a. LAPS

An integral part of the AWIPS, the Local Analysis and Prediction System, integrates data from numerous meteorological observation systems into a very high resolution gridded framework. The LAPS was in part developed to provide real-time forecasts/nowcasts of the preconvective environment. The AWIPS-LAPS produces hourly surface analyses (on a 10-km horizontal grid) of a number of standard meteorological fields and parameters such as the lifted index (LI) and the vertical motion associated with local topography. Data sources for analysis include METAR (a French acronym meaning aviation routine weather report), surface mesonet, rawinsonde, satellite, commercial aircraft, large-scale numerical model, wind profiler, and radar. These data are combined using a modified Barnes (1964) technique, variationally applied splines, and a dynamic adjustment step that forces the winds to satisfy mass conservation (strong constraint) and thermodynamic/momentum equations [weak constraint; McGinley and Smart (2001)]. Upper-level wind analyses first employ nonradar data (profiler, sounding, etc.) and then merge these analyses with Doppler radial winds that have been mapped to the analysis grid. Upper-level moisture (i.e., specific humidity) analyses consist of large-scale model data, Geostationary Operational Environmental Satellite (GOES) satellite water vapor (Birkenheuer 1992, 1996), and modifications based on the LAPS cloud analysis. The LAPS 3D temperature analysis consists of a first-guess large-scale model field and available soundings. An artificial boundary layer (with a depth of 50 hPa) is introduced in order to smoothly couple the surface and upper-level temperature analyses. LAPS cloud analyses (Albers et al. 1996) employ surface-based cloud observations, and satellite and radar data to produce 3D cloud fields. Cloud-top heights are determined using a combination of the analysis temperature field and satellite radiances.

b. MAPS

The Mesoscale Analysis and Prediction System is the development/research version of the Rapid Update Cycle (RUC) model. Because the RUC model provides the first-guess fields for the ADAS analyses produced herein, we feel it is important to provide a description of the MAPS analysis component. The MAPS analysis scheme is based on a multivariate optimum interpolation technique and assimilates a wide range of data including that obtained from aircraft, wind profiler, rawinsonde, surface stations, buoys, radio acoustic sounding system (RASS), Doppler velocity azimuth display winds, GOES and Special Sensor Microwave Imager (SSM/I) precipitable water vapor, and GOES cloud drift winds. Because the analysis is coupled to the RUC, the background field is provided by the previous-hour forecast. The MAPS system has undergone a steady stream of improvements over the course of the past few years, most notably a recent upgrade in horizontal resolution from 40 to 20 km, bringing the system into the realm of the mesoscale. The analysis component of the 40-km operational RUC (RUC-2), used herein to initialize our analyses, modifies background moisture values according to observed precipitable water values. The analysis is constrained to match the shape of the background water vapor mixing ratio but is either moistened or dried out according to the observations (Benjamin et al. 1998). A multivariate height–wind analysis (partially constrained by geostrophy) is performed at all analysis levels. The wind analysis is anisotropic and oriented along the flow, according to the geostrophically derived horizontal covariances of forecast error (Benjamin 1989). The temperature (virtual potential temperature) background is updated taking into account the height and wind observation analysis in the previous steps. A series of univariate analyses are applied to the temperature (using surface, aircraft, rawinsonde, and RASS data), wind at the lowest five levels where the analysis is constrained (nongeostrophic) to closely match the surface wind observation, the surface pressure (fitting to surface pressure observations), and the moisture field (at all levels). The RUC-2 1-h assimilation cycle integrates forecast background fields and observations using a multivariate/univariate two-pass analysis for winds/pressure and produces analyses that are consistent with model dynamics in data-void regions. Other notable RUC-2 analysis features include the use of all station pressure (altimeter) and surface wind observations regardless of the difference between station and model elevation, pressure reduction to the model elevation using the local lapse rate over the lowest five background field levels, a reduction in expected surface observation errors, the use of a minimum topography field to diagnose the surface and temperature and dewpoint (from a higher-resolution topographical grid; see section 6), and the reduction (from model terrain height to actual station elevation) of temperature and dewpoint observations using the local lapse rate. The latter feature allows a high percentage of surface temperature and dewpoint observations over the western United States into the 40-km 3D analysis.

c. Other mesoscale analysis systems

In an extension to their earlier mesoscale work associated with the Program for an Operational Meterological Information System (PROMIS) project, the Swedish Meteorological and Hydrological Institute now has an operational mesoscale analysis system, Mesan (Haggmark et al. 2000). Based on optimal interpolation, Mesan ingests observations from various platforms, including satellite, radar, automatic surface stations, and precipitation gauge measurements. Mesan employs a first-guess field from the High Resolution Limited Area Model (HIRLAM) and has been producing hourly analyses, at 44-km horizontal resolution, since October 1996. The Met Office (UKMO) also operates a mesoscale analysis and forecasting system referred to as the Nowcasting and Initialization for Modeling Using Regional Observation Data (Nimrod; Golding 1998). Nimrod has been operational in the UKMO since 1995. This fully automatic system ingests radar, satellite, surface reports, and numerical weather prediction model products. Used to generate analyses and short-term forecasting, Nimrod linearly extrapolates present features (for short-term forecasts) and incorporates nonlinearities (e.g., flow evolution for longer-term forecasts) from an NWP model. Forecasts of 0–6 h are made with a 30-min update cycle. Nimrod is part of the UKMO Interactive Mesoscale Initialization System (IMI; Wright and Golding 1990) that includes a moisture observation preprocessing system (MOPS; Wright 1993) and a statistical analysis correction scheme (Lorenc et al. 1991).

