Transcript Frequent Subgraph Discovery Mining Scientific & Relational

Chemical Biological
Applications of Mesoscale
Atmospheric Modeling
Presenter:
John Mewes
Department of Atmospheric Sciences
University of North Dakota
Research Team:
Leon Osborne (UND)
John Mewes (UND)
Paul Kucera (UND)
Mark Askelson (UND)
Ben Podoll (GRA UND)
Todd Williams (GRA UND)
Rhesa Freeman (URA UND)
Kaycee Frederick (URA UND)
Outline
Objective Analysis (M. Askelson)
Wind Motion using a Cross-Correlation Analysis
Technique (P. Kucera)
Mesoscale Data Assimilation for Model
Initialization (L. Osborne)
Land-Surface Modeling efforts at the University
of North Dakota (J. Mewes)
UND
Regional Weather Information Center
Objective Analysis
Technique
Mark A. Askelson
University of North Dakota
AHPCRC Annual Review Meeting
August 2003
Background

Challenges


Irregularly-distributed data have deleterious effects on the efficacy
of analysis schemes
Utility of mesoscale NWP forecasts

Depends on accuracy (predictive and scale)


Depends on initialization errors (amounts, variables, etc.)
Need to explore sensitivity to realistic changes in model initialization
variables (land-use, resolution, humidity, etc).
UND
Regional Weather Information Center
Background
Purpose

Alleviate deleterious effects of irregular data
distribution by incorporating the response filter into
LAPS.
Example of Analyses, Amplitude and Phase Modulations for Three Different Analysis
Schemes.
UND
Regional Weather Information Center
Background

Run atmospheric
models at very high
resolution to explore
utility in supporting Army
operations

Evaluate model
sensitivity to model
parameters expected to
cause significant
differences in small-scale
fields



Resolution
Land-use
Physics parameterizations
Simulated rain-water, cloud-water,
and surface flow fields (MM5)
UND
Regional Weather Information Center
Results
Results
Response filter partially incorporated into LAPS





Development and initial testing of 1D and 2D filters complete.
Identified LAPS routines that use empirical weighting techniques.
Identified LAPS routines that generate empirical weights.
Designed interfaces for response filter.
Desired Amplitude Responses vs. Amplitude Response Modulations When The Reference Number is 4
1.2
1
Amplitude Response
0.8
Desired Amplitude Responses
0.6
Amplitude Response Modulations
0.4
Figure that shows that the 1D
response filter can reproduce
the desired amplitude
modulation when data are
irregularly distributed.
0.2
0
0
0.1
0.2
0.3
0.4
0.5
0.6
Frequency
UND
Regional Weather Information Center
Results

Tests of model sensitivity
 Summer Institute


Crystal Paulsen (UND) and Georgette Holmes (JSU)
Experiments run on the Cray X1
Lots of help from Tony Meys (NetASPx)
Hor. grid spacing: 20, 10, 5 and 1 km over large domain.


Q, dx = 20 km, z = 1.6 km
Q, dx = 1 km, z = 1.6 km
UND
Regional Weather Information Center
Results

Land-Use


Changed area of crop land to bare ground
in north central Oklahoma (apparent as red
box in bottom-right image).
Cloudiness over area changed.
Clouds and surface T, original land use
(crop land)
Clouds and surface T, changed land use
(bare ground)
UND
Regional Weather Information Center
Conclusion
Results


Response filter appears to be superior to ‘simple’ schemes.
MM5 shows significant sensitivities to grid spacing, land
use, and physics parameterizations (Students’ presentation
at http://www.ahpcrc.org/~cpaulsen/index.html)
Future Work

Response filter and LAPS



Finish integration
Real-time testing
Model sensitivity tests
Compare with observations
 Perform more tests (e.g., dx = 0.5, 0.25 km)
UND
Regional Weather Information Center

Wind Motion using a
Cross-Correlation Analysis
Technique
Paul A. Kucera
University of North Dakota
AHPCRC Annual Review Meeting
August 2003
Cross-Correlation Analysis (CCA) of
Lower Tropospheric Wind Fields
Motivation:


Provides information about the 3-D wind fields in
regions with very few or no direct observations (i.e.
rawindsondes)
Provides spatial wind estimates for improved
mesoscale model initialization in data sparse regions
Issues:



Assumes cloud and precipitation elements are quasisteady-state between each time interval
Sensitive to spatial and temporal resolution of the
data
CCA is computationally intensive that is well-suited for
high-performance computing
UND
Regional Weather Information Center
CCA Technique
Determine Lagrangian Autocorrelation for horizontal lags α, β between images at
time, t, and t + τ. The parameters S and T are the observations in search window
and surrounding target windows, respectively, and n, m are the dimensions of the
n 1 m 1
windows.
R ,  , 
 S
l 0 k 0
 S
n 1 m 1
l 0 k 0
kl

