An Evaluation of the MM5, RAMS, and Meso-Eta Models at Subkilometer Resolution Using VTMX Field Campaign Data in the Salt Lake Valley (original) (raw)
1. Introduction
Mesoscale meteorological models have been used primarily as research tools to advance our understanding of regional-scale processes and help develop parameterizations of these processes for use in large-scale weather forecast and climate prediction models. Recent advances in computer hardware and parallel-processing algorithms have decreased the computational time required by mesoscale models so that operational forecasts of regional weather (Horel and Gibson 1994; Mass and Kuo 1998), forest fires (Smallcomb 2001), and air pollution (Fast et al. 1995; Sistla et al. 2001) are now common. Several studies have compared the performance of two or more mesoscale models using either long-term routine observations or data from short-term field campaigns. These studies usually focus on a particular type of application including short-term regional forecasting (White et al. 1999), precipitation (Colle et al. 1999), transport and diffusion (Hanna and Yang 2001; Hogrefe et al. 2001), and military operations (Cox et al. 1998).
To date, nearly all mesoscale model intercomparison and evaluation studies have employed horizontal grid spacing on the order of 10 km, and they have attributed part of the model prediction errors to coarse horizontal resolution. Because operational forecasts at a horizontal resolution of 1 km or less will be feasible in the near future, it is very useful to document the relative strengths and weaknesses of mesoscale models at such a small spatial scale.
In this paper we present what is to our knowledge the first intercomparison and evaluation of regional simulations using three state-of-the-art mesoscale models, the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (Penn State–NCAR) Mesoscale Model (MM5; Grell et al. 1993), the Regional Atmospheric Modeling System (RAMS; Pielke et al. 1992), and the National Centers for Environmental Prediction's (NCEP) Meso-Eta (Black 1994), with grid spacing smaller than 1 km. Simulations are carried out for the Salt Lake valley where detailed measurements are available from the Vertical Transport and Mixing (VTMX) field campaign (Doran et al. 2002) conducted during October 2000. The types of atmospheric phenomena we examine differ from previous studies because of the complex topography and the presence of the Great Salt Lake. Interactions between terrain-induced flows and the lake/land breezes produce complicated local circulation patterns. Our model evaluation focuses on these local circulations and the valley boundary layer development. We address three questions: 1) When terrain is well resolved at fine resolution, how well do the models reproduce the observed characteristics of terrain-induced flows and boundary layer structure in the valley? 2) What are the relative strengths and weaknesses of each model? 3) Is one model better suited for the valley environment than others and why?
While there is no doubt that mesoscale model evaluations using long-term data are important, the drawback of these comparisons is their dependence on routine observations that do not provide sufficient spatial and temporal resolution to determine the factors contributing to the forecast errors. Another problem with a statistically based long-term verification is that it tends to incur a large penalty for small forecast errors in timing or position of mesoscale features. Consequently, data from short-term intensive field campaigns, such as VTMX, are important sources for model evaluation and development. Several of the major population centers in the western United States are situated in valleys and basins. High-resolution mesoscale forecasting efforts for these urban centers would likely benefit from the comparisons and evaluations performed in this study using the VTMX data from the Salt Lake valley.
2. The Salt Lake valley terrain and the VTMX field campaign
The Salt Lake valley (Fig. 1) is bordered by high terrain on three sides: the Oquirrh Mountains to the west, the Wasatch Mountains to the east, and the Traverse Range to the south. The highest peak along the Wasatch Mountains has an elevation of 3300 m MSL, approximately 2000 m above the valley floor. There are a number of canyons on the east side of the valley and a gap in the middle of Traverse Range that connects the Salt Lake valley to the Provo valley to the south. The valley opens up to the north with the Great Salt Lake to the northwest. Under weak synoptic forcing, winds in the Salt Lake valley are dominated by thermally induced flows including slope and valley flows, canyon flows, gap winds, and lake/land breezes. In addition to the complex flow patterns, the high terrain surrounding the valley also contributes to the formation of several interesting phenomena such as nocturnal cold air pools, gravity waves, multiple elevated stable layers, and localized shear flows.
The VTMX field campaign, sponsored by the U.S. Department of Energy's Environmental Meteorology Program, took place in the valley during October 2000. The primary goal of this campaign was to collect an extensive dataset to be used to characterize boundary layer processes contributing to vertical transport and mixing in urban basins and valleys. To reach this goal, it was necessary to measure meteorological variables both near the valley floor and in the valley atmosphere at relatively high spatial and temporal resolution. This was achieved during the field campaign by a suite of in situ and remote sensing instruments including rawinsondes, tethersondes, radar wind profilers, sodars, surface meteorological stations, sonic anemometers, a Frequency-Modulated Continuous-Wave (FMCW) radar, a Doppler lidar, a water vapor lidar, aerosol lidars, and an instrumented aircraft. The locations of the measurements employed by this study are shown in Fig. 1.
A total of 10 intensive operation periods (IOPs) were conducted during the month-long campaign. Each IOP started in late afternoon and ended the next morning because the program's focus was on the nocturnal stable boundary layer and the morning and evening transition periods. Details about the field campaign, including the instrumentation and measurement strategies, as well as general weather conditions during all the IOPs, can be found in Doran et al. (2002).
3. Model description and configuration
MM5, RAMS, and Meso-Eta are three-dimensional, primitive equation, nonhydrostatic mesoscale models. MM5 and RAMS have much in common: both use terrain-following vertical coordinates, have nesting capabilities, and include various options for physical parameterizations of subgrid-scale turbulent mixing, radiation, cumulus convection, and land surface processes. The Meso-Eta model, on the other hand, employs a step-mountain vertical coordinate (Mesinger et al. 1988), allows only a single grid, and, as an operational forecast model, has a fixed set of physical parameterizations.
The parameterizations employed by Meso-Eta include a modified Mellor–Yamada level-2.5 turbulence parameterization (Janjić 1994), an explicit cloud water scheme (Zhao and Carr 1997), the Betts–Miller cumulus parameterization (Janjić 1994), the Oregon State University (OSU) land surface model (Chen et al. 1996), and a radiation scheme (Lacis and Hansen 1974; Fels and Schwarzkopf 1975). We selected a combination of physical parameterizations for the other two models that are commonly used. For RAMS, these included the Chen–Cotton (1983) cloud-radiation scheme, a level-2.5 turbulence closure scheme with a prognostic equation for turbulent kinetic energy (TKE) and a diagnostic mixing length scale (Mellor and Yamada 1982), and a diagnostic cloud scheme that allows clouds to be produced whenever supersaturation occurs. For MM5, we used the Dudhia (1989) cloud radiation scheme, the Reisner et al. (1998) mixed-phase cloud scheme, and the Grell et al. (1991) cumulus parameterization for the two outermost grids. Clouds and precipitation were not a major issue because of the mostly clear skies observed during the IOPs; therefore, the most relevant scheme for our purposes is the boundary layer parameterization. We compared four boundary layer schemes in MM5 that represent different types of approaches to treat subgrid turbulent motions and found relatively small differences in the quantities of interest to us, except for mixed-layer depths (Zhong and Fast 2002). As a result, we selected the Gayno–Seaman scheme (Shafran et al. 2000), which is a modified Mellor–Yamada level-2.5 scheme, in order to be consistent with the other two models. We also compared the MM5 results from the use of a simple multilayer soil model with those from a more sophisticated land surface model (Chen and Dudhia 2001) and we chose to use the simple model because it actually produced smaller errors in surface temperatures for the Salt Lake valley.
