2016 L09 MEA716 2 11 PBL5 - Nc State University...Thu 2/11/2016 • Finish turbulence and PBL...
Transcript of 2016 L09 MEA716 2 11 PBL5 - Nc State University...Thu 2/11/2016 • Finish turbulence and PBL...
Thu 2/11/2016• Finish turbulence and PBL closure:
• WRF PBL options• Diffusion and scale issues• Paper presentations (Keith, Laura, Lindsay, Hans)
Reminders/announcements:- PBL paper discussion for today & Tuesday 2/16- Midterm Thu 3/3- Project hypothesis assignment, due (presented) Tue 3/15
- Added a short “progress report”, due on 2/25, to allow feedback
Winter Storm Forecasts at hour 42
RUC, YSU, KF(Lindsay)
Noah, TEMF, KF(James)
Noah-MP, MYJ, BMJ(Keith)
Interesting differences just by altering LSM, PBL/sfc layer
Several only ran postprocessor out 12 hours – storm not yet interesting!
Some didn’t postprocess – please finish (through 84 hours)
All had complete run, some problematic (QNSE PBL not working well)
Micrometeorology and Turbulence Parameterization
A wide variety of PBL schemes are available in WRF
Some are designed for specific situations/phenomena
Some are designed to run in conjunction with a separate shallow mixing scheme (to handle entrainment), others are not
Consider outputting and examining PBL tendencies for various fields in order to assess scheme impact on model atmosphere
For very high resolution, consider diffusion scheme carefully, and consider Shin-Hong scale-aware PBL option (more on this today)
Re-Cap from Tuesday
Outline1.) Review of turbulence and properties
- Characteristics, worksheet
- Definitions, TKE, introduction to closure problem
- Tendencies, and flux divergence
2.) Closure strategies- Bulk aerodynamic
- K-theory (mixing length)
- Local and non-local closures
- WRF schemes
- Scale issues, diffusion
Conclude with presentation/discussion of journal papers describing schemes
WRF PBL Options (partially from Dudhia)bl_pbl Scheme Sfc layer Characteristics Design Cloud mixing
1 YSU 1 Explicit entrainment, first order Local + non-local Qc, Qi
2 MYJ 2 TKE scheme Local, 1.5 order Qc, Qi
4 QNSE 4 TKE, a spectral scheme (quasi-normal scale elimination)
Local, 1.5 order Qc, Qi
5 MYNN2 1,2,5 Improves MY length scale, adds buoyancy effects
Local, 1.5 order Qc
6 MYNN3 1,2,5 Higher order version of MYNN2 Local, 2nd order Qc
7 ACM2 1,7 Combines non-local, eddy diff., asymmetric mixing
Local + non-local Qc, Qi
8 BouLac 1,2 TKE similar to MYJ, Tested for orographic turbulence
Local, 1.5 order Qc
9 UW 9 TKE scheme, for CAM, explicit entrainment
Local, 1.5 order Qc, Qi (?)
10 TEMF 10 Explicit shallow cumulus, considers total turb. Energy
Local + non-local Qc, Qi
11 Shin-Hong 1 + others?
Scale-aware non-local PBL scheme for “gray zone” runs
Non-Local Qc, Qi
12 GBM 9 With entrainment, for coarse vert. resolution (GCM)
Local, 1.5 order Qc, Qi
99 MRF 1 Older version, YSU updates Local + non-local QC, QI
SCM PBL comparisons: Hot August day in NC
Run WRF SCM:CompareRTHBLTENPBLH
Observed PBLH ~ 2.5 kmCooling above
Warming below
SCM PBLH comparisons: Hot August day in NC
YSU
MYJ
YSU
SCM RTHBLTEN comparisons: Hour 11
YSU MYJ Shin-HongQNSE
SCM RTHBLTEN comparisons: Hour 11
BouLac MYJ MYNN3MYNN2.5
SCM RTHBLTEN comparisons: Hour 11
TEMF MYJ Grenier-Bretherton
UW
Compared one time, during development of convective PBL on a hot August day in NC
Remarkable agreement in magnitude of positive potential temperature tendency within PBL (coupled)
Less agreement in depth of tendency, and strength and altitude of cooling tendency near PBL top
Outliers were TEMF, BouLac, and MYNN3
Comparison far from comprehensive; see added PBL comparison papers on www page (e.g., Cohen et al. 2015)
PBL SCM test summary
Generally accepted that higher order closures are better:
- Even if crude closure, added information in non-parameterized equations improves representation
- Actual observations of higher-order turbulent moments (e.g., 4th order) hardly exist, so limited basis for testing such closures
Stull test: “Generally, the higher-order local closures and non-local closures yield more accurate solutions than lower order, but they do so at added expense…”
Available parameterizations generally work well in situations for which they were designed
Additional PBL Considerations
Recall that surface layer in WRF is the first model layer; this forces consideration of vertical level distribution
With YSU, ACM2, GFS and MRF PBL schemes, lowest full level should be .99 or .995 (not too close to 1)
TKE schemes (e.g., MYJ) can tolerate thinner surface layers
Tailor your selection to the research problem at hand; e.g., hurricane modeling, very specific choices to make:
isftcflx = 1, 2 (account for altered exchange coefficients with strong winds over water)
See Braun and Tao (2000), Hill and Lackmann (2009)
Run with OML or sf_ocean_physics, or coupled model?
