Observed and WRF-Simulated Low-Level Winds in a High-Ozone Episode during... Central California Ozone Study

advertisement
2372
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 47
Observed and WRF-Simulated Low-Level Winds in a High-Ozone Episode during the
Central California Ozone Study
J.-W. BAO
NOAA/Earth System Research Laboratory, Boulder, Colorado
S. A. MICHELSON, P. O. G. PERSSON,
AND
I. V. DJALALOVA
NOAA/Earth System Research Laboratory, and Cooperative Institute for Research in Environmental Sciences, University of
Colorado, Boulder, Colorado
J. M. WILCZAK
NOAA/Earth System Research Laboratory, Boulder, Colorado
(Manuscript received 3 July 2007, in final form 23 December 2007)
ABSTRACT
A case study is carried out for the 29 July–3 August 2000 episode of the Central California Ozone Study
(CCOS), a typical summertime high-ozone event in the Central Valley of California. The focus of the study
is on the low-level winds that control the transport and dispersion of pollutants in the Central Valley. An
analysis of surface and wind profiler observations from the CCOS field experiment indicates a number of
important low-level flows in the Central Valley: 1) the incoming low-level marine airflow through the
Carquinez Strait into the Sacramento River delta, 2) the diurnal cycle of upslope–downslope flows, 3) the
up- and down-valley flow in the Sacramento Valley, 4) the nocturnal low-level jet in the San Joaquin Valley,
and 5) the orographically induced mesoscale eddies (the Fresno and Schultz eddies). A numerical simulation using the advanced research version of the Weather Research and Forecasting Model (WRF) reproduces the overall pattern of the observed low-level flows. The physical reasons behind the quantitative
differences between the observed and simulated low-level winds are also analyzed and discussed, although
not enough observations are available to diagnose thoroughly the model-error sources. In particular,
hodograph analysis is applied to provide physical insight into the impact of the large-scale, upper-level
winds on the locally forced low-level winds. It is found that the diurnal rotation of the observed and
simulated hodographs of the local winds varies spatially in the Central Valley, resulting from the combining
effect of topographically induced local forcing and the interaction between the upper-level winds and the
aforementioned low-level flows. The trajectory analysis not only further confirms that WRF reproduces the
observed low-level transport processes reasonably well but also shows that the simulated upper-level winds
have noticeable errors. The results from this study strongly suggest that the errors in the WRF-simulated
low-level winds are related not only to the errors in the model’s surface conditions and atmospheric
boundary layer physics but also to the errors in the upper-level forcing mostly prescribed in the model’s
lateral boundary conditions.
1. Introduction
Winds within and immediately above the atmospheric boundary layer (ABL) impose direct control
over the transport and dispersion of pollutants in the
Corresponding author address: Jian-Wen Bao, NOAA/Earth
System Research Laboratory, Mail Stop PSD3, 325 Broadway,
Boulder, CO 80305.
E-mail: jian-wen.bao@noaa.gov
DOI: 10.1175/2008JAMC1822.1
© 2008 American Meteorological Society
lower troposphere. Accurately simulating these lowlevel winds in areas of complex local surface forcing is
still a major challenge to atmospheric modelers involved in air-quality-control planning. One such area in
the United States is the Central Valley of California.
Meteorological conditions associated with poor air
quality in California’s Central Valley have long been a
research subject because it is widely acknowledged that
meteorological conditions and topography are key factors in episodes of acutely poor air quality in the Cen-
SEPTEMBER 2008
BAO ET AL.
tral Valley. Because of the unique geographical environment surrounding the valley, airborne pollutants are
not easily ventilated out of the valley under light-wind
conditions. Observation studies have shown that the
complex topography around the valley often induces
particular low-level wind systems (such as local lowlevel eddies) that play important roles in the recirculation of pollutants (e.g., Niccum et al. 1995; Tanrikulu et
al. 2000; Zhong et al. 2004). Numerical modeling studies of these wind systems have also been carried out
using mesoscale models (e.g., Lin and Jao 1995; Seaman
et al. 1995; Burk and Thompson 1996; Stauffer et al.
2000).
Mesoscale meteorological models have uncertainties
in the representation of the surface-forcing physics as
well as the spatially varying land surface characteristics
and topographic details. Atmospheric analyses used to
specify the initial and lateral boundary conditions for
the models also have uncertainties in the background,
upper-level pressure, and wind fields. There is no doubt
that these uncertainties limit the accuracy of atmospheric models in simulating the low-level winds in the
Central Valley. To understand the characteristics of the
atmospheric model errors in air-quality-control planning, it is imperative for atmospheric modelers to establish episode cases that represent worst-case weather
conditions and to assess the skills of the atmospheric
models in simulating these conditions. A few episode
cases for air-quality-control planning in the Central
Valley have been selected during the period of the Central California Ozone Study (CCOS). CCOS was a combined observational and modeling program that took
place during the summer of 2000 and was designed to
improve the understanding of the mechanisms of ozone
formation and transport in the Central Valley.
In this study, the 29 July–3 August 2000 high-ozone
episode is investigated using observations collected
during the CCOS field experiment and the advanced
research version of the Weather Research and Forecasting Model (WRF; Skamarock et al. 2005). The reason for choosing this particular episode is that its meteorological conditions are typically associated with
high-ozone events in the region. An analysis of the observations is performed to identify characteristics of the
low-level winds in the Central Valley, and the WRF
simulation is compared with the observations with respect to these characteristics. The purpose of the study
is 1) to understand better the low-level wind variation
that affects the pollution transport and dispersion in the
Central Valley and 2) to assess the skills of the WRF
model in simulating the observed low-level winds for
the researchers who rely on numerical model simula-
2373
FIG. 1. Coverage of the one-way nested meshes of 36- (D1), 12(D2), and 4-km (D3) grid sizes for the WRF simulation.
tions to provide and enhance current knowledge about
the meteorological processes that influence air quality.
The rest of the paper is arranged as follows: The
configuration of the numerical model simulations is
summarized in section 2. The comparisons of the observed and simulated low-level winds with respect to
several mesoscale low-level flow features in the Central
Valley are presented in section 3. The comparisons of
the observed and simulated low-level winds in terms
of hodograph and trajectory analysis are presented in
section 4. The summary and conclusions follow in section 5.
2. Numerical model simulations
The WRF simulation for this case study is run using
a set of 36-, 12-, and 4-km one-way nested grids (see
Fig. 1) that have 50 vertical stretched levels, 30 of which
are within the lowest 2 km and the lowest of which is at
about 12 m above the surface. The 4-km grid encompasses the CCOS area, which covers the entire state of
California. Boundary and initial conditions are prescribed using the 6-hourly 40-km National Centers for
Environmental Prediction (NCEP) Eta analysis. The
sea surface temperature is specified with the information available in the Eta analysis and is kept constant
during the simulation. The simulation begins at 1200
UTC 29 July and is run for 120 h, ending at 1200 UTC
3 August 2000. In all of the simulations, the Eta planetary boundary layer and surface layer schemes are
2374
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 47
used along with the Noah land surface model, the Lin
et al. microphysics parameterization scheme, and the
Dudhia shortwave and Rapid Radiative Transfer
Model (RRTM) longwave radiation parameterization
schemes. The Kain–Fritsch convective parameterization scheme is used on the 36- and 12-km grids. No
convective parameterization scheme is used on the
4-km grid. Details of these physics parameterization
schemes in the WRF model can be found in Skamarock
et al. (2005).