Similar horizontal grid configurations were used for MM5 and RAMS. To resolve the high mountain barriers surrounding the valley and the narrow canyons and mountain gaps, five nested grids were employed with horizontal grid spacing of 45, 15, 5, 1.67, and 0.56 km. The domains for grids 3, 4, and 5 are shown in Fig. 2. Because Meso-Eta does not have nesting capability, it was configured with a domain somewhat larger than grid 4 in Fig. 2 with a horizontal grid spacing of approximately 0.85 km. The vertical grid configurations were similar for the three models; all had a vertical grid spacing of about 30 m adjacent to the surface and 12 levels within the lowest 12 km. The domains of RAMS and MM5 extended up to a height of 12 km MSL, while the top of the Meso-Eta model was at 23 km MSL.
The default land use categories in MM5 and RAMS were based on an older dataset that does not reflect the recent changes resulting from urban sprawl. To adequately define the urban area in the valley, we employed the 2000 U.S. Geological Survey dataset (available online at http://landcover.usgs.gov) for the innermost grid. There are some differences in the characterization of the vegetation types in MM5 and RAMS, but the urban area in our simulations was identical.
Conventional turbulence parameterizations such as the Mellor–Yamada scheme assume that turbulence mixing is dominated by vertical mixing. This assumption may no longer be valid as the role of horizontal mixing increases with reduced horizontal grid spacing in mesoscale model applications. Despite the subkilometer horizontal grid spacing, the aspect ratio of the horizontal to vertical grid spacing near the surface in our model configurations varies on the order of 35:1 to 50:1 so that the horizontal gradients would still be considerably smaller than the vertical gradients. Therefore, the assumption that the vertical mixing is much larger than the horizontal mixing should be valid. Nevertheless, applying conventional turbulence parameterization at small horizontal scales is a research topic that needs to be addressed by the mesoscale modeling community.
4. Case selection and model simulations
The 10 IOPs during the month-long VTMX field campaign have been grouped into two categories based on synoptic and mesoscale conditions (Doran et al. 2002). One category included those with weak synoptic winds and well-developed terrain-induced flows, while the other category included days modulated by synoptic and mesoscale weather systems. For this study, we selected one case from each category to evaluate the model performance for both weak and strong synoptic wind conditions.
The first case was for IOPs 6 and 7, two consecutive days between 2200 UTC 16 October and 1600 UTC 18 October, that were characterized by clear skies and light winds. Wind speeds at the crest level of the Wasatch Mountains (∼700 hPa) were less than 5 m s−1 throughout the 2-day period except near sunrise on 18 October when winds aloft increased to near 10 m s−1 as a result of an approaching upper-level trough. The weak synoptic winds aloft, in combination with strong diurnal heating and cooling in the valley and over the surrounding mountains, allowed the development of pronounced terrain-induced circulations on both days.
The synoptic conditions for the second case, IOP 10 between 2200 UTC 25 October and 1600 UTC 26 October, were significantly different from IOPs 6 and 7. The period began with a weak shortwave ridge and ended with the passage of a shortwave trough. Winds aloft were much stronger (over 10 m s−1) over the entire period and, as a result, terrain-induced local circulations were substantially modulated or suppressed.
For the first case, the model simulations were initialized at 1200 UTC on 16 October and continued for 48 h, ending at 1200 UTC 18 October. The simulations for the second case were initialized at 1200 UTC on 25 October and continued for 36 h, ending at 0000 UTC on 27 October. The model output used for initialization were the NCEP Aviation (AVN) Model analyses for RAMS, the NCEP–NCAR reanalysis for MM5, and the NCEP 32-km operational Eta analyses for Meso-Eta. Some of the differences in the model predictions will result from the different initial and boundary conditions employed by the models. An example of the initial profiles of potential temperature, speed, direction, and specific humidity at the Salt Lake City airport (Fig. 1) along with the observed sounding at 1200 UTC 16 October is shown in Fig. 3. The initial conditions from all of the analyses are nearly identical above the heights of the surrounding mountains and agree very well with the observed soundings. Below the mountain heights inside the valley, there are some differences among the three analyses, but the differences were generally small. Differences between the analyses at other times were similar to those shown in Fig. 3. An examination of the large-scale conditions at the model's lateral boundaries also revealed only small differences. Because conditions within the Salt Lake valley below the ridgetops of the surrounding mountains are highly dependent on local terrain-induced forcing, we expect that the effect of the slightly different initial and lateral boundary conditions will be relatively small.
5. Results and discussions
a. Boundary layer structure and evolution
The vertical profiles of potential temperature and relative humidity simulated by the three models were compared with rawinsonde soundings taken at three sites shown in Fig. 1. The first site was at the Salt Lake City International Airport in the northern part of the valley, where two additional soundings were launched to supplement the standard soundings at 0000 and 1200 UTC (1600 and 0400 LST). The second site, Wheeler Farm, was located in the middle of the valley where soundings were obtained every 1–3 h during IOPs. Rawinsondes were released every 2–3 h during IOPs at the third site, called NCAR, located near the gap in the Traverse Range at the southern end of the valley.
The simulated and observed potential temperature profiles at the three sounding sites for the early morning of IOP 6 and late afternoon of IOP 7 are shown in Fig. 4. Multilayer structures are evident in the early morning profiles with a surface-based radiation inversion between 200 and 300 m above ground. All three models produced a surface-based nocturnal inversion layer, but none captured the details of the multilayer structures aloft. The simulated inversion strength was somewhat weaker than what was observed at Wheeler Farm. At the NCAR site where the observed inversion was weaker compared to that at Wheeler Farm as a result of stronger winds through the gap in the Traverse Range, the agreement between the simulated and observed inversion was better. The observed afternoon profiles show the development of a mixed layer that was about 500–600 m deep at the Wheeler Farm and airport sites and about 800 m at the NCAR site. Near-surface temperatures increased about 10 K from the morning profile at each site and there were relatively small site-to-site differences in mixed-layer temperatures. The simulated mixed-layer depths agreed reasonably well with observations at Wheeler Farm and the airport, but were somewhat lower at the NCAR site. The predicted mixed-layer temperatures were consistently lower than the observed ones by 1–2 K and this cold bias extended beyond the top of the boundary layer to the crest level of the Wasatch Mountains. The cold biases were the smallest in Meso-Eta and largest in MM5 although the differences among the models were relatively small.
The models also failed to accurately produce the details of the multilayer temperature structures during the morning of IOP 10 as shown in Fig. 5. The observed nocturnal surface-based inversion layer reached a height of less than 100 m during this morning, which was much shallower than those that developed during IOPs 6 and 7. The simulated inversion layers were deeper and the inversion strengths were weaker than the observed, a feature consistent with the previous case. Similar to IOPs 6 and 7, the predicted late afternoon potential temperature profiles also showed a cold bias extending from near the surface to over 2000 m, but the biases were smaller than those in the previous case.
The moisture profiles from Meso-Eta were closer to the observed profiles than those from MM5 and RAMS (not shown). The discrepancies were smaller aloft than in the lower levels and smaller in early morning than in the afternoon. MM5 had a wet bias of 1–2 g kg−1 in the lower atmosphere while RAMS showed a dry bias of a similar amount. The discrepancies between simulated and observed moisture profiles were smaller during IOP 10 than during IOPs 6 and 7 for all three models.
The comparison of simulated and observed mixed-layer depths determined using all soundings available during daytime at three sites are shown in Fig. 6 for IOPs 6 and 7. Since the focus of VTMX was the nocturnal boundary layer and most IOPs commenced in late afternoon and continued through the night into the following morning, there were only a few daytime soundings. The limited data points show that all three models predicted the mixed-layer depths that agreed reasonably well with the observed depths at relatively low values, but they all underpredicted the depths at higher values. It is not obvious, from the limited comparison, which model predicted the mixed-layer depths better.
