short-term forecast performance of polar wrf in the antarctic
DESCRIPTION
Short-term forecast performance of Polar WRF in the Antarctic. Francis Otieno 1 , David Bromwich 1, 2 , Keith Hines 1 and Elad Shilo 3. - PowerPoint PPT PresentationTRANSCRIPT
Byrd Polar Research Center, The Ohio State University, Columbus, OH
2Atmospheric Sciences Program, Department of Geography, The Ohio State University, Columbus, OH
3 Israel Meteorological Service
Funded by NSF and NASA
Francis Otieno1, David Bromwich 1, 2, Keith Hines1 and Elad Shilo3
Outline of Talk1. Introduction and background2. Objectives of this study3. Model Domain and
configuration4. Results
a) Data qualityb) Surface variablesc) Upper air variables
5. Work to be completed6. Conclusion
BackgroundMost contemporary forecast models have been developed primarily in
and for the mid latitudes in the Northern Hemisphere.
The Polar Meteorology Group optimized the Penn State NCAR MM5 model for polar applications. In tests over Greenland, SHEBA region, Atqasuk, Barrow and Iceland the model shows significant skill (e.g. Wilson 2010, Hines and Bromwich 2008, Bromwich et. al (2009)
The Antarctic Mesoscale Prediction System (AMPS) project previously used Polar MM5 until June 2008 to make the forecasts but is currently using the new mesoscale model (Polar WRF, 3.0.1)
AMPS archive probably provides the only detailed high resolution four dimensional structure of the Antarctic atmosphere; invaluable tools in providing forecasts needed in support of Antarctic operational activities.
BackgroundIn the 2009-2010 field season, 400 LC-130 missions were conducted by
USAF on the continent (NSF 2010).
Accurate forecasts have played a critical role in previous evacuations from the continent and will continue to do so as the number of research and tourist visits increase.
The model is continuously becoming very sophisticated with new changes every year (PWRF2.0, 2.2, 3.0, 3.0.1.1, PWRF3.2 due Aug 2010)
It is very important to periodically assess the skill of the model as the safety of personnel and economic operations in Antarctica depend on the accuracy these forecasts
More information on Polar WRF (PMG Website)
Antarctic versus Arctic • Antarctica lacks the dense data network needed
to provide Polar WRF with an accurate representation of the large scale circulation.
• For skillful forecasts the model must accurately represent the Antarctic katabatic winds (Bromwich and Liu 1996, Parish and Bromwich 1998, Cassano and Parish 2000) that are governed by the balance of gravity, thermal stability and synoptic forcing (Elevated ice sheet)
• Sea ice impacts the atmosphere ocean interaction
ObjectivesAssess the performance of Polar
WRF3.0.1.1 in short-term forecasts over Antarctica; On an annual time scale
Identify an optimal configuration for short-term forecasts in Antarctica
Identify any deficiencies in the model when used in Antarctica
The DomainTerrain-Radarsat Antarctic Mapping
Project (RAMP-DEM Lui et al. 2001)
SST – Real-time, global, sea surface temperature analyses from NCEP
Sea Ice- SMMR on Nimbus 7 and SMM/I DMSP using bootstrapping (NSIDC)
Lateral boundary Conditions from NCEP Final Analysis
Reconstructed mean surface air temperature (Monaghan et al. 2008) provide the ice temperatures at depth.
Fig. 1: Region of study showing the terrain elevation and location of stations. Red contour marks 0.1 fractional sea ice and shows the winter extent.
All Upper air stations located near the coast (except Vostock and Amundsen-Scott)
Not all stations useable
Model Physics and ConfigurationModel Physics and ConfigurationWSM 5-class schemeGoddard shortwaveRRTM Longwave (G)Noah Land-surface modelMellor Yamada-Janjic (Eta)Grell-Devenyi39 vertical levels Based on Arctic Config.
