1. Introduction
The lower Colorado River basin, located in the center of the semiarid southwest United States (Fig. 1), is characterized by high heterogeneity in topography, weather, and hydrology, providing a difficult challenge for numerical prediction of summer rainfall over the region. From the high terrain of the northeastern half of the basin to the low-elevation desert on the southwest (separated by the Mogollon Rim), the daily average temperature difference can reach 20°C, and the difference in annual precipitation is greater than 500 mm. The precipitation meteorology of the region consists of two distinct seasons: winter and summer. The spring and fall tend to be very dry over the southwestern half of the basin, and to be the transition season over the northeastern half of the basin. Winter weather of the region is dominated by midlatitude synoptic systems, which bring large amounts of snowfall in the high terrain. In the summer, rainfall in the southwestern portion of the basin is subtropical in origin and is strongly influenced by the North American monsoon (NAM; Higgins et al. 1997). Moist southerly and southeasterly winds across the U.S.–Mexico border bring convective instability to the region. Convective systems develop preferentially over high terrain areas and usually produce short-lived, intense thunderstorms (Douglas et al. 1993). In low-elevation areas, the atmospheric buoyancies are generated with strong solar heating, but the synoptic subsidence in the high levels sharply suppresses the development of deep convection. When the midlevel wind profile is favorable, organized mesoscale convective systems (MCSs; Smith and Gall 1989) and thunderstorms can occur, particularly over the southwestern portion of the basin. The resulting summertime rainfall is, therefore, highly localized and strongly influenced by topography and subtropical weather features, with the storms sometimes accompanied by strong winds and flash flooding. In the last three decades, the southwest has experienced the fastest-growing population and economy in the United States (Wallace et al. 1999). Marginal availability of water resources in the semiarid region necessitates accurate meteorology and hydrology predictions, especially for quantitative precipitation forecasts (QPFs).
Many previous studies have contributed to a better understanding of the characteristics of the summer weather in this semiarid region, especially their relationship with NAM. In the summer of 1990, a field experiment of the Southwest Area Monsoon Project (SWAMP; Meitin et al. 1991) was conducted, which launched pilot balloons accompanying ground observations to investigate the wind, moisture, and temperature fields over western Mexico and southern Arizona. The observations showed that moist air from the Gulf of California sometimes surged northward into Arizona at very low levels (<700 hPa). Using SWAMP data, Douglas et al. (1993) and Douglas and Li (1996) analyzed the low-level jet (LLJ) stream over the Gulf of California and the diurnal variation of wind fields over the Arizona low desert. Maddox et al. (1995) summarized 11-yr observations of the North American climates, identifying three synoptic patterns associated with severe summertime thunderstorms in the southwest United States. McCollum et al. (1995) analyzed the precursor and supportive environment of a central Arizona MCS using synoptic data and observations from SWAMP. Higgins et al. (1997) and Higgins and Shi (2000) studied the variability of southwest monsoon rainfall and causative factors. Wallace et al. (1999) examined the use of Tucson sounding data for predicting thunderstorm occurrence over the Arizona low desert and found Tucson sounding data to be nonrepresentative of conditions over most of the Sonoran Desert. They also found that the surges of moist air from the Gulf of California frequently do not occur immediately prior to thunderstorm events in central Arizona. These studies indicated that the high variability of summer rainfall in the region results from the mesoscale circulations due to complicated interactions between the subtropical weather systems and the local geographic and thermodynamic conditions. The mechanisms controlling these circulation patterns in the region need to be better understood and quantified. One way to do this is through numerical modeling studies.
Atmospheric mesoscale models that provide useful knowledge about summer weather have been applied over the region; however, in general, the prediction results verify poorly, especially for predictions of thunderstorms and convective rainfall. Dunn and Horel (1994a,b) found that the operational Eta Model frequently underestimates the precipitation in the region. Schmitz and Mullen (1996) studied the atmospheric water vapor balance of European Centre for Medium-Range Weather Forecasts (ECMWF) analyses (data) and found that most water vapor enters the Sonoran Desert at low levels from the northern Gulf of California and that most upper-level moisture over the same region comes from the Gulf of Mexico. Stensrud et al. (1995) studied the climatology of northern Mexico and southern United States associated with NAM, using the Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model version 4 (MM4) initiated with SWAMP data. They were able to simulate certain meteorological characteristics over this areas, such as an LLJ stream over the Gulf of California. Farfan et al. (1998) established a real-time fifth-generation PSU–NCAR Mesoscale Model (MM5) forecast for the southwest United States and evaluated improvements derived from the inclusion of a nested high-resolution (7 km) grid for southern Arizona. Wallace et al. (1999) reviewed the results of Farfan et al. (1998) and found that MM5 was unable to accurately simulate the thermodynamic structure of the afternoon boundary layer over Arizona. Detailed reviews and research history of studies related to the impacts of NAM on southwest summer weather can be found in Douglas et al. (1993) and Adams and Comrie (1997).
The objectives of this study are to use high-resolution mesoscale modeling to explore the basic features of summertime weather over the semiarid lower Colorado River basin and to examine the accuracy of mesoscale numerical predictions. The methodology compares model results with ground and upper-air observations for selected cases. Two events in July 1999 are studied in this paper: one (5–6 July) for typical summertime weather with mostly clear skies and scattered thunderstorms in the southeastern highland areas of Arizona, and the second (8–9 July) for unusual storms that produced serious flash flooding over Las Vegas. We investigate 1) to what extent the model is capable of reconstructing the dynamic and thermodynamic characteristics of these two weather events over the lower Colorado River basin, and 2) what are the model weaknesses in predicting certain aspects of the system.
