Modifications to the MYNN PBL and Surface Layer Scheme for WRF-ARW Joseph Olson 1,2 John M. Brown 1 1 NOAA-ESRL/GSD/AMB 2 Cooperative Institute for Research in Environmental Sciences
Feb 05, 2016
Modifications to the MYNN PBL and Surface Layer Scheme
for WRF-ARW
Joseph Olson1,2
John M. Brown1
1NOAA-ESRL/GSD/AMB2Cooperative Institute for Research
in Environmental Sciences
Background on Mellor-Yamada-Nakanishi-Niino (MYNN) PBL Scheme
• Implemented into WRF-ARW in 2008 (v3.0). • Main features of the MYNN include:
• Turbulent kinetic energy (TKE)-based local mixing scheme (like MYJ)
• Option to run at level 2.5 or 3.0 closure.
• Liquid water potential temperature, θl (= θ - (θ/T)(Lv/cp)ql), and total water content, qw (= qv + ql), are used as thermodynamic variables.
• Tuned to a database of LES simulations in order to overcome the typical biases associated with other MY-type schemes (insufficient growth of convective boundary layer and underestimated TKE).
• More elaborate mixing length formulations to flexibly change behavior across the stability spectrum.
• Initially adopted a very simple surface layer scheme, taken from an old version of the YSU surface layer scheme, with very little customization to the PBL scheme.
Known Problems with MYNN• warm-bias over desert/bare soil regions • production of negative TKE • high 10- and 60-m wind speed bias • excessive low-level clouds over the ocean/arctic
Wind Vector RMSE
MYNN
MYNN
MYJ
MYJ
6-hr forecasts from MYNN and MYJ between 20120410-20120510 verified against rawinsonde data over CONUS region
Temperature Bias
• Adjustment of closure constants to remove negative TKE problem (Canuto et al. 2008 and Kitamura 2010) but also removes critical Richardson number.
• Subsequent modifications to closure constants C2 and C3 to help reduce over-diffusive behavior
• Further modifications to mixing length formulae (surface layer and buoyancy length scales) to compensate increased diffusivity (shown later).
after
before
Recent changes to MYNN PBL scheme
and the turbulent length scale lt is:
and the buoyancy length scale lb is:
The mixing length is designed such that the shortest length scale among, ls, lt, and lb will dominate:
where the surface layer length scale ls is a function of the stability parameter(ζ=z/L; L in the M-O length):
if 0 ≤ ζ ≤ 1
if ζ < 0
MYNN Mixing Length Formulation
ls ={
Stable Conditions
Unstable Conditions
where qc is a turbulent velocity scale ~O(w*)
Changes to Surface Layer Length Scale (ls)
At z/L = -1, ls ~ 1.0z
At z/L = 1, ls = kz/3.7
At z/L = 1, ls = kz/3.1
At z/L = -1, ls ~ 0.75z
α4 = 100, cns = 2.7 α4 = 20, cns = 2.1
• Iterative - accurate solution of z/L, u* and θ*.• Updated thermal/moisture roughness lengths over land (Zilitinkevich
(1995), Air Pollution III – Vol. I.)
• Updates thermal/moisture roughness lengths over water, taken from the COARE 3.0 bulk algorithm (Fairall et al. 2003, J. of Climate).
• Use of consistent flux-profile relationships with those used to formulate the surface layer length scale in the PBL scheme.
• Relaxed some arbitrary limits, such as lower limit for u* (0.1 0.01).
Changes to MYNN surface layer scheme
2 m Temperature Verification in RAP(all metars across CONUS region)
BIAS RMSEMYJ
MYNN
• Afternoon warm bias reduced by up to 1oC in MYNN compared to MYJ
• Both Biases and RMSEs are comparable at night.
coldwarm
10 m Wind Speed Verification in RAP 9 hr forecasts – 01-12 June 2012
verified against all metars across the CONUS region
MYJMYNN
OBS-MODEL(negative means high wind speed bias)
α4=100 α4=20
Evolution of the Low-Level Jet (3-km nest)Loop was made from 1-hr output intervals between 00-18 UTC 06 Sept 2011.
. .... ....
Wind Speed (m s-1)
Wind speed and θ (Leeds, ND)
TKE and θ (Leeds, ND)
m2 s-2
m s-1
80-m wind speed (color)WFIP-North Profilers
Profile Evolution (Leeds, ND) – Lb Tests
MYJ
MYNNα2=0.53
MYNNα2=0.75
MYNNα2=0.64
• All model profiles are taken from the 3 km simulations.
• MYNN has stronger LLJ but too strong under 100 m AGL.
• MYNN is improved with increased α2.
Mean profiles (LDS) 03-09 UTC (MYNN-lb tests)
• α2 = 0.75 produces a weaker LLJ (~1.4 m s-1 weaker than α2=0.53).