An important quantity governing vertical mixing of near-surface pollutants at night is the strength of the surface-based inversion. As a measure of the inversion strength, we computed the observed and simulated potential temperature gradients between 15 and 150 m for all the sounding times at the Wheeler Farm and NCAR sites. The vertical resolution of the airport soundings was too coarse to allow adequate estimates of the observed inversion strengths. As shown in Fig. 7, the observed gradients fall between 0.02 and 0.06 K m−1 for IOPs 6 and 7. The large values of over 0.05 K m−1 occurred almost exclusively at the Wheeler Farm site over the valley floor in midvalley. At the NCAR site, both MM5 and Meso-Eta underpredicted the inversion strengths all the time while RAMS was closer to the observations. At Wheeler Farm, all three models underpredicted the gradients (with only one exception for RAMS), especially when the observed values were high. For IOP 10, the differences between Wheeler Farm and NCAR were much greater with the observed gradients at Wheeler Farm being 3–4 times higher than those at NCAR. All three models produced weaker inversions at both sites, and the agreements at NCAR were much better than at Wheeler Farm where the simulated values were only a small fraction of the observed values.
b. Terrain-induced circulations
Well-developed slope and valley flows and lake/land breezes were observed by the measurement network during IOPs 6 and 7 as a result of the especially light synoptic winds that prevailed through most of the 2-day period. Figure 8 shows a comparison of simulated surface wind vectors over a portion of the model domain with the surface observations in the afternoon and at night. The 32 surface stations within the subdomain clearly revealed the development and diurnal reversal of the thermally driven flows. In the afternoon (Fig. 8a), the surface winds observed over the valley floor were generally from the north and northwest, representing lake breezes superimposed on upvalley flows. RAMS and MM5 produced a convergence line over the valley resulting from flow around Antelope Island (Fig. 1). Upslope or upcanyon flows developed at the stations near the base of Parleys Canyon and Emigration Canyon as well as the station within Big Cottonwood Canyon. Northerly winds occurred through the gap in the Traverse Range advecting air from the Salt Lake valley to the Provo valley in the south. The wind directions reversed at night (Fig. 8b) with south to southeasterly winds over the valley floor and downslope and downcanyon winds in and near the mouths of the canyons. The downcanyon flows along the Wasatch Mountains were stronger than their daytime counterparts and strong downslope flows also developed over the lower slopes of the Oquirrh Mountains in the southwest. Winds over the valley floor, however, were generally much weaker at night than during the day and in some places they were near calm, which, together with the low near-surface temperature at night, indicated the formation of a cold air pool in the valley at night.
The surface wind fields predicted by RAMS and MM5 were quite similar and were in very good agreement with the observations during both day and night. The fine horizontal resolution allowed the models to reproduce the large spatial variations in the surface wind fields in the valley to a level of detail that was rather impressive, especially at night. The surface wind fields predicted by Meso-Eta were smoother, even though the horizontal grid spacing was only somewhat larger than for MM5 and RAMS, so that the model missed or underestimated the convergence and divergence of the flows in the valley. The largest discrepancy between the observed surface winds and those predicted by Meso-Eta occurred in the southwest part of the valley, where the predicted nighttime winds were predominantly from the south rather than from the southwest down the slope of the Oquirrh Mountains.
Next, we describe the performance of the models in predicting the winds aloft. The simulated boundary layer winds for IOPs 6 and 7 were compared to the hourly wind measurements from two 915-MHz radar wind profilers and the results are shown in Fig. 9. The profiler data were processed with NCAR Improved Moments Algorithm (NIMA), which mimics human experts' ability to identify atmospheric signals in the presence of contaminants (Morse et al. 2002). Located in the middle of the valley, the winds at the Pacific Northwest National Laboratory (PNNL) site (Fig. 1) were dominated by along-valley circulations. The second profiler site, called the Idaho National Engineering and Environmental Laboratory (INEL; Fig. 1), was located north of PNNL and west of the base of Parleys Canyon; therefore, the winds at this site were influenced by the outflow from the canyon at night.
At both sites, the observed wind vectors clearly revealed a pronounced diurnal reversal of winds in the low levels with development of north to northwesterly upvalley flows in the afternoon and southerly downvalley flows at night. The north to northwesterly upvalley flows started in late afternoon around 1600 mountain standard time (MST) on both days and ended about an hour after sunset. The speeds of the upvalley flows were 4–6 m s−1 and they occupied a layer from the surface to about 500 m. Southerly downvalley flows developed after midnight and continued into early morning. At night, the downvalley southerly winds at PNNL were weaker and shallower than their daytime counterparts. The observed nighttime winds at INEL were from the east to southeast and represented a southerly downvalley component superimposed on the easterly downcanyon flows from Parleys Canyon.
All three models were able to capture the diurnal reversal of the low-level winds at the two sites. MM5 and RAMS predicted the afternoon northwesterly upvalley wind speeds within 2 m s−1 of the observed values and the thickness within 100 m of the observed thickness, but the simulated onset of the upvalley flows occurred in midafternoon rather than late afternoon, as observed. The Meso-Eta model also simulated the development of the afternoon upvalley flows, but the speeds of the upvalley winds were much weaker and the layer was also shallower compared to the observations. The simulated southerly downvalley wind speeds agreed reasonably well with the observed at the PNNL site, but the depths of the downvalley flow layers were deeper, especially those of the Meso-Eta model. All models reproduced reasonably well the development of the nighttime easterly flows at INEL between 200 and 600 m AGL from Parleys Canyon. The simulated downcanyon winds were somewhat stronger and deeper in MM5, and weaker in Meso-Eta than the observed. Relatively large discrepancies between simulated and observed winds occurred from time to time in the midvalley atmosphere above 1 km. For example, during the night of 16–17 October, the observed winds between 1 and 2 km at PNNL were northerly at moderate speeds of over 5 m s−1. The simulated winds at these levels during the same time periods were also northerly, but the wind speeds were much weaker than the observed, especially in the case of Meso-Eta where the simulated winds were nearly calm.
The simulated and observed hourly wind profiles at the PNNL site for IOP 10 are shown in Fig. 10. The strong southerly synoptic winds suppressed the development of the terrain-induced flows so that no diurnal low-level wind reversals occurred during this case. Despite the stronger synoptic forcing, relatively large errors were produced by all three models. For example, during the night between 1000 and 2000 m the observed winds were south-southeasterly while the models produced south-southwesterly winds. The simulated wind speeds were generally weaker than observed, with RAMS's predictions being somewhat better than those of MM5 and Meso-Eta.
Figure 11 shows the simulated and observed wind speeds averaged over the lowest 500 m as a function of time for both cases. For IOPs 6 and 7, MM5 and RAMS tracked the observed wind speed changes fairly well, except that the wind speed increase associated with the development of the northwesterly upvalley flows in the afternoons occurred too early in the models. While MM5 and RAMS peak afternoon wind speeds between 4 and 6 m s−1 were close to the observed value of 5 m s−1, the afternoon winds in Meso-Eta were lower than the observed, especially on the second day when the simulated upvalley winds were only 20% of the observed values. The winds from all three models were usually too high during the late evening and early morning between 0600 and 1800 UTC. The relative agreement for IOP 10 was not better than during IOPs 6 and 7 even though the synoptic forcing was much stronger. The predicted low-level winds in MM5 and RAMS followed the observed overall variation in time, while those predicted by Meso-Eta failed to track the rapid increase in the evening and the higher wind speeds during the nighttime hours.
