## 1. Introduction

The Weather Research and Forecasting (WRF) model is a next-generation mesoscale modeling system designed to place new and existing U.S. research and operational models under a common software architecture (Skamarock et al. 2001). Lead institutions involved in the effort include the National Center for Atmospheric Research (NCAR), Air Force Weather Agency (AFWA), National Centers for Environmental Prediction (NCEP), National Oceanic and Atmospheric Administration (NOAA), and other government agencies and universities. In WRF, it is possible to mix and match the dynamical cores and physics packages of different models to optimize performance since each model has strengths and weaknesses in different areas. This feature is particularly advantageous for intermodel comparisons and sensitivity studies.

WRF can be thought of as a software architectural framework in which different dynamical cores (the discretized version of the equations of motion and their coordinate system) and model physics packages are housed under the same code. NCAR has developed two dynamical cores for WRF, one using a height-based terrain-following *σ* coordinate (the so-called NCAR height core, no longer supported at this time) and the other using a pressure-based terrain-following *σ* coordinate (the so-called NCAR mass core). Various model physics packages developed at NCAR, NCEP, and elsewhere have been rewritten into the WRF software architectural framework. Theoretically, there is no limit as to how many dynamical cores can be incorporated into WRF. For example, NCEP has rewritten its Nonhydrostatic Mesoscale Model (NMM) into the WRF software architecture, and it is named the WRF NMM core.

The NOAA Cooperative Institute for Regional Prediction (CIRP) ran a beta version of the WRF model (CIRP WRF) from March 2003 to September 2004 in real time. This study aims to help model developers by evaluating the ability of the WRF model to predict warm season surface sensible weather over the western United States. CIRP WRF’s performance is also compared to that of the Eta Model, the operational mesoscale model developed by NCEP. Surface sensible weather variables (i.e., 2-m temperature, 2-m dewpoint, and 10-m wind) were evaluated because of their importance in power consumption and fire weather prediction, as well as the availability of high-density surface observations provided by the MesoWest cooperative networks^{1} (Horel et al. 2002). Accurate temperature forecasts can save power companies millions of dollars. The incremental benefit of a 1° (Fahrenheit) improvement in forecast accuracy for U.S. power companies as a whole is about $35 million per year (U.S. Department of Commerce 2004). Surface wind and relative humidity (which can be calculated from temperature and dewpoint) strongly influence fire weather conditions and wildfire behavior. For example, at the Salt Lake City Forecast Office (FO), conditions for a red flag warning for forest fires are satisfied when (i) the relative humidity (RH) drops below 15% *and* (ii) the wind gust exceeds 13.4 m s^{−1} at least three times within a 3-h period (M. Jackson, Salt Lake City Forecasting Office, 2005, personal communication).

This note is organized as follows. Sections 2 and 3 provide a description of CIRP WRF and the NCEP Eta Model, respectively. Section 4 explains the methodology used in this study. Section 5 presents the results. Section 6 is the discussion section. Finally, section 7 provides the summary.

## 2. Real-time WRF at CIRP

CIRP WRF used the NCAR mass core, which is based on a hydrostatic, pressure-based, terrain-following *σ*-type vertical coordinate, named the *η* coordinate^{2} by the WRF developers (not to be confused with the *η* coordinate in the Eta Model). The model domain had an area of 4 410 000 km^{2} with 169 by 169 horizontal grid points at 12.5-km grid spacing, covering the entire western United States and portions of Mexico and Canada (Fig. 1a). Thirty-four half-*η* levels^{3} were used in the vertical, with 7 levels below 1.5 km AGL and the lowest level at about 30 m AGL. Two daily 48-h forecasts (0000 and 1200 UTC initialization) were run using the Eta-212^{4} grids (40-km horizontal grid spacing) for initial and lateral boundary conditions (boundary conditions were updated at 3-hourly intervals). Because the warm season is characterized by a lower degree of synoptic variability, the performance of CIRP WRF should be less hampered by the lateral boundary conditions. CIRP WRF used physics packages similar to the real-time WRF at NCAR (NCAR 2003, personal communication): Ferrier (new Eta) microphysics (Chen and Dudhia 2003), the new Kain–Fritsch convective parameterization (Chen and Dudhia 2003; Kain 2004), RRTM longwave radiation (Mlawer et al. 1997), Dudhia shortwave radiation (Dudhia 1989), the fifth-generation National Center for Atmospheric Research–Pennsylvania State University Mesoscale Model’s (MM5) five-layer slab soil model (Chen and Dudhia 2003), the planetary boundary layer (PBL) scheme from the Medium-Range Forecast (MRF) model (Hong and Pan 1996), and the surface layer scheme as in Zhang and Anthes (1982). Soil moisture is not explicitly represented in the slab model, but a fixed moisture availability acts as a proxy for soil moisture (Zhang and Anthes 1982).

