Evaluation of WRF and HadRM Mesoscale Climate Simulations over the United States
Pacific Northwest
Yongxin Zhang, Valérie Dulière, Philip Mote, and Eric P. Salathé Jr.
Climate Impacts Group, Joint Institute for the Study of the Atmosphere and Ocean,
University of Washington, Seattle, U.S.A.
Abstract
This work compares the WRF (Weather Research and Forecasting) simulations at 36km and 12-km resolution and HadRM (Hadley Center Regional Model) simulations at
25-km resolution with the observed daily maximum and minimum temperature (Tmax
and Tmin) and precipitation at HCN (Historical Climate Network) stations over the
United States Pacific Northwest for 2003. The WRF runs were driven by the
NCEP/NCAR reanalysis data while the HadRM runs were drive by the NCEP-DOE
(Department of Energy) reanalysis data. The Tmax simulated by WRF and HadRM
compares well with the observations. Mainly warm biases of Tmax are noted in the WRF
simulations along the coast of Oregon and Washington with cold biases in the interior
during autumn and winter. The HadRM simulations show mainly warm biases in the
interior during the same time period. During summer, the distribution of Tmax biases
shows large variability while during spring, cold biases of Tmax are dominant in both
model simulations. The simulated Tmin compares reasonably well with the observations
although not as well as Tmax for both models. Warm biases of Tmin prevail in both
model simulations when averaged both annually and seasonally. The model biases of
Tmax come mainly from the large-scale driving data while the model biases of Tmin
originate mainly from the regional models. The temporal correlation between simulated
and observed daily precipitation is relatively low. However, the correlation increases
steadily for longer averaging times. Large improvement in the temporal correlation of
precipitation is evident in the 12-km WRF simulations when compared to the 36-km
WRF simulations and 25-km HadRM simulations. Model overestimation of the observed
precipitation is indicated in both model simulations when averaged annually. A large
variability in the spatial distributions of the normalized seasonal biases of precipitation is
evident in both model simulations. Differences in the simulated temperature and
precipitation between the two models can be largely attributed to differences in the largescale driving data.
1. Introduction
The United States Pacific Northwest is characterized by mountainous terrain and
intricate land-sea contrasts (Fig. 1) resulting in a host of fine-scale weather systems such
as sea and land breezes, rain shadows, and downslope windstorms, that define the local
weather and climate (Mass 2008). In a warming climate, such fine-scale weather systems
can significantly alter the local temperature and precipitation trends (Salathé et al. 2008)
and are essential to consider in climate simulations and climate change assessment at
regional and local scales. Global models are generally able to resolve the large-scale
weather systems that affect the Pacific Northwest but not the fine-scale processes
associated with the local terrain. To capture these smaller features, a more realistic
representation of the local complex terrain and the heterogeneous land surfaces is needed
(Mass et al. 2002; Salathé et al. 2008). Therefore, the use of limited-area regional climate
models with horizontal resolutions on the order of tens of kilometers is crucial.
Recently, there have been increasing efforts over the Pacific Northwest in using
limited-area mesoscale models for downscaling reanalysis data (Leung et al. 2003a, b)
and global climate model simulations (Salathé et al 2008) to study regional climate and
climate change. Much of the work has been involved with the fifth-generation
Pennsylvania State University – National Center for Atmospheric Research (NCAR)
Mesoscale Model (MM5) (Grell et al. 1993). As NCAR has been phasing out MM5 and
has released the state-of-the-art, next-generation Weather Research and Forecasting
(WRF) model (http://www.wrf-model.org/index.php), it is timely to switch from the
MM5-based to the WRF-based mesoscale climate modeling and examine its performance
over the Pacific Northwest. HadRM (Hadley Center Regional Model) is another limitedarea regional climate model widely used worldwide as part of the PRECIS (Providing
REgional Climates for Impacts Studies) system, which was developed at the Hadley
Center of the United Kingdom Met Office. The PRECIS system can be easily applied to
any area of the globe to generate detailed climate change projections. Both the WRF and
HadRM models are coupled land-atmosphere modeling systems that can be used to add
fine-scale information to the large-scale projections of global climate models. What
differs between these two models is that WRF, like MM5, was designed for short-term
weather forecasts and refinements have to be applied in the model in order to perform
long-term climate simulations (Salathé et al. 2008).
In this study, we apply the WRF-based and HadRM modeling systems over the Pacific
Northwest in downscaling reanalysis data (Kalnay et al. 1996; Kanamitsu et al. 2002) for
2003. The reanalysis data incorporate all the available observations at the time of
processing in order to best represent the 6-hourly large-scale state of the atmosphere.