Additional PBL Considerations
1 10 100 km
Convective ParameterizationExplicit Convection
LES PBL Parameterization
Two Stream Radiation3-D Radiation
Physics, Resolution, & Parameterization (modified from Jimy Dudhia, NCAR)
Physics “No Man’s Land”
= model grid length
“No Man’s Land” a.k.a. “terra incognita”, “gray zone”…Be careful when running models with these grid lengths
At what grid lengths are parameterizations designed to operate?
PBL scheme assumes turbulent eddies are not resolved; at grid size dx << 1 km, this assumption begins to break down
What factors determine when one enters “terra incognita”?
3-D Diffusion can be used instead of PBL (WRFV3 onward); coupled to surface physics (surface fluxes)
Can also consider new Shin and Hong (2015) scale-aware option for PBL at high resolution; see Honnert et al. (2011, JAS)
Honnert et al. introduce scaling parameter, , ratio of grid length to PBL & cloud layer heights
Additional PBL Considerations
( )c
xh h
Honnert et al. (2011, JAS):
Full range of grid lengths, determined scale at which resolved contribution equaled parameterized
For TKE, = .2 (for 1 km PBLH, 200 m grid length)
For , = .4 (for 1 km PBLH, 400 m grid length)
Additional PBL Considerations
( )c
xh h
( )c
xh h
2
Honnert et al. (2011, JAS), Fig.1: Top row , 2nd is w
Honnert et al. (2011, JAS):
Additional PBL Considerations
resolved
subgrid
High-resolution model runs and the “Gray Zone”
High-resolution model runs and the “Gray Zone”
Slide from Shin WRF workshophttp://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/8.2.pdf
High-resolution model runs and the “Gray Zone”
Slide from Shin WRF workshophttp://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/8.2.pdf
High-resolution model runs and the “Gray Zone”
Slide from Shin WRF workshophttp://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/8.2.pdf
High-resolution model runs and the “Gray Zone”
Slide from Shin WRF workshophttp://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/8.2.pdf
High-resolution model runs and the “Gray Zone”
Slide from Shin WRF workshophttp://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/8.2.pdf
High-resolution model runs and the “Gray Zone”
Slide from Shin WRF workshophttp://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/8.2.pdf
High-resolution model runs and the “Gray Zone”
Slide from Shin WRF workshophttp://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/8.2.pdf
High-resolution model runs and the “Gray Zone”
Slide from Shin WRF workshophttp://www2.mmm.ucar.edu/wrf/users/workshops/WS2014/ppts/8.2.pdf
BAMS February 2011 Issue
bl_pbl_physics=0
Turns off PBL scheme, select diff_opt and km_opt choices to represent turbulence
As discussed, at grid length < 1 km, begin to resolve largest turbulent eddies, must consider diffusion schemes
Separate from PBL and surface layer parameterizations, models include sub-grid scale diffusive effects:
&dynamicsw_damping = 1,diff_opt = 1,km_opt = 4,diff_6th_opt = 0, 0, 0,diff_6th_factor = 0.12, 0.12, 0.12,damp_opt = 0,zdamp = 5000., 5000., 5000.,dampcoef = 0.2, 0.2, 0.2khdif = 0, 0, 0,kvdif = 0, 0, 0,
Additional Considerations
Diffusion & Turbulence without PBL scheme
Can include sub-grid scale mixing effects for scalar variables of form
Diff_opt=1 gives 2nd order diffusion on model coordinate surfaces
Either computes constant K value, or use with PBL
km_opt is used to select K method: 1 for constant, 2 for Smagorinsky (related to deformation of horizontal wind)
z
Kzy
Kyx
Kx vhh
Diffusion in WRF (Dudhia)
mixing