The 3-month verification period extended from 1 June to 31 August 2003, with WRF 1.2 updated to WRF 1.3 on 14 July, halfway through the validation period. Tests showed some quantitative differences in the performance between the two versions, but the magnitude and trend of the performance measures were similar (not shown). Thus, having used two different model versions should not affect the conclusions of this study.

## 3. The Eta Model

The Eta Model configuration featured a similar horizontal grid spacing (12 km) as in CIRP WRF, but with a much larger area, encompassing the entire North American continent and parts of the eastern Pacific. With 60 vertical levels, the Eta Model also had higher vertical resolution. In addition, 24 vertical levels in the Eta Model were below 1.5 km AGL with the lowest level at about 20 m AGL. The primary Eta Model physics packages were the Betts–Miller–Janjić convective parameterization (Janjić 1994), Ferrier (new Eta) microphysics (same as WRF), the Mellor and Yamada (1982) 2.5-level PBL scheme, the Goephysical Fluid Dynamics Laboratory (GFDL) radiation scheme (Lacis and Hansen 1974; Fels and Schwarzkopf 1975), and the Noah land surface model (Ek et al. 2003). The Global Forecast System (GFS) provided the lateral boundary conditions. Verification of the Eta Model was based on the Eta-218 grid (Fig. 1b), which has the same horizontal grid spacing (12 km) as the native Eta Model grid. Again, MesoWest observations were used for verification. A description of the Eta Model can be found in Black (1994).

## 4. Methodology

Forecast cycles were selected for verification only when both the CIRP WRF grids and the corresponding Eta grids were available continuously from forecast hours 0 to 48. Each forecast cycle was verified at 3-hourly intervals. Due to local hardware failure or missing grids, this represented 119 out of a possible 184 cycles.

MesoWest stations within the CIRP WRF domain (Fig. 1a) that reported at least 50% of the possible observations in the 3-month period provided the verification data. A total of 1875 MesoWest stations met these criteria. MesoWest obtains data from a number of networks, each designed to meet the specific needs of its operating agency. As a result, there is considerable diversity in the site characteristics, sensor types and heights, and reporting intervals of the observations. Because of a lack of station metadata, no effort was undertaken to account for variability in sensor height, and it was assumed that large contrasts between the model and observations reflected the characteristics of the model more than the verifying dataset.

Because the model grid point represents a grid average, the model elevation and the station elevation are not exactly the same. Adjustments for differences in model and terrain elevations were not made for the following reasons. First, although the adjustment of the temperature using a standard lapse rate (6.5 K km^{−1}) led to a decrease in temperature BE for both the Eta and CIRP WRF, it also increased the cumulative temperature MAE for both models (not shown). Second, another study also demonstrated that adjusting the temperature using a standard lapse rate does not necessarily lead to an improvement in model skill. In a model verification study for selected sites in northern Utah during the 2002 Winter Olympics, Hart et al. (2004) adjusted the model temperature to the observed elevation using a standard lapse rate (6.5 K km^{−1}). The temperature adjustment lowered the MAE for mountain locations, but degraded or provided no improvement at the Wasatch Front and mountain valley locations. Third, even though one can conceivably use a more advanced technique to adjust the model temperature to the observed elevation, the verification would partially reflect the adjustment technique, rather than the model forecast. Finally, adjustment of wind and moisture is even more complicated than that of temperature. Because of the similar horizontal resolutions between CIRP WRF and the Eta and the similar trends in model terrain error in both models (not shown), not adjusting the forecasts to the observed elevations should not affect the intercomparison between the two models.