One of the prominent features observed during the VTMX field campaign was a strong low-level jet through the gap in the middle of the Traverse Range on the southern end of the Salt Lake valley. The soundings launched at the NCAR site revealed that the jet, which reverses direction diurnally in response to the combined effects of dynamic channeling and lake/land and mountain/valley thermal contrasts, usually reached a maximum speed of 6–10 m s−1 at a height between 100 to 300 m above ground and decreased rapidly to below 2 m s−1 at 1000 m. Figure 12 shows a comparison of simulated vertical profiles of wind speed and direction with soundings taken at the NCAR site in late afternoon and early morning for IOP 7. In the afternoon, RAMS and MM5 produced a northerly low-level wind speed maximum similar in magnitude to that of the observed at a jet height of around 100 m. However, the northerly winds remained strong to over 700 m before they decreased so that the resulting jet profile was too broad. The Meso-Eta model failed to produce the low-level jet at that time; instead, it produced a wind profile that showed little variation in the lowest 1000 m and a wind speed of only about half of the observed low-level maximum speed. The observed early morning profiles showed southerly winds in the lowest 1200 m and a jet with a peak speed of 8 m s−1 about 150 m above ground. All three models accurately predicted the wind reversal in the lowest 1000–1200 m, but the predicted maximum jet speeds were between 4 and 6 m s−1, which were lower than the observed peak speeds.
c. Surface statistical comparison
The VTMX field campaign featured an extensive surface network of instruments that measured temperature, relative humidity, wind speed, and direction. To evaluate how well each model simulated these surface quantities, hourly model outputs at grid points closest to surface stations were compared to observations at all 36 surface stations within the innermost domain. The following statistical measures were computed to quantify the model errors:
where _ϕ_′ is the difference between the simulated and observed variables and N is 36. The model forecast errors include contributions from both systematic and nonsystematic sources. Systematic errors that are represented by the model bias are usually caused by a consistent misrepresentation of geometrical (topography), physical (e.g., radiation, convection), or numerical factors. Nonsystematic errors, which are indicated by the error standard deviation, represent the random error component caused by uncertainties in model initial and boundary conditions or errors in the observations.
Figure 13 shows the bias, rms error, and error standard deviation as a function of time for IOPs 6 and 7. The temperature biases were all negative except at the beginning when MM5 and RAMS were too warm. The biases were larger during daytime and smaller at night. The biases in MM5 and RAMS closely track each other and the values were smaller than the temperature biases produced by Meso-Eta. The wind speed biases among the models were similar in magnitude and were usually between −1 and +1 m s−1. Large wind direction biases usually occurred at the same time when wind speed biases were relatively large. Similar to temperature, the speed and direction biases in MM5 and RAMS become positive or negative at the same times, which were sometimes different from those of Meso-Eta. For wind speed and direction, the model biases and error standard deviations were comparable in magnitudes, which indicates that the systematic and nonsystematic errors contributed more or less equally to the total errors. For temperature, the relative decrease of mean biases, compared to the error standard deviation during nighttime hours, suggested that nonsystematic errors composed the larger portion of the total temperature forecast errors at night.
Similar diurnal variations of the statistical measures for IOP 10 are shown in Fig. 14. The temperature biases in this case were generally larger than in the previous case. Both RAMS and Meso-Eta produced a cold bias throughout most of the simulation. MM5 produced a warm bias during the first 12 h and a cold bias afterward that was similar to the cold bias in RAMS. Unlike IOPs 6 and 7, when larger biases were found during the day, the cold biases in this case were larger at night than during the day and the biases in Meso-Eta were generally higher than the other two models, which was consistent with what happened during IOPs 6 and 7. The model temperature biases were generally larger than the error standard deviation, which indicated that systematic model errors contributed more strongly to the total temperature error. The biases in wind speed were also larger than those during IOPs 6 and 7; however, the observed and simulated wind speeds are higher as well. Except for 2 h, the speed biases from RAMS and MM5 were within ±1 m s−1. The speed biases in Meso-Eta were always positive at values between 1 and 4 m s−1, indicating a consistent overprediction of surface wind speeds in this case by Meso-Eta. The larger contribution to the total error for wind speed evidently is derived from a systematic model error for Meso-Eta. The wind direction biases for this case fall in the same range as that for IOPs 6 and 7. For MM5 and RAMS, the biases and the error standard deviations were comparable, indicating similar contributions from the systematic and nonsystematic errors to the total error. The larger bias values in Meso-Eta indicate that the systematic error component accounted for a larger fraction of the total forecast error.
Tables 1 and 2 summarize the bias, rms error, and error standard deviation averaged over daytime or nighttime periods as well as the entire time period. The mean temperature biases were all negative for the three models in both cases. The mean cold biases were larger during the day than at night for IOPs 6 and 7, but during IOP 10 the largest mean cold bias occurred during the night. In both cases the largest cold bias was produced by the Meso-Eta model. The mean speed biases were small for RAMS and MM5 (0.5–0.7 m s−1) in both cases. The mean speed bias for Meso-Eta was almost negligible for IOPs 6 and 7 (<0.2 m s−1), but for IOP 10, the mean bias was much larger, at over 1.9 m s−1. The mean biases for wind direction were generally small, but the error standard deviations were quite large.
To examine whether the model forecast errors depend on the locations of the sites, we grouped the 36 stations into three different categories representing valley floor (20 sites), slope (9 sites), and mountain (7 sites). We computed the statistical measures for the three subgroups and found only small differences between the subgroups in all measures. The only significant difference existed for wind direction predicted by Meso-Eta when the biases and rms errors over the slopes (bias, 20°; rmse, 82°) and mountains (bias, −16°; rmse, 72°) were much larger than over the valley floor (bias, −4°; rmse, 57°). This finding is consistent with the surface wind vectors shown in Fig. 8. Further dividing into day and night indicated that the large biases in Meso-Eta-predicted surface wind direction at the slopes and mountains occurred primarily during the night (slopes, 56°; mountains, −33°) rather than during the day (slopes, −17°; mountains, 2°). the statistics indicate that MM5 and RAMS better represent the upslope and downslope flows along the Wasatch and Oquirrh Mountains than Meso-Eta.
6. Discussion
All three models evaluated in this study were able to qualitatively reproduce the general features in boundary layer structure and evolution as revealed by the observations such as boundary layer depth, daytime upvalley flow, nighttime downvalley and canyon flows, and the near-surface convergence and divergence over the valley floor. Nevertheless, relatively large errors associated with the timing, dimensions, and magnitude of these local circulations and the magnitude of the boundary layer temperatures still occurred. In this section we discuss some of the errors common to all three models that are relevant to air quality applications.
One interesting aspect of this study is where the largest wind speed and direction errors occurred in the valley atmosphere. While the models captured the observed features of the up- and downvalley flows within 500 m of the ground, large differences between the simulated winds and the observations from the radar wind profilers often occurred between 1 and 2 km AGL. For example, during the evening of IOP 6, Meso-Eta produced light and variable winds between 1 and 2 km AGL (Fig. 9a) where the observed winds were moderate at 4 and 6 m s−1. The winds from MM5 and RAMS were closer to the observations in general, but the errors were still large at times. In the early evening hours of 26 October during IOP 10, the observed winds between 1 and 2 km were from the southeast at speeds greater than 5 m s−1, but the simulated winds were much weaker and were from south to southwest. These results indicate that the dynamic and thermodynamic effects of the mountains on the ambient flow and/or the regional pressure gradients were not simulated adequately. The failure to adequately capture the middle- to upper-valley atmosphere winds will have an effect on the vertical transport and mixing of pollutants from the urban to regional scales. It is worth noting, however, that the profiler winds aloft may not be as accurate as in the low levels because the signal-to-noise ratio usually decreases with height.