diff_opt=1Horizontal diffusion along model levelsOnly neighboring points on same model
level involved
diff_opt=2Horizontal diffusion acts along model levels, but numerical method includes vertical correction term using more grid points
Km_opt:1: use khdif and kvdif (const)2: 1.5 order TKE prediction3: Smagorinsky – deform/stab K4: 2-D Smagorinsky
For diff_opt = 2,
mix_full_fields = .true. (rather than perturbation fields)Recommended, but default is .false. (base-state subtracted before mixing)
For very high resolution:
mix_isotropic = 1 (same length scale for vertical, horizontal diffusion)
For large-eddy simulation (real data): diff_opt = 2, bl_pbl_physics = 0
isfflx = 0 (idealized heat, momentum flux, set in namelist)isfflx = 1 (from physics, sf_sfclay_physics = 1, set sf_surface_physics)isfflx = 2 (momentum only from physics, heat from tke_heat_flux in NL)
still need sf_sfclay_physics = 1km_opt = 2 or 3mix_isotropic = 1 (if vertical, horizontal grid length comparable)
Diffusion in WRF (Dudhia)
Real-data case, PBL physics on:Best is diff_opt=1, km_opt=4 (defaults)Complements vertical diffusion from PBL scheme
High-resolution real-data cases (~100 m grid)bl_pbl_physics = 0diff_opt=2; km_opt=2, or 3 (TKE or Smagorinsky scheme)
Idealized cloud-resolving modeling (smooth or no topography)bl_pbl_physics = 0diff_opt=2; km_opt=2,3
Complex topography with no PBL schemediff_opt=2 is more accurate for sloped coordinate surfaces, and prevents diffusion up/down valley sides
Note: WRF can run with no diffusion (diff_opt=0)
Diffusion in WRF (Dudhia) - recommendations
Turbulence/Surface Representation in Models
– Sub- & surface processes: Land-Surface Model, LSM
– Near-surface processes: Surface layer scheme
– Boundary layer processes: PBL scheme
– Boundary-layer entrainment: PBL, shallow cu, cu scheme
– Clear air turbulence: PBL and diffusion schemes
– Shallow moist convection: PBL, shallow cu, cu schemes
Potential for overlap exists
Be aware of this potential in namelist selections!
• Handled by either (i) convection scheme (e.g., BMJ), (ii) PBL scheme (e.g., YSU), or (iii) in separate routine; can be all of above if not careful!
• If running without convective parameterization, use PBL with entrainment, or run shallow convection scheme
• Our next topic is convective parameterization, so we will spend more time on representation of shallow and deep convection soon
Shallow (non-precipitating) Convection
https://www.youtube.com/watch?v=HTMsWDQwubA
Alexandar A. Baklanov, Branko Grisogono, Robert Bornstein, Larry Mahrt, Sergej S. Zilitinkevich, Peter Taylor, Soren E. Larsen, Mathias W. Rotach, and H. J. S. Fernando
Objective: Review challenges facing PBL schemes
Unknown physical
origins of PBL flows
under weak large-scale
flow
Poor spatial coverage of
observations
TKE often the only prognostic
variable (TPE overlooked)
Unrealistic turbulence decay rates
New theory predicts Ri-
dependency of turbulent Prandtl number
Not all data agrees on its behavior
Observationsinsufficient
PBL affected by surface roughness/heterogeneity,
thermally-driven flows, and combinations of the two
Mean flows vs. turbulence and parameterization of
surface fluxes key problems
Exchange between atmosphere/water
Ice/sea mix
Monin-Obukhov Similarity Theory
(constant flux surface layer)
invalid in urban canopy
Instrumentation networks expensive,
observations inadequate
Microscale effects not fully understood, let alone parameterized
“Further advancements in [the development of high-resolution models] are stymied,
however, as long as … PBL schemes remain uncertain.”
