earth observing laboratory | earth observing laboratory ......– adjoint-sensitivity experiments...
TRANSCRIPT
Assimilation of Mesoscale Observations for use in
Numerical Weather PredictionSteve Koch
Thermodynamic Profiling Technologies WorkshopBoulder 12 – 14 April 2011
Important Questions(from 2003 USWRP Mesoscale Observing Networks Workshop)
• Is it more effective to sample the upper troposphere with fewer observing systems than to sample the boundary layer with more observing systems for mesoscale modeling?
“The Committee envisions a distributed adaptive “network of networks” (NoN) serving multiple environmental applications near the Earth’s surface, jointly provided and used by government, industry, and the public.”
• Is it more cost effective to have intermittent, targeted observations at the mesoscale than to enhance the present operational networks to provide additional data in a continuous manner for improving mesoscale prediction?
“The committee finds that, overall, the status of U.S. surface meteorological observation capabilities is energetic and chaotic, driven mainly by local needs without adequate coordination…an overarching national strategy is needed to integrate disparate systems from which far greater benefit could be derived and to define the additional observations required to achieve a true multi-purpose network that is national in scope”
Important Questions(from 2003 USWRP Mesoscale Observing Networks Workshop)
• What kinds of observations are best for deriving all the other variables not directly observed?
Important Questions(from 2003 USWRP Mesoscale Observing Networks Workshop)
The highest priority observations needed to address current inadequacies are:• PBL height (useful as retrieval constraint)• Soil moisture and temperature profiles• High-resolution vertical profiles of humidity• Measurements of air quality and related chemical composition
above the surface layer
Just below the aforementioned highest priorities are quantities for which some capabilities currently exist but fail to meet a serviceable national standard for one or more reasons:• Vertical profiles of wind (can be used to derive Tv)• Vertical profiles of temperature…
• What mix of radiometric, lidar, interferometric, and active radar systems should be used to obtain the greatest improvement in forecasting severe weather?
Important Questions(from 2003 USWRP Mesoscale Observing Networks Workshop)
“Federal agencies and their partners should deploy lidars and radio frequency profilers nationwide at approximately 400 sites to continually monitor lower tropospheric conditions. Humidity, wind, and diurnal boundary layer structure profiles are the highest priority for a network, the sites for which should have a characteristic spacing of ~125 km but could vary between 50 and 200 km based on regional considerations…Emerging technologies for distributed-collaborative-adaptive sensing should be employed by observing networks, especially scanning remote sensors such as radars and lidars.”
• What role can field experiments play in determining the optimal mix of observations needed to realize the greatest improvements in mesoscale data assimilation and prediction?
Important Questions(from 2003 USWRP Mesoscale Observing Networks Workshop)
“The national network architecture should be sufficiently flexible and open to accommodate auxiliary research-motivated observations and educational needs, often for limited periods in limited regions...federal agencies and partners should employ testbeds for applied research and development to evaluate and integrate national mesoscale observing systems, networks thereof, and attendant data assimilation systems.”
Stan Benjamin: Observation Sensitivity Experiments (OSEs)
• Used state-of-the-art assimilation/modeling system• Used all available observations for relative impacts• 2 types of OSEs
– Data denial experiments (more expensive)• Can show actual effect on forecast skill
– Adjoint-sensitivity experiments (less exp but requires adj)• Provides difference magnitude in forecasts but not actual difference in
forecast skill
• Note: It is easier to show positive impact from certain observation systems when using older assimilation systems or without all available observations. But those results will be misleading.