Reanalysis fields are particularly suitable for driving regional climate models and for
model validation. Here, we present validation results based on comparisons to station
observations over the Pacific Northwest from the Historical Climate Network (HCN;
Karl et al. 1990). Our purpose is to examine the models performance in reproducing
station observations at various time scales. This work is organized in the following.
Section 2 contains a brief description of the models. Experimental design is discussed in
section 3. Section 4 compares model simulations with the observations. Major
conclusions and discussions are presented in section 5.
2. Model Description
2.1 WRF Model
The WRF model is a state-of-the-art, next-generation mesoscale numerical weather
prediction system designed to serve both operational forecasting and atmospheric
research needs (http://www.wrf-model.org). It is a non-hydrostatic model, with several
available dynamic cores as well as many different choices for physical parameterizations
suitable for a broad spectrum of applications across scales ranging from meters to
thousands of kilometers. The dynamic cores in WRF include a fully mass- and scalarconserving flux form mass coordinate version. The physics package includes
microphysics, cumulus parameterization, planetary boundary layer (PBL), land surface
models (LSM), longwave and shortwave radiation (Skamarock et al. 2006).
In this work, the microphysics and convective parameterizations used were the WRF
Single-Moment 5-class (WSM5) scheme (Hong et al. 2004) and the Kain-Fritsch scheme
(Kain and Fritsch 1993), respectively. The WSM5 microphysics explicitly resolves water
vapor, cloud water, rain, cloud ice, and snow. The Kain-Fritsch convective
parameterization utilizes a simple cloud model with moist updrafts and downdrafts that
includes the effects of detrainment and entrainment. The land-surface model used was the
NOAH (National Centers for Environmental Prediction - NCEP, Oregon State
University, Air Force, and Hydrologic Research Lab) LSM 4-layer soil temperature and
moisture model with canopy moisture and snow cover prediction (Chen and Dudhia
2001). The LSM includes root zone, evapotranspiration, soil drainage, and runoff, taking
into account vegetation categories, monthly vegetation fraction, and soil texture. The
PBL parameterization used was the YSU (Yonsei University) scheme (Hong and Pan
1996). This scheme includes counter-gradient terms to represent heat and moisture fluxes
due to both local and non-local gradients. Atmospheric shortwave and longwave
radiations were computed by the NCAR CAM (Community Atmospheric Model)
shortwave scheme and longwave scheme (Collins et al. 2004), respectively.
The design of the WRF-based mesoscale climate modeling over the Pacific Northwest
follows that of the MM5-based mesoscale climate modeling described in Salathé et al.
(2008). Basically, similar refinements as in the MM5 setup are made to the WRF
configuration in order to perform long simulations and fully represent the climate system
response to climate change forcing. Firstly, nudging is applied to the outermost regional
model domain from the forcing fields in order to prevent possible drift of regional model
solution from that of the driving global climate model over long term. Nudging relaxes
the regional model simulations for wind, temperature, and moisture towards the driving
global climate model simulations. The inner nested domains are not nudged, allowing the
mesoscale model to freely develop fine-scale features. Secondly, WRF is modified so that
soil temperatures vary at the lower boundary in accordance with the evolving surface
temperatures predicted by the model.
2.2 HadRM Model
HadRM (Jones et al. 2004) is the third-generation Hadley Center regional climate
model (HadRM3H). It is a limited-area, high-resolution version of the atmospheric
general circulation model HadAM3H which is itself an improved version of the latest
atmospheric component of the atmosphere-ocean coupled general circulation model
(HadCM3; Gordon et al. 2000; Johns et al. 2003). HadRM is a hydrostatic version of the
fully primitive equations. It includes dynamical flow, clouds and precipitation, radiative
processes, land surface and deep soil. An interactive atmospheric sulfur cycle is also
available but was not used in this study.
The latitude-longitude grid is rotated in HadRM so that the equator lies inside the
region of interest to obtain quasi-uniform grid box area over that region. The available
horizontal resolutions are 0.44°0.44° and 0.22°0.22°. For the purpose of this study, we
chose the higher resolution which corresponds to a minimum resolution of ~ 25 km at the
equator of the rotated grid.
HadRM was released as part of the PRECIS package. This package also includes
software to allow display and processing of the model output data
(http://precis.metoffice.com). The PRECIS package is flexible, easy to use and
computationally inexpensive. It can be applied over the U.S. Pacific Northwest to provide
detailed climate information for regional climate studies and climate impact assessment.
3. Experimental Design
WRF was set up by using multiple nests at 108 km, 36 km, and 12 km horizontal grid
spacing (Fig. 1a). The outermost WRF domain covers nearly the entire North American
continent as well as much of the eastern Pacific Ocean and the western Atlantic Ocean.