3. Observed and simulated low-level wind systems
It has been recognized in the mesoscale-modelevaluation community (e.g., Rife and Davis 2005) that
properly choosing measures of performance is important in the assessment of the skills of atmospheric models. For model simulations of transport and dispersion,
the most important measure is whether the simulations
can even qualitatively reproduce observed low-level
winds. In this study, we define the low-level winds as
the winds within the lowest 500 m above ground level
(AGL) that are directly associated with the major
transport and dispersion processes revealed by the observations collected during the CCOS field experiment.
Previous studies (e.g., Niccum et al. 1995; Lin and Jao
1995; Zhong et al. 2004, and references therein) using
observations and/or numerical models have revealed
that there are several low-level flow components of the
low-level winds in the Central Valley: 1) the incoming
low-level marine airflow through the Carquinez Strait
into the Sacramento River delta, 2) the diurnal cycle of
upslope–downslope flows, 3) the up- and down-valley
flow in the Sacramento Valley, 4) the nocturnal lowlevel jet in the San Joaquin Valley, and 5) two orographically induced mesoscale eddies (the Fresno and
Shultz eddies). As revealed in this case study, all of
these components are observed during the 5-day period
and, more important, their evolution is greatly influenced by the upper-level background winds. To our
knowledge, the ability of WRF to simulate the lowlevel winds in complex terrain during the summertime
in central California has not been systematically investigated.
a. Descriptions of observations
The observational datasets used for the meteorological comparison include a network of 24 915-MHz wind
profilers and one 449-MHz wind profiler, along with
the surface data at these sites. The network of wind
profilers (see Fig. 2) is one of the core sets of meteorological instrumentation used for CCOS. The wind
FIG. 2. Map of California, indicating the locations of the 25
profiler sites that operated during CCOS.
profilers provide hourly averages of wind speed and
direction, typically to heights of 3000 m AGL. In addition to winds, the profilers measure the vertical profile
of virtual temperature up to 1000 m AGL using the
radio acoustic sounding system (RASS) technique. The
depth of the daytime convective ABL can also be determined from the wind profiler measurement by visually inspecting values of range-corrected signal-to-noise
ratio, vertical velocity, which is large within the convective ABL, and radar spectral width, which is a measure
of turbulence intensity (White 1993; Angevine et al.
1994; Bianco and Wilczak 2002). Because there are irreconcilable disparities in the quality and representativeness of all of the available surface measurements,
only the quality-controlled 25 profiler and RASS measurements and the collocated surface observations are
used in this study.
b. Synoptic overview of the case
The synoptic meteorological situation during the 29
July–3 August 2000 period, as depicted by the 500-hPa
SEPTEMBER 2008
BAO ET AL.
2375
FIG. 3. The NWS 500-hPa charts of geopotential height (solid contours; dam) and isotherms (dashed contours;
°C) valid at 0000 UTC (a) 30 Jul, (b) 31 Jul, (c) 1 Aug, and (d) 2 Aug 2000.
weather charts from the National Weather Service
(NWS) in Fig. 3, is characterized by a ridge at 500 hPa
that started to retrogress toward the west from the Four
Corners area and strengthen on the first two days. The
ridge remained strong and continued to slowly regress
toward the west until 31 July, when it is centered over
eastern Nevada. On 1 and 2 August, the ridge axis
slowly rotates clockwise, moving from southern to
northern California. During the same time, a largescale trough over the eastern Pacific Ocean slowly
moves eastward around the northwestern periphery of
the high pressure region. The regression of the large-
scale ridge produces large-scale warm subsidence (e.g.,
the 850-hPa temperature over Oakland, California,
reaches as high as 27°C and the 500-hPa height peaks at
5970 m). At the surface (not shown), high pressure is
present over the Great Basin area, with its center located to the northeast of the Central Valley, rendering
a weak offshore pressure gradient between San Francisco, California, and Reno, Nevada, and a weak north–
south gradient from San Francisco to Las Vegas, Nevada. Such a synoptic pattern makes the low-level
winds within the Central Valley relatively weak, a condition conducive to high-surface-ozone events. By the
2376
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
end of the 5-day period, as the large-scale trough moves
onshore, the conditions over the Central Valley become less conducive to poor air quality because of
stronger winds and cooler temperatures.
The fact that the aforementioned large-scale setting
provides a weak large-scale, upper-level forcing that is
favorable for poor air quality has a great implication for
the comparisons of the observed and simulated lowlevel winds in the Central Valley. In principle, the evolution of the low-level winds is affected by many factors
ranging from topography, land surface characteristics
(soil temperature and moisture, vegetation, albedo,
roughness length, etc.), land/water contrast, and cloud
coverage to upper-air conditions, including mesoscale
and synoptic forcing. In particular, mesoscale and synoptic circulations always exert significant influence on
the winds in the ABL and lead to exchange of atmospheric properties between the ABL and the free troposphere. Therefore, although it is extremely important
to evaluate the simulated winds within the lowest 500 m
AGL, the differences between the observed and simulated upper-level winds should also be examined to understand the relative impact of the upper-level winds on
the evolution of the low-level winds.
c. Mesoscale low-level-flow components
1) MARINE
STRAIT
AIRFLOW THROUGH THE
CARQUINEZ
Under the weakly forced synoptic condition mentioned in the previous section, a low-level inflow of
marine air moving through the Carquinez Strait into
the Central Valley is maintained during the day near
the San Francisco Bay Area. This inflow is forced by
the pressure gradient corresponding to the surface thermal contrast, in which the coastal sea surface temperature is cooler than the land surface temperature in the
Central Valley. Once the marine air enters the Central
Valley, it splits into northward flows up the Sacramento
Valley and southward flows up the San Joaquin Valley
because of the blocking effect of the Sierra Nevada.
Because the Carquinez Strait is located closer to the
northern end of the Central Valley (i.e., the Sacramento Valley), the northward flow does not have as far
to travel as the southward flow before it encounters
orographic blocking. This closeness of the blocking topography in the Sacramento Valley to the Carquinez
Strait causes more of the incoming marine flow to move
southward than northward; thus, much of the daytime
incoming marine flow veers up the San Joaquin Valley.
The 5-day-mean observed low-level winds at 0000 UTC
(late afternoon local time) averaged over the first 500 m
AGL shown in Fig. 4a (black wind barbs) depict an
VOLUME 47
example of this daytime behavior of the incoming flow.
The splitting of the flow is clearly seen in the observations. There is a northward flow at Redding (RDG) and
Chico (CCO) in the Sacramento Valley, and there is a
southward flow up the San Joaquin Valley. The intensity of the incoming marine flow is reduced at night
because of the nocturnal cooling at the surface in the
Central Valley, such that a reversed down-valley flow is
dominant in the Sacramento Valley [see further discussion in section 3c(3) and the southerly flow at RDG,
CCO, and Arbukel (ABK) in Fig. 4b]. It has long been
recognized that the inflow of marine air through the
Carquinez Strait into the Central Valley is a major component of the low-level wind system affecting the air
quality in the Central Valley (e.g., Niccum et al. 1995;
Zhong et al. 2004). The impact of this incoming flow on
the air quality in the Central Valley is two sided: although it has a ventilating effect on the existing pollution in the Central Valley, it also can transport pollution emitted in the San Francisco Bay Area into the
Central Valley. The WRF model simulation qualitatively reproduces the aforementioned features of the
incoming marine air. For example, in Fig. 4a, the splitting of the incoming flow through the Carquinez Strait
during the day is clearly seen in the 5-day-mean WRFsimulated low-level winds averaged over the lowest 500
m AGL (red wind barbs, which are plotted at every
fifth grid point) at 0000 UTC. The daytime up-valley
flow in the Sacramento Valley and San Joaquin Valley
are both present in Fig. 4a. The 5-day-mean WRF simulation of the nighttime winds at 1200 UTC averaged
over the lowest 500 m AGL (red barbs in Fig. 4b) shows
that the down-valley flow in the Sacramento Valley is
present, as is the intensification of the low-level flow up
the San Joaquin Valley after 1200 UTC [see further
discussion in section 3c(4)].