 S kl Tkl  Tkl
kl  S kl
  T
2
n 1 m 1
l 0 k 0

kl  Tkl

2
Second Image Overlaid
With Cartesian Grid
First Image Overlaid
With Cartesian Grid
The location of maximum
correlation for lags, α, β at time
lag, τ will determine “best”
direction and speed of the
elements in search window, S
…
…
…
Search Area Around
Corresponding Grid
Point in Second Image
Search Location At
Center of Grid Point
In First Image
UND
Regional Weather Information Center
Example Wind Retrieval
1122 UTC
1142 UTC
6 m/s
12 m/s
UND
Regional Weather Information Center
Verification of CCA Technique
Rawindsonde – 0933 UTC
2 km
Altitude
12 m/s
CCA Technique: RMSE ~15 deg in wind direction
~ factor 2 underestimation in wind speed
UND
Regional Weather Information Center
Error Analysis
0942 UTC
Large errors due
to temporal
evolution of the
storms between
time steps
6 m/s
12 m/s
1152 UTC
0952 UTC
6 m/s
12 m/s
6 m/s
12 m/s
UND
Regional Weather Information Center
Improved Approach: Echo Tracking Combined
with Spatial Decomposition Retrievals
Currently implementing echo tracking (CCA Technique) along with
spatial decomposition algorithms developed by the BMRC Australia
for nowcasting of severe storms (Seed 2003)
The algorithm is computationally efficient and has the ability to
reduce retrieval error significantly (~50% reduction in RMSE)
through the decomposition of various storm scales.
Spatial Decomposition Algorithm:


assumes that storms have a multiplicative structure that are organized
as continuum of scales ranging from 100 m to 100’s of km
Storm structure can be decomposed using a FFT and a bandpass filter
centered each cascade scale based on the following equation:
n
Fi , j t    X k ,i , j t ; i  1,...L; j  1,..., j; L  2n
k 1
where Xk,i,j is the field of the scale k, L is the spatial domain size for each
scale k
UND
Regional Weather Information Center
Spatial Decomposition of Storm Scales
The storm structure at different scales
can be characterized by its
autocorrelation function
The lifetime of a pattern of the
reflectivity field is dependant on its
scale (i.e. small scales are less
correlated)
Use a Autoregressive model order 2,
AR(2), to predict the evolution of the
storm at various scales using the
equation
X k,i,jt  τ   Φk,1 t X k,i,jt   Φk ,2 t X k,i,jt  1
Example Autocorrelation
functions
Where Φk,1(t) and Φk,2(t) are the model
coefficients using the Yule-Walker
equations
UND
Regional Weather Information Center
Example Decomposition of a Storm
Large Scale Features
Original Reflectivity Field
Small Scale Features
Medium Scale Features
UND
Regional Weather Information Center
Current Research Activity
Currently implementing the echo
tracking/spatial decomposition software to
large WSR-88D radar dataset:
 4 months (July-October 2002) WSR-88D
radar data from South Florida (Key
West, Miami, Melbourne, and Tampa
Bay).
 9500 merged radar maps at a 6-min
temporal and 2 km x 2 km horizontal 1
km vertical resolution (1 km – 12 km
altitude)
 Spatial domain: 900 km x 900 km
 Data have been QC’ed by students
Near Term:
 Develop an interface to ingest wind
fields into LAPS
 Parallelize code for implementation on
the AHPCRC computer resources
UND
Regional Weather Information Center
Mesoscale Data
Assimilation for Model
Initialization
Leon F. Osborne, Jr.
Director, Regional Weather Information
Center
Professor, Atmospheric Sciences
University of North Dakota
Grand Forks, North Dakota
Challenges and Relevance of
Investigation
Work focuses on enhancing detail of boundary layer
structure in an operational data assimilation system
(LAPS)


Improving data acquisition of low-atmosphere data
Establishing multiple analysis layers within atmospheric
boundary layer
Challenges


Lack of direct PBL observations
Expanding LAPS code to accommodate
expanded remotely sensed wind observations
Relevance

Provide improved initialization of mesoscale and
CFD models yielding improved chemicalbiological dispersion forecasts
UND
Regional Weather Information Center
Core Analysis Method
3D-Variational adjustment is applied to objectively analyzed
fields containing heterogeneous data types:
J = J B + JO + JC
JB is a weighted fit of the analysis to the
background field
JO is a weighted fit of the analysis to the
observations
JC is a term which can be used to minimize
the noise produced by the analysis (e.g., by
introducing a balance).
UND
Regional Weather Information Center
Three-Dimensional Variational Assimilation
Domain initialized with a
previous forecast for
mass, momentum and
moisture
Utilizes data models i.e.
Doppler radial winds in
data assimilation
3DVar adjustments are
made throughout the
atmosphere including
new data layers in the
PBL
Transformation matrix, K,
is replaced by models
for various remotely observed data
J  JO  JB  JD  JS