The use of this large domain ensures that synoptic weather systems approaching the U.S.
are well represented by the time they reach the region. The 36-km model domain covers
the continental U.S. and part of Canada and Mexico. The innermost model domain is
centered on the Pacific Northwest and includes the states of Washington, Oregon, and
Idaho (Fig. 1b). We used 31 vertical levels in the model with the highest resolution (~ 20
– 100 m) in the boundary layer. The model top was fixed at 50 mb. One-way nesting was
applied in this work.
The domain of HadRM (Fig. 1a) was chosen with the highest available horizontal
resolution of ~ 25 km at the equator of the rotated grid. The HadRM model domain
covers a large part of the eastern Pacific Ocean and part of Mexico and Canada to better
resolve the synoptic weather systems that affect the Pacific Northwest. This model
domain encompasses entirely the states of Arizona, California, Idaho, Nevada, Oregon,
Utah and Washington. There are 19 vertical hybrid levels in HadRM spanning from the
surface to 0.5 mb.
The WRF and HadRM runs were initialized at 0000 UTC December 1, 2002 and ended
at 0000 UTC January 1, 2004. The first one-month simulations by WRF and HadRM
were regarded as model spin-up. The initial and lateral boundary conditions were
interpolated from the NCEP/NCAR Reanalysis Project (NNRP; R1) data (Kalnay et al.
1996) for WRF and from the NCEP-DOE (Department of Energy) AMIP-II
(Atmospheric Model Intercomparison Project) Reanalysis (R-2) data (Kanamitsu et al.
2002) for HadRM. The lateral boundary conditions were updated every six hours for both
models. SST was updated every six hours in WRF using the RTG_SST (Real-Time,
Global, Sea Surface Temperature) analysis (ftp://polar.ncep.noaa.gov/pub/history/sst)
developed and archived at NCEP. In HadRM, SST was taken from a combination of the
monthly HadISST (Hadley Center’s sea ice and sea surface datasets;
http://badc.nerc.ac.uk/data/hadisst) and weekly NCEP observed datasets
(http://www.cdc.noaa.gov/cdc/reanalysis/reanalysis.shtml). The simulations from both
WRF and HadRM models were output every hour.
The R-2 data are an updated and processing error-fixed version of the NNRP data
(Kanamitsu et al. 2002). The R-2 data have been shown to improve soil moisture, snow
cover and radiation fluxes over ocean when compared to the NNRP data. Over the Pacific
Northwest, the R-2 data are about 0.2 ~ 0.9°C warmer and about 5% wetter than the
NNRP data. In terms of other surface fields and upper-air fields (not shown) the
differences between the R-2 and NNRP data sets are negligible.
4. Results
In this section, model simulations from WRF and HadRM are compared with
observations at 63 HCN stations in the states of Washington, Oregon and Idaho. We
select only the stations from which 80% or more of the daily precipitation and
temperature measurements are available during the year 2003 and the stations whose
corresponding model grid points are land grid points in the WRF and HadRM model
domains. The locations of these HCN stations are indicated in Fig. 1b.
Comparisons will focus on daily maximum and minimum temperatures (Tmax and
Tmin) and precipitation with averages over various time scales for the nested WRF and
HadRM domains. The daily maximum and minimum temperatures are found from the
hourly temperature simulations. A lapse rate correction of 6.5°C/km was applied to the
simulated temperature to account for the differences in temperature between the elevation
of the model grid point and the station elevation. No lapse rate was applied to
precipitation as a lapse rate over complex terrain would depend on several factors such as
mountain width, buoyancy and moisture fields (Smith and Barstad 2004) as well as on
winds (Esteban and Chen 2008). Comparisons between the nested WRF domains and
HadRM domain are also presented for daily temperatures and precipitation.
4.1 Maximum Temperature
Figure 2 shows scatter plots of the observed and simulated daily Tmax at all HCN
stations combined and all days of the year. Correlation coefficients between the
observations and simulations are listed in Table 1. Both WRF and HadRM simulate the
observed Tmax well. High correlation coefficients between the observations and
simulations are noted (0.92 for WRF Domain 2 and 0.93 for both WRF Domain 3 and
HadRM Domain). The simulated temperature at the HCN grids is virtually identical for
the two nested WRF domains, with a correlation coefficient of 0.99 and it is 0.93 between
the nested WRF domains and the HadRM domain. One intriguing phenomenon
associated with the HadRM simulations is the parabolic tendency in the scatter plots
especially for Tmax smaller than 20°C (Figs. 2c, e and f). This appears to be inherited
from the large-scale forcing as can be inferred from the difference plot in monthly mean
temperatures averaged over the U.S. Pacific Northwest domain between the two driving
data sets (R2 – NNRP; Fig. 3). This plot shows higher Tmax in R2 than in NNRP during
winter months with relatively small differences between the two driving datasets in the
other months. We will also show later in the model biases of seasonal mean Tmax that
warmer biases are noted in the HadRM simulated Tmax during winter when compared to
the WRF simulations.