The observed and simulated diurnal changes of the
incoming flow through the Carquinez Strait are further
compared using the diurnal cycle of the observed lowlevel winds at Livermore (LVR), averaged over the 5
days from 29 July to 3 August 2000 (Fig. 5). The wind
profiler observations (Fig. 5a) indicate that incoming
flow below 500 m is the weakest during the morning
and intensifies during the afternoon, reaching a maximum about 1 h before sunset (between 0100 and 0200
UTC). The depth of the incoming flow (the westerly
flow) increases during the day. There is also a decoupling of the lower-level winds (below 500 m) from 0600
to 1500 UTC, perhaps a manifestation of the decoupling between the low-level flow channeled by the local
orographic variation and the upper-level flow over the
coastal mountains. The 5-day-mean WRF-simulated diurnal cycle of the winds at LVR, as depicted in Fig. 5b,
SEPTEMBER 2008
BAO ET AL.
FIG. 4. Five-day-mean winds averaged over the lowest 500 m AGL at every five
grid points of the 4-km grid for the WRF simulation (red wind barbs) overlaid with
wind profiler observations (black wind barbs) at (a) 0000 and (b) 1200 UTC.
2377
2378
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 47
slower during the day and faster during the night and
early morning than the observed flow. Comparison of
the local topography near the LVR site with the model
topography strongly suggests that this bias in the simulated incoming marine flow may be attributed to the
difference between the model topography and the real
topography at LVR. Also, the return easterly flow that
is observed between 800 and 1400 m in the late morning
(from 1200 to 1900 UTC) occurs later in the simulation,
is weaker, and is confined to lower levels than the observed flow. Another difference between the observed
and simulated winds is found at levels higher than 1500
m AGL, where the simulated winds are predominantly
southwesterly while the observations show more southerly flow. This indicates a large-scale bias at upper levels that is further discussed in section 4.
2) DIURNAL
CYCLE OF UPSLOPE–DOWNSLOPE
FLOWS
FIG. 5. Five-day-mean (a) wind profiler–observed wind vectors
and (b) WRF-simulated wind vectors at LVR. The (c) bias and (d)
RMSE of WRF-simulated wind speed (m s⫺1) at LVR; SR indicates sunrise, and SS indicates sunset.
is consistent with the observations: there is a daytime
intensification of the incoming flow, the depth of the
incoming flow grows during the day, and there is a
nighttime weakening in the low-level (below 500 m)
winds. However, there are some differences between
the WRF simulation and the observations. An examination of these differences in terms of the bias and
root-mean-square error (RMSE) averaged over the five
days (Figs. 5c and 5d) indicates that the simulated lowlevel incoming marine flow (below 600 m AGL) is
Figure 4 indicates that the low-level winds in the
Central Valley exhibit reversal as the result of the development of the upslope winds during the daytime and
the downslope drainage flow during the night because
of the diurnal surface radiative heating/cooling cycle.
The upslope–downslope flows are especially apparent
at sites along the Sierra Nevada—in particular, Grass
Valley (GVY) and Trimmer (TMR). At TMR, the
5-day-mean diurnal cycle of the observed low-level
winds (Fig. 6a) shows that a westerly wind component,
which at this location is an indication of the development of upslope flow, develops in the early afternoon
(between 2000 and 2100 UTC), whereas an easterly
wind component, which is indicative of a local downslope flow, occurs after sunset (between 0600 and 0700
UTC) and through the night. This diurnal change of the
upslope and downslope flows is known to be one of the
important factors controlling the air quality in the Central Valley. During the day, the upslope flow induces
subsidence in the valley that potentially suppresses the
development of the daytime ABL and thus is conducive
to confinement of the pollution at low levels. During
the night, the downslope drainage flow, along with the
nocturnal low-level jet, affects the air quality in the
Central Valley by circulating and redistributing pollutants within the valley. It is also difficult to ventilate
these pollutants out of the valley during the night because of the nocturnal stable atmospheric boundary
layer. The interaction of the upslope and downslope
flows with the incoming marine flow in the San Joaquin
Valley renders the surface winds diffluent during daytime and confluent during nighttime. These diffluent
and confluent flows within the San Joaquin Valley can
be seen in the observations—in particular, at the Los
SEPTEMBER 2008
2379
BAO ET AL.
tion shows. A noticeable difference in the diurnal cycle
of winds between the observations and the simulation is
that the transition between downslope and upslope
flows in the simulation is not as abrupt as in the observations. The observed winds tend to become very weak
before changing direction, whereas the WRF-simulated
winds do not. The depth of the maximum upslope and
downslope flows in the WRF simulation are less than
the observed by about 200 m. This difference is possibly
due to the differences between the real and model topography. There consequently is a noticeable negative
wind speed bias around sunrise and positive wind speed
bias near the surface around sunset, although the
RMSE near the surface is relatively small. Note also
that the simulated winds have a significant directional
shear at 2500 m AGL, whereas the observed directional
shear is lower, at ⬃1800 m AGL, indicating the bias in
the simulated upper-level winds.
3) UP- AND DOWN-VALLEY
SACRAMENTO VALLEY
FIG. 6. As in Fig. 5, but for TMR.
Banos (LBA), Stevinson (SVS), and Waterford (WFD)
sites in Fig. 4a (diffluent flow) and Fig. 4b (confluent
flow).
The WRF simulation qualitatively reproduces the
aforementioned characteristics of the upslope and
downslope flows at TMR (see Fig. 6b). There is a westerly wind component during the day, which is indicative
of an upslope flow, and easterly flow during the night
after 0500 UTC, which is indicative of a downslope
flow. The 5-day-averaged winds shown in Figs. 6a and
6b along with the bias and RMSE over the five days
(Figs. 6c and 6d) indicate that overall the simulated
nighttime downslope flow is stronger than the observa-
FLOW IN THE
The same mechanism responsible for the formation
of the upslope and downslope flows causes the development of the up-valley and down-valley flows (Egger
1991); that is, anabatic winds form during the day when
hillside slopes are heated more than the valley floor.
The differential heating of contact air causes air to flow
upslope. After sunset, katabatic winds form as the result of radiative cooling of upper slopes lowering the
temperature of air in contact with them and the colder,
denser air then sinking rapidly downslope.
It is found that during this 5-day period the diurnal
rotation of the upslope and downslope flows along the
foothills of the Sacramento Valley has an integral dynamical effect on the overall flow in the Sacramento
Valley, which leads to the development of up- and
down-valley flow on the scale of the entire Sacramento
Valley. The up-valley flow can be most prominently
seen in the observations of the 5-day-mean observed
low-level winds at 0000 UTC (late afternoon local time)
averaged over the lowest 500 m AGL (Fig. 4a) at RDG,
CCO, and ABK. The reversal during the night to downvalley flow is also seen at these sites in the observations
(Fig. 4b). The timing and transition of the up- and
down-valley flows are shown in the 5-day-mean observed (Fig. 7a) and WRF-simulated (Fig. 7b) time–
height wind profiles at RDG. Note that the y axis in Fig.
7 is along the Central Valley for better illustration of
the along-valley flow. It is seen in this figure that the
observed low-level winds, within the lowest 500 m
AGL, are predominantly up or down valley (Fig. 7a).