Little work done over heterogeneous surfaces
Horizontal grid on
order of m necessary
Nature and theory ofturbulent boundary layer
structure and flows
Turbulence closure problem
Stability dependence ofthe Prandtl and critical Richardson numbers
Airflows within and above urban and other
complex canopies
Air-sea-ice interactions
Improvement of PBLschemes in operational
and environmentalsecurity models
Broader Impacts: Recommendations for future investigation
• Need extensive PBL datasets• Analyze data from different sites uniformly• Database center that accommodates
upgrades
• Experiments, numerical simulations, and data analyses to determine dependence ofPrT on Ri
• Utilize TTE (turbulent total energy) concept to eliminate Ri correction coefficients and explicit use of the critical Ri
• Use LES/DNS to analyze idealized complex terrain and urban flows• Consolidate PBL field study data to “develop a unifying theory of mesoscale PBL
flows over complex terrain and to develop PBL schemes necessary for accurate high-resolution operational modeling.”
• Long-term urban testbeds in varying regions, climates, etc.• Develop more accurate mesoscale/microscale coupled models
• New observations and models are necessary to determine marine PBL features
• Flows near land/sea and ice/sea boundaries also need to be studied• Determine at what wind speeds sea spray becomes important
• Turbulence closure, mesoscale convection, and representation of complex terrain/rough surfaces need to be improved
• Use advancing techniques to better model, monitor, and forecast PBL processes affecting toxic plumes
• Develop/refine theory for sub-mesoscale processes
“It is important not to simply adjust existing schemes, but to
develop new ones based on sound PBL physics.”
Nature and theory ofturbulent boundary layer
structure and flows
Turbulence closure problem
Stability dependence ofthe Prandtl and critical Richardson numbers
Airflows within and above urban and other
complex canopies
Air-sea-ice interactions
Improvement of PBLschemes in operational
and environmental security models
Sensitivity of High-Resolution Simulations of Hurricane Bob (1991) to Planetary Boundary Layer Parameterizations
Scott A. Braun and Wei-Kuo TaoMonthly Weather Review, December 2000
Methods/Justification• Assumptions about boundary
layer processes are important for maintenance and evolution of hurricanes
• Sensitivity of high-resolution models to parameterization of PBL processes?
• MM5 V2.5
• 36-km grid, 12-km grid, 4-km grid
• 9 simulations
• Compared to available observations
Schemes Tested• Burk-Thompson PBL scheme• Bulk-aerodynamic PBL scheme• Blackadar PBL scheme• MRF PBL scheme• Blackadar vertical mixing, B-T surface fluxes• Blackadar vertical mixing, bulk surface fluxes, no
wind speed dependence of z0• Blackadar vertical mixing, bulk surface fluxes, wind
speed dependence of z0• Bulk vertical mixing, B-T surface fluxes• MRF vertical mixing, B-T surface fluxes
Results• Burk-Thompson and bulk aerodynamic =
strongest storms, MRF = weakest storm
• MRF had low-level inflow deeper and weaker, strong outflow from eye above PBL was absent, PBL much drier, cloud base much higher—MRF has excessively deep mixing
• Key difference between B-T and Blackadar are surface flux algorithms
Results• Intensity increases as ratio of exchange coefficients
for enthalpy and momentum increase• Blackadar and MRF schemes (low ratio) have
weaker storms, B-T and bulk (higher ratio) have stronger storms
• Hard to determine individual roles of surface fluxes and vertical mixing
• Only change surface flux = higher ratio, higher intensity (if roughness increases with wind speed), but pressure and wind speed don’t vary as expected, probably due to complex interactions
• Horizontal precipitation—extremely varied!
Future Work
• Measurements for exchange coefficients for heat, moisture, and momentum
• Research effects of dissipative heating, sea spray, ocean-atmosphere coupling
A New Vertical Diffusion Package with an Explicit Treatment of Entrainment
Processes Lindsay Blank
February 11, 2016
The Basics
Authors: Song-You Hong1, Yign Noh1, and Jimy Dudhia2
Journal: Monthly Weather Review, September 2006
Scientific Question: Does the YSU PBL Scheme perform better than the MRF PBL Scheme?
Important Advance?: Yes
1. Yonsei University, Seoul, South Korea 2. NCAR, Boulder, Colorado
Yonsei University Scheme
What differentiates the scheme from others?: Handles entrainment explicitlyUses vertically varying parametersNonlocal-K momentum mixing
For what purpose was the scheme designed?: Address deficiencies in MRF PBL scheme
In what circumstances would you use that scheme over others?:Mesoscale modeling
Methods and Results
Idealized and real comparisons.
YSU outperforms MRF in both tests.
Improvement due to explicit treatment of entrainment.
Conclusions and Significance
YSU PBL scheme better represents PBL behavior than MRF PBL scheme.
YSU PBL scheme aids in model representation of convection.
Improved representation of convection/PBL leads to more skillful model behavior.