2. Background
a. Model description
The numerical weather forecasting model used in this study is version 4.2 of the Regional Atmospheric Modeling System (RAMS) developed by Colorado State University (Pielke et al. 1992; Walko et al. 2000). This model has been used for many mesoscale modeling studies. A recent intercomparison (Cox et al. 1998) of four mesoscale models [RAMS; MM5; the Navy Operational Regional Prediction System, version 6 (NORAPS6); Relocatable Window Model (RWM)] forecasting for five regions concluded that RAMS performed best. Version 4.2 was improved with several important physical processes, including radiation transfer, land surface, and cloud microphysics. In this study, the model was run with the nonhydrostatic, compressible set of equations in σz coordinates. The new land surface scheme, the Land Ecosystem Atmospheric Feedback model, version 2 [LEAF-2; Walko et al. (2000)], was utilized to calculate the storage and vertical transfer of water and energy in soil, vegetation, and surface water cover with the patchy method to simulate the subgrid heterogeneity. The Smagoringsky (1963) scheme was used for atmospheric diffusion. The atmospheric scattering and absorption of shortwave and longwave radiation were parameterized using the two-stream radiative transfer model (Harrington 1997) that accounts for the optical properties of clouds, rain droplets, and ice particles separately. The cloud microphysics modeled the clouds and precipitation in each atmospheric column at any grid scale using the bulk, one-moment scheme of Walko et al. (1995), in which eight hydrometeor species and their interactions are described; the mass mixing ratios of rain, pristine ice, snow, aggregates, graupel, and hail are explicitly predicted, and cloud liquid water is diagnosed. According to Zhang et al. (1988) and our tests (discussed later), in order to improve convective rainfall predictions, when conditional instability occurs in atmospheric columns with grid size at 10 km × 10 km or larger, both the cloud microphysics scheme and the cumulus parameterization scheme should be applied. It appears that the strengths and weaknesses of these schemes compensate each other. In RAMS (unlike other mesoscale models such as MM5), only one cumulus parameterization scheme, the modified Kuo scheme of Tremback (1990), is available for use. Previous comparative studies (Molinari and Dudek 1992; Wang and Seaman 1997; Belair et al. 2000) indicated that Kuo's convective scheme works well for large spatial scales (>70 km) but does not work very well for the mesoscale simulations, as in this study at 10–30-km grid scales.
b. Simulation configuration
Multigrid resolutions (Fig. 1) with two-way interaction on grid boundaries were used for the simulations. At the spatial resolution of 40 km × 40 km, grid 1 was used to interact with Eta Model output through its upper and lateral borders. The coverage area of grid 1 (approximately, 20°N, 94°W to 42.5°N, 124°W), including almost the entire southwest United States, northern Mexico, and eastern Pacific Ocean, is much larger than the lower Colorado River basin in order to reduce numerical errors from the boundary interactions. Grid 2 (approximately, 30°N, 107°W to 39°N, 118°W) covered the lower Colorado River basin with finer resolution of 10 km × 10 km. Grids 3 and 4 were designed to simulate individual convective systems with high resolutions of 5 km × 5 km and 2.5 km × 2.5 km, respectively, and their coverage areas and sizes varied according to the events simulated. For grids 1 and 2, both the cloud microphysics and cumulus parameterization scheme were applied to simulate precipitation, but in the domains of grids 3 and 4, only the cloud microphysics scheme was used. The vertical grid spacing started at 100 m and was stretched to 1 km at upper levels to the model top of 19.5 km.
c. Data processing
In the simulations, RAMS (grid 1) was nested into the regional Eta Model of the National Centers for Environmental Predictions (NCEP). To obtain more realistic initial conditions, the starting fields from the Eta Model were adjusted further through the RAMS objective analysis scheme using upper-air sounding and surface meteorological observation data. After initiation, the Eta output [Advanced Weather Interactive Processing System (AWIPS) 40-km grid analysis data] alone provided time-varying forcing through the upper and lateral borders of grid 1. Here the Eta analysis instead of the Eta forecast data is used for boundary conditions to reduce the impacts of the Eta forecast errors in the RAMS modeling. The forcing was nudged over six grid points into the study boundary with a nudging time of 1200 s and the application of the Klemp–Wilhelmson condition for the normal velocity and the zero-gradient inflow–outflow conditions for other variables. The boundary forcing was updated every 12 h at 0000 and 1200 UTC. The RAMS results under this forcing arrangement were used to examine the prediction performance of the model.
The following observed data were used for model initialization and/or verification in this study:
Upper-air sounding data used for initialization and verification were obtained from the National Oceanic and Atmospheric Administration's (NOAA) Forecast System Laboratory Radiosonde Database Access online (raob.fsl.noaa.gov).
Surface meteorological observations used for initialization were downloaded from the NCAR Data Support Section Web site (ncardata.ucar.edu/datasets).
Ground-based multisensor (gauge and radar) precipitation data used for validation were from the NCEP National Stage IV analyses (http://www.scd.ucar.edu/dss).
Surface air temperature and rainfall data for verification were provided by the Colorado River Basin River Forecast Center at Salt Lake City (http://www.cbrfc.gov).
3. Results of case tests
a. Case 1: 0000 UTC 5 July–0000 UTC 7 July 1999 (a 48-h simulation)
During 5–6 July 1999, the weather in the study domain was characterized by heavy rainfall along the northwestern coast of Mexico (from Sinaloa to Sonora), clear skies over most of the southwestern United States, but with thunderstorms occurring over southeastern Arizona and the neighboring New Mexico area. Therefore, the high-resolution grid 3 (5 km × 5 km, nested inside grid 2) was applied to central and southern Arizona. Grid 4 (2.5 km × 2.5 km, nested inside grid 3) was added to southeastern Arizona, as shown in Fig. 1 (the boxes with dashed boundaries). In RAMS, if fine grid boxes are nested inside coarse grid boxes, the model runs at high resolutions, and the results of the fine grids are integrated over the coarse grids to replace the coarse grid values. Meanwhile, the model also runs at the coarse grids, and the results are interpolated along the border of nesting fine grids to provide the forcing for high-resolution runs. This is the so-called two-way interaction between the nested grids. Using the nesting technique, the simulation becomes more detailed and more focused on the selected intensive study area.
