gfs mos wind guidance: problem solved?
DESCRIPTION
GFS MOS Wind Guidance: Problem Solved?. Eric Engle and Kathryn Gilbert MDL/Statistical Modeling Branch 15 May 2012. Overview. Reason for Refresh Development Sample Predictand Definition Predictors and Equation Development Independent Test Systems Verification Impact on Gridded MOS. - PowerPoint PPT PresentationTRANSCRIPT
GFS MOS Wind Guidance: Problem Solved?
Eric Engle and Kathryn GilbertMDL/Statistical Modeling Branch
15 May 2012
Reason for Refresh Development Sample Predictand Definition Predictors and Equation Development Independent Test Systems Verification Impact on Gridded MOS
Overview
GFS Model “bugfix” implemented in May 2011 Set new thermal roughness length to address a low
level warm bias over land Affected the behavior of the low level wind fields
Usually chosen as predictors (10-m u, v, speed)
Reason for Refresh
GFS Vegetation TypeVegetation Type Description7: Groundcover only (perennial)8: Broadleaf shrubs with perennial groundcover9: Broadleaf shrubs with bare soil11: Bare soil Roughness length
change has most impact in these vegetation type areas.
Verifications show this is true (not shown)
VEGETATION TYPES (DORMAN AND SELLERS, 1989; JAM)
GFS Model “bugfix” implemented in May 2011 Set new thermal roughness length to address a low
level warm bias over land Affected the behavior of the low level wind fields
Usually chosen as predictors (10-m u, v, speed) Had a direct impact on GFS MOS wind speed
guidance Large guidance errors (strong positive biases) Western CONUS (low vegetation/desert areas) Most pronounced in warm season, during daytime hours
Reason for Refresh
Implementation 32.2 knot
error
2009
Credit: Dr. Yun Fan
2010
2011 Jan-Apr
2011 May-July
GFS Model “bugfix” implemented in May 2011 Set new thermal roughness length to address a low level warm
bias over land Affected the behavior of the low level wind fields
Usually chosen as predictors (10-m u, v, speed) Had a direct impact on GFS MOS wind speed
guidance Large forecast busts (strong positive biases) Western CONUS (low vegetation/desert areas) Most pronounced in warm season, during daytime hours
Complaints from NWS forecasters and private sector GFS MOS wind guidance “unusable”
Reason for Refresh
“Our workload has increased due to this problem”
“…we continue to deal with serious fire weather conditions…”“The forecaster
stated that this issue makes the point and gridded MOS…typically used to populate GFE…unusable and times.”
“The situation with the MAV guidance winds has become a source of frustration and a workload issue for our office.”
Responses from Users
Turning off partial inflation (PI) Many stations benefit from PI
Development without boundary layer model predictors Many stations benefit from these
Bias correction Big project…significant MOS production overhead
The solution…collect sufficient mixed sample and redevelop Most timely
Solutions Investigated
Previous two warm seasons available April through September, 2010 and 2011.
Comprised of three different versions of GFS model.
Development Sample
Date Range VersionApril – June 2010 Pre-9.0.0 (operational)July – September 2010 9.0.1 (reforecast)April – May 9 2011 9.0.0 (operational)May 10 – September 2011
9.0.1 (operational)
Balance the influence of “new” model data (version 9.0.1) with a sufficient sample size 64% “new” / 36% “old”
Predictands: 10-m U-wind 10-m V-wind 10-m Wind Speed
Derived from hourly observed 10-m wind speed and direction Wind data are quality controlled via MDL software
Regression equations for predictands are developed simultaneously Predictors selected that best fit all 3 predictands
Different coefficient 3-hourly guidance to 192-h 6-hourly guidance from 204-h to 264-h
Predictand Definition
U, V, Speed at 1000, 925, 850, 700, and 500 hPa; and 10-m
Mass divergence, relative vorticity, vertical velocity at 925, 850, 700, and 500 hPa
Mean Layer RH 1000-850; 850-700 hPa; and 1-0.44 sigma Temperature Difference between 1000-925, 1000-800,
and 1000-700 hPa levels K index Sine and cosine DOY (harmonic functions) PBL mixing parameter Bilinear interpolation Observed predictors offered out to 15-h (persistence)
Predictors Offered
Single station equation development Multiple linear regression (forward selection) Maximum number of predictors: 10 (no
forcing) 100 “cases” required equation to be
developed Three independent test systems were
developed Determine the best mix of “old” and “new” GFS
model sample
Regression and Equation Development
TEST1 Only GFS v9.0.1 “new” used July-Sept. 2010 (reforecast); May 10 – Sept. 2011
TEST2 Same as TEST1, but includes April 1 – May 9, 2011
(GFS v9.0.0) TEST3
All data from previous 2 warm seasons included April-Sept. 2010 and 2011
Independent Test Systems
Cross-validation “leave one out” technique used
Each of the 3 test systems comprised of 8 equation sets 7 out of 8 months of “new” GFS model data
included in each equation set Withheld month is used to verify the equation
set in which it was held back. End up with 8 fully independent months of
guidance to verify against
Independent Testing
Compare Equations for KPHX (24-h)00Z
Operational (as of 5/14/2012)Predictor Coeff.
10-m U-wind -0.138210-m V-wind -0.414110-m Wind Speed 1.64321-0.44 sigma Mean Layer RH
0.0321
850 hPa Mass Divergence -0.57821000-850 hPa Mean Layer RH
0.0209
1000-925 hPa Temp difference
-1.6628
Cosine 2*DOY 0.7597500 hPa Wind Speed 0.0734850 hPa Vert. Velocity (GB)
-0.3317
New (awaiting implementation)Predictor Coeff.
925 hPa U-wind -0.0569850 hPa Wind Speed 1.244510-m V-Wind -0.6657Cosine DOY -1.8675700 hPa Wind Speed 0.14481000-850 hPa Mean Layer RH
0.0519
PBL Mixing (low) 5.2159850 hPa Vert. Velocity (GB)
-0.3785
500 hPa Mass Divergence -0.1887K Index (GB) ( ≥ 40) -1.0493
Verifications show TEST3 being the most accurate/skillful system Attribute this to a longer sample
New equations have similar skill in wind direction guidance as the old equations
Guidance in Alaska degraded with test developments Existing equations will remain in place
Verification Summary
Impact on Gridded MOS
GMOS Example
GMOS Example
GMOS Example
GMOS Example
Change in GFS thermal roughness length Negative impact on GFS MOS
Wind equations redeveloped using previous 2 warm seasons Except for Alaska
Large wind speed guidance errors removed Positive impact on Gridded MOS Future work?
Bias correction Redevelop again with longer sample
Implementation in June 2012
Summary
See the New Equations in Action…
http://www.mdl.nws.noaa.gov/~mos/mos/gfsmos_wind/index.php
Questions?