The 5-day mean timing of the reversal of the up- and
down-valley flows lags 7 h behind sunrise and sunset in
2380
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 47
valley flows seen in the observed time–height profile at
RDG (Fig. 7b). However, a comparison of the simulation with the observations indicates that the simulated
vertical extent of the down-valley flows is shallower
than the observed flow, perhaps because of the differences in the observed and WRF-simulated upper-level
wind direction and in the actual terrain versus that used
in WRF, corresponding to the noticeable bias and
RMSE of the simulated wind speed right after sunrise
and sunset (Figs. 7c and 7d). Overall, the intensity of
the simulated up- and down-valley flows is weaker than
the observed, and the timing of the transition from the
up- to down-valley flows is about 4 h earlier in the WRF
simulation than in the observations, rendering the duration of the simulated up-valley flow shorter than the
observed. Also, the WRF-simulated transition is more
gradual than the observed; that is, the observed winds
are more bidirectional than the simulated. All of these
differences may be attributed to the differences between the real valley topography to the north of RDG
and the errors in the simulated large-scale upper-level
flow (further discussion on the latter appears in section
4) over RDG.
4) NOCTURNAL LOW-LEVEL
JOAQUIN VALLEY
FIG. 7. As in Fig. 5, but for along-valley wind vectors at RDG.
the observations. It is interesting to note [as shown by
the observed and simulated time–height wind profiles
in section 3c(4)] that such a diurnal transition from the
up-valley winds to the down-valley winds is not present
in the San Joaquin Valley because of the dominance of
the southward inflow of the marine air through the Carquinez Strait.
The WRF simulation shows the same up-valley winds
in the Sacramento Valley at 0000 UTC (Fig. 4a) and
down-valley winds at 1200 UTC (Fig. 4b). Furthermore,
the WRF simulation qualitatively reproduces the 5-daymean diurnal transition between the up- and down-
JET IN THE
SAN
Figure 8a depicts the 5-day-mean time–height wind
profile over all of the sites [a total of 12: the sites are
(label in parentheses) Angiola (AGO), Bakersfield
(BKF), Fresno (FAT), LBA, Lemoore (LEM), Lost
Hills (LHS), Sacramento (SAC), Visalia (SJV), SVS,
Travis (TRA), WFD, and Tracy (TCY) in Fig. 2] in the
San Joaquin Valley. As in Fig. 7, the y axis in this figure
is along the Central Valley for better illustrating the
along-valley flow. It is clearly seen that there is a diurnal change in the intensity of the low-level southward,
up-valley flow. The low-level flow experiences a nocturnal acceleration, which can be explained as the result
of a geostrophic balance involving the cross-valley Coriolis force and a net cross-valley pressure gradient
force associated with nighttime slope cooling along the
mountains surrounding the Central Valley (Blackadar
1957). In contrast, the southward low-level winds are
weakened during the day as the low-level divergence
develops as a result of the formation of the upslope
winds being forced by horizontal pressure gradients associated with slope heating along the mountains surrounding the Central Valley.
The WRF simulation qualitatively reproduces most
of the mean characteristics of the observed low-level
winds in the San Joaquin Valley (Fig. 8b). The simulated acceleration of the low-level winds up the valley
(i.e., the northerly low-level flow along the axis of the
SEPTEMBER 2008
2381
BAO ET AL.
indicated by the bias and RMSE of the simulated wind
speed (Figs. 8c and 8d). The 5-day-mean maximum
speed of the nocturnal low-level jet is stronger in the
WRF simulation than in the observations by 1–2 m s⫺1.
The wind speed bias during the day indicates that the
incoming flow is too weak. Another difference between
the simulated and observed winds is noticeable at levels
higher than 2000 m AGL, and is most noticeable from
0000 to 1200 UTC. During this time, the observed
winds above 2000 m have more of a southerly component, whereas the winds in the WRF simulation are
more westerly (Figs. 8a and 8b). These differences are
indicative of the errors in the WRF-simulated upperlevel winds. In concept, it is not difficult to understand
that because the pressure gradient force associated with
the upper-level winds exerts a great impact on the lowlevel winds (see, e.g., Blackadar 1957), errors in the
upper-level winds can lead to errors in the low-level
nocturnal jet.
5) THE FRESNO
FIG. 8. As in Fig. 7, but averaged over the San Joaquin Valley.
Twelve sites (labeled AGO, BKF, FAT, LBA, LEM, LHS, SAC,
SJV, SVS, TRA, WFD and TCY in Fig. 2) are included in the
average.
San Joaquin Valley) during the night is in qualitative
agreement with the observations. The timing of the
simulated nocturnal low-level jet is also in reasonable
agreement with the observations. The shutdown of the
ABL mixing and subsequent acceleration of the lowlevel winds occur at nearly the same time, between 2300
and 0000 UTC, in the simulation as they do in the observations. However, quantitative differences between
the simulation and the observations are noticeable, as
AND
SCHULTZ
EDDIES
The profiler wind measurements also reveal two major eddies in the Central Valley. In the Sacramento
Valley, interaction between the northward marine inflow and the nocturnal down-valley flow often leads to
the formation of a counterclockwise local eddy to the
north or northwest of Sacramento, known as the Shultz
eddy. The other eddy is associated with the aforementioned nocturnal low-level jet in the San Joaquin Valley. At night, as the along-valley nocturnal low-level jet
develops, the cross-valley Coriolis force makes the jet
veer toward the western side of the valley. Meanwhile,
the downslope flows in the eastern side of the valley
intensify, and a horizontal shear forms between the
nocturnal jet and the foothills to the east. With the
blocking effect of the Tehachapi Mountains at the
southern end of the San Joaquin Valley, this shear often gives rise to a local eddy in the San Joaquin Valley,
known as the Fresno eddy. This eddy circulates counterclockwise and plays a part in redistributing local polluted air. The dynamical conditions favorable for the
formation of both the Fresno and Shultz eddies are
investigated and discussed by Lin and Jao (1995).
The WRF simulation qualitatively reproduces both
the Schultz and Fresno eddies. The Schultz eddy can be
seen in both the observations and the WRF simulation.
At 0900 UTC (Fig. 9a), the observations of the 5-daymean winds averaged over the lowest 500 m AGL
(black wind barbs) indicate wind shear between ABK,
TRA, SAC, and Pleasant Grove (PSG), whereas the
WRF-simulated winds (red wind barbs) show similar
shear in this area of the Sacramento Valley. However,
there are some differences between the simulated and
2382
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
FIG. 9. As in Fig. 4, but valid at (a) 0900 and (b) 1500 UTC.
VOLUME 47
SEPTEMBER 2008
BAO ET AL.
observed winds. These differences lead to the direction/
orientation of the simulated flow around the Schultz
eddy being different from the observations. In particular, the simulated Schultz eddy is farther north in the
Sacramento Valley in the WRF simulation than is observed.
The Fresno eddy tends to form later in the morning
than the Schultz eddy and can be seen in both the observations and the WRF simulation (Fig. 9b) at 1500
UTC. The Fresno eddy is depicted in the network of
profilers as the shear between the profiler locations at
FAT and SJV and the profiler location at LHS. The
WRF-simulated winds (Fig. 9b) also indicate shear between these sites and clearly show the eddy at this time,
but the simulated eddy is farther to the east than is
observed. In addition, the WRF-simulation shows that
the Fresno eddy extends along a large portion of the
eastern side of the southern San Joaquin Valley. However, because there is a gap in the profiler network in
this area, the horizontal extent of the Fresno eddy in
the WRF-simulation cannot be verified.
Figure 10 shows the 5-day-mean winds at SJV and
FAT for the wind profiler observations and from the
WRF simulation. It indicates that the winds at these
locations are affected by the Fresno eddy. It is seen that
after sunset the observed 5-day-mean winds at SJV
clearly show the rotation associated with the change of
the wind direction from up to down valley as the Fresno
eddy forms, whereas after sunrise, the winds turn back
to up valley as the Fresno eddy dissipates (Fig. 10a).