1
m,n
JO   m,n CVrm,n  Vrob
2 m,n

2
1
2
2
2 
JB    ub  u  ub    vb v  v b    wb w  w b   
2  i , j ,k
i , j ,k
i , j ,k

1
JD   D D 2
2 i , j ,k
1
2
2
2 
JS    us  u    vs  v    ws  w   
2  i , j ,k
i , j ,k
i , j ,k

UND
Regional Weather Information Center
Data Assimilation Activities
Observed Data Sources

In Situ


METAR, SYNOP, Mesonet, Aircraft, Rawinsonde
Remote Sensing

GOES, POES, NEXRAD
Model Backgrounds

Meso-ETA

Provides background field for observation refinement
Data Volume (all domains)


Input: 425 Mbytes each hour
Output: 1,439 Mbytes each hour
Frequency

Hourly across 3 domains
Grid Spacing:


Vertical: 35-40 levels (maximized for PBL support)
Horizontal: 5-kilometers
UND
Regional Weather Information Center
UND LAPS Data Assimilation Support
Provides primary data support for UND
AHPCRC atmospheric science research
activities
Prepares a 3-Dimensional representation of
atmospheric structure and conditions

Includes parameterizations for depicting the
presence of clouds within moisture fields
Initialization data for MM5 and WRF
mesoscale modeling



Provides cold-start initialization as default
Hourly data assimilation provides inputs for
FDDA initializations (warm-start)
Supports diabatic initialization (hot-start) for
MM5 and WRF with proper adjustments to
mesoscale model initialization codes
UND
Regional Weather Information Center
Accomplishments:
Modification of LAPS code to support non-uniform vertical levels
 Expansion of LAPS levels within atmospheric boundary layer to
provide 10 hPa resolutions
Incorporated multiple Doppler radar into LAPS momentum analysis routines
using CRAFT provided data
 real-time data processing of WSR 88-D Level II data to produce 3-D
volumes of cloud and wind information
Boundary layer enhancements to LAPS to permit
Development of an interactive LAPS profile retrieval system and interactive
LAPS display capability for researchers (next slide)
UND
Regional Weather Information Center
Interactive retrieval of LAPS data
Permits researchers to download
location specific data regions and
profiles for use in model testing.
Interactive LAPS data viewer
LAPS does not have an inherent
visualization toolkit as released by FSL.
A java-based visualization system has
been developed at UND that permits
researchers to selectively view 2-D and
3-D datasets. A web-based applet has
been developed for offsite users.
UND
Regional Weather Information Center
AHPCRC Chem-Bio
LAND-SURFACE
MODELING
efforts at the University of North
Dakota
Dr. John J. Mewes
Associate Professor
Atmospheric Sciences
AHPCRC Annual Review Meeting
August 27th, 2003
Goal
To improve analyses and short-term forecasts of the lower atmospheric
stability structure by coupling an advanced Land Surface Model (LSM) to
the Local Analysis and Prediction System (LAPS)
Why
The stability structure of the lower atmosphere is of primary importance in
modulating both its dispersive properties and the effects it has on the
propagation of electromagnetic radiation.
Critically Important
• Latent heat fluxes
• Sensible heat fluxes
• Emission, absorption and reflection of radiation
UND
Regional Weather Information Center
How
Chose the “NOAH” Land-Surface
Model because of its present
sophistication and potential for
further enhancements by the LSM
community.
• Embedded the NOAH LSM within
the LAPS framework, using LAPS
analyses of temperature, winds,
humidity, cloud cover (to calculate
radiation), and precipitation as
forcing.
• Added a ‘tiling’ feature to instill the
effects of sub-grid scale land surface
variations into the atmospheric
analyses.
UND
Regional Weather Information Center
Tiling
Basic idea is that the fluxes over
one grid cell are a weighted
aggregation of the fluxes from
each ‘tile’ of unique soil /
vegetation type pairing within
the cell:
Fcell=F1A1+F2A2+….+FNAN
where each cell (1..N) has a
unique pairing of soil and
vegetation characteristics and an
area (A) that is representative of
their actual distribution within
the cell.
UND
Regional Weather Information Center
Current Status
LSM is operational and undergoing operational testing in several domains.
D.C. / Baltimore Corridor
Oklahoma City
Wichita
Tulsa
• Primary verification efforts are being conducted in the Southern Plains to
take advantage of vast ARM & Oklahoma Mesonet observational resources.
UND
Regional Weather Information Center
Immediate Research Plans
• Continue LSM verification, tuning, and enhancement
efforts.
• Begin utilizing the LAPS LSM heat and radiative fluxes
to improve LAPS analyses of the lower atmospheric
stability structure.
•
Parameterize stability structure in terms of fluxes and ambient
atmospheric characteristics?
• Drive a 1-D PBL model?
• Use LSM fields to initialize a short-term mesoscale model (that
also uses NOAH) forecast, which can then serve as the
background field for the next analysis?
• Other possibilities?
UND
Regional Weather Information Center