• Benjamin et al. (2010) results for 8 observing systems to follow…
RUC Hourly Assimilation CycleCycle hydrometeor, soil temp/moisture/snow plus atmosphere state variables
Hourly obs in 2009 NCEP RUCData Type ~NumberRawinsonde (12h) 80NOAA profilers 30 VAD winds 110-130 PBL – profiler/RASS ~25Aircraft (V,temp) 1400-7000 TAMDAR (V,T,RH) 0-1800Surface/METAR 1800-2000 Buoy/ship 100- 200 GOES cloud winds 1000-2500 GOES cloud-top pres 10 km res GPS precip water ~300Mesonet (temp, dpt) ~7000Mesonet (wind) 2000-4000METAR-cloud-vis-wx ~16003-d Radar reflectivity 2km
11 12 13 Time (UTC)
1-hrfcst
BackgroundFields
AnalysisFields
1-hrfcst
3dvar
Obs
1-hrfcst
3dvar
Obs
Obs sensitivity exps
RUCWind forecastAccuracy
Sept-Dec2002
Verification against rawinsonde data over RUC domainRMS vector difference (forecast vs. obs)
RUC is able to use recent obs to improve forecast skill down to 1-h projection for winds
1 3 6 912
Analysis~ ‘truth’
Retrospective runs – an excellent test bed for measuring the impact of
observing systems• All RUC data were
saved for two 10-day period
• Winter– 26 Nov - 5 Dec 2006
• Summer– 5-15 August 2007
WINTER
SUMMER
RH - national – 1000-400 hPa#1 obs type = Raobs#2 = GPS-PW
No-aircraft - controlNo-profiler - controlNo-VAD - controlNo-RAOB - controlNo-surface - controlNo-GPS-PW – controlNo-mesonet – controlNo-AMV - control
WINTER
SUMMER
Temp - national - 1000-100 hPaTie for #1 = Aircraft, RAOBsAircraft more at 3h, RAOB-12h
No-aircraft - controlNo-profiler - controlNo-VAD - controlNo-RAOB - controlNo-surface - controlNo-GPS-PW – controlNo-mesonet – controlNo-AMV - control
WINTER
SUMMER
Wind - national - 1000-100 hPa#1 = Aircraft#2 = RAOBs
No-aircraft - controlNo-profiler - controlNo-VAD - controlNo-RAOB - controlNo-surface - controlNo-GPS-PW – controlNo-mesonet – controlNo-AMV - control
Dave Turner, et al.: Observing System Simulation Experiment (OSSE) Study of Impact of
Lower Tropospheric Temperature, Moisture, and Winds
• Observing System Simulation Experiment (OSSE) of a single wintertime case of 4 observing systems:– Doppler Wind Lidar (DWL)– Microwave Radiometer (MWR)– Atmospheric Emitted Radiance Interferometer (AERI): infrared– Scanning Raman Lidar (SRM): a research-only system
• Synthetic ground-based remote sensors placed at each of the 140 existing WSR-88D radar sites (to minimize installation and operation costs) – not the 400 sites recommended by NAS report
• Used 18-km WRF model and DART DA system• Results limited to just this one case, and did not consider
wind profilers or other proven systems
OSSE Thermo/Wind Profiler Study Results
• The best analysis was achieved when both DWL wind observations and thermodynamic (temperature and moisture) profile observations from the SRM, AERI, and MWR were assimilated simultaneously
• Impact of these systems was limited to ~4 km (PBL)• Joint AERI+MWR approach recommended
– AERI provides needed vertical resolution, but MWR provides both all-weather operations and a reasonable “first guess” for AERI retrievals
• Assimilating thermodynamic data alone without DWL data did not produce strong enough moisture transport, thus failed to predict the heaviest precipitation
Turner et al. (2011)
4-km forecasts initialized using radar observations yield improved short-range forecasts of convective activity (Kain et al. 2010).
Particularly helpful for looking at convective mode and evolution.
Courtesy Jack Kain and Ming Xue
CAPS: Value of Radar Reflectivity and Radial Velocity Data Assimilation for QPF
© Patrick Marsh
7:21pm (0021Z)Lawton Tornado
Minco Tornado 10:54pm (0354Z)
Tornadoes of 8-9 May 2007 El Reno tornado
Union City tornado
Need to sample the PBL fully (75% lost by WSR-88D): Collaborative Adaptive Sensing of the Atmosphere (CASA) X-Band Radar Network
30 km range
EnKF analysis and ensemble forecasts for May 8-9 2007 tornadic mesoscale convective system (MCS)
• Experiment contained 40 Ensemble members. Reflectivity and radial velocity observations from 5 WSR-88D radars as well as the 4 CASA radars were assimilated every 5 minutes over a 1 hour window.
• Analysis and probabilistic ensemble forecasts were generated for three experiments to test effect of assimilated CASA data and use of a mixed-microphysics ensemble using three single moment ice microphysics schemes and 2-moment scheme.
0:00Z 0:30Z 1:00Z 1:30Z 2:00Z 2:30Z 3:00Z 3:30Z
1 hr. spin-up period
4:00Z
Deterministic forecast
Ensemble forecast
4:30Z 5:00Z
Reported tornadoes
Assimilation period
Snook, Jung and Xue 2011a,b. Putnam et al. (mostly CASA supported)
Effects of assimilated CASA data and mixed-microphysics ensemble
• Analyzed reflectivity fields using CASA and WSR-88D radar data (top left) compare well with radar observations (top right); reflectivity structure of the main convective line is well-captured.