One might expect improved performances of mesoscale models under conditions of stronger synoptic forcing. This was not true when comparing the two cases in this study. During IOPs 6 and 7, when the ambient winds were the weakest, the average low-level winds predicted by RAMS and MM5 at the PNNL site were usually within 1 or 2 m s−1 of the observed values and both models predicted the peak wind speeds correctly (Fig. 11). When the ambient winds were stronger during IOP 10, the RAMS- and MM5-simulated peak wind speeds between 0700 and 1200 UTC on 26 October were 1–4 m s−1 too low compared to the observed peak wind speed (∼11 m s−1). Meso-Eta did not predict an increase in the nighttime wind speed for this case at all. The predicted lower than observed wind speeds in this case would result in pollutants not being transported far enough downwind and in errors in vertical dispersion associated with mechanical mixing.
A cold bias extending from the surface to the top of the valley atmosphere was a common error that occurred in all three models and for both cases. One possible cause for this could simply be a cold bias in the model initial conditions, but a comparison of the model initial temperature profiles with the 1200 UTC temperature soundings taken at the airport revealed no initial cold bias in MM5 and RAMS for either cases (Fig. 3 for IOPs 6 and 7). The comparison of observed and simulated near-surface temperatures (Figs. 13 and 14) also showed either a warm bias or a slight cold bias for MM5 and RAMS initially at the surface. Therefore, a cold bias in the initial conditions was unlikely to explain the relatively large cold biases that developed subsequently in the RAMS and MM5 simulations. For Meso-Eta, a cold bias existed initially both in the atmosphere and near the surface for IOPs 6 and 7, which may have contributed partially to the cold biases seen in this case. However, in the case of IOP 10, the Meso-Eta initial temperature profile was nearly identical to the airport sounding and the initial surface temperatures were even slightly higher than observed. Yet, a cold bias in the valley atmosphere subsequently developed during the simulation. The cold biases in the simulated daytime boundary layer temperatures were found to persist throughout the night, leading to a nighttime cold bias in the valley atmosphere. For example, temperatures within the residual layer between 0.3 and 1.0 km AGL were 2°–3°C too low throughout the night during IOP 6 (i.e., 1200 UTC 17 October; Fig. 4), which was clearly a carryover of the cold bias in the mixed layer, which developed during the previous afternoon.
Another common model error, previously reported by Hanna and Yang (2001), was that the predicted strengths of nocturnal surface-based inversions were generally too weak compared to the observations. This problem was most severe adjacent to the valley floor where the observed surface inversion was particularly strong. This implies that the predicted vertical mixing and diffusion would be stronger than observed at night. If the model output were used to drive an air quality model, pollutants released over the valley floor would be mixed through a deeper layer and consequently be transported farther away from their source areas.
The relatively low vertical resolution may be one factor contributing to the underestimation of the near-surface vertical temperature gradients at night. To investigate this possibility, additional RAMS simulations were performed for the two cases that doubled the number of vertical levels in the lowest 1000 m from 12 to 24, with the height of the first grid point reduced from 15 to 7 m above ground. An example of the vertical profiles of temperature, wind speed, and direction at the Wheeler Farm site at 1200 UTC 17 October during IOP 7 is shown in Fig. 15. The predicted potential temperatures from this simulation were closer to the observed profile than the original RAMS simulation using coarser vertical resolution. As shown in Tables 1 and 2, statistics associated with the surface temperature and wind speeds improved with the higher vertical resolution. Nevertheless, the near-surface potential temperature gradient of 0.04 K m−1 was still much less than the observed value of 0.1 K m−1 even though the fine vertical resolution should have been sufficient to resolve this feature. Vertical resolution, therefore, is unlikely to be the only reason for the predicted errors within the valley atmosphere.
Another potential cause for the cold biases and low near-surface vertical temperature gradients may be inadequate representation of the vertical mixing in the models. To examine the sensitivity of temperature biases to model boundary layer parameterizations, simulations of IOPs 6 and 7 were repeated using either the Blackadar (Blackadar 1976, 1979) or the Medium-Range Forecast (MRF; Hong and Pan 1996) model boundary layer parameterizations in MM5. These two schemes use somewhat different parameters for representing boundary layer turbulence, and they differ considerably from the Gayno–Seaman scheme employed in the original MM5 run in that they are first-order closure schemes based entirely on the predicted mean variables. Despite the differences in their treatment of boundary layer turbulence, simulations with these two schemes also produced cold biases at the surface and in the valley atmosphere with magnitudes similar to those in the original MM5 simulation using the Gayno–Seaman scheme (Zhong and Fast 2002). Obviously, more analyses are required in order to gain insight into why all three schemes failed to reproduce the observed temperature values, but the sensitivity test suggests that factors other than errors in turbulent mixing may have played a larger role in producing the cold bias in the model simulations.
To further evaluate the representation of vertical mixing in the models, the observed and predicted TKE for IOPs 6 and 7 at the PNNL and NCAR sites are shown in Fig. 16. The simulated values were obtained by averaging the nine grid points closest to the observation site. It is interesting to note how much the predicted TKE values differed among the three models at each site and how much their agreement with observations changed from one site to another. At PNNL, the best agreement between predicted and observed TKE values came from MM5, while at NCAR, Meso-Eta showed much better agreement than the other two models. Both RAMS and Meso-Eta captured the large increase in TKE from PNNL to NCAR that was a result of much larger wind shears at NCAR, while MM5 failed to predict this increase. Another problem with MM5 is that turbulence was completely shut down at night, while in reality significant mixing still occurred, especially at the NCAR site where large wind shears existed as a result of the strong gap flow at night. For example, the predicted TKE values at night by RAMS and Meso-Eta were higher than the observed at both sites, which may partially explain why the simulated nocturnal temperature gradients in the surface-based inversion layer were too weak. This, on the other hand, could not be an explanation for why the MM5-predicted inversion strengths were also too weak because turbulent mixing was mostly eliminated at night.
In the Mellor–Yamada turbulence parameterization, the vertical eddy exchange coefficients for heat, K h, and momemtum, K m, are determined from
where S m and S h are nondimensional eddy exchange coefficients and l is a length scale. If TKE is too large, then the vertical eddy exchange coefficients will be too large as well. The measurements at the Arizona State University (ASU) site (Fig. 1) were employed to evaluate the eddy exchange coefficients predicted by RAMS (K h and K m are not saved by MM5 and Meso-Eta). At this site, sonic anemometers and other measurements (Fig. 1) were made on a tower at two levels, at about 5 and 14 m AGL, so that the observed eddy exchange coefficient could be computed from
Observed K h ranged from 0.05 to 0.25 m2 s−1 during the daytime, but the simulated K h from RAMS was between 1 and 4 m2 s−1. Observed K h decreased to 0.05 m2 s−1 or lower at night when downslope wind speeds of 2 m2 s−1 occurred. While the simulated nighttime temperatures and winds were close to the observed values, the simulated K h values were several orders of magnitude higher than observed between 0.5 and 2 m2 s−1. This comparison further demonstrates that predicted vertical mixing near the surface was too strong, in general.