To perform the verification, 2-m temperature, 2-m dewpoint, and 10-m wind forecasts from the CIRP WRF and the Eta models were first bilinearly interpolated to the MesoWest station locations. Note that forecast–observation pairs for wind were excluded for observed winds of 2.5 m s^{−1} or less because anemometers are not very accurate at low wind speed. MAE and BE scores were calculated for the aforementioned variables (bias was not computed for wind direction).

*F*and

_{i}*O*are the forecast and observed values for the

_{i}*i*th station, respectively, and

*N*is the number of observation–forecast pairs. The MAE is defined by

The numbers of observation–forecast pairs for temperature, dewpoint, and wind were 3 587 164, 2 658 297, and 934 844, respectively.

## 5. Results

### a. 2-m temperature

The 3-month cumulative temperatures MAE for CIRP WRF (the Eta) were 3.3°C (2.8°C) for the 0000 UTC cycle and 3.1°C (2.8°C) for the 1200 UTC cycle (Table 1). For the 0000 UTC cycle, as a function of forecast hour, the CIRP WRF and the Eta temperature MAE exhibited maxima at forecast hours 12 and 36, which correspond to early morning (1200 UTC), and forecast hours 24 and 48, which correspond to late afternoon (0000 UTC) (Fig. 2). Maxima in temperature MAE for CIRP WRF were particularly pronounced in the early morning when they exceeded 4°C and were 1°C greater than those of the Eta.

For the 1200 UTC cycle, the CIRP WRF temperature MAE peaked near 3.7°C at forecast hours 12 and 36, which corresponds to late afternoon (0000 UTC) (Fig. 2). This contrasts with the 0000 UTC CIRP WRF, which featured peak temperature MAEs in the morning. The 1200 UTC cycle Eta temperature MAE featured weak maxima in late afternoon (forecast hours 12 and 36) and early morning (forecast hours 24 and 48), similar to the 0000 UTC cycle, with some evidence of error growth with increasing forecast projection. During some forecast hours, the Eta MAE was at least 0.5°C lower than that of WRF.

The bias analysis provides insight into the MAE characteristics of the two models. CIRP WRF 3-month cumulative temperature bias varied from 1.5°C for the 0000 UTC cycle to −1.2°C for the 1200 UTC cycle (Table 1). In contrast, the Eta Model exhibited a weak warm bias in both forecast cycles (0.9° and 0.6°C for the 0000 and 1200 UTC cycles, respectively). Both the CIRP WRF and the Eta temperature biases varied diurnally, with maximum warm bias in the morning hours [forecast hours 12 and 36 (24 and 48) for the 0000 (1200) UTC cycle] (Fig. 3). The amplitude of the diurnal variability in the bias error was particularly pronounced for the 0000 UTC CIRP WRF cycle.

The temperature bias in CIRP WRF can be explained partly by the procedure of the soil temperature initialization in the WRF slab-soil model. The slab model was designed to handle global model grids with only two layers in the land surface model (J. Dudhia 2004, personal communication). For example, the NCEP global reanalysis has two layers in its land surface model (0–10 and 10–200 cm). By default, the soil reservoir temperature^{5} from the slab model (centered at 25 cm below ground) is initialized with the top-layer soil temperature (the only layer closest to the reservoir depth). The WRF developers have not yet accounted for instances where soil temperature data are available in more than two layers as with our Eta Model soil temperature initialization. This caused a bias in the reservoir temperature initialization, and this is reflected in the initial warm (cold) 2-m temperature bias in the 0000 (1200) UTC cycle. Note the shift in the peak of the 2-m temperature bias in the WRF 0000 (1200) UTC cycle to the positive (negative) side of the ordinate due to the bias in the soil temperature initialization (Fig. 3). However, the shape of the bias time series is the same for both CIRP WRF and the Eta regardless of the initialization time. The surface sensible heat flux depends partly on the temperature gradient between the land surface and the overlying air. With everything else being equal, a higher surface soil temperature would enhance the surface sensible heat flux from the surface to the overlying air, warming the overlying air. An erroneous soil temperature initialization can give rise to an erroneous surface sensible heat flux, leading to an erroneous 2-m temperature. We will discuss in the section 5d how a better soil temperature initialization can improve the CIRP WRF forecast.