1) Large-scale setting
In Fig. 2, the predicted (left column) and sounding observed (middle column) fields of geopotential height and wind at 500hPa, 700 hPa, as well as sea level pressure and the differences between the observations and the predictions of these fields (right column) at 0000 UTC 7 July (48-h predictions) are plotted. The point-sounding data were first interpolated through the RAMS Isentropic Analysis (ISAN) scheme into continuous fields. At 500 hPa, a dominant subtropical high (anticyclone) was centered northeast of the Four Corners area accompanying another high over the Pacific Ocean west of the northern Baja Peninsula. This synoptic pattern was categorized in Maddox et al. (1995) as type I of the three distinct recurring Arizona storm patterns in the region. The winds were characterized by westerly flow on the north and easterly flow on the south (Fig. 2a). There was an important trough along the Gulf of California to Arizona, which allowed the moisture to be transported northward. In the fields of sea level pressure (Fig. 2g), because of the strong near-surface heating of the atmosphere, the synoptic pattern of the wind was quite different from the one at 500 hPa. The winds were dominated by the northerly flow over the ocean and southerly flow over the continent. A trough was visible along the western border of Arizona, that is, the valley of the Colorado River, which connects the Gulf of California to the lower Colorado River basin. At 700 hPa (Fig. 2d), the wind fields were under the transition from the upper-level pattern to the low-level pattern: a southeasterly flow from the Gulf of Mexico passed the northern Mexican plateau and entered the southwest United States. These wind patterns are consistent with previous analyses (Douglas et al. 1993; Douglas and Li 1996; Stensrud et al. 1995). A comparison of the modeled atmospheric fields with the ones derived from the sounding data shows that they are similar, but the gradients of the predicted atmospheric fields seem smaller than those of the observations. The effect of the “thermal low” over the Sonoran Desert due to the strong surface heating was not clear in the predicted fields at the 700-hPa level, but showed up at the sea surface level. The sounding fields included several closed isobars in Mexico, which were from sounding station 76458 (23.18°N and 106.42°W) and seemed like thunderstorm highs, but did not match well with the local wind flow; therefore, these closed isobars could be errors resulting from the interpolation of coarse-resolution sounding data to the finescale analysis grid.
Furthermore, in the predicted sea level pressure shown in Fig. 2, there are closed isobars along the Mogollon Rim in central Arizona, while the sounding results do not have the closed isobars. This is because higher horizontal resolution was used in this area in modeling while the sounding stations are too coarse. The high was likely from the downdraft and outflow of model-predicted thunderstorms in that area. The model rainfall, discussed later, shows that the storms occurred mainly from 1800 to 0600 UTC in this case. From the differences in Fig. 2, generally, the height was overestimated (Fig. 2c), and sea level pressure was underestimated in southern California and central Mexico and overestimated in northwestern Mexico (Fig. 2i).
Figure 3 shows the predicted (left column) and observed (middle column) fields of mixing ratio at 500 and 700 hPa, precipitable water, and the differences between sounding observations and the model predictions of these fields (right column) at 0000 UTC 7 July. At the 500-hPa level, the high-moisture (>3 g kg–1) region spread northward from the most moist (>4 g kg–1) region, which consists of the extreme eastern Pacific, Baja Peninsula, Gulf of California and western coast of Mexico, to northern Mexico, Arizona, and western New Mexico (Fig. 3a). The mixing ratio at 700 hPa (Fig. 3d) and precipitable water fields (Fig. 3g) showed that the moisture axis (and its high-level extension) pattern stretched from the Gulf of California and the western coast of Mexico northward into Arizona. This indicated the major pathway of moisture transport to the southwest United States. Comparing with the sounding observations, the moisture distribution at 500 hPa was well predicted (Fig. 3c). However, at 700 hPa, atmospheric moisture was underestimated more than 3 g kg–1 over Baja California Norte and central Arizona, overestimated about 2 g kg–1 over southern California and southern Arizona, and matched well with the observations over the other area (Fig. 3f). The precipitable water in Fig. 3i also shows the same characteristics: the modeling results matched the sounding results well over most of the grid 1 area, except over Baja California Norte and central Arizona, where the model seriously underestimated precipitable water by more than 10 mm.
To further illustrate the three-dimensional features of the wind and moisture fields, the modeled wind component and the mixing ratio along two vertical sections intersecting at 32.6°N, 114.5°W (see Fig. 1) are plotted in Fig. 4. Notice that, in order to enhance the vertical wind fields, in Fig. 4, the scale of the vertical wind component is different from the scale of the horizontal wind component. From Fig. 4a, that is, the west–east section, it is apparent that strong atmospheric buoyancy developed over Arizona due to the surface heating related to the thermal low; however, this vertical flow did not exceed 3-km height and was “overlaid” by strong sinking at the border between the North American continent and the Pacific Ocean. Notice that the contours of the mixing ratio suddenly drop downward at the edge of the continent. This reflects the strong sinking and drying that is being forced over the Pacific Ocean. The plot also indicates that deep convection was probably favored over elevated terrain, that is, the area around the Arizona–New Mexico border. In Fig. 4b, that is, the south–north vertical section, a strong southerly low-level jet stream blew from the Gulf of California into the valley of the Colorado River, bringing large amounts of water vapor to Arizona. Clearly, for the lower Colorado River basin, the dominant water vapor supply for this event was the southerly winds from the Gulf of California. Figure 4a also indicates that, at high levels (>700 hPa), easterly winds from the southern plains are advecting slightly drier air toward the southwest United States. The analysis by Schmitz and Mullen (1996) using the ECMWF results reached a similar conclusion.