The Navy Operational Regional Prediction System, version 6 (NORAPS6), uses a multivariate technique to blend observations and background data obtained from a large-scale model (Cox et al. 1998). Designed to replace NORAPS, the Coupled Ocean–Atmosphere Mesoscale Prediction System (COAMPS; Hodur 1997) specifies 3D atmospheric fields either by interpolating large-scale fields to the COAMPS grids or by blending observations with the first-guess fields using a multivariate optimum interpolation analysis scheme. The Meso-Eta (Black 1994) uses a 3D variational assimilation system (Eta Data Analysis System; Parrish et al. 1996) that includes the ingest of satellite radiances, and the Regional Atmospheric Modeling System (RAMS) objectively analyzes large-scale model data from the National Centers for Environmental Prediction (NCEP). As previously mentioned, the ARPS (Xue et al. 2001) employs an analysis scheme, ADAS, that uses a “hybrid” successive correction technique. By hybrid we are referring to elements of the analysis scheme that reflect a combination of successive correction and optimum interpolation methodologies (see next section). The advantage of analysis packages such as LAPS or ADAS is that they can be run independently as nowcasting tools or easily configured to initialize any mesoscale model.

a. Analysis variables

ADAS is based on a successive correction method (SC; e.g., Bergthorsson and Doos 1955; Cressman 1959; Barnes 1964) developed by Bratseth (1986). The method, which converges to optimum interpolation (OI), is amenable to high-resolution analyses and large and diverse data streams because, unlike more sophisticated techniques (e.g., variational techniques), it does not require large matrix inversions. Adjustments in the analyzed temperatures and wind field are relegated to the postiteration step and, like OI, the technique accounts for the relative error between the observations and background field. Here, “background” refers to a first-guess field provided by a model. The Bratseth technique differs from its SC counterparts in several important respects: 1) in the presence of observation error, the analysis does not converge to the data; 2) it accounts for background field errors; and 3) the analysis has been formulated to mitigate the impact of spatial inhomogeneities in the data. ADAS analyzes five variables: wind (u and υ components separately), potential temperature, pressure, and moisture.²

The vertical velocity w is estimated using a first guess from the O'Brien (1970) method with input winds obtained from the analysis horizontal components. The horizontal wind components are then “adjusted” (assuming that the 3D divergence is zero everywhere in the analysis domain) using the following functional J:

View Expanded

where the weights α₁ and α₂ are assumed to equal 1; u, υ, and w are the analysis/O'Brien wind components; ρ = ρ(z) is the air density; the superscript a denotes the adjusted wind component; λ is a Lagrange multiplier; ξ, η, and ζ are the ADAS curvilinear coordinates; and dA = dξdη. Taking the variation of Eq. (1) with respect to u^a and υ^a and applying the appropriate boundary conditions yields three Euler–Lagrange equations (one for each of the adjusted components u^a, υ^a, and mass conservation). The three equations and three unknowns (u^a, υ^a, λ) can then be determined (e.g., Sherman 1978). Because the application of Eq. (1) is two-dimensional, it is efficient and inexpensive. However, errors in the analyzed wind field will feed into the O'Brien estimate of w, which in turn will affect the adjusted wind fields (for additional discussion see section 7).

In addition to the standard meteorological observations, ADAS ingests Doppler radial velocities and reflectivity. ADAS analyzes radial wind “increments” and not the radial winds themselves. Increments are created for the u- and υ-component directions separately by first subtracting the observed radial wind υ^obs_r from the analysis radial wind υ^anl_r, the latter of which is given by

View Expanded

where w^anl is initially assumed to be zero [for the small elevation angles (θ) of 0.5°, 1.5°, 2.4°, and 3.4° associated with the Next-Generation Weather Radar (NEXRAD) Information Dissemination Service (NIDS; Baer 1991) data used here, w^anl sinθ is generally negligible], r is the range, ϕ is the azimuth, and ds/dr is the local derivative of the beam path assuming a 4/3 effective earth's radius model [i.e., Eq. (2.28) of Doviak and Zrnic 1984]. The increments are then created by projecting the difference between the observed and analysis radial wind (i.e., υ^obs_r − υ^anl_r) onto the u- and υ-component directions, respectively (ADAS is also configured to ingest data from single- and dual-Doppler wind retrievals). In addition to the analysis of the radial wind increments, the radar reflectivity can be used to increase the relative humidity in regions of enhanced reflectivity. For relative humidities under 90% and radar reflectivity greater than 40 dBZ the Utah ADAS creates pseudo-observations of relative humidity equal to 90%. The choice of a 90% threshold is somewhat arbitrary; however, because the majority of precipitation over NW Utah falls during the winter season where there is little probability of hail, we believe that this value is reasonable. Often, for synoptically forced wintertime events, the background (RUC-2) RH is >90%. The relatively high reflectivity threshold was selected to keep the number of these somewhat ad hoc adjustments to a minimum while allowing for some modification of the background RH in the presence of intense convection. Ultimately, it would be useful to examine the impact of this adjustment on the analyses.