The WRF-simulated winds at SJV show a similar
change of up-valley flow (Fig. 10b). At about 1300
UTC, the winds in the lowest few hundred meters shift
from northwesterly to southwesterly as the Fresno eddy
starts to form, which is a little earlier than in the observations. The WRF-simulated winds then turn from
southwesterly to westerly and remain westerly until
about 2100 UTC, when they become northwesterly (up
valley) once the Fresno eddy has dissipated.
The 5-day-mean winds at FAT, only about 60 km
north-northwest of SJV, also illustrate the development
of the Fresno eddy (Figs. 10c and 10d). The observed
low-level winds (Fig. 10c) are generally up valley
(northwesterly), except after 1300 UTC at which time
there is a shift to southerly, then southeasterly, and
then to easterly as the Fresno eddy develops. A downvalley component of the winds remains in place in the
observations until 2000 UTC. The Fresno eddy circulation extends to between 700 and 800 m in the observations. The 5-day-mean winds at Fresno in the WRF
simulation show a similar pattern (Fig. 10d). The simulated winds in the lowest model levels start to turn from
up valley to down valley at 1300 UTC, which is com-
2383
parable to observations. The circulation in the WRF
simulation extends to about 500 m, which is less than
what was observed by a few hundred meters. Correspondingly, there are significant bias and RMSE in the
simulated wind speed at both SJV and FAT (Figs. 10e–
h). First, the simulated winds at SJV are stronger at
some times and weaker at other times than is observed
during the daytime; at FAT, the simulated winds are
generally weaker during the morning and most of the
day. Second, the upper-level flow (⬎800 m) in the simulation is overall faster than is observed during both day
and night at both sites.
6) SUMMARY OF THE LOW-LEVEL
IN THE CENTRAL VALLEY
WIND SYSTEMS
In the previous sections, analysis and comparisons of
the observed and simulated low-level winds show that
there are five important local wind systems in the Central Valley: 1) incoming low-level marine flow through
the Carquinez Strait into the Sacramento River delta,
2) diurnal cycle of upslope–downslope flows, 3) up- and
down-valley flow in the Sacramento Valley, 4) nocturnal low-level jet in the San Joaquin Valley, and 5) orographically induced mesoscale eddies (the Fresno and
Schultz eddies).
The connection of these aforementioned processes
can be summarized in a conceptualization shown in Fig.
11. During the daytime, the land–sea surface thermal
contrast maintains a low-level inflow of marine air moving through the Carquinez Strait into the Central Valley. Once the marine air enters the Central Valley, it
splits into northward and southward flows because of
the blocking effect of the Sierra Nevada. There is a
diurnal change in the intensity of the incoming marine
flow, leading to the diurnal variation of both the northward and southward flows. During the night, the southward flow experiences a nocturnal acceleration, which
leads to the formation of the nocturnal low-level jet in
the San Joaquin Valley. The Fresno eddy forms when
the low-level jet is forced to veer toward the western
side of the valley by the cross-valley Coriolis force, the
downslope flows on the eastern side of the valley intensify and form a horizontal shear between the jet and
the foothills to the east, and the winds in the southern
San Joaquin Valley are blocked by the Tehachapi
Mountains.
In the Sacramento Valley, the interaction between
the northward marine inflow and the nocturnal upvalley flow often leads to the formation of the cyclonic
Schultz eddy during the night and early morning to the
north and northwest of Sacramento. During the entire
5-day episode, the Sacramento Valley is under the influence of the rotation of up- and down-valley flow
2384
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 47
FIG. 10. Five-day-mean (a) wind profiler–observed and (b) WRF-simulated wind vectors at SJV. Five-day-mean (c) wind profiler–
observed and (d) WRF-simulated wind vectors at FAT. The (e) bias and (f) RMSE of WRF-simulated wind speed (m s⫺1) at SJV. The
(g) bias and (h) RMSE of WRF-simulated wind speed (m s⫺1) at FAT. Again, SR indicates sunrise and SS indicates sunset.
while the overall surface winds in the San Joaquin Valley are predominantly southward. All of these flow processes are important because they interact with each
other to exert temporal and spatial control over the
transport and dispersion of pollutants in the Central
Valley.
4. The observed and simulated hodographs and
trajectories
Comparisons of the observations and the WRF simulation indicate that the WRF model captures the major
characteristics of the observed low-level winds in the
Central Valley, but quantitative differences are noticeable. These differences can, in theory, be attributed to
the errors in the simulated topographically induced local forcing (largely because of the errors in the atmospheric boundary layer physics) and the simulated interaction between the upper-level winds and the aforementioned low-level flows. Given that there are only
wind profiler observations available and there are no
observations for us to verify the model-simulated nearsurface forcing such as the surface momentum and heat
SEPTEMBER 2008
BAO ET AL.
2385
FIG. 11. Conceptualization of the daytime and nighttime low-level wind regimes during the 5-day episode.
fluxes, it is impossible to actually attribute the errors in
the simulated low-level winds to errors in the model
physics associated with topographically induced local
forcing. Thus, in this section, further analysis of the
model errors is carried out using only the wind profiler
observations.
As discussed in section 3, some of the quantitative
differences in the low-level winds between the observations and the simulation may arise from the differences in the observed and simulated upper-level winds
that are mainly controlled by the upper-level, largescale forcing. Although it is not always possible to
separate the large-scale forcing from the local forcing, it
is useful to use hodograph analysis to see the relative
impact of the differences in the large-scale forcing
versus local forcing on the observed and simulated diurnal cycles of the low-level winds in the Central Valley. Because the low-level large-scale pressure gradient
is strongly affected by the upper-level pressure gradient
and because the displacement of the hodograph center is controlled by the low-level, large-scale pressure gradient, the hodograph analysis can be used to
assess the impact of the upper-level, large-scale pressure gradient. It is also worth examining the differences in terms of transport processes in the Central
Valley. In the following sections, comparisons of
the observed and simulated low-level winds are performed using both hodograph analysis and trajectory
analysis.
a. Hodographs
Figure 12 shows the comparisons of the 5-day-mean
observed and simulated diurnal cycles of the low-level
winds, which are averaged within the lowest 500 m
AGL at several sites in the Central Valley, using hodographs. The vector RMSE is also labeled in each panel
to provide a measure of the differences between the
observed and simulated hodographs. It is clearly seen
that there are differences in the diurnal evolution of
the low-level winds. At RDG, the observed winds
are southerly during the evening (from 0100 to 0400
UTC), whereas the WRF-simulated winds are westerly,
suggesting a wind bias (Fig. 12a). The hodograph of the
observed incoming marine flow, represented by the
winds at LVR (Fig. 12b), has a greater enclosed area
than does the simulated hodograph. Also note that
the observed hodograph at LVR rotates counterclockwise throughout the entire day, whereas the WRFsimulated hodograph at LVR rotates clockwise from
2000 to 0900 UTC and has no rotation between 0600
and 1900 UTC. At SAC, the observed background
flow on which the diurnal cycle is superposed is westerly to southwesterly, but the simulated background
flow is northwesterly (Fig. 12c). Furthermore, the observed flow is weaker than the WRF-simulated flow.