Annual mean model biases of Tmax at HCN stations are presented in Fig. 4. For both
WRF Domain 2 and Domain 3, similar but small biases on the order of -1 ~ 1ºC are
identified along the coast of the Pacific Northwest while cold biases around 3ºC are noted
in the interior. It is possible that these cold biases are partially inherited from the largescale driving data and partially related to the modeled terrain effect as eastward moving
air masses are subject to large modification by the Cascade and Rocky Mountains. Cold
biases fluctuating around the mean of -1.9ºC are also noted in high-resolution MM5
simulations over the Pacific Northwest driven by the NNRP data (Salathé et al. 2008).
The annual mean biases of Tmax in the HadRM simulations range predominantly
between -1 ~ 1ºC at HCN stations and are smaller in magnitude when compared to the
WRF simulations. As mentioned previously, the R-2 data that drive HadRM are about 0.2
~ 0.9°C warmer than the NNRP data (see Fig. 3). This difference in large-scale driving
data may account for the smaller magnitude of the Tmax biases in the HadRM
simulations when compared to the WRF simulations.
Figure 5 shows scatter plots of the observed and simulated monthly mean Tmax at all
HCN stations combined. Correlation coefficients are listed in Table 1. Nearly a perfect
relationship is noted in Fig. 5 between the observations and model simulations with a
correlation coefficient of 0.97 for both models. This is an appreciable improvement of the
model performance over the daily time scale (see Figs. 5 and 2 and Table 1). Table 1 also
shows the correlation coefficients of 10-day mean Tmax between the observations and
model simulations. The correlation coefficients change little from the 10-day mean to the
monthly mean for Tmax.
Figure 6 shows seasonal mean model biases of Tmax at HCN stations. Model biases
from both models are generally smaller than 5ºC in magnitude for all seasons. During
winter and autumn, the nested WRF domains exhibit predominantly warm biases along
the coast of Oregon and Washington and cold biases in the interior. During spring,
virtually all stations show cold biases in the nested WRF domains. Summer is
characterized by mainly cold biases in the nested WRF simulations with scattered warm
biases noted along the coast of Oregon and Washington and the southern part of Idaho.
For HadRM, the distributions of model biases appear similar to those of WRF except in
the interior where mainly warm biases are noted in winter, summer and autumn.
Similar seasonal distributions in the model biases of Tmax between WRF and HadRM
seem to suggest that the large-scale driving data play a role in causing the distributions.
This is supported by examining the model biases of Tmax in the WRF Domain 1 at 108km resolution that exhibit similar distributions to the nested WRF domains and HadRM
domain although in winter the WRF Domain 1 shows large cold biases at more stations
than the other domains (Fig. 6). In our model runs nudging is applied throughout the
WRF Domain 1 and this domain basically represents the large-scale weather conditions.
We further performed sensitivity study by running the WRF Domain 1 using the same
initial and boundary conditions as before but without using nudging. The results (not
shown) exhibit rather similar distributions to the WRF Domain 1 simulations with
nudging. This illustrates that the model biases of Tmax are largely inherited from the
forcing data. We also examined the radiation budget at surface around the time of the
maximum temperature from both models. We noted similar radiation budget in both
models without significant differences.
4.2 Minimum Temperature
Figure 7 shows scatter plots of the observed and simulated daily Tmin at all HCN
stations combined. Correlation coefficients between the observations and simulations are
presented in Table 1. A good correspondence is evident between the observations and
model simulations with the correlation coefficients being 0.85 for both WRF domains
and 0.86 for HadRM domain. It is evident that the correlation coefficients of Tmin
between the observations and simulations are always smaller than those of Tmax (see
Table 1). Two reasons might be responsible for these discrepancies. Firstly, the 108-km
WRF Domain 1 simulations show a correlation coefficient of 0.87 for minimum
temperature and 0.94 for maximum temperature, suggesting that the large-scale driving
data at a 6-hourly temporal resolution do not represent the observed daily minimum
temperatures as well as the maximum temperature. Secondly, WRF and HadRM may
contain model deficiencies in the parameterization of nocturnal boundary layer physics
and dynamics that have a significant impact on nighttime (Tmin) temperatures.
Figure 8 presents the annual mean model biases of Tmin at HCN stations. In contrast
to Tmax, the nested WRF simulations show predominantly warm biases on the order of
0.5 ~ 2.5ºC. Two stations along the coast of Washington exhibit small cold biases (-1.0 ~
-0.5ºC) in the WRF simulations, likely due to the modeled influence of marine air. As in
the case for Tmax, the model biases in WRF are generally smaller in magnitude along the
coast than in the interior.