• α2 = 0.64 produces a weaker LLJ (~0.7 m s-1 weaker than α2=0.53).
• Largest differences in TKE & lm are above the jet max.
Profile Evolution (Leeds, ND) – Ls Tests
MYJ
MYNNcns=2.7
MYNNcns=1.5
MYNNcns=2.1
• All model profiles are taken from the 3 km simulations.
• Parameter cns has a powerful impact on the LLJ strength.
• MYNN is improved with decreased cns..
Mean profiles (LDS) 03-09 UTC (MYNN-ls tests)
• cns = 1.5 produces a weaker LLJ (~1.8 m s-1 weaker than cns = 2.7).
• cns = 2.1 produces a weaker LLJ (~0.9 m s-1 weaker than cns = 2.7).
• Large differences in TKE & lm are below the jet max.
Wind Speed (m s-1)
Wind speed and θ (Ainsworth, NE)
TKE and θ (Ainsworth, NE)
m2 s-2
m s-1
80-m wind speed (color) WFIP-North Profilers
Evolution of the Low-Level Jet (3-km nest)Loop was made from 1-hr output intervals between 00-23 UTC 10 June 2012.
... ..
..
..MYNN PBL scheme configured with: α2 = 0.64, cns = 2.1, and α4 = 20.
Profile Evolution (Ainsworth, NE)
MYNN
MYJ
Profiler Winds
6-hr forecasts from RAP with MYNN and MYJ (including all modifications) between 20120607-20120612
verified against rawinsonde data over CONUS region
Wind Speed Bias Relative Humidity BiasTemperature Bias
Wind Vector RMSE Relative Humidity RMSETemperature RMSE
MYJMYNN
MYJMYNN
Summary
• Important modifications were made to improve the MYNN PBL scheme: (1) modified lb (α2), (2) modified ls (cns) , and (2) modified ls (α4).
• A small change to all 3 parameters can collectively reduce the high wind speed bias in the lowest 100 m, while maintaining better forecast skill of wind speed in the rest of the PBL.
• Many improvements to the surface layer scheme reduced the daytime warm bias, while remaining competitive with the MYJ at night.
• Too much tuning to improve the low-level wind speeds at 13 km grid spacing may degrade wind forecasts at higher resolution (or other altitudes). Cliff Mass will discuss this in the next talk.
Not Shown
• Hybrid PBLH: uses θ-based PBLH in neutral/unstable conditions and TKE-based PBLH in stable conditions.
• TKE budgets are available output fields by configuring Registry.EM.
•Make 3D TKE-budgets output fields dependent on namelist option, to reduce memory useage when not needed.
•Implement changes into v3.4.1.
Future Work
Extra Slides
Background on Rapid Refresh (RAP)• Hourly data assimilation system which uses the WRF-ARW as the
model forecast component (Weygandt et al., 2.1). • Parent model of the High-Resolution RAP (HRRR). • Testing different parameterizations within the WRF-ARW against
our current configuration. • For wind energy application, comparison of current PBL scheme,
Mellor-Yamada-Janjic (MYJ) to Mellor-Yamada-Nakanishi-Niino (MYNN). Main features of the MYNN include:
• Turbulent kinetic energy (TKE)-based local mixing scheme (like MYJ)
• Option to run at level 2.5 or 3.0 closure.
• Tuned to a database of LES simulations in order to overcome the typical biases associated with other MY-type schemes (insufficient growth of convective boundary layer and underestimated TKE).
• More elaborate mixing length formulations to flexibly change behavior across the stability spectrum.
Hybrid PBLH in MYNN Coastal Jet CaseWind speed parallel to the barrier (color), potential temperature (red contours) and TKE (black contours).
ziTKE
ziθv
zihybrid
zihybrid + 0.3zihybrid
22 UTC 00 UTC23 UTC
• If we take the maximum wind speed of the jet as the PBL top, then ziθv significantly underestimates the PBL height.
• ziTKE overestimates the PBL height, especially during the period of elevated mixing associated with the strong vertical wind shear on the outer edge of the jet.
• The hybrid PBLH, zihybrid, best follows the level of maximum wind speeds, but is also shallow-biased prior to the period enhanced mixing.
• The height, zihybrid + 0.3zihybrid, is used as the level for which the turbulent length scale is integrated to (between the surface and zihybrid + 0.3zihybrid). This allows the TKE within the “entrainment layer” to be accounted for when determining the turbulent length scale.
Mt.
Fairw
eath
er
Mt.
Fairw
eath
er
Mt.
Fairw
eath
er
2 m Temperature Verification(8-day means of 9 hr forecasts – valid at 21 UTC)
MYJ MYNN
East Bias: 1.78East MAE: 2.69West Bias: 0.22West MAE: 2.33
East Bias: 2.41East MAE: 3.03West Bias: 1.33West MAE: 2.38