In addition to its mesoscale analysis component, ADAS also has a cloud analysis system (Zhang et al. 1998; Lazarus et al. 2000). The mixing ratios for rain, snow, and graupel are computed as part of the cloud analysis package using the following relationship (Kessler 1969):

Zaρq^b(3)

where Z is the reflectivity factor, ρ is the air density, q is the condensate (i.e., rain, graupel, or snow), and a and b are coefficients [for snow/graupel (rain) we use a = 38 000 (17 300), b = 2.2 (1.75)]. Mixing ratios for cloud ice and cloud water are determined assuming moist-adiabatic ascent in areas analyzed with cloud cover. Cloud cover is obtained from the ADAS cloud analysis using a combination of RUC-2 relative humidity, infrared (IR) and visible (VIS) satellite data, NIDS radar data, and METAR observations. An initial cloud fraction is estimated using

View Expanded

where RH₀ is a user-specified height-dependent relative humidity, RH is the analysis relative humidity, and b is an empirical constant (set to 2 here). Following the specification of the background cloud fraction above, vertical cloud soundings are generated from the METAR observations with cloud thickness assumed to be a function of cloud-base height (Albers et al. 1996). Cloud amount from this step is then added or subtracted depending on the difference between ADAS and observed (i.e., from 10.7-μm satellite radiances) brightness temperatures. NWS Weather Surveillance Radar-1988 Doppler (WSR-88D) data are remapped (using the maximum reflectivity in a grid box) to the ADAS grid—with holes between scans filled using bilinear interpolation. Cloud cover is inserted if the radar echo is above the lowest METAR cloud base and the reflectivity is greater than 20 dBZ. Cloud albedo derived from visible radiances is then used to compare the observed with the analyzed vertically integrated cloud cover (if the latter is larger, the ADAS cloud cover is reduced). While the order of application of the various data streams for the cloud analysis is not unique, it is logical with the METAR data providing cloud base, the IR data yielding cloud top, the WSR-88D filling “in between,” and the VIS data refining the analysis.

b. Analysis data and components

The ADAS consists of three preanalysis software components that map data from the source to the ADAS grid. The remap algorithms include large-scale model, radar, and satellite data (Brewster et al. 1995). The background field is created by interpolating large-scale model data onto the ADAS grid. To avoid extrapolation, the interpolation grid (i.e., large-scale model domain) is selected so that it contains, in entirety, the analysis grid (ADAS). The interpolation is biquadratic in the horizontal and linear in the vertical where the natural log of pressure is used in place of pressure. ADAS can directly accommodate a variety of datasets/formats generated by NCEP including forecast model output from RUC-2 and the Eta Model. Because of its modularity, ADAS can be easily adapted to ingest other grids as well. The radar processing routines ingest either NIDS data or WSR-88D level-II (Golden 1990) data (via tape or a live WSR-88D data feed). The radar data are mapped onto a Cartesian grid by averaging all radar range gates that fall within an ADAS grid box. Depending on the grid and/or radar resolution, data are averaged to create a single value for each grid volume (this thins the data to match a coarser-resolution analysis grid, i.e., creates “superobservations”). The remapped data are identified by latitude, longitude, and radar location—thereby allowing the radar data to be used on a grid different from the one on which they were created (Brewster et al. 1995). Note that this procedure is a preanalysis step because the averaged data, which do not generally coincide with an analysis grid point, are then input into the Bratseth scheme. The scheme has additional correlation terms that take into account the fact that when the radar points north, it has no knowledge of the u component. Man–Computer Interactive Data Access System (McIDAS; Krauss et al. 1972) satellite data are used in the ADAS cloud analysis (Zhang et al. 1998) only. The satellite data are averaged to fill the grid cells. When there are no data in a grid cell (i.e., the grid is finer than the satellite resolution), bilinear interpolation is used to fill the missing grid points.

c. Analysis description

ADAS employs a successive correction method based on a univariate analysis technique developed by Bratseth [for more detail see Bratseth (1986) or Sashegyi (1993)]. The Bratseth method iteratively solves for two equations: one for an “updated” estimate of the gridpoint correction for the analysis variable ϕ (i.e., pressure, temperature, horizontal wind components, or moisture),