Warm biases are noted at each HCN station in the HadRM simulations (Fig. 8). The
biases range from 0.5 to 3.0ºC with large biases located mainly along the coast of
Oregon, the northern parts of Oregon and Washington, and the southeastern part of Idaho.
These warm biases cannot be explained by the large-scale forcing data as we will show
later that the driving data appear to contain cold biases in Tmin as indicated by the WRF
Domain 1 simulations with nudging. This suggests that the model deficiency must play a
role. As has been mentioned before, the R-2 data are about 0.2 ~ 0.9ºC warmer than the
NNRP data over the Pacific Northwest (see Fig. 3) and Fig. 8 indicates that the warm
biases in the HadRM simulations are about 1°C higher than in the nested WRF
simulations.
Figure 9 shows scatter plots of the observed and simulated monthly mean Tmin at all
HCN stations combined with the correlation coefficients given in Table 1. Nearly a linear
correspondence is noticed between the observations and simulations with the correlation
coefficients being 0.95 for WRF Domain 2 and 0.94 for both WRF Domain 3 and
HadRM Domain. This is an appreciable improvement of the model performance over the
daily time scale (Table 1). Note also in Table 1 that the correlation coefficients of 10-day
mean Tmin between the observations and simulations are very close to those of the
monthly mean Tmin. Warm biases are suggested in the scatter plots especially for the
HadRM simulations in that the majority of points are located above the 1:1 line (Fig. 9).
Figure 10 shows seasonal mean model biases of Tmin at HCN stations. Warm biases
on the order of 0 ~ 5ºC prevail in both model simulations throughout the year. Cold
biases are noted along the coast, eastern part of Washington and Snake Plain.
The WRF Domain 1 simulated Tmin exhibits predominantly cold biases throughout
the year (Fig. 10), in contrast to the nested WRF and HadRM simulations that show
mainly warm biases during each season. This discrepancy implies that the model biases
of Tmin in the nested WRF and HadRM simulations cannot be attributable to the driving
data, which suggests that the warm biases are likely associated with the regional models.
Our sensitivity study by running the WRF Domain 1 without using nudging (not shown)
exhibits mainly warm biases of Tmin during each season, similar to the nested WRF and
HadRM simulations. Further analyses indicated that excessive downward longwave flux
at surface during nighttime appears to be responsible for the warm biases of Tmin in the
simulations.
4.3 Precipitation
At daily time scale, a large spreading is observed in the scatter plots of the observed
and model simulated precipitation (not shown) with the correlation coefficients being
0.51 for WRF Domain 2, 0.56 for WRF Domain 3 and 0.52 for HadRM Domain (Table
1). The correlation coefficients increase steadily as daily values are averaged over
increasing number of days (Fig. 11). For each time scale, the observed and simulated
precipitations were averaged over the specified time period and the spatial correlation for
all stations and all times was computed. For example, the correlation coefficients of 5day mean precipitation are 0.69 for WRF Domain 2, 0.76 for WRF Domain 3 and 0.71
for HadRM Domain and they become 0.78, 0.87 and 0.78 for 20-day mean precipitation.
The improvement in correlation over the range of the averaging days is the largest
between 1-day mean and 20-day mean but slow down substantially after 20-day mean
(Fig. 11). Averaging was not performed beyond 90 days since the correlation coefficients
may become misleading beyond 90-day mean due to short-term (one-year in this case)
model simulation. Note that WRF Domain 3 consistently shows the highest correlation
coefficients among all the model domains due to the improved representation of the local
terrain at higher resolutions. Figure 12 shows scatter plots of monthly mean precipitation
at all stations combined. A good relationship is observed between the observations and
simulations with the correlation coefficients being very close to those for 20-day mean
(see Table 1).
To establish how well the models capture the observed daily weather sequence spatial
correlation coefficients of daily precipitation at HCN stations are illustrated in Fig. 13.
Relatively high correlation coefficients (≥ 0.6) are noted along the coast of Oregon and
Washington and over the northern part of Washington in the simulations for both models.
Relatively low correlation coefficients (0.2 ~ 0.4) are identified over the southeastern part
of Washington, the eastern part of Oregon and the southern part of Idaho in both models.
These regional differences in model performance may be related to the modeled terrain
effects as air masses move across the Cascade and Rocky Mountains and are subject to
larger modifications in the interior domain than the coastal areas.
Figure 14 shows the model biases of annual mean precipitation normalized by the
observed annual mean precipitation at each HCN station. Normalization was employed
on precipitation in order to make meaningful comparisons between dry and wet stations.