Polar WRF 3.0.1.1 (PWRF3.2 Aug 2010)
Polar Stereographic centered at S. Pole
60 km; 120 x 120 grid sizeForecast mode 48 hour run ; analyzed
last 24 hrs
SST updated every six hoursSea ice concentrations specified
FNL lateral forcing every six hours
Observed Data AWS Univ. Wisconsin Automatic Weather
Project Surface temperature, pressure, wind and RH
NCDC Surface archive (overlap AWS) BAS British Antarctic Survey (Radiosonde) Upper Air soundings IGRA-Integrated
Radiosonde Archive BSRN Baseline Rad. Network –Downwelling
LW and SW
DATA QUALITY Spikes in data not real Do not occur at the same
time Very difficult to quality
control wind information (spike may be an actual gust)
Could have 98% availability for one variable and not the other
Station may be a summer only
Initially looked at 1993 (Madison 2009); excessive thinning
Looking at 2007 additional AWS
DATA QUALITY
• Less than 50% availabilityOutliers 3σ, deviation from a
running meanAutomating QC is a difficult
challengeLinear interpolationJust set to missing
Surface Pressure•Model shows good skill (correlations > 0.9) for a number of stations
•There are problems with some locations with larger disparity between model and station elevation
•Not all stations have data for the whole period
•Some stations are not well represented e.g. Durmont D’Uvrille
January temperature correlations vary a lot; Results improve substantially with more rigorous quality control.
The direction of flow has to be considered in the quality control; A coastal station may be more representative of a maritime rather than land depending on flow direction
Some station reflect mostly local and not synoptic scale variability
Temperature
Bias RMSD CorrJanuary (18) -2.1 3.4 0.61
July (25) -1.1 3.8 0.64
Results from the Peninsula
Large differences in July days 1-5 and 20-25
Pressure variability simulated better than that of temperature
At 60 km the model misses some of the island stations
Three stations show similar SFCP behavior
Upper Air ResultsAnnual range of temperature observations is captured well; range is larger near model top
Polar WRF shows a summer warm bias in the upper troposphere (~300 hPa)
Winter and summer temperature profiles are simulated well except at Marambio where model shows a warm bias.
Temperature Wind Speed
BIAS RMSD CORR BIAS RMSD CORR
850 1.04 2.57 0.94 1.42 5.44 0.69
700 0.87 2.46 0.92 1.27 4.78 0.68
500 0.03 1.77 0.96 0.51 4.95 0.77
400 0.07 1.7 0.88 0.38 5.93 0.77
300 0.9 2.01 0.65-
0.24 6.5 0.80
250 0.76 2.28 0.86 0.04 5.48 0.83
200 0.04 1.96 0.93 0.45 4.19 0.86
•Climatological features captured (colder, lower dew point temps-July; strong circumpolar winds•Upper air annual statistics are from the nine IGRA stations •Model wind speeds are slightly stronger than observed. •U and V statistics suggest that upper level wind direction reasonably simulate
Geopotential heights simulated well in the lower and mid troposphere
Differences get larger near the model top
Possible fix (RRTMG)
Polar WRF 3.0.1.1 vs 9 IGRA Upper Air Stations for 2007
Temperature Geopotential Wind Speed U-Wind V-Wind
BIAS RMSD CORR BIAS RMSD CORR BIAS RMSD CORR BIAS RMSD CORR BIAS RMSD CORR
850 1.04 2.57 0.94 -6.26 23.67 0.96 1.42 5.44 0.69 0.24 5.39 0.69 0.45 5.34 0.65
700 0.87 2.46 0.92 0.63 26.17 0.89 1.27 4.78 0.68 0.96 4.90 0.69 -1.76 6.58 0.57
500 0.03 1.77 0.96 -1.27 29.56 0.97 0.51 4.95 0.77 0.83 6.07 0.67 -0.14 6.43 0.69