View Expanded

and the other for the “updated” observation estimate,

View Expanded

where ϕ^obs_i is the value of the ith observation; ϕ_x(k) and ϕ_i(k) are the gridpoint estimate and observation estimate for the kth iteration, respectively; and n_obs is the total number of observations. The gridpoint analysis weight α_xi is given by

View Expanded

and the observation analysis weight α_oi is

View Expanded

where ρ_xi and ρ_oi are the gridpoint-to-observation and observation-to-observation spatial correlation functions, respectively; σ² is the ratio of the observation error variance-to-forecast error variance; δ_oi is the Kronecker delta; and m_i is a “normalization factor,” which is a function of the observation density around observation points,

View Expanded

Similar to its SC counterparts, the number of observations n_obs will depend on both the observation distribution around the analysis point and radius of influence (ROI). The Bratseth equations [i.e., Eqs. (5) and (6)] yield different solutions—with the observation estimate [Eq. (5)] converging toward the observations and the gridpoint estimate converging to an analysis value that minimizes the expected error variance considering the background field and observations. The degree to which the Bratseth analysis converges toward the observations depends upon the ratio of the observation-to-forecast error variance σ. Upon convergence, the analysis value is the same as that which would be produced by optimum interpolation (Sashegyi 1993). The difference between the final analysis (i.e., gridpoint) value and observation will depend on the error variance ratio σ². Also note that once the observation estimate converges, the gridpoint estimate becomes stationary. The spatial correlation functions ρ_xi and ρ_oi in Eqs. (7) and (8) are assumed to be Gaussian,

View Expanded

where r_ij and Δz_ij are the horizontal and vertical distances, respectively, and R and R_z are scaling factors. For ADAS, the ROI is a function of the distance scaling parameters and thus varies for each iteration. The scale factors are adjusted to draw for greater detail with each successive analysis pass. Values of R and R_z used for the operational analyses are given in Table 1. The operational scaling distances, reduced from their default values, reflect the average Utah domain station separation, which is on the order of 10 km.

Estimates of the observation error are specified in Table 2 and Fig. 1 and are a function of the data type. The background error profiles are also given in Fig. 1. Default error profiles are denoted by circles (RUC-2) and squares (raob/wind profiler). Included in Table 2 are the assumed error levels for aircraft, ship, buoy, WSR-88D, METAR, and MesoWest (Horel et al. 2002b) observations with default values listed first. The “default” values are for the most part educated guesses for the values of the various parameters that work at 10-km resolution for the Midwest and are thus based on “tuning” for Oklahoma. The values reflect instrument error, precision of reporting error, and error of representativeness. Even if an instrument is precise, the measurement would still only record the temperature at a single point while we seek a value representative of the entire grid cell. Obviously, the error of representativeness would increase for complex topography and coarser resolution. The large values for the surface pressure errors reflect the typical problems with complex terrain pressure observations, most notably the mismatch between the observation elevation and the grid elevation. Note also that the “precise” elevations of some of the pressure sensors are an issue. The slightly increased surface values (over that of the default) for the relative humidity, and winds, directly reflect grid-box variability. (Even at 1 km, the steep gradients in the topography tend to dominate.) Ultimately the operational error values were chosen in a series of trial and error. We assume that the error profiles in the radiosonde winds (Fig. 1d) are the same as those of the profiler data and that METAR and MesoWest error levels are also the same. The 3 m s⁻¹ error in the WSR-88D data is due to the coarse-resolution NIDS data and could be decreased to near 1 m s⁻¹ if the higher-resolution level II data were used. Because of the high uncertainty, we are not currently using ACARS relative humidity in the analyses. In general, we assume that the background errors are larger than the observation errors (Fig. 1). The default values for the RUC-2 were obtained from the National Oceanic and Atmospheric Administration (NOAA)/Forecast System Laboratory's (FSL) statistics for the 60-km RUC model and increased slightly (subjectively) to account for representativeness error, that is, the inability of a 60-km model to forecast the motions resolved on a 10-km grid. Because of the variability associated with the complex terrain, we have increased the expected errors near the surface. It is also important to point out that error levels in the background surface field may be larger because the ADAS remapping algorithm uses the 3D RUC-2 40-km pressure grid instead of the RUC-2 surface fields. The RUC-2 diagnosed surface variables are likely more representative of the true surface fields—a point to which we return in section 6. Adjustments in the RUC-2 default profiles are confined to the lowest 2 km and are made to all fields. Errors in the background wind speed are large near the surface and then increase again aloft—in part a result of the increased wind magnitudes aloft. Sonde RH, temperature, and wind error profiles have been increased from their default values to mitigate analysis bull's-eyes that often accompany the sounding ingest at 1200 and 0000 UTC (the sounding is limited to the first pass to further reduce this effect on the analysis). The sonde RH error profile increases with height between 7.5 and 10 km, reflecting the increased uncertainty due to the sonde hygristor's inability to accurately measure low relative humidities (Brousaides 1975).