The WRF simulations exhibit wet biases at almost every station. The wet normalized
biases in the WRF simulations are generally smaller than 100% along the coast and are
larger than 150% in the interior. The wet normalized biases are also larger in WRF
Domain 2 than WRF Domain 3 at the majority of stations. HadRM exhibits small model
normalized biases (-50 ~ 50%) at stations along the borders of Oregon and Washington
and over the northern part of Idaho with wet biases (≥ 100%) over the southern part of
Idaho.
The model biases of seasonal mean precipitation normalized by the observed seasonal
mean precipitation at each HCN station are shown in Fig. 15. The WRF and HadRM
simulations show a large variability in the spatial distributions of the seasonally
normalized biases. This reflects the strong influence of complex terrain on local
precipitations. During winter, spring and autumn, mainly wet biases are noted in the
nested WRF simulations while in summer, more stations show dry biases when compared
to the other seasons. For HadRM, neither dry biases nor wet biases are dominant although
wet biases are persistently identified over the southern parts of Idaho.
For precipitation, the WRF Domain 1 simulations with nudging exhibit predominantly
dry biases close to 100% in all seasons (Fig. 15), in contrast to the nested WRF and
HadRM simulations that show both wet and dry biases. This implies that the precipitation
in the driving data is rather small since the model biases are computed as the differences
between the modeled and observed precipitation normalized by the observed
precipitation. Our sensitivity study by running WRF Domain 1 without nudging shows
similar distributions in model biases of precipitation to the nested WRF domains but with
larger model biases when compared to the nested WRF and HadRM domains (not
shown).
5. Conclusions and Discussion
This work examines the performance of two regional models, WRF and HadRM in
simulating the observed Tmax, Tmin and precipitation at HCN stations over the United
States Pacific Northwest. The analyses are focused mainly on daily, monthly and
seasonal time scales for the nested WRF and HadRM domains, and in terms of
correlation coefficients and model biases.
The simulated Tmax from WRF and HadRM compare well with the observations as
reflected by high spatial correlation coefficients at daily and monthly time scales. When
averaged annually, cold biases ranging around -3°C are associated with the WRF
simulations in the interior model domain and small biases on the order of -1 ~ 1°C are
noted in the HadRM simulations and along the coast of Oregon and Washington in the
WRF simulations. For WRF, mainly warm biases of Tmax are noted along the coast of
Oregon and Washington with cold biases in the interior during autumn and winter. The
HadRM runs show mainly warm biases in the interior during the same period of time.
During summer, there is a large variability in distribution of Tmax biases while during
spring, cold biases of Tmax are dominant in both model simulations. It appears that the
model biases of Tmax come mainly from the large-scale driving data.
The correlation coefficients of Tmin between the observations and model simulations
are reasonably good although not as high as those of Tmax for both models. The
differences in model performance for Tmin and Tmax are likely due to a combination of
deficiencies in the large-scale driving data and model parameterizations of nocturnal
physics and dynamics, which specifically affect Tmin. Predominantly warm biases of
Tmin are noted in both model simulations when averaged annually and seasonally with
larger warm biases in the HadRM simulations than in the WRF simulations. Model biases
of Tmin tend to be small along the coast of Oregon and Washington in the WRF
simulations when compared to the model interior. It appears that the model biases of
Tmin come mainly from the regional models.
The correlation coefficients between the simulated and observed daily precipitation are
relatively low. However, they increase steadily as daily precipitations are averaged over
increasing number of days. The rising rates in the correlation coefficients level out
considerably around 20 days. The temporal correlation between the observed and
simulated precipitation is higher at stations along the coast of Oregon and Washington
and over the northern part of Washington than elsewhere in both models. Model
overestimation of the observed precipitation is indicated in the WRF simulations when
averaged annually, but with considerable reduction in the biases for the 12-km domain
relative to the coarser 36-km domain. The HadRM simulations display small biases at
stations along the borders of Oregon and Washington and over the northern part of Idaho
with wet biases over the southern part of Idaho when averaged annually. A large
variability in the spatial distributions of the seasonally normalized biases of precipitation
is evident in both model simulations.
Our results suggest that the WRF simulation at 12-km grid spacing does not clearly
outperform 36-km simulation in terms of Tmax and Tmin. Variations in local
temperatures are dictated by large-scale weather systems with mesoscale processes
associated with the underlying surface playing a secondary role, thus grid spacing is not
critical to the temperature simulation even over complex terrain. The primary local effect,
namely decreasing temperature with elevation, has been taken into account through the
lapse rate correction applied to model output. The value of fine grid spacing may be more
significant in simulating climate variability and climate change, which we shall address
in future work. Over interannual and decadal time scales, regional alterations in snow
cover, soil parameters, cloudiness, and circulation associated with interactions between
the large-scale climate change and the regional topography and land-water contrasts
become increasingly important (Salathé et al., 2008). In contrast to the temperature
results, a large improvement is found in the precipitation simulation for the 12-km WRF
when compared to the 36-km WRF and 25-km HadRM simulations. Precipitation in the
Pacific Northwest is generally localized over complex terrain due to the interaction of
airflow with local topography. Thus, the magnitude, distribution, and timing of
precipitation are closely related to the topography, and accurate simulation depends on a
realistic representation of topographic effects.
