1. Jimy Dudhia NCAR MWFR-WG
Major Physical Processes that Influence the Forecast Accuracy for the Very Short Term
Jimy Dudhia
NCAR
MWFR-WG

2. Use of NWP Models in Nowcasting
Cloud-resolving models are increasingly being used to extend nowcasting from 0-1 hrs out to 6 hours
Initialization techniques are specialized for this application to minimize spin-up, e.g.,
Rapid-update cycling (minutes to one hour cycle)
Warm or hot start for convection
Cloud analysis
Dynamical spin-up (e.g. via digital filter)

3. NWP Models and Nowcasting
Linear development is well handled by non-modeling nowcasting techniques
Diurnal cycle
Advection-dominated processes
Linearity breaks down for a variety of reasons
Convective triggering and development
Surface-gradient effects (sea-breezes, topography)
Cloud-cover changes and gradients
Value of NWP models potentially increases as development becomes more nonlinear

4. Outline Physical processes associated with
Convection and Heavy Precipitation
Surface and Topography
Cloud Cover

5. Convection Probably the highest priority in improving nowcasting
Convective systems may be self-organized (squall lines or supercells) or associated with fronts or tropical storms
Nowcasting involves individual storms but deterministicness is limited by cloud life cycle and generation of new clouds often requiring merging to probabilistic methods by 6 hours

6. Convection
Supercell storms have a coherent dynamical structure – major challenge for initialization

7. Convection Prediction
Supercells are associated with the most severe weather (hail, wind, tornadoes, lightning)
Predicting exact formation location often depends on surface heterogeneities (see next topic) or pre-existing storm outflows
Storm motion vector is not with mean wind but is well understood (e.g. right-movers)
Propagation of existing storms is quite predictable with nowcasting techniques

8. Convection Modeling
Cloud-resolving models (1-3 km) capture supercell structures and motions realistically

9. Convection Simulation
However, success in cloud modeling is not sufficient to assume that such models will forecast convective systems well
Depends on two major challenges
Initialization of the correct structure at time zero via data assimilation (far from a solved problem) imbedded in correct large-scale sounding (CAPE, shear)
Even with perfect initial conditions, physics parameterizations especially microphysics are not (and cannot be) perfect within the bulk parameterization framework

10. Microphysics Parameterization

11. Microphysics Parameterization
Bulk schemes
Limited mass variables
Cloud water
Rain
Snow
Ice crystals
Graupel/hail
Some also predict N
Double moment

12. Microphysics Parameterization
Because of feedback to the dynamics via latent heating and drag effects, storm simulations require accurate
precipitation production mechanisms
particle sizes and types (fall-speeds)
evaporation of precipitation
Studies have shown large sensitivity of storm motion and development to variations in parameters
Second-generation storms developing on pre-existing storm outflows will have even greater uncertainty
However, models show some skill in storm type (multicells versus supercells) and consequent propagation direction

13. Microphysics Parameterization
Uncertainties in all aspects
Ice nucleation (dust aerosols)
Rain-formation (cloud condensation nuclei)
Size distributions have to be assumed, affecting
microphysical growth/evaporation rates
fall speeds
collision rates and efficiencies
These uncertainties will remain in any foreseeable microphysics scheme that is to be run in real-time cloud-resolving models
Tuning can mitigate errors for specific purposes (e.g. supercells in midwest) but aerosol amounts vary day-to-day adding inherent uncertainty in the microphysics parameters

14. Microphysics
Parameterization also may subtly affect location and consequent prediction of other high-impact events such as
Flash floods requiring pin-pointing convection’s effect on specific rivers
Heavy local rainfall or snowfall depending on slow-moving systems or multiple systems affecting one location

15. Surface and Topography
Several nowcasting problems relate to the surface
Convective initiation
Sea-breeze development
Wind-storms related to orography
Flash floods from watershed basins
Wildland fire forecasting

16. Surface and Topography
There is potential for models to add value in complex interactions between the large scale flow and
Coastal gradients
Mountains
Also in weak-forcing situations
Convection often evolves from local heterogeneities in fairly flat terrain
Mountain-valley circulations dominate in complex terrain

17. Surface Physics Primary outputs of surface physics are
Sensible heat flux
Latent heat flux
Surface drag
Necessary resolved inputs at initial time of forecast
Surface vegetation characteristics
Vegetation fraction, height, evapotranspiration properties
Soil moisture and temperature
Affected by prior rainfall and soil characteristics
Snow cover

18. Typical Land Surface Model

19. Surface Physics
Uncertainties come from both initial state and parameterization of vegetation effects and soil water and heat budgets
Initial state
Example, soil moisture affects partitioning of heating into sensible and latent heat fluxes which
impacts PBL growth and development of convection
possibly leads to gradients due to heterogeneity in vegetation or soil moisture

20. Surface Physics
Even with corrected initial state using a land data assimilation procedure, the physics has uncertainties in
Vegetation and soil parameters
Albedo, root zone, soil texture
Treatment of snow cover in forests or urban areas
Coupling strength of land to atmosphere
Exchange coefficients, roughness lengths, stability functions

21. Surface Physics Effects of uncertainties
Impact on convective triggering location and timing in weak forcing
Bias in diurnal cycle magnitude

22. Complex Terrain

23. Complex Terrain
Complex terrain often challenges model resolution, weakening or not resolving drainage flows in valleys, for example
Additional heterogeneity effects due to solar radiation on slopes may be included in model, but only for well resolved slopes
Drag effects due to unresolved terrain may not be accounted for leading to wind bias

24. Cloud Cover
Initialization and development of the non-convective cloud field is a challenge for nowcasting and short-range prediction
Initialization of clouds is tied to the data assimilation but also relies on consistency with the microphysics and dynamics to maintain them

25. Cloud Cover

26. Cloud Cover Four types of error in short-range forecasting
Maintain initial clouds too long
Dissipate initial clouds too soon
Fail to form clouds during forecast
Form spurious clouds during forecast
All may be spatially dependent errors and affect the surface temperature and visibility and ceiling forecasts
Related problem is the effect of dust and other aerosols on radiation

27. Cloud Cover
Non-convective clouds interact directly with the radiation in the model having both local and nonlocal (shading) effects
Cloud-radiation interaction depends on how radiation physics interacts with microphysics information
Uncertainties come from assumed ice or water particle properties to provide optical depth, etc.
The radiative heating may directly impact cloud lifetime (dissipation or maintenance)

28. Cloud Cover
Short-range prediction of cloud cover is not often verified in models
Skill depends heavily on initial analysis (e.g. relative humidity layer structure) and on model’s ability to vertically resolve thin cloud layers
Cloud-resolving models do not often use cloud-fraction concepts which may be a drawback for representing fine-scale clouds such as cumulus and their radiative impact

29. Concluding Remarks 1
Physical processes that affect short-range forecasts
Convective storms sensitive to microphysics in models and to land-surface for triggering in weakly forced conditions
Surface properties in land-surface model affect evolution of boundary layer
Orographic effects may not be completely resolved
Large-scale cloud prediction relies on initialization of humidity and cloud-radiation interaction

30. Concluding Remarks 2
Physics is limited by its simplicity in real-time forecast models making specialized tuned physics necessary for certain applications/seasons/regions
Spin-up issues occur where physics interacts with dynamics
Convective initialization
Land-surface initialization
Orographic flow initialization
Cloud initialization
These can be addressed by using cycled forecasts or prior initialization and/or digital filtering in combination with data assimilation
Models may help in nonlinearly developing situations but these are often not deterministically predictable and so ensemble methods should be considered to provide probabilities