d. Quality control

For surface observations, ADAS employs both spatial and temporal quality control measures. The temporal quality control compares consecutive hourly surface files and discards data if the difference between the observations and the average difference (computed from all surface observations) is greater than four standard deviations. Because of steep gradients in the complex terrain, the ADAS two-dimensional spatial quality control, which used the Barnes (1964) technique to compare observations with other nearby station data, has been discarded and replaced by an external quality control procedure. This procedure is applied as a regular part of the processing of the MesoWest data stream (Splitt and Horel 1998). The technique uses statistical linear regression to evaluate the quality of the surface temperature, dewpoint, and pressure. Data-quality flags are generated for data in the MesoWest in a real-time fashion based on the agreement between the current observations and the regression (the regression coefficients are estimated using observations taken over the previous 6-h period). The data are separated into “good,” “questionable,” and “bad” observations based on departures from their regression values. We are currently using both the good and questionable observations.

An additional quality control measure compares the background field (i.e., RUC-2) to the observations by interpolating the former to observation locations. If the differences are larger than a specified threshold level, the observations are rejected. ADAS uses the standard error variance to define a tolerable threshold τ,

View Expanded

where β is a proportionality factor (ranging from 2 to 4 depending on the source), and ɛ_b and ɛ_s are the background and observation errors, respectively (see Figs. 1a–d and Table 2). We have removed the trilinear interpolation (reducing it to bilinear) when mapping the background temperature and relative humidity to the observation locations to arrive at a first-guess background-to-observation difference field. (Note that we are not neglecting “vertical” gradients because the bilinear interpolation is along the ADAS terrain-following coordinate.) In steep terrain, errors in the station elevation and/or discrepancies between the actual station elevation and gridpoint elevations can cause excessively large first-guess difference fields due to the introduction of “free atmosphere” background values from elevated analysis points surrounding the terrain.

Background-to-observation differences are also used to quality control the upper-air data (i.e., profiler, sounding, aircraft) and the Doppler radial winds. As part of the preprocessing, the radial wind data are examined for aliasing, and are also discarded for large variances and/or lack of coverage within an analysis grid cell.

a. Terrain factor

During the course of our early work with the ADAS, we encountered two commonly observed features: 1) the free-atmosphere analysis was often observed to be anomalously warm/cold relative to the ambient atmosphere, and 2) in the absence of observations, summertime near-surface analyses over high-terrain regions were actually cooler than the ambient air (i.e., the opposite of what one might expect in the presence of an elevated heat source during the day). By “free atmosphere” we refer to grid points that are relatively high above (but in relatively close horizontal proximity to) the terrain, while by “ambient” we mean the grid points that are relatively far away from any topography. To explain the first of these analysis features, Fig. 2 illustrates the analysis relationship between a surface observation denoted by point A, grid points (1 and 2), and the vertical distance between the observation and grid point (ΔZ_ij) [see Eq. (10)]. For this simple case, the contribution (of the observation) to the weight ρ_ij due to ΔZ_ij is identical for grid points 1 and 2, despite the fact that grid point 2 is located much higher above the underlying terrain (and therefore should typically be influenced less by surface heating/cooling). In terms of the latter problem, analyses in high-elevation data-void regions are dominated by the background field, which is essentially an interpolation of the RUC-2 temperature lapse rate onto the ADAS terrain. Because of this, the ADAS tends to produce anomalously cool high-elevation analyses where there is little or no data. In response to these issues, we attempt to more realistically capture the impact of elevated terrain on the analyses by introducing an additional term (hereinafter referred to as the terrain factor) to Eq. (10) such that the observation-to-gridpoint Gaussian correlation function becomes

View Expanded

where z_x is the height of the grid point at location x, T_x is the height of the terrain at location x, and R_T is the scaling distance for the terrain factor. Note that Eq. (12) is applied to surface data only. We have set R_T to be 500 m in order to lessen the impact of observations on analysis points high above the surface. The impact of the terrain factor is twofold: to spread high-elevation observation information to data-sparse regions at similar elevations, and to reduce the impact of observations on analysis grid points high above the terrain. In the absence of local upper-air data, the ADAS analysis matches that of the background field in free-atmosphere locations.

b. Minor additions/modifications

ADAS employs the Benjamin and Miller (1990, hereinafter BM) sea level reduction technique as part of its diagnostic package. We have replaced BM with the Mesinger and Treadon (1995, hereinafter MT) pressure reduction technique, which uses the virtual temperatures along the terrain slopes as boundary conditions for a two-dimensional Poisson equation. One advantage of the MT technique is that it replaces the standard lapse rate assumption of BM with “approximate” lapse rates estimated from the “interpolated” (belowground) virtual temperatures. The technique is two-dimensional and thus relatively inexpensive computationally.

