Russell Danielson
Numerical Weather Prediction
Homework #2
Evaluation of Three Planetary Boundary Layer Schemes in the WRF Model
By Xiao-Ming Hu, John Nielsen-Gammon, and Fuqing Zhang
1.)
The planetary boundary layer and the processes associated with it play a crucial role in
many weather phenomena. For example, in the Houston-Galveston area, ozone frequently
exceeds the National Ambient Air Quality Standard leading to negative health implications. To
better forecast these pollution events, one needs to better simulate the meteorological processes
within the planetary boundary layer. However, this is not easy because of a multitude of different
issues such as inadequate initial conditions and too few observations in the boundary layer.
Therefore, many PBL schemes have been developed, all having different assumptions for
parameters such as moisture, transport of mass, and energy. These differences can change the
whole model domain meaning it is important to research the effectiveness of these PBL schemes.
In this study, the authors will evaluate the performance of three PBL schemes using the WRF
model. The three schemes include the newly updated Yonsei University (YSU) scheme, the
newly developed asymmetric convective model, version 2 (ACM2), and the Mellor-YamandaJanjic (MYJ) scheme. The end goal is to identify differences in model performance and comment
on possible consequences for air quality simulations while seeking to find the cause for these
differences.
2.)
Some qualities of the PBL schemes are that the MYJ scheme is a local closure model,
while the YSU and ACM2 schemes are nonlocal models. The MYJ scheme uses the 1.5-order
turbulence closure model of Mellor and Yamada (1982) to represent turbulence above the
surface layer and determines eddy diffusion coefficients from prognostically calculated TKE. It
is argued that the scheme is adequate for stable and slightly unstable flows while errors are more
likely to occur as the flow approaches the free-convection limit. The YSU scheme is a first-order
nonlocal scheme, with a countergradient term in the eddy-diffusion equation. In WRF, the
scheme is modified to increase the critical bulk Richardson number from 0 to 0.25 over land. The
ACM2 scheme includes a first-order eddy-diffusion component in addition to the explicit
nonlocal transport of the original ACM1 scheme and it shuts off nonlocal transport and uses
local closure for stable and neutral conditions.
In this study the WRF, version 3.0.1, is centered over southeastern Texas to simulate the
meteorological conditions during the summer of 2005, which coincides with the Second Texas
Air Quality Study (TexAQS2). The observations collected during the study allow for a
comprehensive validation dataset that has additional observations to the regularly collected ones.
The model setup includes four model domains with two-way nesting with grid spacing of 108,
36, 12, and 4km. All model domains have 43 vertical levels and cumulus parameters are only
used on the 108, 36km fine domains, not the 12, and 4km. The GFS final operational global
analyses are used for the initial conditions and the boundary conditions. Three 36-hour forecasts,
one for each PBL scheme, are initiated at 00Z every day from July 1 through September 30 to
compare how the different schemes modulate the same initial and boundary conditions. The first
12 hours of the simulations were not used in the analysis because they are treated as spinup,
while the remaining 24 hours are used for evaluation. This evaluation will focus on the 12km
domain, despite the 4km domain having finer resolution. The larger domain incorporates the
information from the 4km field.
Russell Danielson
Numerical Weather Prediction
Homework #2
3.)
The authors initially focus on temperature and dew point predictions to compare them to
the observations. At the 211 NWS-FAA sites, the temperatures predicted by the models are
similar to the observations during the night through midday but cold biases are seen during the
day for all models. This is consistent with other research that noticed a cold bias using the MM5
model with the three PBL schemes. For the diurnal cycle as a whole, the MYJ, YWU, and
ACM2 schemes produce -1.25, -0.63, and -0.9°C mean biases, respectively. For dew point, the
three schemes perform relatively well at night through midday but vast dry biases are detected
during the afternoon hours. The worst performing model, the MYJ scheme, overpredicts the dew
point with a 0.86°C mean bias throughout the entire day. Errors in these parameter fields can
lead to fictitious biases when compared to observations for the vertical transformation from
surface to atmospheric values. However, the authors note that biases are largely a result caused
by error in the physical processes simulated by the model rather than the error in diagnosing 2-m
values. Spatial distribution of temperature indicates the YSU and ACM2 schemes predict higher
temperatures during the afternoon than the ACM2 scheme, while at night the YSU scheme
predicts higher temperatures than the other two schemes. Further investigations reveals that the
PBL schemes may struggle to accurately predict the effects of cloud cover and/or soil moisture,
which may lead to daytime temperature biases. In an attempt to understand why the models are
in error, the authors looked at sensible heat fluxes, the Bowen ratio, and incoming solar
radiation. However, the differences in performance between the schemes arise directly within the
PBL schemes themselves, instead of the differences in the surface heat fluxes or partially
external feedbacks like cloud cover.
PBL heights are the next focus in this paper. The results from looking at a 1.5-thetaincrease method to define the top of the PBL show that the ACM2 predicts the highest PBL
height and the MYJ predicts the lowest with the YSU being in the middle but closest to the
ACM2 scheme. A reason for the underprediction in the MYJ scheme is that it lacks sufficient
heat and moisture transport to entrain the warmer and drier air from just above the top of the
PBL. Attempts have been made to increase the vertical mixing in the MYJ scheme, which will
improve the temperature and moisture biases. During the nighttime, all of the schemes tend to
produce low-level wind shear that is too large and this may have implications for the simulated
horizontal dispersion of pollutants in air quality modeling. Comparing the three schemes, the
authors note that the MYJ and ACM2 schemes have weaker vertical mixing and wind shear than
the YSU scheme.
4.)
There are many directions for future work to expand on this research. One option is to
use the same schemes but for a different period of time and a different set of observations. Air
pollution is a problem in many other cities around the world and running the model over Los
Angeles or Beijing may provide insight into how the models predict the air pollution. If
additional observational studies are undertaken that provide quality information, a similar study
can be performed to compare the results with and see if there are any major differences
depending on year or season of the study. The issues within the boundary layer don’t just end
with air pollution, however. Other issues such as marine boundary layers, frontal structure within
boundary layers, and the effect of snow cover on the boundary layer still exist and must be
researched.
In addition, this study only examined three different schemes. There are many other
schemes that are run for many different models. A study can investigate these other schemes
with the same dataset for comparison. Also, other models than the WRF can be used to feed
atmospheric variables into the domain the researcher is focusing on.
Russell Danielson
Numerical Weather Prediction
Homework #2
Finally, a difficult mechanism for meteorologists to predict is lake-effect snow and
boundary layer processes play a crucial role during lake-effect snow. Therefore, accurate
prediction of the boundary layer is essential for forecasters to make quality forecasts. If I were to
setup a numerical experiment to explore this issue, I would first start by obtaining an
observational dataset that has many boundary layer observations during lake-effect snow. Then I
would choose PBL schemes that I am particularly interested in, ones that are used in models that
forecasters regularly use for this phenomenon. Then, I would only select a single model, like this
study, to run the same exact time period for each scheme so I can compare them. I would make
sure that the model I chose depicts lake-effect snow relatively well, such that when there is a
significant lake-effect snowstorm, the model captures it. The results will describe how well the
schemes represent the observations and how the schemes differ from each other while briefly
explaining why these differences might occur.