2) Surface temperature
In Fig. 5, the predicted surface air temperature (1.5-m height) at 0000 UTC 7 July is validated with surface measurements. The patterns of surface air temperature agree with the observations and show high coherence to the regional geography and topography: the hottest zone begins at the Gulf of California, extending along the Colorado River valley, first northward (following the western border of Arizona), then northeastward; the warm atmosphere also spreads from the valley across nearby low-elevation areas. The temperature gradient across the basin was large: temperature in the eastern highland (23°C) was more than 20°C lower than that in the western low-elevation area (43°C). The Sonoran Desert exhibited characteristics of a thermal low. The distinct high-temperature zone from southern Nevada down to the Baja Peninsula was responsible for the generation of the low-level trough shown in Fig. 2. Notice that the pattern of temperature distribution in the basin resembles that of the moisture distribution, with more moist air present to the east of the axis of hottest air. In the differences of observed and predicted surface temperature given in Fig. 5c, the RAMS temperature forecast exhibited warm biases. One overly warm area was near the Four Corners area, and the other was near the Sonoran Desert area. The possible reason for this is related to the selection of soil type. In standard RAMS version 4 or lower, the soil type is selected by users, but only 1 of 12 soil types can be used in the whole domain. This seems like a design limitation.
3) Rainfall
In Fig. 6, modeled 24-h (from 0000 UTC 5 July to 0000 UTC 6 July) and 48-h rainfall accumulations (from 0000 UTC 5 July to 0000 UTC 7 July) in grid 1 (40-km resolution) at the lower Colorado River basin are compared with the Eta Model predictions (∼40 km resolution) and the gauge observations. The station gauge data are interpolated to the grid points using an inverse square interpolation method. There are about 650 gauge stations over the lower Colorado River basin. The Eta Model forecasted little rainfall, possibly because of the coarse spatial resolution used to represent the mountainous topography in the area, which is considered the main storm trigger mechanism. The RAMS model predicted the rainfall along the Mogollon Rim across the border between Arizona and New Mexico, as well as rainfall over the joint area of southern Arizona and northern Mexico, which agrees quite well in patterns with the observations. However, the intensive centers of rainfall during the 24–48 h were not simulated well in either location or intensity. Further discussion will be given later. Notice that there was no rainfall in the moisture-rich Colorado River valley and the Sonoran Desert, which, according to the model forecasts (Fig. 4), was due to the air subsidence occurring in middle levels.
Figure 7 shows RAMS 48-h rainfall accumulation data over grids 3 and 4 and the corresponding observations. In the figure, the gray shadows represent the heights of topography. The observations data are still the gauge data interpolated to grids 3 and 4, respectively. There are about 590 gauge stations in grid 3 and 159 stations in grid 4. Notice that the observed rainfall has a slight difference in the two grids because of the interpolated resolutions. In grid 4 of Fig. 7, the model predicted three rainfall distribution areas corresponding to the observations. One heavy rainfall distribution area was near the border of Arizona and Mexico. Labels A, B, C, D, E, and F in the prediction (Fig. 7c) and labels A′, B′, C′, D′, and E′ in the observation (Fig. 7d) were the rainfall “centers” or “cores” in the area. The model results show that two centers E and F were over 40 mm and that center E was closely located at the place where the E′ center was observed with the amount of about 45 mm. The predicted rainfall was approximately along the border, that is, east to west distribution, while the observed rainfall was nearly orthogonal to the border, that is, south-southeast to north-northwest distribution. By carefully comparing the predicted and the observed centers, similar distribution of rainfall centers can be seen: centers A′, B′, C′, D′, and E′ in the observations can find their counterpoints A, B, C, D, and E in the predictions, but centers A′ and B′ were overestimated and centers C′, D′, and E′ were underestimated.
The second rainfall core area was in the Mount Lemmon area. Generally, the prediction underestimated this rainfall distribution and missed the southern part of the “rainfall belt.” However, the model caught two rain centers, a and b, which approximately correspond to the observations a′ and b′ in locations and intensities.
The third rain cores were located over the mountainous area of the northern border of grid 4. Comparing to the observations, the prediction only caught one of the four rainfall centers and missed at least two of them. However, the rainfall was relatively small in this area. It is not easy to predict specific rain cores using the mesoscale model, given the present accuracy of initial and boundary values.
In Fig. 7a (grid 3), the rainfall in the southern area has already been discussed relative to grid 4. Here discussion is focused on the rainfall along the Mogollon Rim, that is, the high-elevation terrain in the northeastern portion of grid 3. Generally, the model predicted storms along the Mogollon Rim, which agrees with the observations. However, predicted storm d was about 66 mm, while the observed storm d′ was about 56 mm, and predicted storm c had 70 mm in amount, while storm c′ in the observations was less than 30 mm. Additionally, the model missed the rainfall along the eastern border of grid 3. Further discussion of the reasons why the rainfall was overestimated along high-elevation mountains is presented in the next case.
