Model Parameterizations: Issues important for heavy rainfall forecasting Mike Baldwin NSSL SPC CIMMS
Model Parameterizations: Issues important for heavy rainfall forecasting
Mike Baldwin
NSSL SPC CIMMS
Skill of current operational models
• To put it kindly…
• Current NCEP models do a poor job forecasting heavy rainfall
• AVN (red), Eta (green), NGM (blue) for Jan-Sept 2000
Ingredients needed by a model in order to predict some phenomena of interest
• Adequate grid spacing– To be able to resolve the feature
• Physical processes– All those important in the development,
maintenance, and decay of the feature
• Dynamics– Accuracy, hydrostatic/non-hydrostatic
• Adequate initial/boundary conditions– To be able to capture important forcing
Ask yourself:• “Does the model I’m using have the necessary
ingredients to predict the feature(s) that I’m considering or expecting?”
YES: model guidance taken literally can be useful
NO: model by itself is of little value (but not worthless)
Either way: knowledge of model characteristics will increase the value of NWP guidance
What are Model Parameterizations?
• Techniques used in NWP to predict the collective effects of physical processes which cannot be explicitly resolved
• Sub-grid scale or perhaps near-grid scale processes:
• For example; cloud physics, convection, turbulent mixing, radiation, surface exchanges
Interaction between different processes is critical
• Especially for mesoscale models• Not only important to do a good job with
a specific physical process• All pieces have to work well together in
order for model to perform well• Several studies have shown great
forecast sensitivity to subtle changes to a parameterization
Outline
• Quick overview of convective parameterizations currently available in NWP models (both operational and research)
• Look at some research/case studies of NWP performance in heavy rain events
• Talk about future
Current EMC models use different approaches
• RUC II: Grell scheme
• Eta: Betts-Miller-Janjic (Kain-Fritsch used experimentally at NSSL & SPC)
• MRF/AVN: Grell-Pan scheme– Grell, Grell-Pan, and Kain-Fritsch schemes are Mass-Flux schemes, meaning they use simple cloud models to simulate rearrangements of mass in a vertical column
– Betts-Miller-Janjic adjusts to “mean post-convective profiles” based on observational studies
“Mass-flux” parameterization
MM5 Model Physics Options• Precipitation physics
– Cumulus parameterization schemes: • Anthes-Kuo • Grell • Kain-Fritsch • Fritsch-Chappell• Betts-Miller • Arakawa-Schubert
– Resolvable-scale microphysics schemes:• Removal of supersaturation• Hsie's warm rain scheme• Dudhia's simple ice scheme• Reisner's mixed-phase scheme • Reisner's mixed-phase scheme with graupel• NASA/Goddard microphysics with hail/graupel • Schultz mixed-phase scheme with graupel
MM5 Model Physics Options• Planetary boundary layer parameterization
• Bulk formula • Blackadar scheme • Burk-Thompson (Mellor-Yamada 1.5-order/level-2.5 scheme) • Eta scheme (Janjic, 1990, 1994) • MRF scheme (Hong and Pan 1996)• Gayno-Seaman scheme (Gayno 1994)
• Surface layer process parameterization• fluxes of momentum, sensible and latent heat• ground temperature prediction using energy balance equation • variable land use catagories (defaults are 13, 16 and 24)• 5-layer soil model• OSU land-surface model (V3 only)
MM5 Model Physics Options
• Atmospheric radiation schemes:• Simple cooling • Dudhia's long- and short-wave radiation scheme • NCAR/CCM2 radiation scheme • RRTM long-wave radiation scheme (Mlawer et al., 1997)
(copied from MM5 web page: www.mmm.ucar.edu/mm5/)
What do all convective parameterizations do?• Predict convective precipitation• Feedback onto larger scales the effects of
transports, mixing, circulations, etc. found within convective elements
• Change vertical stability• Redistribute and generate heat• Redistribute and remove moisture• Make clouds that affect surface heating and
atmospheric radiation
How do convective schemes accomplish these tasks?
• convective triggering (yes/no)
• convective intensity (how much rain?)
• vertical distribution of heating
• vertical distribution of drying
Triggering/activating
• CAPE (ALL)
• mass or moisture convergence exceeding a certain threshold (Kuo)
• positive destabilization rate (Grell)
• perturbed parcels can reach their level of free convection (KF)
• sufficient cloud layer moisture (BMJ)
Convective intensity
• proportional to mass or moisture convergence (Kuo)
• sufficient to offset large-scale destabilization rate (Grell)
• sufficient to eliminate CAPE, constrained by available moisture (KF)
• proportional to cloud layer moisture (BMJ)
Vertical distribution of heating and drying
• determined by adjusting to empirical reference profiles (BMJ, Kuo)
• estimated using a simple 1-D cloud model to satisfy the constraints on intensity (Grell, KF)
Different ways to classify convective schemes
• Molinari and Dudek (1992)• Traditional
• clear separation between convective and stratiform or grid-scale precipitation
• Hybrid• direct interaction between convective and grid-scale physics
• Fully Explicit• grid-scale cloud and precipitation physics ONLY
Fully explicit
??? Hybrid Traditional
0.1 6050403020101Grid Spacing (km)
Example: Betts-Miller-Janjic (BMJ) scheme – Eta Model
• BMJ scheme requires some CAPE• “Equilibrium-type” scheme• Deep and shallow components
– Deep = precipitating– Shallow = non-precipitating
• Critical factor in determining yes/no/amount of precipitation is the cloud layer moisture
A few quick points…
• Deep convection is given first priority
• Deep convection will fail if – cloud layer is too dry – cloud is too shallow
• Scheme defers to shallow convection if deep convection fails
• No feedback of cloud water/ice
Deep convection example
– KSBN 18h forecast from 00Z 31 May 2000 Eta run
Cloud depth
Feedback from BMJ scheme
Td
Tsounding 3h earlier
What does BMJ deep convection do?