400 0.07 1.70 0.88 2.2 33.13 0.98 0.38 5.93 0.77 0.43 7.11 0.67 0.24 7.74 0.70
300 0.90 2.01 0.65 5.38 36.26 0.98 -0.24 6.50 0.80 0.21 7.67 0.67 0.44 8.39 0.70
250 0.76 2.28 0.86 22.47 41.3 0.98 0.04 5.48 0.83 0.08 6.73 0.68 0.39 7.47 0.70
200 0.04 1.96 0.93 36.98 48.74 0.99 0.45 4.19 0.86 0.12 5.29 0.70 0.45 6.04 0.70
Biases in Polar WRF simulations are small between 850 and 300 hPa
Geopotential errors become larger above 300 hPa (small percentage differences)
Downwelling Radiation SW and LW
The variability and amplitude of the downwelling longwave radiation (responsible for surface heating during the long Antarctic winters) is forecast well
But the Model allows excessive downwelling shortwave (~40 Wm-2) at the surface during the Austral summer The excess shortwave appears in the hours following the local noon
Wind speed ANN DJF MAM JJA SON
BIAS RMSD CORR BIAS RMSD CORR BIAS RMSD CORR BIAS RMSD CORR BIAS RMSD CORR
850 1.4 5.4 0.69 0.4 3.6 0.83 1.7 5.1 0.74 1.8 6.1 0.65 1.0 5.4 0.7
700 1.3 4.8 0.68 0.1 2.8 0.64 1.2 4.5 0.68 1.5 5.2 0.67 0.4 4.2 0.72
500 0.5 5.0 0.77 0.5 3.5 0.81 0.5 5.2 0.76 0.4 5.3 0.75 -0.1 4.2 0.81
400 0.4 5.9 0.77 0.7 4.7 0.80 0.3 6.0 0.78 0.4 6.2 0.77 -0.5 5.3 0.8
300 -0.2 6.5 0.8 0.2 6.3 0.80 -0.2 6.7 0.81 -0.2 6.4 0.81 -0.8 5.8 0.82
250 0.0 5.5 0.83 0.1 4.5 0.86 1.1 5.8 0.81 -0.5 5.4 0.84 -0.9 5.2 0.85
200 0.5 4.2 0.86 0.5 2.7 0.89 1.5 4.1 0.84 -0.3 4.7 0.83 0.1 4.0 0.88
Meridional
850 0.5 5.3 0.65 0.5 3.6 0.74 0.6 5.2 0.69 0.9 5.9 0.62 0.2 4.9 0.62
700 -1.8 6.6 0.57 0.2 3.6 0.77 -0.9 4.7 0.83 -2.8 7.6 0.56 -0.2 4.3 0.81
500 -0.1 6.4 0.69 0.0 4.7 0.67 0.5 5.8 0.79 -0.5 7.7 0.64 -0.6 5.7 0.69
400 0.2 7.7 0.7 -0.2 6.3 0.67 0.9 7.1 0.78 -0.2 9.0 0.64 0.2 6.6 0.7
300 0.4 8.4 0.7 -0.9 7.6 0.75 1.3 7.4 0.77 0.1 9.7 0.63 0.7 7.3 0.71
250 0.4 7.5 0.7 -0.5 5.7 0.74 0.9 6.6 0.73 0.3 8.9 0.64 0.7 6.6 0.72
200 0.5 6.0 0.7 -0.2 4.4 0.71 0.6 5.1 0.74 0.9 7.5 0.65 0.5 5.3 0.72
Zonal
850 0.2 5.4 0.69 0.1 3.7 0.84 0.7 5.1 0.71 -0.1 6.0 0.66 -0.3 5.2 0.74
700 1.0 4.9 0.69 0.6 3.8 0.85 1.9 5.1 0.71 0.7 5.2 0.68 0.1 4.4 0.83
500 0.8 6.1 0.67 0.1 4.5 0.84 1.4 5.5 0.69 0.9 6.6 0.67 1.0 6.5 0.62
400 0.4 7.1 0.67 -0.2 5.6 0.86 0.8 6.5 0.7 0.5 7.6 0.66 0.6 7.7 0.62
300 0.2 7.7 0.67 -0.2 6.4 0.87 0.8 7.2 0.7 0.1 8.1 0.65 0.5 8.2 0.63
250 0.1 6.7 0.68 -0.3 5.1 0.87 1.1 6.4 0.72 -0.3 7.2 0.68 0.3 7.3 0.63
200 0.1 5.3 0.7 -0.2 3.5 0.88 1.2 4.8 0.75 -0.4 6.0 0.68 0.4 5.7 0.64
Little variation in seasonal WSP statistics
No obvious height dependence
Bias in wind components generally small (~1 m/s)
Infer that wind direction probably okay
DJF Zonal wind statistics slightly better (strong; JJA?)
SummaryA number of other factors could influence the forecast performance of Polar WRF e.g.
– Lateral boundary data (ERA-40, ERA-INTERIM, NNR or FNL– Interannual variability 1993 versus 2007– Model physics and configuration (WRFV2.0 to WRF3.2)
For stations that have been adequately scrubbed, the preliminary surface verification show good skill ( surface data are more problematic)
Upper air statistics don’t show any major problems (300 hPa and at the model top)
Overall Polar WRF shows skill levels in the SH that are comparable to those in the NHTBD Observations– Scrub and repeat; Only use AWS data from AMRC for overlaps with NCDCTBD –Model May ultimately use Polar WRF3.2Bracket the impacts of interannual variability and lateral boundary data