In this case, the rainfall occurred along the mountains or mountainous slope areas. If the rainfall timing is not considered, RAMS can predict these kinds of storms in patterns and locations; however, RAMS tends to overestimate rainfall amounts. The statistics of rainfall accumulation over the grid 4 area (mean, standard deviation, and spatial correlation coefficient) in comparison with observations at different grid levels are listed in Table 1. In the grid 4 area, there are 159 gauge stations. For this analysis, the predicted grid rainfall values were first linearly interpolated into each gauge station; then, the statistics from the predicted and observed rainfall data at these stations were calculated. As shown in Table 1, during the first 18 h (starting from 0000 UTC 5 July), both the predicted and observed rainfalls were very small, and the spatial correlation coefficients between these two datasets are low. After 18 h, the correlation coefficients could be as high as 0.35–0.45 in grids 1 and 2, and slightly lower in grids 3 and 4. Furthermore, the observed rainfall shown in Table 1 occurred mainly during 1800 UTC 5 July–0600 UTC 6 July; there was little rainfall in the other time periods. The model results showed that, in grids 1 and 2, there were relatively large amounts of rainfall in this period [notice that there are two rainfall schemes in grids 1 and 2: convective parameterization scheme (CPS) and cloud microphysics scheme], but in grids 3 and 4, only small amounts of rainfall occurred during the time period 1800–2400 UTC 5 July, and large amounts of rainfall were produced during the period 0000–0600 UTC 6 July. Hence, in the first time period, the rainfall in grids 1 and 2 was likely mainly from CPS. This indicates that it is important to use CPS in the coarse grids and that the rainfall could have a time delay with only microphysical processes even at a resolution as high as a few kilometers. Molinari and Dudek (1992) also discussed this issue. Case 1 is a usual and recurring event in the NAM season. From the above analyses, it appears that RAMS can reproduce the basic characteristics of rainfall at this synoptic scale.
b. Case 2: 0000 UTC 8 July–1200 UTC 9 July 1999 (a 36-h simulation)
On 8 July 1999, Las Vegas, Nevada, experienced one of its worst flash flood events of the twentieth century. As a result, much of the Las Vegas valley experienced 35%–70% of its annual rainfall (∼105 mm) in a period of only 60–90 min, causing unprecedented flash flooding across the valley, with two deaths and $20 million in property damage (Haro et al. 1999). Because the operational Eta Model forecasted severe storms over the Four Corners area, rather than over Las Vegas, we focus on two questions: 1) can a nested, fine-resolution version of the RAMS model perform better for this unusual case, and 2) what are the possible error sources in the prediction processes? The RAMS simulation started at 0000 UTC 8 July, about 10 h prior to the onset of the severe storms and was integrated for 36 h (12 h after the end of the storms). In the simulation, two finer grids (grids 3 and 4) were applied to the area where the severe storms occurred: grid 3 covered an area including northwestern Arizona and southern portions of California, Nevada, and Utah (approximately 33°N, 112°W; 38°N, 117°W) with 5 km × 5 km resolution (Fig. 1), and grid 4 covered the Las Vegas area (approximately 35°N, 114°W; 37°N, 116°W) with 2.5 km × 2.5 km resolution (Fig. 1).
1) Large-scale setting
In Fig. 8, the predicted (left column) and observed (middle column) fields of geopotential height and winds and the differences (right column) between sounding observations and predictions are plotted at 1200 UTC 8 July (12-h prediction at the onset of the severe rainfall) for 500- and 700-hPa levels, respectively. The synoptic weather pattern had evolved considerably during the period following case 1. At 500 hPa, two typical systems are indicated in Fig. 8: one was a Pacific anticyclone located over the ocean well to the west of the grid, and the other was an anticyclone of the western extension of the Atlantic subtropical high located over New Mexico. A zone of deformation at 500 hPa had developed over the Las Vegas region. Convective storms were focused in this zone. At 700 hPa, the wind fields show a similar pattern to 500 hPa. The 700-hPa wind field pattern over the Gulf of California favored movement of moist, subtropical air northward into the lower Colorado River basin and provided the source of moisture for the Las Vegas flooding.
The right column in Fig. 8 shows the differences of geopotential height between observed (sounding) and modeled fields at 500 and 700 hPa. The model generally simulated the height pattern well. However, there are differences in the Mexico area. In the southern part of the Gulf of California, the model overestimated heights, while the model underestimated heights over central Mexico, especially at 700 hPa. A possible reason for this underestimation is that 700 hPa is near the surface in the central Mexican Sierra Madre Occidental (refer to Fig. 1) and is greatly affected by surface characteristics such as soil type and land use (as mentioned in case 1). Some of these problems likely resulted from inaccurate lateral boundary conditions.
In Fig. 9, the 12-h predictions of mixing ratio at the 500- and 700-hPa levels and the differences between the sounding observations and predictions are plotted. Figure 9 shows a similar pattern of moisture distribution as in Fig. 3 for case 1; that is, a monsoon boundary (with sharp moisture gradients) divided the air mass over the region into two sectors: one was the dry air mass over the Pacific Ocean, and the other was the moist air mass over the continent. In case 2, the high moisture area (>3 kg kg–1 at 500 hPa) extended farther northward, and the mixing ratio at 700 hPa over the southwest, from southern California to southern Colorado (including both Las Vegas and the Four Corners area), was greater than 9 g kg–1. The occurrence of convective rainfall was favored in the western edge of the moist air mass during case 2. Comparing with the sounding observations, the moisture distribution at the high level was predicted well (see the difference of mixing ratio between the sounding observed and modeled in Fig. 9c); however, at the low level (Fig. 9f), moisture distribution over the (northern) extratropical region did not verify as well as that over the subtropical region. The possible reason relates to inaccurate lateral boundary condition. The other area, where the model predicted moisture incorrectly, is near the junction between the Gulf of California and the Colorado River. The reason for this is not clear.
2) Rainfall
In Fig. 10, the predicted 6-h rainfall fields from 1200 UTC 8 July to 0600 UTC 9 July (the main precipitation period) over the domain of grid 1 across the lower Colorado River basin are compared with radar–rain gauge composite data and the operational Eta prediction results. The observed data (right column) show that most rainfall fell in southern Nevada, southern Utah, and northwestern Arizona, and scattered thunderstorms were distributed over the Mogollon Rim from central Arizona to the Arizona–New Mexico Border and the low-elevation regions of southwestern Arizona.