• Stabilize the cloud layer
• Typically heats mid/upper cloud
• Dries lower part of cloud
• Does not modify the sub-cloud layer
• Feedback reduces CAPE and precipitable water
Shallow convection example
– KOKC 4h forecast from 12Z 1 Jun 2000 Eta run
Cloud depth
Feedback from BMJ scheme
T Td
sounding 4h earlier
What does BMJ shallow convection do?
• Mixes moisture up from cloud base to cloud top
• Mixes heat down from cloud top to cloud base (destabilizes the cloud layer)
• Location of cloud base & top critical for determining impact on forecast fields
• Could affect lapse rates, cap strength• Does not affect precipitable water
How to recognize BMJ shallow convection
• No convective precipitation• Forecast sounding has a smoothly
varying moisture profile up to ~200mb deep (usually concave shape)
• “Straight-line” temperature profile over the same layer, just above LCL
• Base of unusual mid-tropospheric inversion indicates cloud top
Gallus (1999) Weather and Forecasting p. 405-426
• “Eta simulations of three extreme precipitation events: Sensitivity to resolution and convective parameterization”
• Ran Eta Model at four different horizontal resolutions (78, 39, 22, and 12km) and with two convective schemes (BMJ vs. KF)
• Variations in precipitation forecasts were found to be highly case dependent
Gallus (1999) Figure 1a
• 16-17 Jun 1996• MCS formed over
central IA in warm sector ahead of sfc low
• Heavy rains also occurred north of warm front in WI
Gallus (1999) Figure 3
Gallus (1999) Figure 10
• BMJ runs
• 78, 39, 22, and 12km res
• Contours at 5mm, then every 25mm
Gallus (1999) Figure 13
• KF runs
• 78, 39, 22, and 12 km res
• Contours at 5mm, then every 25mm
Items to note for this case
• BMJ runs DO NOT produce higher precip amounts as resolution increases
• KF runs DO produce higher precip amounts as resolution increases
• BMJ produced a broad area of precipitation that covered observed region for many hours
• Peak amounts in high-res KF runs produced mainly by grid-scale precipitation scheme
Gallus (1999) Figure 1b
• 16-17 Jul 1996
• MCS/MCC developed north of warm front
• Training cells found in region of peak rain
Gallus (1999) Figure 6
Gallus (1999) Figure 7
• Omaha sounding 00 UTC 17 Jul 96
• 2000 J/kg CAPE, less than 25 J/kg CIN (above inversion)
• Example BMJ reference profiles (dashed)
Gallus (1999) Figure 15
• BMJ runs• Peak ppt
decreases as resolution increases
• Areal coverage increases with resolution
Gallus (1999) Figure 16
• KF runs
• Location errors are large, too far to the north
• Peak amounts are more reasonable and increase with resolution
Gallus (1999) Figure 1c
• 27 May 1997• Jarrell TX tornado
outbreak• Supercell-type
heavy rain event• Boundary
interaction important
Gallus (1999) Figure 8
Gallus (1999) Figure 9
• Estimated sounding near Jarrell at 18 UTC 27 May 97
• CAPE near 5000 J/kg, no CIN
Gallus (1999) Figure 19
• BMJ runs• Peak ppt
increases with resolution
• Heaviest ppt produced mainly by grid-scale scheme in high-res runs
Gallus (1999) Figure 21
• KF runs
• Peak ppt increases slightly with resolution
Gallus (1999) Figure 20
• BMJ moisture divergence at 15 UTC
• Outflow-type circulation initiated in NW Texas
• Moved to the southwest
Lessons learned…
• This study shows great sensitivity of the QPF to different convective schemes and horizontal resolutions
• No consistent behavior by either scheme from case to case
• Should expect great difficulty in developing a model to predict heavy rainfall with accuracy and consistency
Sensitivity, continued…
• Not only are models sensitive to different parameterizations
• Changes to a single parameterization can also produce significant differences in model QPF
• Spencer and Stensrud (1998) MWR for example
Future: Eta Model
• 10km Eta is coming in 2001
• should include precipitation data in data assimilation (initial conditions)– helps mainly during early part of forecast
• no major changes to model parameterizations are expected for 10km Eta implementation
Future: short-range ensemble forecasting
• varying both initial conditions and model configuration
• goal: to predict the range of possible scenarios and forecast uncertainty
• problems: – low correlation between spread and errors– difficult to produce much spread among
ensemble members
Stensrud et al (2000) MWR
• Found that varying initial conditions produced the “best” ensemble when the large-scale forcing was strong
• When the large-scale forcing was weak, varying model physics produced the “best” ensemble
MM5 ensemble (NSSL) for 3 May 1999 case (24h fcst)
• 3 different convective schemes (KF, Grell, BMJ)
• 2 different PBL schemes (BL, BT)
MM5 ensemble for 3 May 1999 case
MM5 ensemble (NSSL) for 3 May 1999 case (30h fcst)
Future: local modeling
– workstation Eta (Bob Rozumalski)• easy to set up and run• can run at high resolution in near real-time on a
cheap Linux box
– WRF (Weather Research and Forecasting) model
• next-generation NWP modeling system for research AND operations
• large development effort• test versions currently available
Future: neural networks
• what is a neural network?
• model that “learns” the relationship between input (observations, NWP output) and output (QPF)
• Hall et al (1999) for example show incredible results for local QPF (DFW)