Because the ADAS is not directly coupled to a forecast model, observations that enter the ADAS at or near the surface can produce superadiabatic layers. This happened frequently during the warm season because the RUC-2 tends to be relatively cool just above the surface and ADAS lacks sufficient upper-air observations to warm the background field. To correct this, ADAS previously adjusted these superadiabatic layers by increasing or decreasing temperatures aloft depending on a user-specified direction (i.e., top-down or vice versa). In an effort to better represent the interface between the background and surface data, we have replaced the ADAS adjustment with a standard dry convective adjustment scheme (e.g., Haltiner and Williams 1980, p. 312).

a. The background field

Because the RUC-2 background field that we employ is created by mapping data from the 3D 40-km RUC-2 pressure grid to the ADAS grid, we are not necessarily comparing/initializing ADAS analyses with the best possible RUC-2 first-guess field at the surface. The RUC-2 diagnoses a surface temperature and dewpoint field using local lapse rates from model layers above the surface. To arrive at surface temperatures and dewpoints close to station elevation, which is assumed to be in the valleys, the RUC-2 uses a minimum topography field obtained from a high-resolution 10-km topographic grid. The MAPS terrain is a “slope envelope” topography field (mean plus standard deviation over a grid box) that is derived from a 5′-resolution topography dataset. The envelope topography field is subjected to one smoothing and one desmoothing pass from the filter developed by Shapiro (1970). The standard deviation is calculated using “subgrid” terrain data (e.g., the 40-km grid uses the terrain variance estimated from a 10-km grid).

In order to better understand the differences between the RUC-2 surface data (RUC2S) and that generated by ADAS using the 3D RUC-2 pressure grid (RUC2P) we present temperature time series for 28 April 2001 (Figs. 5 and 6). The first time series (Fig. 5) is that for a high-elevation (at ∼3351 m) station located at Hidden Peak, Utah (HDP, denoted by filled squares). Also shown are several nearby lower-elevation stations. As one might expect, the RUC2S (long-dash line) is too warm at HDP because (in part) it is extrapolated to an elevation approximately 1200 m below that of HDP (see Table 3). A back-of-the-envelope calculation indicates observed lapse rates near the dry adiabatic that would account for approximately 12°C of the 18°C difference (due to extrapolation alone) between the 2000 UTC HDP observation and the RUC2S estimate. Note that the RUC2P at HDP (dash–dot line) is also too warm but is closer to the observations at HDP than either the RUC2S or ADAS 10-km grid (dotted line). Ostensibly, the warm surface observations at nearby (i.e., within a few kilometers) lower-elevation stations Alta-Collins (CLN; diamond) and Alta-Guard House (AGD; triangles) have a greater impact on the ADAS 10-km analysis than on the coarser-resolution RUC-2. A summary of the analysis, RUC2S, and actual elevations at HDP (and valley station MS8) and surrounding stations, are given in Table 3. Differences between the actual elevation and background elevation are large (500–1200 m) for the high-elevation stations while the 1-km ADAS grid elevations are significantly closer to the true elevations. As a result, it is no surprise that the ADAS 1-km analysis (solid line) is closest to the observed temperature time series at HDP. At the Tooele Valley station MS8 (Fig. 6), the RUC2S estimate (long-dash line) is quite good—better than the first-guess temperature field produced from the RUC2P grid (dot–dash line). Unlike HDP, the RUC-2 extrapolation to the lowest elevation at MS8 is relatively close to the actual station elevation. Note also that, in relatively flat terrain, there is little difference between the 1- and 10-km ADAS analyses (solid and dotted lines, respectively).

b. Summer case

Analysis products during 8 September 2000 are presented here. We chose this day because it was convectively active, producing thunderstorms along and east of the Wasatch Front (Fig. 7). As previously mentioned, a key product in our day-to-day operations is the display of RUC-2-only analyses, that is, the RUC-2 3-D gridded forecasts mapped to the high-resolution ADAS grid. This product allows us to evaluate whether the small-scale detail that we observe is an artifact of better resolving the terrain or the “true” value added by the analysis of local data. Figures 8a and 8b depict surface temperature (°C) and streamlines for the 2100 UTC 8 September RUC-2 (interpolated to the ADAS grid) and 2200 UTC 8 September ADAS analysis, respectively. There are typically on the order of 150–200 surface observations inside our 1-km grid; however, for display purposes, we show a reduced set of surface observations. Although the terrain impact is evident, the surface observations increase the temperatures in the southwest portion of the analysis domain where the RUC-2 background is too cool in comparison with observations. The ADAS analysis also increases temperatures in the area along and west of the Wasatch Front range (i.e., to the east of the Great Salt Lake, hereafter GSL) and southeast of the GSL in response to the relatively warm surface observations in these areas. There is little correlation between the RUC-2 streamlines and observed surface winds. The ADAS analysis resolves the mesoscale southerly flow over the western half of the domain, and the diffluent flow centered over the southeastern portion of the GSL. There are noticeable differences in wind speed as well (not shown) in the SW portion of the domain where the impact of the Dugway network is to increase the wind speeds in the analysis over that of the RUC-2 (the same is true for the Toelle Valley where the RUC-2 is too low) and a general decrease in wind speed along the western edge of the Wasatch mountain range (with the decrease extending along the entire N–S Wasatch corridor).

