Presentation on theme: "Problems With Model Physics in Mesoscale Models Clifford F. Mass, University of Washington, Seattle, WA."— Presentation transcript:
Problems With Model Physics in Mesoscale Models Clifford F. Mass, University of Washington, Seattle, WA
Major Improvements in Mesoscale Prediction Major improvements in the skill of mesoscale models as resolution has increased to 3-15 km. Since mesoscale predictability is highly dependent on synoptic predictability, advances in synoptic observations and data assimilation have produced substantial forecast skill benefits. Although model physics has improved there are still major weaknesses that need to be overcome.
Important to Know the Strengths and Weaknesses of Our Tools
Very Complex Because Model Physics Interaction With Each OtherAND Model Dynamics
Some Physics Issues with the WRF Model that Are Shared With Virtually All Other Mesoscale Models
Overmixing in Mesoscale Models Most mesoscale models have problems in maintaining shallow, stable cool layers near the surface. Excessive mixing in the vertical results in excessive temperatures at the surface and excessive winds under stable conditions. Such periods are traditionally ones in which weather forecasters can greatly improve over the models or models/statistical post-processing
Time series of bias in MAX-T over the U.S., 1 August 2003 – 1 August Mean temperature over all stations is shown with a dotted line. 3-day smoothing is performed on the data. Cold spell
Shallow Fog…Nov 19, 2005 Held in at low levels for days. Associated with a shallow cold, moist layer with an inversion above. MM5 and WRF predicted the inversion…generally without the shallow mixed layer of cold air a few hundred meters deep MM5 or WRF could not maintain the moisture at low levels
High-Resolution Model Output
So What is the Problem? We are using the Yonsei University (YSU) scheme in most work. We have tried all available WRF PBL schemes…no obvious solution in any of them. Same behavior obvious in other models and PBL parameterizations. Doesnt improve going from 36 to 12 km resolution, 1.3 km slightly better. There appears to be common flaws in most boundary layer schemes especially under stable conditions.
Problems with WRF surface winds WRF generally has a substantial overprediction bias for all but the lightest winds. Not enough light winds. Winds are generally too geostrophic over land. Not enough contrast between winds over land and water. This problem is evident virtually everywhere and appears to occur in all PBL schemes available with WRF. Worst in stable conditions.
Northeast U.S. from SUNY Stony Brook (Courtesy of Brian Colle): hr wind bias for NE US: additive bias (F-O)
SUNY Stony Brook: Wind Bias over Extended Period for One Ensemble Member
U.S. Army WRF over Utah
Cheng and Steenburgh 2005 (circles are WRF)
UW WRF km: Positive Bias Change in System July 2006Now
Wind Direction Bias: Too Geostrophic
MAE is something we like to forget…
Surface Wind Problems Clearly, there are flaws in current planetary boundary layer schemes. But there also be another problem?the inability to consider sub-grid scale variability in terrain and land use.
The 12-km grid versus terrain
A new drag surface drag parameterization Determine the subgrid terrain variance and make surface drag or roughness used in model dependent on it. Consulting with Jimy Dudhia of NCAR came up with an approachenhancing u* and only in the boundary layer scheme (YSU). For our 12-km and 36-km runs used the variance of 1-km grid spacing terrain.
38 Different Experiments: Multi- month evaluation winter and summer
Some Results for Experiment71 Ran the modeling system over a five-week test period (Jan 1- Feb 8, 2010)
10-m wind speed bias: Winter Original
MAE 10m wind speed
Case Study: Original
During the 1990s it became clear that there were problems with the simulated precipitation and microphysical distributions Apparent in the MM5 forecasts at 12 and 4-km Also obvious in research simulations of major storm events.
Early Work (mainly MM5, but results are more general) Relatively simple microphysics: water, ice/snow, no supercooled water, no graupel Tendency for overprediction on the windward slopes of mountain barriers. Only for heaviest observed amounts was there no overprediction. Tendency for underprediction to the lee of mountains
MM5 Precip Bias for 24-h 90% and 160% lines are contoured with dashed and solid lines For entire Winter season
Testing more sophisticated schemes and higher resolution ~2000 Testing of ultra-high resolution (~1 km) and better microphysics schemes (e.g., with supercooled water and graupel), showed some improvements but fundamental problems remained: e.g., lee dry bias, overprediction for light to moderate events, but not the heaviest. Example: simulations of the 5-9 February 1996 flood of Colle and Mass 2000.
5-9 February 1996 Flooding Event
MM5: Little Windward Bias, Too Dry in Lee Bias: 100%-no bias Windward slope Lee
IMPROVE Clearly, progress in improving the simulation of precipitation and clouds demanded better observations: – High quality insitu observations aloft of cloud and precipitation species. – Comprehensive radar coverage – High quality basic state information (e.g., wind, humidity, temperature) The IMPROVE field experiment (2001) was designed and to a significant degree achieved this.