However, both the WRF simulation and the observations at SAC have the maximum wind speed at 0300
UTC. In the San Joaquin Valley, as shown by the
2386
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 47
FIG. 12. Wind profiler–observed and simulated 5-day-mean
hodographs averaged from the surface to 500 m AGL at (a) RDG,
(b) LVR, (c) SAC, (d) WFD, and (e) LEM. Blue lines are the
observed hodographs, and red lines are hodographs from the
WRF simulation. Every third hour on each curve is labeled with
the time (UTC). Vector RMSE at each location is indicated by the
RMSE number.
hodograph at WFD (Fig. 12d), the WRF-simulated
nocturnal jet is stronger than the observed one in the
San Joaquin Valley. In addition, the rotation direction of the simulated hodograph at WFD is in agree-
ment with the observations, but the simulated diurnal cycle lags the observed one by about 2 h. Both
the observed and WRF-simulated hodographs at
LEM (Fig. 12e) have a varying vector rotation, al-
SEPTEMBER 2008
BAO ET AL.
though the timing of the vector rotation is different by
a few hours.
The most important result of the hodograph comparisons is the variation in the wind vector rotation and
wind speed at various locations in the Central Valley,
which corresponds to the different magnitudes of the
vector RMSE at individual sites. The differences in the
rotation of the observed and simulated hodographs can
be explained by the errors in the simulated large-scale
pressure gradient, large-scale flow, and local topographically induced pressure perturbation. The mechanism for the diurnal rotation of the low-level winds in
the Central Valley is the same as in a sea breeze. Haurwitz’s (1947) theory predicts that the winds on a flat
plane driven by a surface thermal gradient, which
changes sign diurnally, rotate clockwise in the Northern
Hemisphere. As shown in theoretical and modeling
studies by Kusuda and Alpert (1983), Kusuda and Abe
(1989), and Steyn and Kallos (1992), the imbalance between the local topographically induced pressure perturbation and the background pressure gradient is responsible for the lack of the clockwise (in the Northern
Hemisphere) wind rotation or for the counterclockwise
wind rotation in locations with significantly complex
topography and surface thermal contrast (such as in
parts of the Central Valley). It is also reasonable to
assume that a similar imbalance in pressure forcing can
also be responsible for an abrupt change, rather than
gradual rotation, of wind direction.
The displacement of the WRF-simulated and observed hodograph centers at some of the locations in
the Central Valley, as shown in Fig. 12 (see Figs. 12c
and 12d in particular), is further evidence that the cause
of the overall errors in the simulated low-level winds
does not lie only in the errors of the model’s local thermal forcing. Previous analysis of the dynamics of diurnally varying local winds (e.g., Rotunno 1983; Bao et al.
2005) can be used to show that an error in the local
thermal forcing does not displace the center of the
hodograph; it only changes the area that the hodograph
curve encloses. The errors in the large-scale pressure
gradient and winds are directly responsible for the displacement of the simulated hodograph centers relative
to the observations. Therefore, the displacement between the observed and simulated hodographs indicates that there are errors in the model-simulated upper-level winds, which are mostly governed by the
simulated large-scale forcing.
b. Interaction between the lower-level and
upper-level winds
Previous studies (e.g., Pielke and Segal 1986; Banta
et al. 2004) have shown that the upper-level pressure
2387
gradient and winds on the larger scale closely interact
with the evolution of the lower-level winds within a
mountain basin complex. Those studies are consistent
with the discussion in section 4a on hodograph comparisons, in which some of the differences between
the simulated and observed low-level winds, in particular the discrepancy between the up- and down-valley
winds in the Sacramento Valley and the nocturnal
low-level jet in the San Joaquin Valley, can be attributed in part to the differences in the simulated and
observed synoptic and mesoscale upper-level winds resulting from the intrinsic dynamical link between the
low-level and upper-level winds. Examples of these differences are shown in Fig. 13, which illustrates the
simulated winds and geopotential height at 700 hPa and
the counterparts from the NCEP Eta analysis at 0000
UTC 1 and 2 August 2000. In comparing Figs. 13a and
13b, it is obvious that on 1 August 2000 the major differences between the analysis and simulation are in the
location of the large-scale trough just offshore and the
displacement of the high to the southeast of the trough.
On the next day, the analysis (Fig. 13c) indicates that
while the San Joaquin Valley is still under the influence
of the upper-level high the large-scale trough to the
north of the Sacramento Valley has intensified and is
approaching the Sacramento Valley. In contrast, although the simulated large-scale trough is approximately along the northwest coast of California, its orientation is different from the analysis, so that the simulated 700-hPa flow over the Central Valley is more
westerly (Fig. 13d). It is these types of differences on
the larger scales that can contribute significantly to the
differences between the observed and simulated local
low-level winds that are evident at many locations
within the Central Valley and that are discussed in the
previous sections.
One way to confirm whether the upper-level pressure
gradient and winds on the larger scale contribute significantly to the evolution of the low-level winds is to
compare the sensitivities of the simulated low-level
winds to the uncertainties in the lateral boundary and
surface forcing, assuming that the uncertainties in the
atmospheric forcing embedded in the lateral boundary
conditions and the soil initialization can be well approximated by two operational analyses. As shown in a
companion study by Michelson and Bao (2008), the
sensitivity analysis for this case indicates that the WRFsimulated low-level winds in the Sacramento Valley are
more sensitive to atmospheric forcing than to soil initialization. The simulated low-level winds in the southernmost part of the San Joaquin Valley are more sensitive to the soil initialization than they are in the northern San Joaquin Valley. In the mid-Central Valley,
2388
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
VOLUME 47
FIG. 13. The 700-hPa winds and height (m) at 0000 UTC 1 Aug from (a) the NCEP Eta analysis and (b) the WRF simulation. (c),
(d) As in (a) and (b), but for 0000 UTC 2 Aug.
where the winds are more directly affected by the incoming marine flow, the winds are overall more sensitive to the atmospheric forcing than to the soil initialization. However, there is more sensitivity to the
soil initialization in this area than in the Sacramento
Valley. Because the change in atmospheric forcing affects the WRF simulation through upper-level forcing
on larger scales through the lateral boundary conditions, the results of this study indicate the large impact of the upper-level winds on the low-level winds,
which is not uncommon in the Central Valley during
winter (e.g., Pauley et al. 1996). To our knowledge,
such an impact in the summertime has hardly been
studied for air-quality applications. Although the details of the interaction belong to future studies, we wish
to make the point that, because of the interaction, the
errors in the model-simulated low-level winds are related not only to the errors in the model’s surface conditions but also to the errors in the large-scale, upperlevel forcing.
SEPTEMBER 2008
BAO ET AL.
c. The trajectory analysis
The differences between the observed and simulated
low-level winds can also be illustrated in terms of transport and dispersion by using a trajectory model. The
trajectory model used in this study, which was originally
developed by A. B. White (2006, personal communication), takes as input the measured or simulated winds
at all 25 wind profiler sites located in California during CCOS to calculate forward or backward particle
trajectories. A simple advection scheme is used in this
trajectory model for calculating the positions of air parcels using estimated wind speed and direction for the
time period prior to the ending time. The wind distribution required for advection calculation is generated
by applying objective analysis on wind profiler measurements. Despite the simplicity in the trajectory calculation, such a trajectory model has been applied successfully for analyzing source regions for transportrelated pollution phenomena in which observed winds
are used to track the path of parcels of air arriving at a
particular monitoring site over a period of hours or
days.
To be consistent, the comparison of simulated trajectories with those derived from the wind profiler measurements is made in such a way that only the simulated
horizontal winds extracted at the wind profiler sites are
used in the trajectory model. The trajectory model provides an effective alternative to depicting the sensitivity
in terms of transport and dispersion. By examining
dominant air trajectories during the day and night preceding the measurements of the worst air-quality conditions in the Central Valley during the 5-day period,
source regions influencing those conditions associated
with both observed and simulated winds can be identified.