A problem in the models' surface energy balance is likely to be the major factor contributing to the cold bias in the near-surface temperatures. Unfortunately, there were no complete measurements of the surface energy budget to permit us to examine this issue. We did, however, evaluate components of the surface energy balance using measurements from sonic anemometers and radiometers at several locations in the valley. Figure 16 also depicts the observed and simulated sensible heat fluxes for IOPs 6 and 7 at the PNNL and NCAR sites. The simulated sensible heat flux values were obtained by averaging the nine grid points closest to the observation site. At PNNL (Fig. 16a), the predicted downward sensible heat fluxes at night from all three models were slightly larger than the observed, and during the day the predicted upward fluxes from RAMS and MM5 were generally too high compared to the observed, while the fluxes from the Meso-Eta model were in good agreement with the observations. The predicted near-surface daytime temperatures at this site were usually lower than the observed by 1°–3°C for RAMS and Meso-Eta, while the temperatures from MM5 were close to the observations. The observed fluxes at NCAR (Fig. 16b) were 50–70 W m−2 higher than at PNNL and the agreement between modeled and observed fluxes were generally better. The predicted afternoon temperatures from RAMS and Meso-Eta, however, again were lower than observed. Comparisons of measured and modeled sensible heat fluxes at three other sites were similar to those at the PNNL site with the predicted sensible heat fluxes from RAMS and MM5 being too high during the day and night.
Another component of the surface energy balance we can examine is the radiative fluxes. The predicted solar and net radiative fluxes were compared with observations at the NCAR site for IOP 6 and 7. The predicted incoming solar radiation from MM5 and RAMS lagged behind the observations by 15–30 min so that the morning values were generally lower and the afternoon values were higher than observed. But as shown in Table 3, the integrated solar energy from all three models was adequately handled with errors of less than 3%. A test simulation in RAMS indicated that the lag in solar radiation was due to the zenith angle calculation. A more robust zenith angle calculation eliminated the lag in solar radiation, but there was only a minor improvement in other predicted quantities. The net radiation during the day, however, was underpredicted by approximately 6% in MM5 and 12% in RAMS, indicating that either reflection of the incoming solar energy was too large or the net longwave radiative fluxes were too small. The smaller net radiation in MM5 and RAMS was consistent with the cold biases produced by the two models. The Meso-Eta, on the other hand, overpredicted incoming solar radiation by about 12% and net radiation by a similar amount, which was inconsistent with the overall cold bias in its predicted temperature. During nighttime, all three models overpredicted the longwave radiation loss by a substantial amount (>100% for RAMS, 45%–65% for MM5, and 65%–90% for Meso-Eta), which might be a significant factor for the cold surface temperature biases at night. It would be useful to know whether the large longwave radiation loss was the result of too much outgoing radiation at the surface or too little incoming longwave flux, but there were no direct measurements of downward longwave radiation and surface radiative temperatures during VTMX to separate the two components.
Finally, the urban area itself probably needs to be more accurately represented by the land use parameterizations of mesoscale models. There is only one land use category for urban areas in these models and the values for albedo, emissivity, leaf area index, vegetation fraction, roughness length, and displacement height may not be representative of Salt Lake City. The actual land use distribution across the valley is complex. A dense urban core consisting of buildings up to 120 m tall is located in the northeastern end of the valley, which is surrounded by open commercial areas, vegetated suburban areas, and semirural development at the edge of the city. The differences in the surface properties among these urban land use types will likely have an effect on the heating and cooling of the boundary layer that is not accounted for by the models. For example, a small urban heat island effect was detected in the vicinity of the downtown area during the VTMX field campaign (Allwine et al. 2001), which none of the models was able to reproduce.
7. Summary
We have presented, what is to our knowledge, the first detailed intercomparison and evaluation of three mesoscale models, RAMS, MM5, and Meso-Eta, applied to an urban valley at a horizontal grid spacing of less than 1 km. Data from the VTMX field campaign in the Salt Lake valley were used for the evaluation. Simulations were carried out for two cases with light (IOPs 6 and 7) and strong (IOP 10) winds above the crest of the mountains surrounding the valley.
The intercomparison and evaluation focused on boundary layer development and local circulations in the valley. All three models were able to capture reasonably well the general features revealed by the observations such as boundary layer depth, daytime upvalley flow, nighttime downvalley and canyon flows, and the near-surface convergence and divergence over the valley floor. Still, relatively large errors associated with the timing, dimensions, and magnitude of these local circulations and the boundary layer temperatures and vertical temperature gradients were produced.
In both cases, a cold bias of 1–3 K was found for both day and night from the floor of the valley to the top of the valley atmosphere. The biases at the surface were larger during the day in the case of weak synoptic winds and larger at night when the synoptic winds were stronger. Overall, the largest biases were produced by Meso-Eta. In the valley atmosphere, the cold biases were slightly smaller in the strong synoptic wind case and the largest biases were from MM5, although the differences among the models were generally small.
The overall colder predicted daytime temperatures in the valley did not correspond to generally higher predicted surface sensible heat fluxes, but the nighttime cold biases may be partially explained by too great a predicted downward sensible heat flux at night when compare to observed fluxes. Although total incoming solar radiation was adequately predicted by MM5 and RAMS, an underestimation in net radiation during the day may be one of the reasons for the daytime cold bias in the two models. The errors in the radiative flux predictions, however, could not explain the daytime cold bias in the Meso-Eta model because the radiative fluxes were overpredicted by Meso-Eta during the day. At night, all three models overestimated substantially the longwave radiation loss, which may have played a major role in producing the cold bias at night, and which indicates a need for a careful examination and evaluation of the longwave radiation schemes in these models.
Mixed-layer depth during the day and the surface-based inversion strengths at night are two important factors in determining air quality conditions in the valley. The predicted mixed-layer depths agreed reasonably well with the observations when the mixed layers were low (<600 m), but the predictions were generally too low for deeper mixed layers. The predicted nocturnal surface-based inversion layers were generally deeper than the observed, especially for the case of strong winds when the observed inversions were quite shallow (∼50 m). At the midvalley floor site where the observed inversion strengths were strong, the predicted inversion strengths of all three models were much too weak. The simulated inversion layers were closer to the observed at the site near the gap in the Traverse Range where the winds were usually stronger than over the middle of the valley. Increased vertical resolution in the low levels resulted in a slightly improved inversion prediction, but the errors were still substantial.
The low-level flow patterns in the Salt Lake valley were dominated by thermally induced local circulations when the synoptic winds were weak, and all three models were able to capture the characteristics of these local circulations. RAMS and MM5 adequately predicted the depths and speeds of the afternoon upvalley flows, but the predicted onset time was somewhat too early. Meso-Eta predicted weaker and shallower upvalley winds. Nighttime downvalley flows were predicted reasonably well by all three models. All models successfully predicted the channeling and the directional reversal of the winds through the mountain gap in the southern end of the valley, but they failed to reproduce the low-level jet profile during the day and predicted a low-level jet at night that was too weak. Forecast biases of the surface winds were relatively small. Relatively large discrepancies between the observed and simulated wind speed and direction were found at times between 1 and 2 km AGL, and the errors were not necessarily smaller for stronger ambient winds.
Despite the differences in the numerics and physical parameterizations in the three models, their forecast errors were surprisingly similar in nature. RAMS and MM5 outperformed the Meso-Eta in that they better captured terrain-induced circulations and the associated convergence and divergence in the valley.
With the development of parallel versions of these mesoscale models and the availability of relatively inexpensive Linux clusters, mesoscale numerical forecasting at subkilometer resolution will become a reality in the near future. This study shows that although significantly improved horizontal as well as vertical (as in the case of RAMS) resolution may allow mesoscale models to produce more detailed boundary layer structures and allow local circulation patterns to be used for local forecasting, relatively large forecast errors may still exist, which suggests that the future for accurate mesoscale forecasting still lies in improved parameterizations of the surface energy budget, especially longwave radiation and turbulent mixing.