Our analyses also suggest that HadRM and WRF are generally comparable in their
performance in resolving the observed Tmax, Tmin and precipitation at horizontal
resolutions on the order of tens of kilometers although there are improved skills in
precipitation for the 12-km WRF when compared to the 25-km HadRM. The skill shown
here suggests both models are suitable for climate simulations. Reduced annual mean
diurnal cycle in temperature is indicated in both model simulations. For WRF, this
reduced diurnal cycle results from a modest cold bias in Tmax and warm bias in the
Tmin; for HadRM, it results from a large warm bias Tmin. Note that HadRM is a
hydrostatic model with less flexibility in horizontal resolution but is computationally
efficient while WRF is a complex modeling system with a broad spectrum of applications
across scales ranging from meters to thousands of kilometers but is computationally
expensive to run. Nevertheless, differences in the simulated temperature between the two
models are comparable to differences in the large-scale driving data. Consequently, the
quality of the driving data is as much a limiting factor in climate simulation as the
regional models.
Acknowledgements:
This work is funded by an Environmental Protection Agency STAR Grant; by the
National Science Foundation (ATM0709856); and by a Microsoft Corporation gift to the
Climate Impacts Group. We thank the PRECIS team especially Richard Jones, David
Hein, David Hassell, and Simon Wilson for providing the PRECIS package and for their
guidance in using the package. The WRF simulations were performed at the National
Center for Atmospheric Research (NCAR) Computational and Information System
Laboratory (CISL). NCAR is sponsored by the National Science Foundation. This
publication is partially funded by the Joint Institute for the Study of the Atmosphere and
Ocean (JISAO) under NOAA Cooperative Agreement No. NA17RJ1232, Contribution
No. 1602.
References:
Collins, W. D., P. J. Rasch, B. A Boville, J. J. Hack, J. R. McCaa, D. L. Williamson, J. T.
Kiehl, B. Briegleb, C. Bitz, S.-J. Lin, M. Zhang, and Y. Dai, 2004: Description of the
NCAR Community Atmospheric Model (CAM 3.0). NCAR Tech. Note, NCAR/TN464+STR, 226 pp.
Esteban, M. A., and Y.-L. Chen, 2008: The Impact of trade wind strength on precipitation
over the windward side of the Island of Hawaii. Mon. Wea. Rev., 136, 913-928.
Gordon, C., C. Cooper, C. A. Senior, H. Banks, J. M. Gregory, T. C. Johns, J. F. B.
Mitchell, and R. A. Wood, 2000: The simulation of SST, sea ice extents and ocean heat
transports in a version of the Hadley Center coupled model without flux adjustments.
Clim. Dyn., 16, 147-168.
Grell, G., J. Dudhia, and D. R. Stauffer, 1993: A description of the fifth generation Penn
State/NCAR mesoscale model (MM5). NCAR Tech. Note, NCAR/TN-398 + IA,
National Center for Atmospheric Research, Boulder, CO, 107 pp.
Hong, S.-Y., J. Dudhia, and S.-H. Chen, 2004: A revised approach to ice microphysical
processes for the bulk parameterization of clouds and precipitation. Mon. Wea. Rev., 132,
103-120.
Kalnay, E., M. Kanamitsu, R. Kistler, W. Collins, D. Deaven, L. Gandin, M. Iredell, S.
Saha, G. White, J. Woollen, Y. Zhu, M. Chelliah, W. Ebisuzaki, W. Higgins, J. Janowiak,
K. C. Mo., C. Ropelewski, J. Wang, A. Leetmaa, R. Reynolds, R. Jenne, and D. Joseph,
1996: The NCEP/NCAR 40-year reanalysis project. Bull. Amer. Meteor. Soc., 77, 437471.
Kanamitsu, M., W. Ebisuzaki, J. Woollen, S.-K. Yang, J. J. Hnilo, M. Fiorino, and G. L.
Potter, 2002: NCEP-DOE AMIP-II Reanalysis (R-2). Bull. Amer. Meteor. Soc., 83, 16311643.
Leung, L. R., Y. Qian, and X. Bian, 2003a: Hydroclimate of the western United States
based on observations and regional climate simulation of 1981-2000. Part I: Seasonal
statistics. J. Climate, 16, 1892-1911.