Similarly, the surface dewpoints for the 2100 UTC RUC-2 and 2200 UTC ADAS (Figs. 9a and 9b) indicate that the RUC-2 is especially dry over the GSL. The RUC-2 is also too dry along the Wasatch Front (east and south of the GSL) with RUC-2 dewpoints ranging from −5° to 0°C while ADAS and the observations range between 5° and 10°C. The bull's-eye in both the temperature and dewpoint analyses over the southwest portion of the GSL is on the order of the ROI for the last pass (12 km) as ADAS attempts to analyze for the relatively cool/moist air in that region.

Because 8 September was a convective day, we also show the lifted index (LI) for both the RUC-2 (2100 UTC) and ADAS (2200 UTC; Figs. 10a and 10b). Estimates of the RUC-2 and ADAS LI are computed using an average of temperature and mixing ratio over the lowest three grid levels (approximately 150 m). RUC-2 LIs are significantly larger over most of the domain (except for the southeast). Over the southern portion of the GSL, LIs approach −6 in the 2200 UTC ADAS analysis while the RUC-2 indicates stable conditions with LIs on the order of 4–6. Although we have reduced some of the sensitivity of the LI estimate to surface conditions (by vertically averaging the dewpoint and temperature), the differences between the two are still quite large. These differences are consistent with the surface temperature and dewpoint temperature analyses in Figs. 8 and 9, reflecting the drier and cooler RUC-2. It is also important to point out that LI estimates will be affected by the use of the previous-hour RUC-2 especially during transition periods when surface cooling/warming is significant or during rapidly changing weather conditions. In Fig. 11, we show the 2200 UTC sounding taken at a point near the Salt Lake City airport. It is evident that the low-level RUC-2 is too cool and too dry. Note also that the RUC-2 low-level winds are southwesterly while ADAS indicates southeasterly winds, as was observed over the Salt Lake valley (SLV) at 2200 UTC. In the absence of data aloft, the upper-air ADAS analysis matches that of the RUC-2.

c. Winter case

A winter weather event on 16 March 2001 is examined over the northwest Utah domain. The event is characterized by the passage of a strong cold front and subsequent valley/mountain snowfall. Visible satellite imagery for 1700 UTC (Fig. 12) indicates extensive cloud cover south and east of the GSL decreasing to the west and southwest of the GSL. Figures 13a and 13b depict surface temperature (°C) and streamlines at 1800 UTC for 16 March 2001 RUC-2 (interpolated to the ADAS grid) and at 1900 UTC for 16 March 2001 ADAS analysis, respectively. Unlike the previous case (section 6b), data from a dense network of surface stations are absent over the southwest portion of the Utah domain. As a result, the ADAS analysis is close to that of the background in this region. The surface observations increase the temperatures over the north arm of the GSL (by 2°–4°C) and the Utah Lake area in the southeastern portion of the domain (by 1°–2°C). The ADAS analysis also increases temperatures over the elevated terrain in the eastern third of the domain with RUC-2 high-elevation temperatures between −6° and −10°C whereas the observations are warmer (between −5° and 0°C). Absent in the RUC-2, the analysis resolves the small-scale southeastly flow south of the GSL, the westerly flow to the west of Utah Lake, and the cyclonic circulation over the southeast arm of the GSL. Comparison of surface dewpoint observations with the 1800 UTC RUC-2 and 1900 UTC ADAS (Figs. 14a and 14b) indicates that the RUC-2 is on the order of 5°C too dry over most of the northwest Utah domain. For the most part, ADAS moistens the domain, including both high elevation and valleys.

In the presence of clouds and precipitation, the WSR-88D has a significant impact on the upper-air analyses via the ingestion of radial wind data, which is the primary source of nonsurface data for ADAS. However, for widespread precipitation events, the NIDS data stream can be computationally expensive, involving (at times) upward of 200 000 individual radial wind observations. In order to produce timely analyses, we reduce the ROI (for the radar data stream only), thereby limiting the number of surrounding observations that impact the analysis at the observation point in question. Because of the lack of smoothing with neighboring observations, this effectively results in a direct “overwrite” (instead of analysis) of the background radial wind with the observed.

To illustrate the impact of the WSR-88D data on the upper-air analysis, we run ADAS both with and without NIDS data. Figure 15 is a horizontal cross section of the 600-mb difference wind field (m s⁻¹) (ADAS with and without the NIDS radial winds) and the column-integrated total water, q_t (g kg⁻¹), for a winter precipitation event at 1700 UTC 16 March 2001. The q_t is the sum of cloud ice, cloud water, rain, snow, and graupel mixing ratios over the vertical extent of the analysis domain (approximately 11 km). As previously mentioned, the mixing ratios and cloud cover are computed as part of the cloud analysis package. It can be seen from Fig. 15 how the NIDS data affects the upper-level wind analysis. The difference vectors indicate that, directly to the west of the GSL, the winds have shifted from a westerly component (ADAS without radar) to a more northerly direction (ADAS with radar). Noticeable changes (1–3 m s⁻¹) in the analyzed wind field are also present over Utah Lake in the southeastern portion of the domain and northeast of the GSL where the winds have a more northerly component as a result of the NIDS ingest. A broad area of smaller wind adjustments is also present both south and east of the GSL. WSR-88D data at 1700 UTC (not shown) indicate that these regions are associated with locally enhanced reflectivity. Radial wind data (Fig. 16) from the 0.5° KMTX tilt are consistent with the analysis adjustments—indicating flow toward the radar north of Promontory Point, flow away from the radar to the west, and a strong inbound–outbound wind couplet in the southern portion of the domain. The radar data affect a relatively deep layer extending from 500 to 700 mb (not shown). The analyzed q_t is in relatively good agreement with the visible satellite image (Fig. 12), which indicates overcast conditions to the east and south of the GSL with areas of broken clouds southwest and northwest of the GSL.