Acknowledgments
This work was supported by the U.S. Department of Energy, under the auspices of the Atmospheric Sciences Program of the Office of Biological and Environmental Research. We would like to thank John Leone of Lawrence Livermore National Laboratory for providing a version of the USGS land use dataset and Matthew Pyle of NCEP for his assistance with the Meso-Eta model. We would also like to thank the participants of the VTMX field campaign whose datasets were used in this study, including but not limited to Will Shaw, John Hubbe, Chris Doran, John Horel, Dave Parsons, Kirk Clawson, Lacey Holland, and Mike Splitt.
REFERENCES
- Allwine, K. J., J. D. Fast, and J. C. Torcolini, 2001: Measured surface temperature distribution across an urban area. .
Allwine, K. J., J. D. Fast, and J. C. Torcolini, 2001: Measured surface temperature distribution across an urban area. Preprints, Third Int. Symp. on Environmental Hydraulics, Tempe, AZ, Arizona State University.
)| false - Black, T. L., 1994: The new NMC mesoscale Eta Model: Description and forecast examples. _Wea. Forecasting, 9 , 265–278.
Black, T. L., 1994: The new NMC mesoscale Eta Model: Description and forecast examples. Wea. Forecasting, 9, 265–278.
)| false - Blackadar, A. K., 1976: Modeling the nocturnal boundary layer. .
Blackadar, A. K., 1976: Modeling the nocturnal boundary layer. Preprints, Third Symp. on Atmospheric Turbulence, Diffusion, and Air Quality, Raleigh, NC, Amer. Meteor. Soc., 46–49.
)| false - Blackadar, A. K., 1979: High resolution models of the planetary boundary layer. .
Blackadar, A. K., 1979: High resolution models of the planetary boundary layer. Advances in Environmental Science and Engineering, Vol. 1, J. Pfafflin and E. Ziegler, Eds., Gordon and Breach, 50–85.
)| false - Chen, C., and W. R. Cotton, 1983: A one-dimensional simulation of the stratocumulus-capped mixed layer. _Bound.-Layer Meteor., 25 , 289–321.
Chen, C., and W. R. Cotton, 1983: A one-dimensional simulation of the stratocumulus-capped mixed layer. Bound.-Layer Meteor., 25, 289–321.
)| false - Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. _Mon. Wea. Rev., 129 , 569–585.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585.
)| false - Chen, F., and . 1996: Modeling of land-surface evaporation by four schemes and comparison with FIFE observations. _J. Geophys. Res., 101 , 7251–7268.
Chen, F., and Coauthors. 1996: Modeling of land-surface evaporation by four schemes and comparison with FIFE observations. J. Geophys. Res., 101, 7251–7268.
)| false - Colle, B. A., K. J. Westrick, and C. F. Mass, 1999: Evaluation of MM5 and Eta-10 precipitation forecasts over the Pacific Northwest during the cool season. _Wea. Forecasting, 14 , 137–154.
Colle, B. A., K. J. Westrick, and C. F. Mass, 1999: Evaluation of MM5 and Eta-10 precipitation forecasts over the Pacific Northwest during the cool season. Wea. Forecasting, 14, 137–154.
)| false - Cox, R., B. L. Bauer, and T. Smith, 1998: A mesoscale model intercomparison. _Bull. Amer. Meteor. Soc., 79 , 265–283.
Cox, R., B. L. Bauer, and T. Smith, 1998: A mesoscale model intercomparison. Bull. Amer. Meteor. Soc., 79, 265–283.
)| false - Doran, J. C., J. D. Fast, and J. Horel, 2002: The VTMX 2000 campaign. _Bull. Amer. Meteor. Soc., 83 , 537–551.
Doran, J. C., J. D. Fast, and J. Horel, 2002: The VTMX 2000 campaign. Bull. Amer. Meteor. Soc., 83, 537–551.
)| false - 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.
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.
)| false - Fast, J. D., R. P. Addis, and B. L. O'Steen, 1995: Advanced atmospheric modeling for emergency response. _J. Appl. Meteor., 34 , 626–649.
Fast, J. D., R. P. Addis, and B. L. O'Steen, 1995: Advanced atmospheric modeling for emergency response. J. Appl. Meteor., 34, 626–649.
)| false - Fels, S. B., and M. D. Schwarzkopf, 1975: The simplified exchanged approximation: A new method for radiative transfer calculations. _J. Atmos. Sci., 32 , 1475–1488.
Fels, S. B., and M. D. Schwarzkopf, 1975: The simplified exchanged approximation: A new method for radiative transfer calculations. J. Atmos. Sci., 32, 1475–1488.
)| false - Grell, G. A., Y-H. Kuo, and R. Rasch, 1991: Semi-prognostic tests of cumulus parameterization schemes in the midlatitudes. _Mon. Wea. Rev., 119 , 5–31.
Grell, G. A., Y-H. Kuo, and R. Rasch, 1991: Semi-prognostic tests of cumulus parameterization schemes in the midlatitudes. Mon. Wea. Rev., 119, 5–31.
)| false - Grell, G. A., J. Dudhia, and D. R. Stauffer, 1993: A description of the fifth generation Penn State/NCAR mesoscale model. .
Grell, G. A., J. Dudhia, and D. R. Stauffer, 1993: A description of the fifth generation Penn State/NCAR mesoscale model. NCAR Tech. Note 398 + 1A, 122 pp. [Available from National Center for Atmospheric Research, P.O. Box 3000, Boulder, CO 80303.].
)| false - Hanna, S. R., and R. Yang, 2001: Evaluations of mesoscale models' simulations of near-surface winds, temperature gradients, and mixing depths. _J. Appl. Meteor., 40 , 1095–1104.
Hanna, S. R., and R. Yang, 2001: Evaluations of mesoscale models' simulations of near-surface winds, temperature gradients, and mixing depths. J. Appl. Meteor., 40, 1095–1104.
)| false - Hogrefe, C., and . 2001: Evaluating the performance of regional-scale photochemical modeling systems: Part I—Meteorological predictions. _Atmos. Environ., 35 , 4159–4174.
Hogrefe, C., and Coauthors. 2001: Evaluating the performance of regional-scale photochemical modeling systems: Part I—Meteorological predictions. Atmos. Environ., 35, 4159–4174.
)| false - Hong, S-H., and H-L. Pan, 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. _Mon. Wea. Rev., 124 , 2322–2339.
Hong, S-H., and H-L. Pan, 1996: Nonlocal boundary layer vertical diffusion in a medium-range forecast model. Mon. Wea. Rev., 124, 2322–2339.
)| false - Horel, J. D., and C. V. Gibson, 1994: Analysis and simulation of a winter storm over Utah. _Wea. Forecasting, 9 , 479–494.
Horel, J. D., and C. V. Gibson, 1994: Analysis and simulation of a winter storm over Utah. Wea. Forecasting, 9, 479–494.
)| false - Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. _Mon. Wea. Rev., 122 , 927–945.
Janjić, Z. I., 1994: The step-mountain eta coordinate model: Further developments of the convection, viscous sublayer, and turbulence closure schemes. Mon. Wea. Rev., 122, 927–945.
)| false - Lacis, A. A., and J. A. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth's atmosphere. _J. Atmos. Sci., 31 , 118–133.
Lacis, A. A., and J. A. Hansen, 1974: A parameterization for the absorption of solar radiation in the earth's atmosphere. J. Atmos. Sci., 31, 118–133.
)| false - Mass, C. F., and Y-H. Kuo, 1998: Regional real-time numerical weather prediction: Current status and future potential. _Bull. Amer. Meteor. Soc., 79 , 253–263.