### b. 2-m dewpoint

The 3-month cumulative dewpoint MAEs for CIRP WRF (the Eta) were 3.6°C (3.3°C) for the 0000 UTC cycle and 3.6°C (3.4°C) for the 1200 UTC cycle (Table 1). As a function of forecast hour, the dewpoint MAE followed a diurnal pattern for both models (Fig. 4). For CIRP WRF, excluding the large initial MAE in the 1200 UTC cycle, the dewpoint MAE maxima (3.8°–4.1°C) generally occurred in the nighttime to morning period from 0300 to 1200 UTC (in the morning at 1200 UTC) for the 0000 (1200) UTC cycle, with MAE minima (3°–3.5°C) in the morning to late morning period from 1500 to 1800 UTC for both the 0000 and 1200 UTC cycles. The large initial dewpoint MAE in the 1200 UTC cycle for CIRP WRF will be explained later. As for the Eta Model, the dewpoint MAE maxima generally occurred during the 0000–1200 UTC window for both forecast cycles, and the minima occurred during the 1500–1800 UTC window (at 1500 UTC) for the 0000 (1200) UTC cycle.

CIRP WRF (the Eta) featured a 3-month cumulative dewpoint bias of 1.1°C (−0.8°C) for the 0000 UTC cycle and −0.8°C (−0.9°C) for the 1200 UTC cycle (Table 1). Except for the CIRP WRF 0000 UTC cycle, as a function of forecast hour, both models were drier than observed, with the dry bias being more pronounced in the Eta Model (Fig. 5). In particular, the CIRP WRF 1200 UTC cycle had an initial 2-m dewpoint bias of −3°C, much drier than the Eta. Consistent with the MAE analysis, the CIRP WRF and the Eta dewpoint biases also exhibited a diurnal pattern. For CIRP WRF, in terms of magnitude, dewpoint bias maxima generally occurred in the 0600–1200 UTC window and at 1200 UTC for the CIRP WRF 0000 and 1200 UTC cycles, respectively. As for the Eta, in terms of magnitude, dewpoint bias maxima generally occurred at 1200 UTC for the Eta for both the 0000 and 1200 UTC cycles.

The large initial dewpoint BE and MAE in the CIRP WRF 1200 UTC cycle may be partly attributed to the erroneous soil temperature initialization, although it is possible for other factors to contribute to the problem such as atmospheric initialization procedures and the method of calculating the 2-m water vapor mixing ratio (Figs. 4 and 5). For CIRP WRF, the 2-m water vapor mixing ratio (*q*_{2m}) is a weighted sum of the surface water vapor mixing ratio (*q*_{sfc}) and the water vapor mixing ratio at the lowest model level above ground (*q _{a}*). The weights depend on the Richardson number and the moisture availability. In fact,

*q*

_{sfc}is the saturation water vapor mixing ratio corresponding to the ground surface temperature. From the Clausius–Clapeyron equation, a positive (negative) bias in the ground surface temperature leads to a (negative) positive bias in

*q*

_{sfc}and 2-m dewpoint. Note the correspondence between the initial negative 2-m temperature bias (partly a reflection of the erroneous ground temperature initialization) and the initial negative 2-m dewpoint bias in the CIRP WRF 1200 UTC cycle (Figs. 3 and 5). The correspondence between the initial positive dewpoint bias and the initial positive temperature bias can also be seen in the 0000 UTC CIRP WRF cycle.

### c. 10-m wind

The 3-month cumulative wind speed MAEs for CIRP WRF (the Eta) were 1.7 (1.6) m s^{−1} for the 0000 UTC cycle and 1.8 (1.6) m s^{−1} for the 1200 UTC cycle (Table 1). As a function of forecast hour, the 0000 UTC CIRP WRF wind speed MAE varied between 1.6 and 1.9 m s^{−1} with no regular diurnal variability (Fig. 6). The 0000 UTC Eta Model wind speed MAE showed a more regular diurnal pattern with maxima in the afternoon (forecast hours 24 and 48) and minima at night. This diurnal pattern also was evident for the 1200 UTC Eta cycle. Unlike the 0000 UTC cycle, the 1200 UTC CIRP WRF wind speed MAE showed some diurnal structure, with the largest errors during the late night or early morning hours (forecast hours 18–24 and 42–48).

