LAPS Technical Overview

1
LAPS Technical Overview
2
LAPS Mission

• A system designed to
• Exploit all available data sources
• Create analyzed and forecast grids with analysis
  systems and numerical models
• Build products for specific forecast applications
• Provide reliable forecast guidance
• Use advanced display technology
• All within a local weather office, forward site,
  or in fully deployed mode

3

• LAPS Basic Structure

Univariate Analysis
Analysis Merging/ Balancing
NWP Model Initialization/ Prediction
Data
Very diverse Force geometric,
Reconcile gridded Generate forecasts data
sets smoothing constraints
fields force and user-specific
to interpolate data to
consistency based products
high resolution grids on
atmospheric scale
4
LAPS Components - Stage 1

• Data Acquisition and Quality Control
• Univariate Analysis of the Following Fields
• Temperature
• Winds
• Water Vapor
• Clouds
• Microphysical variables
• Vertical motions

5
Data Acquisition and Quality ControlLAPS
supports a wide range of data types
6
LAPS data ingest strategy
7
Multi-layered Quality Control

• Gross Error Checks
• Rough Climatological Estimates
• Station Blacklist
• Dynamical Models
• Use of background and mesoscale models
• Standard Deviation Check
• Statistical Models
• Buddy Checking

8
Standard Deviation Check

• Compute Standard Deviation of observations-backgro
  und
• Remove outliers
• Now adjustable via namelist

9
LAPS Radar Ingest
10
Remapping Strategy

• Polar to Cartesian
• 2D or 3D result (narrowband / wideband)
• Average Z,V of all gates directly illuminating
  each grid box
• QC checks applied
• Typically produces sparse arrays at this stage

11
Remapping Strategy (reflectivity)

• Horizontal Analysis/Filter (Reflectivity)
• Needed for medium/high resolutions (lt5km) at
  distant ranges
• Replace unilluminated points with average of
  immediate grid neighbors (from neighboring
  radials)
• Equivalent to Barnes weighting at medium
  resolutions (5km)
• Extensible to Barnes for high resolutions (1km)
• Vertical Gap Filling (Reflectivity)
• Linear interpolation to fill gaps up to 2km
• Fills in below radar horizon visible echo

12
Mosaic Strategy (reflectivity)

• Nearest radar with valid data used
• /- 10 minute time window
• Final 3D reflectivity field produced within cloud
  analysis
• Wideband is combined with Level-III
  (NOWRAD/NEXRAD)
• Non-radar data contributes vertical info with
  narrowband
• QC checks including satellite
• Help reduce AP and ground clutter

13
Radar Mosaic
14
Analyzed Reflectivity (800 hPa)
15
Surface Precipitation Accumulation

• Algorithm similar to NEXRAD PPS, but runs
• in Cartesian space
• Rain / Liquid Equivalent
• Z 200 R 1.6
• Snow case use rain/snow ratio dependent on
  column maximum temperature
• Reflectivity limit helps reduce bright band
  effect

16
Storm-Total Precipitation (wideband mosaic)
17
Storm-Total Precipitation
18
Storm-Total Precipitation (RCWF narrowband)
19
Precip type and snow cover
20
Future Cloud / Radar analysis efforts

• Account for evaporation of radar echoes in dry
  air
• Sub-cloud base for NOWRAD
• Below the radar horizon for full volume
  reflectivity
• Continue adding multiple radars and radar types
• Evaluate Ground Clutter / AP rejection

21
Future Cloud/Radar analysis efforts (cont)

• Consider Terrain Obstructions
• Improve Z-R Relationship
• Convective vs. Stratiform
• Precipitation Analysis
• Improve Sfc Precip coupling to 3D hydrometeors
• Combine radar with other data sources
• Model First Guess
• Rain Gauges
• Satellite Precip Estimates (e.g. GOES/TRMM)

22
The LAPS Analysis
23
Three-dimensional Analysis

• Looking for a function that is the best fit of
  weather through backgrounds and observations in
  3D.
• Data assimilation techniques Barnes, 3DVar,
  4DVAR, KF. Differences.Barnes is a point-wise
  fitting 3DVAR, 4DVar. KF are global fitting
  schemes.