Vertical cross sections across the northern Utah domain are part of our regular operational output. We show north–south vertical cross sections (as indicated by the line AB in Fig. 3) along the Wasatch Front for the RUC-2 and ADAS at 1800 and 1900 UTC, respectively (Figs. 17a and 17b). The 0°C contour, which is nearly absent on the RUC-2 cross section (the RUC-2 is too cool), is depicted by the “thick” dashed line while the potential temperature is contoured by the thin lines. Here, q_t is shaded and the horizontal winds are displayed with a full barb equal to 5 m s⁻¹. The impact of the observations (radar and surface) on the analysis (Fig. 17b) is evident with the 0°C isotherm sloping downward toward the surface in the northern portion of the domain (associated with the surface cold front). The potential temperature analysis also supports the presence of a front in the analysis with a more stable low-level profile to the north (left) in Fig. 17b. There are also large differences in the wind field between the RUC-2 and ADAS analyses over the lowest 4 km. The winds have backed (from northwest to west) over the central Wasatch Front while the near-surface ADAS (RUC-2) winds in the SLV are from the southwest (east). The q_t field (ADAS only) indicates that a relatively deep cloud layer exists over this portion of the analysis domain.

d. Spring case

The terrain of the 10-km-resolution ADAS surface analysis is shown in Fig. 18. Superimposed upon the terrain are the locations of the surface observations available between 2000 and 2100 28 April 2000, which are used in the ADAS analysis valid at 2100 UTC (Fig. 19). The distribution of stations in the western United States is highly variable: the average distance from an analysis grid point to the nearest observations is less than 10 km across much of northern Utah and yet is over 100 km in the Navajo Nation of northeastern Arizona.

As an illustration of the utility of the ADAS surface analyses over the western United States, we begin by showing in Fig. 20 the NCEP automated surface analysis for 2100 UTC 28 April 2000. The NCEP analysis helps to define a large trough that extends from central Montana through northwestern Utah to southern Nevada and into Arizona. The ADAS sea level pressure analysis (not shown) is quite similar to the NCEP analysis. The gross features of the pre- and postfrontal environment are defined by the NWS–Federal Aviation Administration (FAA) observations. A well-defined wind shift across Nevada and Utah is apparent in the ADAS surface analysis (Fig. 19) as well as a strong baroclinic zone with temperature decreasing from greater than 25°C over the deserts to the southwest of the GSL to less than 15°C over eastern Nevada. Channeling of the flow up the Snake River valley in Idaho (a feature that was present in the RCU2 analysis) is particularly striking as are many other local weather features associated with the passage of the strong cold front.

To assess the impact of the MesoWest surface observations upon the analysis, Fig. 21 shows the difference between the ADAS surface analysis and the background fields of temperature and wind (i.e., the RUC-2 3D analysis interpolated to the ADAS grid). Overall, the RUC-2 provides a very good background field with adjustments in temperature and wind speed made by ADAS of only 1°–2°C and 1–2 m s⁻¹. The RUC-2 captured the location of the baroclinic zone and wind shift across Nevada and Utah quite well, with large adjustments required only over the deserts of western Utah where the frontal position remained farther west of that specified by the RUC-2 (i.e., the ADAS analysis is warmer with more southerly wind near the point labeled A). Although not exclusive, some of the largest adjustments are evident in mountain valleys and the relatively low-lying areas adjacent to major mountain ranges, for example, near Moscow, Idaho (labeled B); north of Redding, California (labeled C); and along the Dolores River, in Colorado (labeled D). The observed temperatures at these and other locations where the discrepancies between the RUC-2 and ADAS are large are close to those analyzed by ADAS. Hence, the large adjustments to the background field in these regions indicate that interpolation of the RUC-2 3D temperature field to the higher-resolution ADAS grid is not able to capture the local heating under way without the additional constraint provided by the local surface observations.

Since variations in terrain can mask significant variations in temperature associated with weather events, Fig. 22 shows the 24-h change in surface temperature following 2200 UTC 28 April as the cold front continued to progress eastward. In this instance, temperatures rebounded sharply in the Cascades of Oregon and Sierras of California while temperatures dropped significantly over Wyoming and Utah.