 14 1.02
RUC_Humidity_mixing_ratio 15 0.95
f107TB_40Mean 16 0.94 RUC_Mean_sea_level_press
ure 17 0.90 RUC_derived_Saturated_richards
18 0.88 Top 19 0.87 RUC_derived_Ellrod_1
20 0.84 Longitude 21 0.75
RUC_derived_Structure_function 22 0.75
RUC_derived_NGM2 23 0.70 RUC_derived_Totals_in
dices 24 0.64 VIL_15_0 25 0.64 Latitude
26 0.64 f107TB_10Max 27 0.63
RUC_derived_Showalter_index 28 0.63
RUC_derived_Severe_weather_thr
29 0.54 VIL_30_0 30 0.54
RUC_derived_Richardson_number 31 0.51
f107TB_40Npts 32 0.51 RUC_derived_Lapse_rate
33 0.49 RUC_derived_Vertical_shear 34 0.47
RUC_Potential_temperature 35 0.45
RUC_derived_Channel_4_infrared 36 0.45
f107TB_20Max 37 0.44 RUC_Precipitable_water
38 0.44 f107TB_40StdDev 39 0.41
f107TB_40Max 40 0.41 RUC_derived_Vorticity
41 0.39 RUC_derived_NGM 42 0.38
OvershootMask_80fract 43 0.36 f107TB_20Npts
44 0.35 RUC_Northerly_wind 45 0.35
OvershootMask_160fract 46 0.34
RUC_Convective_inhibition 47 0.32
RUC_derived_K_index 48 0.31
RUC_derived_Diagnostic_turbule 49 0.31
RUC_Convective_available_poten 50 0.30
RUC_derived_Precipitabl_wWater 51 0.30
f107TB_10Npts 52 0.29 RUC_derived_Lifted_index
53 0.29 f107TB_20StdDev 54 0.28
RUC_Easterly_wind 55 0.28 RUC_Pressure_vertica
l_velocity 56 0.26 RUC_derived_Absolute_xxx_ad
vec
57 0.24 RUC_derived_Horizontal_shear 58
0.24 RUC_derived_Divergence 59 0.24
RUC_Ice_mixing_ratio 60 0.23 f107TB_10StdDev
61 0.22 RUC_derived_Bulk_richardson_nu 62
0.21 RUC_Soil_temperature 63 0.19
OvershootMask_40fract 64 0.19
RUC_Turbulent_kinetic_energy 65 0.18
RUC_Convective_precipitation 66 0.18
RUC_derived_Inverse_stability 67 0.18
RUC_Ice_particle_number_concen 68 0.10
RUC_derived_Laikhman_alter_zal 69 0.08
RUC_derived_Colson_panofsky 70 0.08
f107TB_15mindiff_20Mean 71 0.08
f107TB_15mindiff_20Max 72 0.08
f107TB_15mindiff_20Min 73 0.06
f107TB_15mindiff_20Npts 74 0.04
OvershootMask_20fract 75 0.02
f107TB_15mindiff_10Mean 76 0.02
f107TB_15mindiff_10Npts 77 0.01
OvershootMask_10fract 78 0.01
f107TB_15mindiff_10Min 79 0.01
f107TB_15mindiff_10Max 80 0.01
f107TB_15mindiff_10StdDev 81 0.00
f107TB_15mindiff 82 0.00 OvershootMask
37
Initial DCIT performance results

• ROC curves (PoD vs. FAR) within 40 nmi of
  convection, above 15,000 ft

RUC, satellite, radar, and lightning-based
diagnostics
RUC model-based diagnostics only (like GTG)
?? Real-time T-storm observations add significant
skill
38
Case study 27 June 2007 2300 UTCProbability
of MoG turbulence
Composite Reflectivity
GOES visible
GOES IR
39
FL210

• RUC only ProbabilityMoG

DCIT ProbabilityMoG (uncalibrated)
40
FL240

• RUC only ProbabilityMoG

DCIT ProbabilityMoG (uncalibrated)
41
FL270

• RUC only ProbabilityMoG

DCIT ProbabilityMoG (uncalibrated)
42
FL300

• RUC only ProbabilityMoG

DCIT ProbabilityMoG (uncalibrated)
43
FL330

• RUC only ProbabilityMoG

DCIT ProbabilityMoG (uncalibrated)
44
FL360

• RUC only ProbabilityMoG

DCIT ProbabilityMoG (uncalibrated)
45
FL390

• RUC only ProbabilityMoG

DCIT ProbabilityMoG (uncalibrated)
46
DCIT real-time prototype example08-11-2008
2215Z
NTDA MoG Tops
NEXRAD Echo Tops
GOES visible
GOES IR
47
08-11-2008 2215Z DCIT probability MoG turbulence
with overlaid in situ EDR tracks
48
Regionalization using RF importance
Frntgth,-SatRi
Frntgth Ellrod1
VWS Frntgth
Ellrod1, -SatRi
-Ri,-SatRi
Slide courtesy of Jennifer Abernethy, NCAR
49
Summary

• RUC climatology and two new simulation studies
  have given additional insight into CIT mechanisms
• Physical insight plus case study analyses has led
  to the development of relevant satellite-based
  feature identification and CIT indicators
• A large database with collocated in situ EDR and
  candidate predictors has been assembled
• A prototype DCIT data fusion algorithm has been
  developed to utilize observation data features
  and model-based diagnostics to produce
  probabilistic turbulence assessments
• Statistical evaluation suggests substantial
  improvement over a comparable RUC-only turbulence
  diagnosis in airspace near thunderstorms
• The DCIT prototype will be an important input to
  the GTG-N system, expected to be the NextGen SAS
  for turbulence nowcasts

50
(No Transcript)