The operational Eta Model (middle column) forecasted rainfall over three main precipitation areas. One area was near the Four Corners area. The second was located in northwestern Mexico and southern California, and the third area was located along the southeastern California–southern Nevada border and the border of Nevada, Arizona, and Utah. Comparing to the observations, Eta overestimated rainfall in the Four Corners area and seriously underestimated the rainfall near the border of the three states (Nevada, Arizona, and Utah). The Eta Model clearly did not predict the disastrous flooding rainfall in the Las Vegas area.
Over the major rainfall area, RAMS predicted (left column) two separate rainfall events. One event occurred early during 1200–1800 UTC along the borders of Utah, Arizona, and Nevada. The peak rainfall amount was about 18 mm, which was underestimated relative to the observed value of 31 mm. The second flood-producing event occurred 4 h later over the mountainous area northwest of Las Vegas, then shifted to the valley. In the city of Las Vegas, the predicted 24-h rainfall was about 50 mm on grid 3 (5 km × 5 km) and 80 mm on grid 4 (2.5 km × 2.5 km), close to the gauge values of 1.5–3 in. with a maximum of 3.25 in. (Haro et al. 1999). However, over the mountainous western area of Las Vegas, the modeled rainfall was more than 100 mm, severely overestimated, the reason for which will be discussed later.
The hourly rainfall during the major rainfall events predicted by the microphysics scheme on the high-resolution grid 4 is compared with the 4 km × 4 km stage IV multisource observed data given in Figs. 11 and 12. In Fig. 11, the rainfall (both observations and predictions) for the first event over the border area of Nevada, Utah, and Arizona is shown at 1300–1400, 1400–1500, and 1500–1600 UTC. Unfortunately, there were no hourly rainfall data available before 1400 UTC. In the modeling results, the event began about 0700 UTC (not shown) 8 July 1999 centered at 36.7°N, 114.35°W (the line intersection points in Figs. 11a,b). Then the modeled rainfall system evolved NNE and SSW separately, while, from the available data, the observed rainfall moved westward over Gass Peak (Fig. 11d). The predicted rainfall ending time was close to the observed, but the area was shifted about 40 km eastward. The predicted rainfall also missed the light precipitation north of Charleston Peak.
Figures 12a and 12b show rainfall predictions and observations for the second event. The prediction of the second event was delayed by about 4 h compared to the observations, which shifted the morning storms to the afternoon. The triggering mechanism for the Las Vegas flooding rainfall was possibly outflows from the first storm event that moved to the Las Vegas area and forced lifting of the low-level wet air. The modeled rainfall was delayed in this event possibly due to three reasons: 1) The model underestimated the intensity of the first event and did not generate strong outflows to trigger the second event; 2) the modeled location of the first event was shifted eastward, causing any outflows from the first event to fail to reach the area where the second storm occurred; and 3) the second event was unusual in that it occurred in the morning. RAMS does not possess a good physical mechanism to trigger that kind of convection. As Pastor et al. (2002) discussed, the “orographic trigger mechanism is favored by the RAMS model in its current configuration.” The model convection was finally triggered when there was strong enough local upslope wind generated by the solar heating near the mountainous area later in the afternoon. This delayed the unusual precipitation event. Molinari and Dudek (1992) discussed the reason for this delay in onset time from aspects of grid scale and more generally dynamics, thermodynamics, and physical processes when a fully explicit microphysics scheme is used to predict precipitation.
If the time delay is ignored, the rainfall core evolutions can be discussed. First, both observed and modeled rainfall cores began over the mountainous area (see the A and A′ panels in Figs. 12a and 12b). Then the rainfall centers extended along the mountain slope (see B, C, D and B′, C′, D′ in Fig. 12). However, from 2000 UTC 8 July in observation and 0000 UTC 9 July in simulation, the observed rainfall moved northward and eastward over Las Vegas, while the simulated rainfall extended southward mainly along the mountain slopes (see E and F in Fig. 12a and E′ and F′ in Fig. 12b). The possible reason for this was discussed earlier. For rainfall over Las Vegas, the maximum was observed at 1700 UTC, while the predicted maximum occurred at 2300 UTC. Comparing Figs. 12a and 12b, the predicted rainfall was greater in amount than observed over the mountainous area, where the upward vertical velocity in the model caused the air mass to be lifted to the level of free convection more easily than over lower elevations.
Table 2 shows the spatial correlation between the RAMS–Eta-predicted and gauge-observed rainfall accumulations over the grid 4 area in case 2. As in case 1, for comparison, the predicted grid rainfall was first linearly interpolated to the locations of 95 gauge stations in the area. Operational Eta predicted that there would be rainfall in the first 12 h and little rainfall during other times. As shown in Fig. 10, the Eta caught the first rainfall event and missed the second one. The spatial correlation coefficients show that the Eta Model did not perform well. The RAMS prediction did not perform very well in the first 12 h, but then did much better afterward. The spatial correlation coefficients could be more than 0.30 in different grids. Clearly, the nesting of high-resolution grids in limited domains has improved the rainfall predictions relative to the large region, especially for the locations and intensities of the heavy rainfall cores.