24
3DVAR Cost Function
B is the model error covariance matrix O Is the
observational error matrix (diagonal) x is the
control variable xb is the background field H
is the observation operator (maybe nonlinear) y
is the observation.
25
LAPS Analysis Philosophy

• Focus on the mesoscale - here model error
  covariances are poorly known - 3-D var is not an
  optimum approach
• Let data define structure
• Use successive corrections (Barnes) exponential
  weight with collapsing radius of influence
• Blend with model background
• Generate smooth fields that will be reconciled in
  stage 2
• Ensure rapid computation

26
(No Transcript)
27
LAPS Analysis Process and Data Structure
LAPSPRD Directory
LSX
Surface Fields
LT1
3-D Temp
L1S
Prcp Accum
LQ3
3-DHumid
LAPS Inter. Data Files
LW3
Wind
Cloud
LC3
LCP
Derived Pds
LM1
Soil
28
Sfc ToverTaiwan
29
CAPE
30
3-D Temperature

• First guess from background model
• Insert RAOB, RASS, and ACARS if available
• 3-Dimensional weighting used
• Insert surface temperature and blend upward
• depending on stability and elevation
• Surface temperature analysis depends on
• METARS, Buoys, and Mesonets (LDAD)

31
Successive correction analysis strategy

• 3-D weighting
• Successive correction with Barnes weighting
• Distance weight e-(d/r)2 applied in 3-dimensions
• Instrument error reflected in observation weight
• Wo e-(d/r)2 / erro2
• Each analysis iteration becomes the background
  for the next iteration
• Decreasing radius of influence (r) with each
  iteration
• Each iteration improves fit and adds finer scale
  structure
• Works well with strongly clustered observations
• Iterations stop when fine scale structure fit
  to obs become commensurate with observation
  spacing and instrument error

32
Successive correction analysis strategy (cont)

• Smooth blending with Background First Guess
• Background subtracted to yield observation
  increments or innovation (uo)
• Background (with zero increment) has weight at
  each grid point
• Background weight proportional to inverse square
  of estimated error
• wb 1 / errb2
• For each iteration, analyzed increment (u) is as
  follows
• ui,j,k ? (uowo) / ( ? (w o ) wb )

33
LAPS Wind Analysis
34
X-sect T / Wind
35
700 hPa Winds and Geopotential - Stage 1
36
Products Derived from Wind Analysis
37
LAPS 3-D Water Vapor (Specific Humidity) Analysis

• Interpolates background field from synoptic-scale
  model forecast
• QCs against LAPS temperature field (eliminates
  possible supersaturation)
• Assimilates all appropriate LAPS upper air data
• Assimilates boundary layer moisture from LAPS Sfc
  Td analysis

38
LAPS 3-D Water Vapor (Specific Humidity)
Analysis continued

• Scales moisture profile (entire profile excluding
  boundary layer) to agree with derived GOES TPW
  (processed at NESDIS)
• Scales moisture profile at two levels to agree
  with GOES sounder radiances (channels 10, 11,
  12). The levels are 700-500 hPa, and above 500
• Saturates where there are analyzed clouds
• Performs final QC against supersaturation

39
Gradient-based Functional
40
Total Preciptiable Water Taiwan
41
3-D Clouds

• Preliminary analysis from vertical cloud
  soundings derived from METARS, PIREPS, and CO2
  Slicing
• IR used to determine cloud top (using temperature
  field)
• Radar data inserted (3-D if available)
• Visible satellite can be used

42
Cloud/Satellite Analysis Data

• 11 micron IR
• 3.9 micron data
• Visible (with terrain albedo)
• CO2-Slicing method (cloud-top pressure)

43
CloudSchematic
44
Cloud Analysis Flow Chart
45
Cloud Coverage without/with visible data

No vis data
With vis data
46
Cloud cover (fraction) with surface stations only
47
Cloud fraction with surface stations and radar
48
Cloud fraction with surface stations, radar, and
IR
49
Cloud Coverage without/with visible data

No vis data
With vis data
50
Cloud Isosurfaces
51
Cloud/precip cross section
52
Cloud Radar X-sect (Taiwan)
53
Derived products flow chart
54
Cloud Radar X-sect (wide/narrow band)
55
Adjustments to cloud and moisture scheme

• Originally cloud water and ice estimated from
  Smith-Feddes parcel
• Model this tended to produce too much moisture
  and ice
• Adjustments
• Scale vertical motion by diagnosed cloud amount,
  extend below cloud base
• 2. Reduced cloud liquid consistent with 10
  supersaturation of diagnosed water vapor and
  autoconversion rates from Schultz

56
Cloud vertical motions
57
Some dependence on cloud type, Updraft goes to
top of cloud
CS
Strongest updrafts in regions of high reflectivity
CB
Downdrafts in stratiform region
Randomness in broad convective regions
Updated CWB/ FSL scheme (cloud derive subr)
58
Cloud type diagnosis
Cloud type is derived as a function of
temperature and stability
59
Case Study Example How LAPS is used in the
National Weather Service

• Utility of LAPS analysis-only for nowcasting
• A Convective Event on 14 May 1999
• Location DEN-BOU WFO

60
Case Study Example

• On 14 May, moisture is in place. A line of storms
  develops along the foothills around noon LT (1800
  UTC) and moves east. LAPS used to diagnose
  potential for severe development. A Tornado Watch
  issued by 1900 UTC for portions of eastern CO
  and nearby areas.
• A brief tornado did form in far eastern CO west
  of GLD around 0000 UTC the 15th. Other tornadoes
  occurred later near GLD.