Olympic Mts. British Columbia Washington Cascade Mts. Oregon California Orographic Study Area Washington Oregon Coastal Mts. S-Pol Radar Range Santiam Pass OSA ridge crest Cascade Mts. < 100 m m m m m m > 3000 m Terrain Heights Portland Salem Newport Medford UW Convair-580 Airborne Doppler Radar S-Pol Radar BINET Antenna NEXRAD Radar Wind Profiler Rawinsonde Legend Ground Observer km WSRP Dropsondes Columbia R. Rain Gauge Sites in OSA Vicinity Santiam Pass SNOTEL sites CO-OP rain gauge sites 50 km Orographic Study Area S-Pol Radar Range Olympic Mts. S-Pol Radar Range Westport 90 nm (168 km) Offshore Frontal Study Area Paine Field Univ. of Washington Area of Multi- Doppler Coverage Special Raingauges PNNL Remote Sensing Site Two IMPROVE observational campaigns: I. Offshore Frontal Study (Wash. Coast, Jan-Feb 2001) II. Orographic Study (Oregon Cascades, Nov-Dec 2001)
The NOAA P3 Research Aircraft Dual Doppler Tail Radar Surveillance Radar Cloud Physics and Standard Met. Sensors Convair 580 Cloud Physics and Standard Met. Sensors
PARSL Site Terrain ht. (m) Distance (km) S-POL Radar Santiam Junction Santiam Pass Camp Sherman inches/year inches/year inches/year inches/year > 100 inches/year < 20inches/year 60 km100 km Slope matches that of an ice crystal falling at 0.5 m/s in a mean cross-barrier flow of 10 m/s, which takes ~3 h. Total flight time: 3.4 h Convair-580 Flight Strategy
We now had the microphysical data aloft to determine what was happening Model Observations
The Diagnosis Too much snow being produced aloft Too much snow blowing over the mountains, providing overprediction in the lee Too much cloud liquid water on the lower windward slopes Too little cloud liquid water near crest level. Problems with the snow size distribution (too few small particles) Several others!
Problems and deficiencies of boundary layer and diffusion schemes can significantly affect precipitation and microphysics Boundary layer parameterizations are generally considered one of the major weaknesses of mesoscale models Deficiencies in the PBL structures were noted during IMPROVE. Errors in boundary layer structure can substantially alter mountain waves and resultant precipitation.
Impacts of Boundary Layer Parameterization on Microphysics Snow-diffCLW-diffGraupel-diff Microphysics Differences ETA - MRF
Lots of activity in improving microphysical parameterizations New Thompson Scheme for WRF that includes a number of significant improvements. Higher moment schemes are being tested. (e.g., new Morrison two-moment scheme) Microphysical schemes are being modified to consider the different density and fall speed characteristics of varying ice habits and degrees of riming.
Convective Parameterization The need for convective parameterization declines at models gain enough resolution to explicitly model convection. Appears that one starts getting useful explicit convective predictions at 4-km grid spacing. In the future, they is one problem that will go away as we move to sub-4km grid spacing.
Real-time 12 h WRF Reflectivity Forecast Composite NEXRAD Radar 4 km BAMEX forecast Valid 6/10/03 12Z 10 km BAMEX forecast 22 km CONUS forecast
Example: Radar reflectivity, 24 h fcst vs obs, valid 0000 UTC May 13, 2005 WRF 4km WRF 2km NMM 4.5km observed
Hurricane Rainbands Ultra high resolution (< 2 km grid spacing) result in better structures and intensity predictions. 15-km grid spacing1.67 km grid spacing
More Physics Issues Serious deficiencies in many land surface modeling schemes, particularly in the areas of snow physics and soil moisture Need to characterize uncertainties in physics schemes and the development of stochastic physics. Require physics schemes applicable to a wide range of resolutions for the next generation of unified models.
Resolution Was Easy We have had a lot of fun increasing resolution over the past few decades. Now we have to put much more emphasis on doing the research and operational testing required to improve model physics and describing the uncertainties in our schemes. This work is made more difficult by the interactions among the physics parameterizations.
Garvert, Mass, and Smull, 2007 Improve-2 Dec13-14, 2001 Changes in PBL schemes substantially change PBL structures, with none bein correct.
An Issue Our method appears to hurt slightly during strong wind speeds and near maximum temperatures in summer.
With Sub-grid drag
Improvement? Next stepcould have the parameterizaton fade out for higher winds speeds and lower stability, possibility by depending on Richardson number. Actually, this makes some sense…sometimes the atmosphere is well-mixed, and at these times variations in sub-grid roughness would be less important.