4. Summary and discussion
Two case studies using RAMS nested inside the operational Eta Model were conducted to test the prediction of summertime weather over the lower Colorado River basin. The results not only simulated the severe rainfall over the Las Vegas area (8 July 1999), which was missed in the operational Eta Model forecast, but also provided useful insights regarding the summertime meteorological characteristics in the region. First, the variation of synoptic patterns strongly affects the weather situation in the basin. In the study, the synoptic patterns of geopotential height in the region show a pair of subtropical highs: one over the eastern Pacific Ocean and the other over the continent. The variation in location and strength of these highs indicates complex interactions among the three major wind systems (i.e., Pacific westerly winds, southerly winds from the Gulf of California, and easterly/southeasterly winds from the Gulf of Mexico) and results in very different rainfall patterns. The model results illustrate that the midlevel sinking over the low elevation of the southwest area of the basin (Fig. 4) “capped” the development of deep convection in case 1; meanwhile, in case 2 (8 July), midlevel deformation zone and low-level convergence over the Las Vegas valley stimulated intense convective storms in the region. As shown by the results, a distinct moisture boundary divided the large study domain (grid 1) into dry and wet sectors. The LLJ stream from the Gulf of California was the major source of atmospheric moisture for the basin.
Local topography and thermodynamics play a significant role in the formation of the weather features in the basin. The “thermal low” resulting from the strong surface heating over the Sonoran Desert is considered to be responsible for the northward LLJ stream from the Gulf of California, which made the valley of the Colorado River the warmest and moistest area in the basin. By nesting fine-resolution grids (grids 3 and 4), the representation of local topography in that specific region was improved in the RAMS model, compared with that in the relatively coarse resolution Eta Model. This is the major reason why the RAMS model could predict the intense convective storms in the grids where finer resolution is used, such as over the Las Vegas valley and the Arizona mountains.
However, the higher resolution (Figs. 6 and 7 and Figs. 10, 11, and 12) is needed to predict rainfall in southern Arizona. In case 1, the 5-km resolution (grid 3) was used in central Arizona (Mogollon Rim), and the mountainous storms were predicted, although amounts were overestimated. When the 2.5-km (grid 4) resolution was used in the Tucson area (southern Arizona), the storms were relatively well predicted. In the case 2 simulations, only 5-km resolution was used in northwestern and southwestern Arizona, the rainfall over mountainous northwestern Arizona was predicted, and the rainfall over the desert area (southwest Arizona) was missed. Poor definition of the land surface parameter and horizontal resolution might be related to these problems. Further study is required to understand these deficiencies.
Finally, the study indicates that the physical simulations of heavy convective rainfall centers for the RAMS model rely more on the cloud microphysics scheme than on the cumulus parameterization scheme, given model resolution high enough to resolve convective storms. However, the cloud microphysics scheme always misses a large number of medium-to-small convective storms. In addition, RAMS has weaknesses in predicting storms that evolve over lower-elevation areas. The use of the cumulus parameterization scheme can compensate for this weakness of the microphysics scheme to a certain extent when coarse grid resolutions are used.
Acknowledgments
This research was partially funded by NASA/GAPP Grant NA16GP1605, NASA EOS-IDS Grant NAG5-3640, NASA TRMM Grant NAG5-7716, and the NSF STC Program (Agreement EAR-9876800). We would like to thank Dr. S. Shumate of the NWS/Colorado River Forecast Center for providing the surface temperature data. The first author also would like to thank Dr. Jaime E. Combariza and other staff at the Computer Center and Information Technology, The University of Arizona, for their help and support. Thanks are also extended to Ms. Corrie Thies for her careful reading and editing of the manuscript.
REFERENCES
Adams, D. K., and A. C. Comrie, 1997: The North American monsoon. Bull. Amer. Meteor. Soc., 78 , 2197–2213.
Belair, S., A. W. Methot, J. Mailhot, and B. Bilodeau, 2000: Operational implementation of the Fritsch–Chappell convective scheme in the 24-km Canadian regional model. Wea. Forecasting, 15 , 257–274.
Cox, R., B. L. Bauer, and T. Smith, 1998: A mesoscale model intercomparison. Bull. Amer. Meteor. Soc., 79 , 265–283.
Douglas, M. V., and S. Li, 1996: Diurnal variation of the lower-tropospheric flow over the Arizona low desert from SWAMP-1993: observation. Mon. Wea. Rev., 124 , 1211–1224.
Douglas, M. V., R. A. Maddox, K. Howard, and S. Reyes, 1993: The Mexican monsoon. J. Climate, 6 , 1665–1677.
Dunn, L. B., and J. D. Horel, 1994a: Prediction of central Arizona convection. Part I: Evaluation of the NGM and Eta Model precipitation forecasts. Wea. Forecasting, 9 , 495–507.
Dunn, L. B., and J. D. Horel, 1994b: Prediction of central Arizona convection. Part II: Further examination of the Eta Model forecasting. Wea. Forecasting, 9 , 508–521.
Farfan, L. M., D. R. Bright, and J. A. Zehnder, 1998: Numerical simulation of mesoscale weather system in Arizona. Preprints, 12th Conf. on Numerical Weather Prediction, Phoenix, AZ, Amer. Meteor. Soc., J13–J16.
Haro, J. A., H. R. Daley, and K. J. Runk, 1999: The Las Vegas flash floods of 8 July 1999: A post-event summary. NOAA Western Region Tech. Attachment, 99-26, 15 pp.
Harrington, J. Y., 1997: The effects of radiative and microphysical processes on simulated warm and transition season Arctic status. Ph.D. dissertation, Colorado State University, Fort Collins, CO, 289 pp.
Higgins, R. W., and W. Shi, 2000: Dominant factors responsible for interannual variability of the summer monsoon in the southwestern United States. J. Climate, 13 , 759–776.
Higgins, R. W., Y. Yao, and X. L. Wang, 1997: Influence of the North American monsoon system on the U.S. summer precipitation regime. J. Climate, 10 , 2600–2622.
