ENVIRON_UNC_WRF_20131101

Report
November 1, 2013
3SAQS WRF Modeling
Recommendations
Bart Brashers, ENVIRON
Jared Heath Bowden, UNC
Where We Left Off at the Last Meeting
• Week long simulations for beginning of January
and July
• Tests included:
–
–
–
–
–
ECMWF vs. NAM Initial and Boundary Conditions
USGS vs. NLCD land-use datasets
PX vs. NOAH land-surface model
TOPO option
Reduced vertical levels
2
Initial Conclusions from Last Meeting
• Our conclusions from model metrics (T,Q,WS,WD)
Keep NAM instead of ECMWF
NLCD with NOAH requires more development
TOPO option created larger negative wind speed bias
Do not reduce number of vertical levels, especially given
the desire to model temperature inversions
– PX LSM/ACM2 vs. NOAH/YSU – found that PX crashed for
January; preliminary comparison for July indicated PX was
also comparable to our BASE configuration
–
–
–
–
3
Where We Went: Several Blind Alleys
• PX LSM/ACM2 would run with 27 levels, crash
consistently with 37 levels
– wrfhelp and Jon Pleim consulted
– Coding error caused crash instead of graceful stop
• Nudging with MESONET data (part of MADIS)
caused some runs to crash (segmentation fault)
– Could not find bad observation
– Did not improve WRF performance noticeably
• New “snow” data from Polar WRF turned out to be
only over ice, not over land
– Good for them, but not for us
4
Revised “Base Case”
• Do not use objectively analyzed fields (metoa files
created by OBSGRID.EXE) for any domains
– NAM already analyzed at 12km (c.f. 36 & 12km domains)
– 4km domain has no analysis nudging, so uses metoa file
only for initial conditions, which should have no effect
after spin-up
– Using metoa files produced far to much precipitation in
previous ENVIRON work near Four Corners
• Use OBS nudging every hour
– Was set to every 3 hours
– Retain nudging coef’s used in ENVIRON Four Corners WRF
5
PRISM Precipitation Comparison
•
•
•
•
•
PRISM is gridded observational data
Sophisticated interpolation techniques
4km resolution, monthly total precip
PRISM data is interpolated to each WRF nested grid
WRF output is simply summed
6
WRF Base Case vs. PRISM, January 2011
WRF
PRISM
7
WRF Base Case vs. PRISM, July 2011
WRF
PRISM
8
WRF Base Case vs. PRISM, January 2011
WRF
PRISM
9
WRF Base Case vs. PRISM, July 2011
WRF
PRISM
10
Preliminary Conclusion
Base Case precipitation performance is
acceptable. Let’s look at the rest of the
typical performance metrics (METSTAT).
11
Introduction to Soccerplots
• Plots of model
Bias vs. Error
for each month
ERROR
• Performance
envelopes from
survey of many
MM5 and WRF
runs
BIAS
• How many months score a goal?
GOOOOAAALLL!
12
BASE
EPA
13
BASE
EPA
14
BASE
EPA
15
BASE
EPA
16
Preliminary Conclusion
We’re doing about as well as EPA’s WRF
modeling, which they use for CMAQ. But
can we do better with stronger nudging?
17
Nudging Case
•
•
•
•
•
•
•
•
New test case, following Kristi Gebhart/RoMANS II
Analysis (3D) nudging the same as Base case
OBS nudging on 4km domain stronger
Nudge Winds and Temperature, but not Humidity
Winds nudging coefficient = 1.2E-3
Temperature nudging coefficient = 6.0E-4
Radius of influence = 60km
Do not nudge to ACARS, SatWind, or Profiler data
18
19
20
Preliminary Conclusion
Stronger nudging reduces model error.
At least when you verify using the same
data you used for nudging...
21
Verifying Against MESONET Data
• MADIS data contains many “data streams”
–
–
–
–
–
–
–
ACARS
HDW
MARITIME
METAR
RAOB
SAO
MESONET
Aircraft
Satellite winds
Ship reports
ASOS, AWOS, more.
Radiosondes
Canadian stations
Smaller station
No thanks
No thanks
Not relevant for 4km
Yes please!
Not relevant for SFC
Not relevant for 4km
Interesting…
• Could we nudge with METAR & RAOB, and verify
against MESONET data?
22
4 WRF Runs vs. 3 Verification Sets
23
4 WRF Runs vs. 3 Verification Sets
24
4 WRF Runs vs. 3 Verification Sets
25
4 WRF Runs vs. 3 Verification Sets
26
Preliminary Conclusion
Stronger nudging did not change model
performance, when using independent
verification data (MESONET). Precip?
27
Base Case vs. Nudging Case, July 2011
Base Case
Inches
Nudging Case
28
New Case: No Temperature Nudging
Base Case
No Temp Nudging
29
Preliminary Conclusion
OBS nudging too strongly produces too
much precipitation, without changing
performance vs. independent data.
30
ENVIRON/UNC WRF Recommendations
•
•
•
•
•
•
•
Taken from Base Case
37 Vertical Levels
NOAH LSM
YSU PBL
RRTMG shortwave & longwave radiation
Thompson Moist Physics
KF Cumulus on 36 & 12km
– kfeta_trigger = 2
– No shallow Cu (not tied to deep convection)
• Light OBS nudging of Winds, Temp, and Humidity
31
Thanks for listening!
Questions? Discussion?
Bart Brashers
[email protected]
Jared Heath Bowden
[email protected]

similar documents