61
NOWRAD and METARS with LAPS surface CIN 2100 UTC
62
Examine soundings near CAPE max at points B, E
and F 2100 UTC
63
Soundings near CAPE max at B, E and F 2100 UTC
64
CIN minimum in area of CAPE max 2200 UTC
65
Point E, CAPE has increased to 2674 J/kg 2200 UTC
66
Radar with METARS and LAPS surface helicity 2100
UTC
67
Radar with METARS and LAPS surface helicity 2300
UTC A brief tornado did form in far eastern CO
west of GLD around 0000 UTC on 15 May. Other
tornadoes later near GLD area.
68
LAPS winds every 10 km, RUC winds every 80
km 2100 UTC
69
Case Study Example (cont.)

• Another field that can be monitored with LAPS is
  helicity.
• A storm motion is derived from the mean wind
  (sfc-300 mb) with an off mean wind motion
  determined by a vector addition of 0.15 x Shear
  vector, set to perpendicular to the mean storm
  motion
• Next well examine some helicity images for this
  case. Combining CAPE and minimum CIN with
  helicity agreed with the path of the supercell
  storm that produced the CO tornado.

70
NOWRAD with METARS and LAPS surface helicity
1900 UTC
71
NOWRAD with METARS and LAPS surface helicity
2000 UTC
72
NOWRAD with METARS and LAPS surface helicity
2100 UTC
73
NOWRAD with METARS and LAPS surface helicity
2200 UTC
74
NOWRAD with METARS and LAPS surface helicity
2300 UTC
75
LAPS Components - Stage 2

• Merge cloud and moisture analyses
• Balance wind and temperature (mass) fields
• Account for terrain
• Prepare model initial condition

76
Key element Water-In-All-Phases Analysis

• Merging
• Cloud Analysis satellite, radar, surface,
  aircraft observations
• Water Vapor Analysis variational method
• Recovery of microphysical variables using simple
  cloud model
• Precipitation Total precipitable
  water
• Water vapor Integrated liquid
  water
• Cloud water Cloud cover
• Cloud Ice Ice Crystals

77
LAPS Cloud Analysis Scheme
Radar/SatelliteAircraft/Surface
Ice/snow Mixing Ratios Rainwater mixing
ratio Cloud water mixing ratio
Cloud grid Synthesizer
Background Analyses
Cloud Typing Algorithm
Environmental conditions (T,rh)
1-D Cloud Model
Interpolation to model gid
? cloud
Var Balance and Continuity Scheme
Cloud Table Database
Adjusted U, V, ??,T
Constraints
78
Derived cloud products flow chart
79
The LAPS Diabatic Initialization Technique

• LAPS includes
• Improved cloud analysis
• Dynamic balance using variational scheme
• Coupling to most mesoscale models
• Links to display devices (AWIPS, others)
• Sustain the operational tradition of LAPS
• Robust data ingest, QC, and fusion
• Platform and model independence
• Computationally inexpensive

80
LAPS Analysis Process and Data Structure (cont)
LAPSPRD Directory
Surface Fields
Balance sub dir
3-D Temp
LT1
Wind
QBAL
LQ3
Cloud
LW3
3-DHumid
Soil
LAPS Prep
FUA -3D
Model
FSF - sfc
81
LAPS Diabatic Initialization
82
Equations
Dynamic Balancing and Continuity Formalism
( ) b are background quantities () are
solution increments from background ( ) are
observation differences from background
83
Equations
Minimize the functional
( ) b are background quantities () are
solution increments from background ( ) are
observation differences from background
84
Equations
Adjusted U, V, ???T
Dynamic Balancing and Continuity Formalism
? cloud
U, V, Momentum Eqn Continuity Constraints
Background error weights
( ) b are background quantities () are
solution increments from background ( ) are
observation differences from background
85
Cloud, Wind and Mass Dynamic Adjustment
FH FL
??c

Tgt 0
q qs
86
700 Hpa Balanced Winds and Geopotential - Stage 2
87
700 Hpa Unbalanced Winds and Geopotential -
Stage 1
88
LAPS balanced vertical motion and cloud (a) and
eqn of motion residual (ms-1) (b)
89
Flow Channeling Streamlines and Isotachs at
2500m over mountains
90
                 
0-Hr
5 min forecast
Hot Start With LAPS
Cloud only Cold start Background only
91

Forecast Evolution
Snow
Cloud ice
Graupel
Rain
Cloud water
Initial Condition on MM5 Grid 1-Hr Forecast on
MM5 Grid Note Graupel not in initial condition,
forms rapidly in model
92
MODEL NOISE dp/dt
Balanced Field
Unbalanced Field
93
Model Start Options
Time-n Time
MM5 Forecast

• Cold Start
• clouds spin up 3 hr
• Warm Start
• nudging to LAPS
• clouds grow in n hours
• Hot Start
• clouds appear instantly

Eta
LAPS Analyses
MM5 Nudging
MM5 Forecast
MM5 Forecast
LII
Dynamically balanced, Cloud-consistent LAPS
Eta TDBC for all runs
94
Why Run Models in the Weather Office?