Maddox, R. A., D. M. McCollum, and K. W. Howard, 1995: Large-scale patterns associated with severe summertime thunderstorms over central Arizona. Wea. Forecasting, 10 , 763–778.
McCollum, D., R. A. Maddox, and K. Howard, 1995: Case study of a severe mesoscale convective system in central Arizona. Wea. Forecasting, 10 , 643–665.
Meitin, J. G., K. W. Howard, and R. A. Maddox, 1991: The southwest area monsoon project: Daily operations summary. 25 pp. [Available from NSSL, 1313 Halley Circle, Norman, OK 73069.].
Molinari, J., and M. Dudek, 1992: Parameterization of convective precipitation in mesoscale numerical models: A critical review. Mon. Wea. Rev., 120 , 326–344.
Pastor, F., M. J. Estrela, D. Penarrocha, and M. M. Millan, 2002: Torrential rains on the Spanish Mediterranean coast: Modeling the effects of the sea surface temperature. J. Appl. Meteor., 41 , 1180–1195.
Pielke, R. A,, and Coauthors. 1992: A comprehensive meteorological modeling system—RAMS. Meteor. Atmos. Phys., 49 , 69–91.
Schmitz, J. T., and S. L. Mullen, 1996: Water vapor transport associated with the summertime North American monsoon as depicted by ECMWF analyses. J. Climate, 9 , 1621–1634.
Smagoringsky, J., 1963: General circulation experiments with the primitive equations. Part 1: The basic experiment. Mon. Wea. Rev., 91 , 99–164.
Smith, W. P., and R. L. Gall, 1989: Tropical squall lines of the Arizona monsoon. Mon. Wea. Rev., 117 , 1553–1569.
Stensrud, D. J., R. Gall, S. Mullen, and K. Howard, 1995: Model climatology of the Mexican monsoon. Mon. Wea. Rev., 125 , 417–437.
Tremback, C. J., 1990: Numerical simulation of a mesoscale convective complex: Model development and numerical results. Ph.D. dissertation, Atmospheric Science Paper 465, Colorado State University, Fort Collins, CO, 247 pp.
Walko, R. L., W. R. Cotton, M. P. Meyers, and J. Y. Harrington, 1995: New RAMS cloud microphysics parameterization. Part 1: The single-moment scheme. Atmos. Res., 38 , 29–62.
Walko, R. L,, and Coauthors. 2000: Coupled atmosphere–biophysics–hydrology models for environmental modeling. J. Appl. Meteor., 39 , 931–944.
Wallace, C. E., R. A. Maddox, and K. W. Howard, 1999: Summertime convective storm environments in central Arizona: Local observations. Wea. Forecasting, 14 , 994–1006.
Wang, W., and N. L. Seaman, 1997: A comparison study of convective parameterization schemes in a mesoscale model. Mon. Wea. Rev., 125 , 252–278.
Zhang, D. L., E. Hsie, and M. W. Moncrieff, 1988: A comparison of explicit and implicit predictions of convective and stratiform precipitating weather systems with a meso-beta-scale numerical model. Quart. J. Roy. Meteor. Soc., 114 , 31–60.
Study area and grid divisions
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Predicted and observed fields of geopotential height (contour: 12.5 m), sea level pressure (contour: 2.5 hPa) and winds (m s–1) at 500 hPa, 700 hPa, and sea level for 0000 UTC 7 Jul 1999, and the differences of geopotential height (contour: 10 m) between the observations and predictions
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Predicted and observed fields of mixing ratio at 500 and 700 hPa (contour: 1 g kg–1), precipitable water (contour: 10 mm), and the differences between the observations and predictions at 0000 UTC 7 Jul 1999.
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Distributions of wind component (m s–1) and mixing ratio (g kg–1) on two vertical cross sections (intersection: 26.5°N, 115°W) at 0000 UTC 7 Jul 1999, for the (a) west–east section and (b) south–north section
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Predicted and observed fields of surface temperature (contour: 3°C) and the difference between the observations and predictions in the lower Colorado River basin (grid 2) at 0000 UTC 7 Jul 1999
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
The (a) 24- and (b) 48-h rainfall accumulations from RAMS, (c), (d) Eta, and (e), (f) observations starting at 0000 UTC 5 Jul 1999 (contours. 5, 15, 30, 50, and 70 mm)
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Predicted and observed 48-h rainfall accumulations in grid 3 and grid 4 (contours: 5, 15, 30, 50, 80, 120, and 150 mm). The gray shadow represents the topography height and labeled letters indicate rainfall cores.
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Predicted and observed fields of geopotential height (contour: 10 m) and winds (m s–1) at 500 and 700 hPa, for 1200 UTC 8 Jul 1999 and the differences of geopotential height between the observations and predictions
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Predicted and observed fields of mixing ratio (contour: 1.0 g kg–1) at 500 and 700 hPa, for 1200 UTC 8 Jul 1999, and the differences between the observations and the predictions
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
The 6-hourly rainfall (1200–1800, 1800–0000, 0000–0600 UTC) from RAMS, Eta, and observations (contours: 1, 5, 10, 20, 30, 50, and 80 mm)
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Hourly rainfall predicted and observed over grid 4 for the first event starting at 1300 UTC 8 Jul 1999 (contours: 1, 5, 10, 20, 30, 50, and 80 mm).
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
(a) Hourly rainfall predicted over grid 4 for the second event: the Las Vegas flooding (contours: 1, 5, 10, 20, 30, 50, and 80 mm). (b) The same as in (a) except for the observed rainfall
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
(Continued)
Citation: Monthly Weather Review 131, 3; 10.1175/1520-0493(2003)131<0521:SWSFTS>2.0.CO;2
Statistical comparisons for the predicted and observed rainfall in case 1
Statistical comparisons for the predicted and observed rainfall in case 2