Figures 14 and 15 show the results of 24-h forwardtrajectory analyses in which air parcels are released at
LVR, RDG, SAC, and SVS (representative of the flow
conditions in the San Francisco Bay Area, the northern
Sacramento Valley, the southern Sacramento Valley,
and the San Joaquin Valley, respectively) at 0000 and
1200 UTC 31 July 2000. It is seen that, although the
WRF simulation produces the overall qualitative characteristics of the observed low-level winds associated
with the dominant airflows in the Central Valley well,
quantitative differences between the simulated and observed flows are clearly shown in the differences of the
locations of each trajectory.
The 24-h trajectories released in the lower layers at
LVR at 0000 UTC 31 July 2000 in the WRF simulation
are in better agreement with the observations (the red
and black trajectories in Figs. 14a and 14b) than those
2389
released above 1 km. The trajectories in the lowest layers (the red trajectories in Figs. 14a and 14b) indicate
that the incoming flow is stronger in the WRF simulation [which has been seen in the discussions in sections
3c(1) and 4a]. In the higher layers, there is a definite
bias in the WRF simulation, as both the two higherlayer trajectories (the blue and green trajectories) in
the WRF simulation are farther east than the observed
trajectories. These differences in the higher-layer trajectories suggest that the errors in the large scale (and
at upper levels) can impact the dispersion and transport
into the Central Valley. The 700-hPa heights and winds
shown in Fig. 13 indicate that the high in the WRF
simulation is farther to the south than in the analysis,
which means that winds at 700 hPa are more westerly in
the WRF simulation than is observed, and thus the
higher-layer trajectories move more toward the east
than do the observed trajectories. All of these indicate
that differences between the observations and the WRF
simulation in the large scale affect the transport into the
Central Valley from the LVR site.
The agreement between the WRF-simulated trajectories and the observed trajectories is better (Figs. 15a
and 15b) for the forward trajectories released for 24 h
at LVR at 1200 UTC 31 July 2000 than when the trajectories are released at 0000 UTC 31 July 2000. However, as in the case with the trajectories released at 0000
UTC 31 July 2000, the trajectory released at the highest
layer in the WRF simulation (blue trajectory) ends up
noticeably to the east of the observed trajectory. Again,
this difference is indicative of the large-scale errors in
the WRF simulation.
In the Sacramento Valley, as seen in the daily-averaged profile of the winds (Fig. 7) and the hodograph at
RDG (Fig. 12a), the low-level flow in the observations
at RDG is bidirectional, whereas the switch between up
valley and down valley is more gradual in the WRF
simulation. The observed trajectories released at 0000
UTC 31 July 2000 (Figs. 14c and 14d) initially indicate
up-valley flow, which is typical for this time of day.
With the onset of the down-valley flow later in the
evening, all of the observed trajectories then turn down
valley. The trajectories in the WRF simulation released
at RDG at lower layers exhibit the same behavior as
the observations; that is, they initially move up valley
and then turn down valley. However, the trajectories at
the higher layers do not turn down valley, as seen in the
observations. The differences between the observed
and WRF-simulated trajectories aloft are indicative of
the errors of the WRF simulation at upper levels and on
the large scale.
The observed trajectories released from the RDG
2390
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
FIG. 14. Twenty-four-hour forward trajectories released at 0000 UTC 31 July
2000 within four different layers: 200–600 m MSL (red trajectories), 600–1000 m
MSL (black trajectories), 1000–1400 m MSL (green trajectories), and 1400–1800 m
MSL (blue trajectories). (a) Observed and (b) WRF-simulated trajectories released from LVR. (c) Observed and (d) WRF-simulated trajectories released from
RDG. (e) Observed and (f) WRF-simulated trajectories released from SAC. (g)
Observed and (h) WRF-simulated trajectories released from SVS.
VOLUME 47
SEPTEMBER 2008
BAO ET AL.
FIG. 15. As in Fig. 14 except that the trajectories are released at 1200 UTC 31
Jul 2000.
2391
2392
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
(Figs. 15c and 15d) site in the morning (1200 UTC 31
July 2000) move down valley the entire 24 h, with a
noticeable shift from cyclonic curvature to anticyclonic
curvature in all four trajectories (although the 600–
1000-m black trajectory has less curvature than the
other trajectories). In the WRF simulation, the lowestlayer trajectory (red) most closely follows the observations. The highest-layer trajectory (blue) and the second lowest-layer trajectory (black) appear to move far
enough to the east that they undergo the diurnal change
from upslope to downslope flow along the slopes of the
Sierra Nevada.
The observed 24-h forward trajectories released from
SAC at 0000 UTC 31 July 2000 (Fig. 14e) show the
incoming marine flow and veering into the San Joaquin
Valley in all four layers. The trajectories from the WRF
simulation (Fig. 14f) at the lowest layers (red, black,
and green trajectories) agree well with the observations, although the fast bias is evident. The WRFsimulated trajectory at the highest layer (blue trajectory), as at the LVR site, moves toward the Sierra Nevada, as opposed to moving southward, up the San
Joaquin Valley, which is what occurs in the observations. Again, as at the LVR site, this difference at the
highest layer is indicative of the large-scale errors in the
WRF simulation. In a similar way, the 24-h forward
trajectories released at 1200 UTC 31 July 2000 from
SAC (Figs. 15e and 15f) also indicate a direction bias at
the highest layer and a speed bias at the lower layers.
The locations of all four of the observed and WRFsimulated 24-h forward trajectories released at SVS at
0000 UTC 31 July (Figs. 14g and 15h) are in fairly good
agreement; however, the WRF simulated flow is
slightly faster than the observed flow. The 24-h forward
trajectories released 12 h later (Figs. 15g and 15h) from
SVS indicate that the highest-layer trajectory (blue) in
the WRF simulation is farther to the east than is observed and that the flow in the WRF simulation is faster
than the observed.
Overall, the simulated paths of the trajectories are
longer and are more spread out than the paths of the
observed trajectories—in particular, in the San Joaquin
Valley (characterized by the parcels released at SAC).
At most of the locations at which trajectories are released, there are more differences aloft than at lower
levels. This is consistent with the major quantitative
differences between the observed and simulated lowlevel winds in the Central Valley that are discussed in
the previous section; that is, the incoming flow and the
nocturnal low-level jet are faster in the WRF simulation
than in the observations, and there are large-scale errors/biases in the WRF simulation.
VOLUME 47
5. Summary and conclusions
To understand better the transport and dispersion
processes in California’s Central Valley, the 29 July–3
August 2000 case is investigated using observations collected during the CCOS field experiment and the advanced research version of the WRF model. The choice
of this particular episode is motivated by the fact that
the meteorological conditions associated with it are
typically in favor of trapping surface pollutants (such as
ozone) in the Central Valley. The investigation is focused on the agreement, as well as the differences, between the observed and simulated diurnal variation of
local winds in terms of 5-day-mean winds within the
lowest 500 m AGL. The comparisons of the observed
and simulated low-level winds are made with respect to
the local wind systems identified from the observations.
Analyses of hodographs and transport trajectories are
performed to help to identify possible error sources of
the simulated low-level winds.
Two major conclusions are drawn from this investigation. First, both the observed and simulated winds
show that there are five important circulation processes
affecting the low-level winds in the Central Valley: 1)
the incoming low-level marine flow through the Carquinez Strait in the Sacramento River delta, 2) the diurnal cycle of upslope–downslope flows, 3) the up- and
down-valley flow in the Sacramento Valley, 4) the nocturnal low-level jet in the San Joaquin Valley, and 5)
the orographically induced mesoscale eddies (the
Fresno and Schultz eddies). More important, the comparisons of the WRF-simulated low-level winds with
the observations indicates that overall the WRF is capable of realistically reproducing all of the observed
mean characteristics of the low-level winds and transport processes in the Central Valley.