Leung, L. R., Y. Qian, X. Bian, and A. Hunt, 2003b: Hydroclimate of the western United
States based on observations and regional climate simulation of 1981-2000. Part II:
Mesoscale ENSO anomalies. J. Climate, 16, 1912-1928 .
Mass, C. F., 2008: The Weather of the Pacific Northwest. University of Washington
Press, 336 pp.
Mass, C. F., D. Ovens, K. Westrick, and B. A. Colle, 2002: Does increasing horizontal
resolution produce more skillful forecasts? Bull. Amer. Meteor. Soc., 83, 407-430.
Salathé, E. P., R. Steed, C. F. Mass, and P. H. Zahn, 2008: A high-resolution climate
model for the United States Pacific Northwest: Mesoscale feedbacks and local responses
to climate change. J. Climate, in press.
Smith, R. B., and I. Barstad, 2004: A linear theory of orographic precipitation. J.
Atmospheric Sciences, 61, 1377-1391.
Table 1. Correlation coefficients of daily, 10-day mean and monthly mean Tmax, Tmin
and precipitation between observations and model simulations from WRF Domain 2,
WRF Domain 3 and HadRM.
Daily
OBS-D02
10-day
Monthly
Daily
OBS-D03
10-day
Monthly
Daily
OBS-HadRM 10-day
Monthly
Tmax
Tmin
Precip
0.92
0.97
0.97
0.93
0.97
0.98
0.93
0.97
0.97
0.85
0.93
0.95
0.85
0.93
0.94
0.86
0.93
0.94
0.51
0.74
0.79
0.56
0.82
0.87
0.52
0.76
0.78
Figure 1. (a) WRF model domains with two nests (Domain 1, 2, 3) and HadRM domain,
and (b) WRF innermost domain (Domain 3) and terrain height (m). Shadings in (a)
represent terrain height (m) for the corresponding WRF domain. Grid spacing for each
domain is: WRF Domain 1, 108 km; WRF Domain 2, 36 km; WRF Domain 3, 12 km;
and HadRM Domain, 25 km. HCN stations in the states of Washington, Oregon and
Idaho are represented by filled brown circles in (b).
Figure 2. Scatter plots of daily Tmax (°C) at HCN stations (a) observations versus WRF
Domain 2 simulations, (b) observations versus WRF Domain 3 simulations, (c)
observations versus HadRM simulations, (d) WRF Domain 2 versus WRF Domain 3, (e)
WRF Domain 2 versus HadRM, and (f) WRF Domain 3 versus HadRM. Correlation
coefficient is shown on the upper right corner of each plot.
Figure 3. Monthly mean differences of Tmax, Tmin and T (°C) between the two driving
data (R2-R1) averaged over the United States Pacific Northwest. Dashed line represents
the 0°C reference line.
Figure 4. Annual mean model biases of Tmax (ºC) at HCN stations for (a) WRF Domain
2, (b) WRF Domain 3, and (c) HadRM. Color and size of the filled squares correspond to
the sign and magnitude of the represented model biases, respectively. Cross represents a
magnitude range of -0.1 ~ 0.1°C.
Figure 5. Scatter plots of monthly averaged Tmax between observations and model
simulations from (a) WRF Domain 2, (b) WRF Domain 3, and (c) HadRM.
Figure 6. Seasonal mean model biases of Tmax (°C) for WRF Domain 1 (a), WRF
Domain 2 (b), WRF Domain 3 (c), and HadRM (d). Each four-panel corresponds to
model biases for one season and is arranged from DJF (December-January-February) on
the top to SON (September-October-November) on the bottom. Color represents the
magnitude of the biases.
Figure 7. Same as Fig. 2 except for daily Tmin (°C).
Figure 8. Same as Fig. 4 except for Tmin (°C).
Figure 9. Same as Fig. 5 except for Tmin (°C).
Figure 10. Same as Fig. 6 except for Tmin (°C).
Figure 11. Correlation coefficients of precipitation between observations and model
simulations on various time scales.
Figure 12. Scatter plots of monthly averaged precipitation between observations and
model simulations from (a) WRF Domain 2, (b) WRF Domain 3, and (c) HadRM.
Figure 13. Correlation coefficients of daily precipitation between observations and model
simulations at HCN stations for (a) WRF Domain 2, (b) WRF Domain 3, and (c) HadRM.
Magnitude of the correlation coefficients are represented by the size of the filled squares.
Figure 14. Annually averaged model biases of precipitation (mm) normalized by annual
mean precipitation at each HCN station for (a) WRF Domain 2, (b) WRF Domain 3, and
(c) HadRM. Cross represents a magnitude range of -0.1 ~ 0.1 mm.
Figure 15. Same as Fig. 6 except for seasonal mean model biases of precipitation (mm)
normalized by seasonal mean precipitation at each HCN station.