In terms of bias, the 3-month cumulative wind speed biases for CIRP WRF (the Eta) model were 0.5 (−0.4) m s^{−1} for the 0000 UTC cycle and 0.4 (−0.5) m s^{−1} for the 1200 UTC cycle (Table 1). Thus, CIRP WRF tended to overpredict the wind speed, whereas the Eta Model tended to underpredict the wind speed. In both models, however, the wind speed bias varied according to the time of day (Fig. 7). CIRP WRF exhibited pronounced wind speed bias maxima at night that reached 0.8–1.2 m s^{−1}, but a relatively small bias (magnitude < 0.3 m s^{−1}) in the afternoon. The Eta Model produced relatively small biases in the 2100–0000 UTC window (afternoon), but generated a large negative (often <−0.8 m s^{−1}) bias (i.e., underprediction) in the 0300–1500 UTC period (nighttime to early morning).

Wind direction is perhaps one of the most difficult variables to forecast. For both initialization times, the 3-month cumulative wind direction MAE for CIRP WRF (the Eta) was 61° (41°) (Table 1). Thus, the Eta Model typically produced a more accurate wind direction forecast. Just as in other variables, the wind direction MAE varied according to the time of the day (Fig. 8). For CIRP WRF (the Eta), the wind direction MAE began to increase at night around 0300 (0300) UTC until it reached a maximum in the morning at 1200–1500 (0900–1200) UTC.

### d. Sensitivity experiments

The CIRP WRF 2-m temperature bias can be reduced by better defining the initial temperature in the slab-soil model. As mentioned previously, the default slab-soil model used the top (0–10 cm) layer soil temperature from the Eta as the WRF reservoir temperature, which represents a temperature centered 23 cm below ground. This is clearly not appropriate, as the 0–10-cm soil temperature is subject to diurnal variations. To improve the soil temperature initialization in the slab-soil model, the WRF code was reconfigured to use the Eta soil temperature at 10–40 cm below ground (i.e., centered 25 cm below ground) for initializing the CIRP WRF reservoir temperature. The 0000 UTC and 1200 UTC 1 July 2003 forecasts^{6} were rerun with this change (experiment FIXEDSLAB). Experiments were performed to examine whether using the more sophisticated Oregon State University (OSU) land surface model (LSM) would improve the WRF forecasts (experiment OSULSM). The soil moisture and temperature from the Eta Model were used to initialize OSULSM. A control experiment (CTL) was also performed in which the standard configuration of the slab- soil model was used.

Although FIXEDSLAB reduced the 2-m temperature MAE by 0.3°C for the 0000 UTC cycle, it did not reduce the temperature MAE for the 1200 UTC cycle (Table 2). However, FIXEDSLAB reduced the warm (cold) bias in the 0000 (1200) UTC cycle compared with CTL from 1.3° to −0.1°C (from −0.9° to −0.6°C). OSULSM had larger temperature MAEs than CTL, about 0.3°–0.6°C. In addition, OSULSM had mixed results in affecting the temperature BE. While OSULSM showed improvement in reducing the warm temperature bias in the 0000 UTC cycle (from 1.3° to −0.8°C), the cold temperature bias became more negative in the 1200 UTC cycle (from −0.9° to −1.2°C). The Eta Model had better temperature MAEs and BEs than all of the WRF experiments except for the BE in FIXEDSLAB for the 0000 UTC cycle (Table 2).

As for dewpoint, FIXEDSLAB outperformed OSULSM and CTL for the 0000 UTC cycle in terms of MAE and BE, but did not perform as well as the Eta Model (Table 2). FIXEDSLAB decreased the dewpoint MAE and BE by 0.3° and 0.7°C for the 0000 UTC cycle, respectively, and increased both the dewpoint MAE and BE by 0.1°C in the 1200 UTC cycle.

The WRF experiments had very similar wind speed MAEs (1.8–2.0 m s^{−1}) and BEs (0.6–0.7 m s^{−1}). On the other hand, the Eta Model had smaller wind speed MAE (1.6 m s^{−1}) and BE (−0.4 m s^{−1}). As for wind direction MAE, the WRF experiments had MAEs ranging from 60° to 63°, and the Eta Model had MAEs between 38° and 39°.