Second, the quantitative differences between the
simulated and observed low-level winds at the wind
profiler sites vary from one area to another in the Central Valley. Overall, the simulated incoming marine
flow is slower during the day and faster during the night
and early morning than is observed. The transition between the simulated upslope and downslope flows
along the foothills of the Central Valley is not as abrupt
as shown in the observations. The depth of the simulated upslope and downslope flows also appears to be
shallower than is observed. The intensity of the simulated up- and down-valley flows in the Sacramento Valley is weaker, the duration of the up-valley flow is
shorter, and the diurnal transition between up- and
down-valley flows is more gradual than is observed.
The simulated nocturnal low-level jet in the San
Joaquin Valley is faster than is observed. The simulated
SEPTEMBER 2008
BAO ET AL.
Schultz eddy is farther north than is the observed,
whereas the simulated Fresno eddy is shallower and
farther to the east than is observed. A close examination of the differences between the simulated and observed winds using the hodograph and trajectory analysis indicates that the upper-level winds affect the lowlevel winds because of the intrinsic dynamic link
between them. It is strongly suggested that the errors in
the model-simulated low-level winds result from the
errors in the model’s surface conditions as well as the
errors in the simulated upper-level forcing. Because the
low-level winds are so crucial in numerical modeling for
air-quality applications—they control the transport and
dispersion of anthropogenic pollutants from the surface—understanding the interaction between the upper
and lower levels and how errors on these scales affect
the low-level flow is important. However, it is difficult
to quantitatively diagnose and partition these errors using the available observational datasets. That is why
methods such as the sensitivity study done by Michelson and Bao (2008) can be vital in understanding why
the upper-level large-scale forcing plays an important
role in controlling the low-level winds in the Central
Valley of California.
Note that although fully understanding the interaction between the errors in the large-scale, upper-level
winds and those in the local-scale, low-level winds remains a research subject, the so-called four-dimensional data assimilation (FDDA) technique can be
practically used to reduce the errors in the simulated
low-level winds where level-wind observations are
available. A separate study (P. O. G. Perrson et al.
2008, unpublished manuscript) of the same case reported here shows that FDDA is effective in significantly improving the accuracy of the simulated lowlevel winds.
Acknowledgments. The authors gratefully thank
A. B. White for providing the software for the trajectory analysis and three anonymous reviewers for their
constructive comments.
REFERENCES
Angevine, W. M., A. B. White, and S. K. Avery, 1994: Boundarylayer depth and entrainment zone characterization with a
boundary-layer profiler. Bound.-Layer Meteor., 68, 375–385.
Banta, R. M., L. S. Darby, J. D. Fast, J. Pinto, C. D. Whiteman,
W. J. Shaw, and B. D. Orr, 2004: Nocturnal low-level jet in a
mountain basin complex. Part I: Evolution and effects on
local flows. J. Appl. Meteor., 43, 1348–1365.
Bao, J.-W., S. A. Michelson, S. A. McKeen, and G. A. Grell, 2005:
Meteorological evaluation of a weather-chemistry forecasting
model using observations from the Texas AQS 2000 field
2393
experiment. J. Geophys. Res., 110, D21105, doi:10.1029/
2004JD005024.
Bianco, L., and J. M. Wilczak, 2002: Convective boundary layer
depth: Improved measurement by Doppler radar wind profiler using fuzzy logic methods. J. Atmos. Oceanic Technol.,
19, 1745–1758.
Blackadar, A. K., 1957: Boundary layer wind maxima and their
significance for the growth of nocturnal inversions. Bull.
Amer. Meteor. Soc., 38, 283–290.
Burk, S. D., and W. T. Thompson, 1996: The summertime lowlevel jet and marine boundary layer structure along the California coast. Mon. Wea. Rev., 124, 668–686.
Egger, J., 1990: Thermally forced flows: Theory. Atmospheric Processes over Complex Terrain, Meteor. Monogr., No. 45, Amer.
Meteor. Soc., 43–58.
Haurwitz, B., 1947: Comments on the sea-breeze circulation. J.
Meteor., 4, 1–8.
Kusuda, M., and P. Alpert, 1983: Anti-clockwise rotation of the
wind hodograph. Part I: Theoretical study. J. Atmos. Sci., 40,
487–499.
——, and N. Abe, 1989: The contribution of horizontal advection
to the diurnal variation of the wind direction of land–sea
breezes: Theory and observations. J. Meteor. Soc. Japan, 67,
177–184.
Lin, Y. L., and I. C. Jao, 1995: A numerical study of flow circulations in the Central Valley of California and formation
mechanisms of the Fresno eddy. Mon. Wea. Rev., 123, 3227–
3239.
Michelson, S. A., and J.-W. Bao, 2008: Sensitivity of low-level
winds simulated by the WRF model in California’s Central
Valley to uncertainities in the large-scale forcing and soil
initialization. J. Appl. Meteor. Climatol., in press.
Niccum, E. M., D. E. Lehrman, and W. R. Knuth, 1995: The influence of meteorology on the air quality in the San Luis
Obispo County–southwestern San Joaquin Valley region for
3–6 August 1990. J. Appl. Meteor., 34, 1834–1847.
Pauley, P. M., N. L. Baker, and E. H. Barker, 1996: An observational study of the “Interstate 5” dust storm case. Bull. Amer.
Meteor. Soc., 77, 693–720.
Pielke, R. A., and M. Segal, 1986: Mesoscale circulations forced
by differential terrain heating. Mesoscale Meteorology and
Forecasting, P. S. Ray, Ed., Amer. Meteor. Soc., 516–548.
Rife, D. L., and C. A. Davis, 2005: Verification of temporal variations in mesoscale numerical wind forecasts. Mon. Wea. Rev.,
133, 3368–3381.
Rotunno, R., 1983: On the linear theory of the land and sea
breeze. J. Atmos. Sci., 40, 1999–2009.
Seaman, N., D. R. Stauffer, and A. M. Lario-Gibbs, 1995: A multiscale four-dimensional data assimilation system applied in
the San Joaquin Valley during SARMAP. Part I: Modeling
design and basic performance characteristics. J. Appl. Meteor., 34, 1739–1761.
Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M.
Barker, W. Wang, and J. G. Powers, 2005: A description of
the Advanced Research WRF version 2. NCAR Tech. Note
NCAR/TN-468⫹STR, 88 pp. [Available from UCAR Communications, P.O. Box 3000, Boulder, CO 80307.]
Stauffer, D. R., N. L. Seaman, G. K. Hunter, S. M. Leidner, A. M.
Lario-Gibbs, and S. Tanrikulu, 2000: A field-coherence technique for meteorological field-program design for air quality
studies. Part I: Description and interpretation. J. Appl. Meteor., 39, 297–316.
2394
JOURNAL OF APPLIED METEOROLOGY AND CLIMATOLOGY
Steyn, D. G., and G. Kallos, 1992: A study of the dynamics of
hodograph rotation in the sea breezes of Attica, Greece.
Bound.-Layer Meteor., 58, 215–228.
Tanrikulu, S., D. R. Stauffer, N. L. Seaman, and A. J. Ranzieri,
2000: A field-coherence technique for meteorological
field-program design for air quality studies. Part II: Evaluation in the San Joaquin Valley. J. Appl. Meteor., 39, 317–
334.
VOLUME 47
White, A. B., 1993: Mixing depth detection using 915-MHz radar
reflectivity data. Preprints, Eighth Symp. on Observations
and Instrumentation, Anaheim, CA, Amer. Meteor. Soc.,
248–250.
Zhong, S. Y., C. D. Whiteman, and X. D. Bian, 2004: Diurnal evolution of three-dimensional wind and temperature structure
in California’s Central Valley. J. Appl. Meteor., 43, 1679–
1699.
Download