Mass, C. F., and Y-H. Kuo, 1998: Regional real-time numerical weather prediction: Current status and future potential. Bull. Amer. Meteor. Soc., 79, 253–263.
)| false - Mellor, G. L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. _Rev. Geophys. Space Phys., 20 , 851–875.
Mellor, G. L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys., 20, 851–875.
)| false - Mesinger, F., Z. I. Janjić, S. Ničković, D. Gavrilov, and D. G. Deaven, 1988: The step-mountain coordinate: Model description and performance for cases of Alpine lee cyclogenesis and for a case of an Appalachian redevelopment. _Mon. Wea. Rev., 116 , 1493–1518.
Mesinger, F., Z. I. Janjić, S. Ničković, D. Gavrilov, and D. G. Deaven, 1988: The step-mountain coordinate: Model description and performance for cases of Alpine lee cyclogenesis and for a case of an Appalachian redevelopment. Mon. Wea. Rev., 116, 1493–1518.
)| false - Morse, C. S., R. K. Goodrich, and L. B. Corman, 2002: The NIMA method for improved moment estimation from Doppler spectra. _J. Atmos. Oceanic Technol., 19 , 274–295.
Morse, C. S., R. K. Goodrich, and L. B. Corman, 2002: The NIMA method for improved moment estimation from Doppler spectra. J. Atmos. Oceanic Technol., 19, 274–295.
)| false - Pielke, R. A., and . 1992: A comprehensive meteorological modeling system—RAMS. _Meteor. Atmos. Phys., 49 , 69–91.
Pielke, R. A., and Coauthors. 1992: A comprehensive meteorological modeling system—RAMS. Meteor. Atmos. Phys., 49, 69–91.
)| false - Reisner, J., R. T. Bruintjes, and R. Rasmussen, 1998: An examination on the utility of forecasting supercooled liquid water in a mesoscale model. _Quart. J. Roy. Meteor. Soc., 124 , 1071–1107.
Reisner, J., R. T. Bruintjes, and R. Rasmussen, 1998: An examination on the utility of forecasting supercooled liquid water in a mesoscale model. Quart. J. Roy. Meteor. Soc., 124, 1071–1107.
)| false - Shafran, P. C., N. L. Seaman, and G. A. Gayno, 2000: Evaluation of numerical predictions of boundary layer structure during the Lake Michigan Ozone Study (LMOS). _J. Appl. Meteor., 39 , 412–426.
Shafran, P. C., N. L. Seaman, and G. A. Gayno, 2000: Evaluation of numerical predictions of boundary layer structure during the Lake Michigan Ozone Study (LMOS). J. Appl. Meteor., 39, 412–426.
)| false - Sistla, G., W. Hao, J-Y. Ku, G. Kallos, K. Zhang, H. Mao, and S. T. Rao, 2001: An operational evaluation of two regional-scale air quality modeling systems over the United States. _Bull. Amer. Meteor. Soc., 82 , 945–964.
Sistla, G., W. Hao, J-Y. Ku, G. Kallos, K. Zhang, H. Mao, and S. T. Rao, 2001: An operational evaluation of two regional-scale air quality modeling systems over the United States. Bull. Amer. Meteor. Soc., 82, 945–964.
)| false - Smallcomb, C., 2001: Using an operational MM5 in the fire weather forecast process for west Texas and southeastern New Mexico. .
Smallcomb, C., 2001: Using an operational MM5 in the fire weather forecast process for west Texas and southeastern New Mexico. Preprints, 11th PSU/NCAR Mesoscale Model User's Workshop, Boulder, CO, PSU/NCAR, 32–35.
)| false - White, B. G., J. Paegle, J. W. Steenburgh, J. D. Horel, R. T. Swanson, L. K. Cook, D. J. Onton, and J. G. Miles, 1999: Short-term forecast validation of six models. _Wea. Forecasting, 14 , 84–108.
White, B. G., J. Paegle, J. W. Steenburgh, J. D. Horel, R. T. Swanson, L. K. Cook, D. J. Onton, and J. G. Miles, 1999: Short-term forecast validation of six models. Wea. Forecasting, 14, 84–108.
)| false - Zhao, Q., and F. H. Carr, 1997: A prognostic cloud scheme for operational NWP model. _Mon. Wea. Rev., 125 , 1931–1953.
Zhao, Q., and F. H. Carr, 1997: A prognostic cloud scheme for operational NWP model. Mon. Wea. Rev., 125, 1931–1953.
)| false - Zhong, S., and J. D. Fast, 2002: An evaluation of boundary layer and land surface parameterizations using data from the VTMX field campaign in the Salt Lake Valley. .
Zhong, S., and J. D. Fast, 2002: An evaluation of boundary layer and land surface parameterizations using data from the VTMX field campaign in the Salt Lake Valley. Preprints, 15th Symp. on Boundary Layers and Turbulence, Wageningen, Netherlands, Amer. Meteor. Soc., 560–563.
)| false
Fig. 1.
Topography (100 m contours) in the vicinity of the Salt Lake valley and the location of VTMX instrumentation employed by the present study. Squares denote upper-air stations, circles denote special surface meteorological stations, and triangles denote operational mesonet stations
Citation: Monthly Weather Review 131, 7; 10.1175/1520-0493(2003)131<1301:AEOTMR>2.0.CO;2
Fig. 9.
(a) Observed winds from the PNNL radar wind profiler during IOPs 6 and 7, and the predicted winds from the RAMS, MM5, and Meso-Eta models. (b) Same as in (a) except for the INEL radar wind profiler. Gray shading denotes nighttime periods
Citation: Monthly Weather Review 131, 7; 10.1175/1520-0493(2003)131<1301:AEOTMR>2.0.CO;2
Fig. 13.
(top) Bias, (middle) rmse, and (bottom) error standard deviation of the (left) simulated temperature, (center) wind speed, and (right) wind direction at 36 stations in the Salt Lake valley and over the foothills during IOPs 6 and 7. Gray shading denotes nighttime periods
Citation: Monthly Weather Review 131, 7; 10.1175/1520-0493(2003)131<1301:AEOTMR>2.0.CO;2
Fig. 15.
Observed (left) potential temperature, (center) wind speed, and (right) direction profiles at Wheeler Farm at 1200 UTC 17 Oct (IOP 7) and predicted profiles from RAMS and another RAMS simulation with a smaller vertical grid spacing
Citation: Monthly Weather Review 131, 7; 10.1175/1520-0493(2003)131<1301:AEOTMR>2.0.CO;2
Fig. 16.
Observed (dots) and simulated (lines) turbulent kinetic energy, sensible heat flux, and temperature at the (a) PNNL and (b) NCAR sites during IOPs 6 and 7. Simulated values obtained by averaging the nine grid points closest to the observational site. Gray shading denotes nighttime periods.
Citation: Monthly Weather Review 131, 7; 10.1175/1520-0493(2003)131<1301:AEOTMR>2.0.CO;2
Table 1.
Model bias, rmse, and standard deviation (std dev) of the errors for (top) surface temperature, (middle) wind speed, and (bottom) wind direction based on 36 stations in the vicinity of Salt Lake City during IOPs 6 and 7. Number of half-hourly observations is shown in parentheses in the time column. “RAMS high” refers to an identical simulation of RAMS, except with twice the number of grid points within 1 km of the surface
Table 2.
Same as Table 1, except for IOP 10
Table 3.
Comparison of integrated downward shortwave and net radiative fluxes (MJ m−2 ) at the NCAR site for IOP 6 and 7 (16–18 Oct)