The issue of altering the moisture availability in the slab model was not explored. There is no doubt that the moisture availability can affect the 2-m temperature and dewpoint as much as the soil temperature (or perhaps more), due to its strong influence in the partitioning of surface sensible and latent heat fluxes. Zhang and Anthes (1982) found that the increase in the moisture availability led to more (less) absorbed shortwave radiation to be partitioned into surface latent (sensible) heat flux, leading to a lower (higher) surface air temperature (dewpoint).

## 6. Discussion and summary remarks

Overall, CIRP WRF produced comparable MAEs to the Eta for 2-m dewpoint and 10-m wind speed. The 2-m temperature forecasts produced by CIRP WRF were, however, much worse. For the CIRP WRF 0000 UTC cycle, the 2-m temperature forecast valid in the early morning (1200 UTC) tended to be too warm partly because of the inadequate development of the nocturnal inversion layer at many locations and partly because of the initial positive 2-m temperature bias. Hart et al. (2004) reported similar problems in the inadequate development of the nocturnal inversion layer with a version of the MM5 that was run over the Intermountain West for the 2002 Olympic Winter Games and featured surface and boundary layer parameterizations similar to those of CIRP WRF. In addition, the magnitude of the morning warm bias in the 0000 UTC CIRP WRF 2-m temperature forecasts has been exacerbated further by the erroneous soil temperature initialization (Fig. 3).

In contrast, the 1200 UTC CIRP WRF cycle featured a more pronounced late afternoon (0000 UTC) cold 2-m temperature bias and little bias in the morning. This shift in the bias characteristics reflects the *initial negative temperature bias* (−1°C) in the 1200 UTC WRF cycle as opposed to the *initial positive temperature bias* (2°C) in the 0000 UTC WRF cycle. As a result, the 2-m temperature in the 1200 UTC CIRP WRF cycle tended to be too cool.

Because of the relative humidity’s dependence on the temperature, there was a 180° phase difference between the temperature bias and the RH bias (not shown). A warm (cold) temperature bias coincided with a negative (positive) RH bias. Overall, both CIRP WRF and the Eta had a negative bias in RH. Given this fact and the fact that CIRP WRF (the Eta) overpredicted (underpredicted) the wind speed (Table 1, Fig. 7), it can be deduced that CIRP WRF (the Eta) would overpredict (underpredict) potentially hazardous wildfire conditions as defined by red flag warning criteria. The reason that CIRP WRF (the Eta) overpredicted (underpredicted) the wind speed might have been due to the coarser (finer) vertical resolution of CIRP WRF (the Eta). In a hurricane simulation using MM5 with the finest nest at 6-km horizontal grid spacing, Zhang and Wang (2003) found that the surface wind speed was reduced for a smaller vertical grid spacing in the lowest model level above ground. This result is consistent with the notion that frictional effects are more pronounced with a thinner layer of air interacting with the bottom surface.

From the results in the sensitivity experiments, the minor change in the initialization of the slab-soil model reservoir temperature appears to improve the surface temperature forecast, especially in reducing the temperature BE, albeit less so in the MAE, but without adversely affecting the BE and MAE in the dewpoint and the wind. However, using the OSU LSM (experiment OSULSM) did not result in superior performance in the 2-m temperature forecast as compared to FIXEDSLAB. The results in the sensitivity experiments are similar to the findings of Zhong and Fast (2003), who found decreased forecast accuracy in mesoscale simulations using more sophisticated LSMs. Some possible reasons for this are (i) the uncertainty in defining the parameters in LSMs such as thermal conductivity of the surface, (ii) not accounting for subgrid-scale heterogeneities in the land surface, and (iii) lack of accurate soil temperature and moisture in initializing the LSM. Therefore, it is not surprising to find that forecast accuracy was not improved using the OSU LSM.