• Take advantage of all local data sources for
  initialization
• Diagnose local weather features having mesoscale
  forcing
• sea/mountain breezes
• modulation of synoptic scale features
• Take advantage of high resolution terrain data to
  downscale national model forecasts
• orography is a data source!

95
Diabatic principle achieved shortest length
forecasts are the most accurate
Hot Start
Typical Cold Start
Accuracy Gain
Results from Road Weather demonstration
Winter-Spring 2004
96
6-H Precipitation verification for all tropical
systems in 2003 over Taiwan (Jian and McGinley,
JMSJ 10/2005)
Hot Start Cold Start
97
12-H Precipitation verification for 4 tropical
systems over Taiwan (Jian and McGinley)
Hot Start Cold Start
98
WRF Run for Typhoon Mindulle using NCEP GFS/NF-15
backgrounds
Cold -GFS
Hot - NFS
99
Some results with Hot Start WRF for Taiwan -
Typhoon Longwang Oct 1, 2005
Thanks to Yun-Tsai Lin and Chris Anderson
100
Ensemble Applications
101
Benefit From Ensembles, to Improved Analyses,
to Better Forecasts, to Probabilities, to
Yes/No Decision Support
Location, Time
Cost/Loss Thresholds
Decision Engine
6-D model grids (variable,x,y,z,t,prob)
Ensemble
Yes or No
Error Covariances
3-D Var Initial Conditions
Customer
102
3DVAR Cost Function
B, O are covariance matrices x is the control
variable xb is the background field H is the
observation operator (maybe nonlinear) y is the
observation. The key is better knowledge of the
structure of B B is situationally dependent -
ensemble approach can help define B. Most 3D var
have simple Bs that over smooth.
103
Ensemble 1-hr precip gt 0.01 in
Ensemble Member
3hr 6hr 9hr 12hr 15hr 18hr 21hr
24hr
104
MM5-Eta
Time-Phased Ensemble an efficient way to get
many members in limited computing environments
t0
H H1 H2 H3 H4 H5
WRF-Eta
Time
Each pair of runs Has a unique Initial condition
based on LAPS
(Number of members) (Number of models)
x (Length of Forecast) / (Start Interval)/2
N
Time weighting is applied to each member
Ensemble at time t0
105
Case April 11, 2005 02z Occluded low precip
across NE, CO, SD.
106
(No Transcript)
107
(No Transcript)
108
Ensemble Post Processing
Six-hour Ensemble forecasts at 2hr
W2 W3 W4 W5
W6
2h 3h 4h
5h 6h
Verification at 2hr
N H
N H
Pi,j,h ?????Wn,t pn,i,j,t / ????Wn,t
n1 th
n1 th
Each forecast is verified at hour, h and weights
computed for each model, n, and for each time, t.
These are updated over time and used to compute
ensemble mean
109
(No Transcript)
110
Precipitation Probabilities on the AWIPS
Workstation
111
Reflectivity at 18GMT 19 Oct 2005 Thanks to NCAR
RAP
112
(No Transcript)
113
(No Transcript)
114
(No Transcript)
115
Future Efforts
116
QPE QPF blending for short-range precipitation
forecasts
117
Vision

• Combine QPE, Extrapolation, and NWP in a seamless
  product extending backwards A hours and forward P
  hours from observation time

QPE
Numerical Weather Prediction
Extrapolation
A hour
Current Time 1hr 2hr
P hr
Time
118
Designing a Forecast/ Observation, QPE/ QPF
Blending Scheme
Forecasts 0H 1H
2H 3H
Forecast Set For New Event
wi
wi
wi
wi
Coefficients/ Weights from Training Set
Post- processor
Correlations
Optimum Forecast Set
Observations 0H 1H
2H 3H
119
Provide LAPS a 3-D Variational Analysis Option

• Implement NCEP GSI and ESRL STMAS3D to take
  advantage of ensemble derived error covariances

120
LAPS III Configuration
Slide 0
Data
Data Ingest
Intermediate data files
Error Covariance
Trans
LAPS
GSI
STMAS3D
Trans
Post proc1
Post proc2
Post proc3
Model prep
WRF-ARW
MM5
WRF-NMM
Verification
Forecast
121
Develop Super-Ensemble

• Combine ensemble members with weights that vary
  over domain.

122
SUPER ENSEMBLING
Wn,91,63,t
Wn,91,34,t
123
The End