• Inclusion of CASA data improves representation of a low-level mesoscale vortex and gust front observed by CASA and WSR-88D.
Final analysis (0200 UTC ) ReflectivityCASA + WSR-88DEnKFComposite Radar Reflectivity Analysis
WSR-88DObservedComposite Radar Reflectivity
Near-surface winds and potential temperature (0140 UTC)
CASA + WSR-88D WSR-88D Only
KCYR Vr – 0141 UTC
Forecast of Minco Mesovortex at 400m resolution
•
Radial velocity at 0340 UTC
OKC TDWR obs Predicted Vr with CASA Predicted Vr with 88D only
Schenkman et al. (2011b MWR)
Hybrid Ensemble Kalman Filter (EnKF) – 4DVAR Data Assimilation
• Xue et al. (2006) and Yussouf and Stensrud (2010) demonstrated the benefit of rapid scan radar data assimilated via EnKF
• But, high-frequency EnKF Data Assimilation is costly• Data I/O can cost 80% of total CPU to read & write ensembles
• 4D extension of EnKF requires fewer cycles while still using observations at their correct times
• Asynchronous EnKF can achieve this• Hybrid 4D Variational and Ensemble Data Assimilation
allows VAR methods to use flow-dependent background error covariance and dynamical constraints
Sampling the nocturnal stable boundary layer(Bob Banta, NOAA/CSD)
• Must have good enough Δz throughout SBL to define its structure, and to determine depth, strength, and rate of growth of SBL
• SBL depth h, a fundamental quantity – difficult to measure– Problem – coarse vertical resolution, precision of
available measurements– POSTER [Banta, Pichugina, et al]: hi-res wind profile
data from Doppler lidar able to address this issue, significantly reduce uncertainty in h estimates
– Use of velocity variance profiles to estimate SBL instead of more traditional aerosol concentration profiles
Time-height cross sections of measured HRDL mean wind U(z) and turbulence σu
2(z) profile data @ 1 min
LLJ structure to winds (* = max), symbols indicate top of SBL using several indicators
NOTE:Consistency among diagnostics, continuity in time ( confidence in estimates of h)
Examples of hourly profiles of LLJ development for one night using NOAA High-Resolution Doppler Lidar (HRDL)
HRDL measurements of thenocturnal Stable Boundary Layer
Thermodynamic retrieval from HRDL wind measurements of the nocturnal SBL
• Techniques to retrieve 3D wind and thermodynamic fields from scanning Doppler lidar are similar to those developed for Doppler radar (e.g., Sun and Crook (1997, 1998).
• 4DVAR technique is used to fit the output of a prognostic model (dry, shallow Boussinesq) to the lidar measurements, which requires development of an adjoint model to compute the gradient of the cost function with respect to the initial state of the forward model.
• Because of poor temporal sampling of eddies, and the disparity with excellent spatial sampling by HRDL, the measured values are not interpolated to the model grid to avoid smoothing.
• Retrievals are quite sensitive to changes in the gradients of the base-state virtual temperature profile
Newson and Banta, 2004a,b JAOS
Evolution of a density current into a bore
Evolution of an undular bore from an advancing density current. Fluid enveloping the current is similar to what happens to warm air underneath a low-level inversion
as a density current intrudes into the stable layer.
Inversion surface
Density currentBore
Evolution of a bore into a soliton
Amplitude-ordered solitary waves
Lidar-Radar Analyses of Convection Initiationby Gravity Currents, Bores, and Solitons
Two-dimensional circulation system relative to the bore and gravity current-like cold front derived from 915 MHz wind profiler shows need for dual lifting to initiate convection
Max = 1 m s-1
Bore lifting
Gravity Current lifting
Koch and Clark, 1999 JAS
Lifting by bore and gravity current-like cold front destabilizes and moistens sounding: strong convection was initiated
Max displacement = 1.25 km
Lifting depth = 2.5 km
(bore + front combined)
Koch and Clark, 1999 JAS
AERI time-height displays show sudden and deep moistening and adiabatic cooling aloft following bore passages
A B
Watervapor
Potentialtemperature
Koch et al. 2008 MWR
Some Issues for this Workshop
• The NAS report states that the boundary layer is critically under-observed, but how strong is the scientific support for this assertion, and what cost-effective technology solutions are available to fill this apparent gap?
• Existing studies that have systematically studied the relative impacts of various observing systems are incomplete – in terms of seasonal coverage, phenomena predicted, number and type of observing systems, and DA technology. What specific recommendations can be made to address this?
• What techniques are available to estimate impact of changes to both current observing system configurations and future combinations of observing systems?