The Noah LSM, a more advanced version of the OSU LSM, has been released recently in the latest version of WRF. The Noah LSM in WRF is more tightly coupled to the radiation and PBL schemes. Although this may improve WRF’s performance, the analysis of the Eta Model (which uses the Noah LSM) presented in this study suggests that improvement in LSM physics alone is insufficient. Improvements in LSM initialization may be equally as important as (or perhaps more important than) improvements in LSM physics. Yang et al. (1994) found that improvement in the initialization of a global model’s soil wetness can reduce the 5-day MAE of the surface temperature forecast. The modeling community should consider improving the LSM initialization and parameterization of coupled land surface–boundary layer processes as a top priority in order to generate better surface sensible weather forecasts.

## Acknowledgments

This research was funded by the National Weather Service through a series of grants and the NOAA Office of Global Programs under Grant GC04-281. We thank Todd Foisy and Matt Masarik for retrieving the Eta-218 and the WRF grids from the CIRP archive. We also thank Dr. Jason Knievel and the WRF Help Desk at NCAR for their help in working with WRF, Dr. Judy Pechmann for providing the scripts to access the MesoWest database, and Mr. Mike Splitt for clarification of the MesoWest quality control procedures. We appreciate comments from Prof. John Horel, Drs. Jason Knievel and Jason Nachamkin, and three anonymous reviewers, which helped to improve the manuscript. Finally, we thank the Center for High Performance Computing at the University of Utah for maintaining the Linux cluster on which WRF was run.

## REFERENCES

Black, T., 1994: The new NMC mesoscale Eta model: Description and forecast examples.

,*Wea. Forecasting***9****,**265–278.Chen, S-H., and Dudhia J. , cited. 2003: Annual report: WRF physics. Air Force Weather Agency, 38 pp. [Available online at http://www.mmm.ucar.edu/wrf/users/docs/wrf-phy.html.].

Dudhia, J., 1989: Numerical study of convection observed during the winter monsoon experiment using a mesoscale two-dimensional model.

,*J. Atmos. Sci.***46****,**3077–3107.Ek, M. B., Mitchell K. E. , Lin Y. , Rogers E. , Grunmann P. , Koren V. , Gayno G. , and Tarpley J. D. , 2003: Implementation of the upgraded Noah land-surface model in the NCEP operational mesoscale Eta model.

,*J. Geophys. Res.***108****.**8851, doi:10.1029/2002JD003296.Fels, S. B., and Schwarzkopf M. D. , 1975: The simplified exchange approximation: A new method for radiative transfer calculations.

,*J. Atmos. Sci.***32****,**1475–1488.Hart, K. A., Steenburgh W. J. , Onton D. J. , and Siffert A. J. , 2004: An evaluation of mesoscale-model-based model output statistics (MOS) during the 2002 Olympic and Paralympic Winter Games.

,*Wea. Forecasting***19****,**200–218.Hong, S-Y., and Pan H-L. , 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model.

,*Mon. Wea. Rev.***124****,**2322–2339.Horel, J., and Coauthors, 2002: Mesowest: Cooperative mesonets in the western United States.

,*Bull. Amer. Meteor. Soc.***83****,**211–225.Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further development of the convection, viscous sublayer, and turbulence closure schemes.

,*Mon. Wea. Rev.***122****,**928–945.Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update.

,*J. Appl. Meteor.***43****,**170–181.Lacis, A. A., and Hansen J. E. , 1974: A parameterization of the absorption of solar radiation in the earth’s atmosphere.

,*J. Atmos. Sci.***31****,**118–133.Mellor, G. L., and Yamada T. , 1982: Development of a turbulence closure model for geophysical fluid problems.

,*Rev. Geophys. Space Phys.***20****,**851–875.Mlawer, E. J., Taubman S. J. , Brown P. D. , Iacono M. J. , and Clough S. A. , 1997: Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-

*k*model for the long-wave.,*J. Geophys. Res.***102****,**16663–16682.Skamarock, W. C., Klemp J. B. , and Dudhia J. , 2001: Prototypes for the WRF (Weather Research and Forecasting) model. Preprints,

*Ninth Conf. on Mesoscale Processes,*Fort Lauderdale, FL, Amer. Meteor. Soc., CD-ROM, J1.5.U.S. Department of Commerce, cited. 2004: Economic statistics for NOAA. 3d ed. National Oceanic and Atmospheric Administration, 50 pp. [Available online at http://www.publicaffairs.noaa.gov/pdf/economics-statistics2004.pdf.].

Yang, R., Fennessy M. J. , and Shukla J. , 1994: The influence of initial soil wetness on medium-range surface weather forecasts.

,*Mon. Wea. Rev.***122****,**471–485.Zhang, D-L., and Anthes R. , 1982: A high resolution model of the planetary boundary layer—Sensitivity tests and comparisons with SESAME-79 data.

,*J. Appl. Meteor.***21****,**1594–1609.Zhang, D-L., and Wang X. , 2003: Dependence of hurricane intensity and structures on vertical resolution and time-step size.

,*Adv. Atmos. Sci.***20****,**711–725.Zhong, S. Y., and Fast J. , 2003: An evaluation of the MM5, RAMS, and Meso-Eta models at subkilometer resolution using VTMX field campaign data in the Salt Lake valley.

,*Mon. Wea. Rev.***131****,**1301–1322.

Average temperature MAE (°C) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta). Note that the 1200 UTC cycle results are plotted at a 12-h lag relative to the 0000 UTC results. The first (second) number in the parentheses in the abscissa corresponds to the forecast hour of the 0000 (1200) UTC cycle forecast.

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average temperature MAE (°C) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta). Note that the 1200 UTC cycle results are plotted at a 12-h lag relative to the 0000 UTC results. The first (second) number in the parentheses in the abscissa corresponds to the forecast hour of the 0000 (1200) UTC cycle forecast.

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average temperature MAE (°C) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta). Note that the 1200 UTC cycle results are plotted at a 12-h lag relative to the 0000 UTC results. The first (second) number in the parentheses in the abscissa corresponds to the forecast hour of the 0000 (1200) UTC cycle forecast.

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average temperature BE (°C) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average temperature BE (°C) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average temperature BE (°C) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average dewpoint MAE (°C) as a function of forecast hour (UTC) for the 0000 UTC (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average dewpoint MAE (°C) as a function of forecast hour (UTC) for the 0000 UTC (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average dewpoint MAE (°C) as a function of forecast hour (UTC) for the 0000 UTC (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average dewpoint BE (°C) as a function of forecast hour (UTC) for the 0000 UTC (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average dewpoint BE (°C) as a function of forecast hour (UTC) for the 0000 UTC (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average dewpoint BE (°C) as a function of forecast hour (UTC) for the 0000 UTC (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind speed MAE (m s^{−1}) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind speed MAE (m s^{−1}) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind speed MAE (m s^{−1}) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind speed BE (m s^{−1}) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind speed BE (m s^{−1}) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind speed BE (m s^{−1}) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind direction MAE (°) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind direction MAE (°) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Average wind direction MAE (°) as a function of forecast hour (UTC) for the 0000 (black curves) and 1200 UTC (gray curves) cycles. Circles (squares) represent results from CIRP WRF (the Eta).

Citation: Weather and Forecasting 20, 5; 10.1175/WAF885.1

Cumulative MAE and BE from Jun to Aug 2003.

Cumulative MAE and BE for 1 Jul 2003 for experiments CTL, FIXEDSLAB, OSULSM, and the Eta Model.

^{1}

MesoWest, maintained by CIRP, provides surface observations from over 3000 stations in the western United States.

^{2}

*η*is defined by

*p*represents pressure, the superscript

*h*means hydrostatic, and the subscripts

*s*and

*t*refer to surface and top, respectively.

^{3}

*η*-levels defined as

*η*-level always resides at the ground level, and the lowest half

*η*-level is always above ground level. The number of half

*η*-levels is always one less than the number of full

*η*-levels.

^{4}

NCEP offers its original Eta forecast grids (at 12-km horizontal grid spacing) at several resolutions and in a reduced domain size. The Eta-212 is one example.

^{5}

The reservoir in the slab-soil model is located at a depth deep enough such that it is not affected by the diurnal cycle of temperature. Thus, for short-term simulations, the reservoir temperature is fixed.

^{6}

The 0000 and 1200 UTC 1 July 2003 forecast cycles were chosen because their 48-h cumulative MAE and BE scores for the 2-m temperature were close to those of the 3-month cumulative values.