Journal of Geophysical Research: Atmospheres

Representation of the Sierra Barrier Jet in 11 years of a high-resolution dynamical reanalysis downscaling compared with long-term wind profiler observations

Authors


Corresponding author: M. Hughes, PSD, ESRL, NOAA, 325 Broadway, R/PSD2, Boulder, CO 80305–3328, USA. (mimi.hughes@noaa.gov)

Abstract

[1] We investigate the representation of the Sierra Barrier Jet (SBJ) in four numerical models at different resolutions, primarily documenting its representation within a high-resolution (6 km), 11-year WRF reanalysis downscaling (WRF-RD). A comprehensive validation of this dynamical downscaling is undertaken during 11 cool seasons (water years 2001–2011, October to March) using available wind profiler data at Chico, CA (CCO). We identify SBJ cases in the observed CCO wind profiler data, as well as in WRF-RD at the closest grid point. WRF-RD's representation of the SBJ at CCO is compared with that of other reanalysis products with coarser horizontal resolutions (i.e., the North American Regional Reanalysis (NARR), the California Reanalysis downscaling, and the NCEP/NCAR Reanalysis) to assess whether downscaling is necessary to correctly capture this topographically induced low-level jet. Detailed comparisons across California between WRF-RD and NARR suggest downscaling is necessary: Only WRF-RD at 6 km resolution is well-capturing this dynamical feature. A catalog of modeled SBJ events that have significant timing overlap with observations is created and used to further assess WRF-RD's representation of SBJ events. In addition, observation-model comparisons of other meteorologically important variables (e.g., precipitation melting level, wind profiles, temperature, and relative humidity) are performed in order to evaluate WRF-RD's ability to capture the dynamical evolution of the SBJ. The detailed, case-by-case comparisons reveal WRF-RD accurately represents 56 percent of the 256 observed SBJ cases occurring during these 11 cool seasons, albeit with a weak wind bias that increases with jet maximum wind strength.

1. Introduction

[2] The Sierra Barrier Jet (SBJ) is a low-level jet situated above the western foothills of California's Sierra Nevada, which is dynamically generated as onshore flow impinges on stably stratified air within the Central Valley [Parish, 1982; Marwitz, 1983; Smutz, 1986; Neiman et al., 2010]. Wintertime cyclones, which cause strong onshore flow and thus also cause SBJ conditions, generate the majority of California's precipitation [Dettinger et al., 2004; Pandey et al., 1999] – up to 60% of cool season (October–March) precipitation fell during SBJ conditions at Chico, CA (CCO) [Neiman et al., 2010]. Its prevalence during stormy conditions, influence on moisture advection, and potential to act as a virtual barrier causing precipitation to fall upstream of elevation gradients [Neiman et al., 2002; Kingsmill et al., 2006; Hughes et al., 2009] lead to the SBJ strongly influencing California's precipitation distribution [Kim and Kang, 2007]. Lundquist et al. [2010] showed that orographic enhancement of Northern Sierra precipitation was sensitive to SBJ altitude, while Neiman et al. [2010] showed sensitivity of precipitation in the northern Central Valley to SBJ strength. Understanding the SBJ's influence on precipitation is particularly relevant given California's propensity for extremes in its hydrological cycle and the implications those have for water management [Dettinger et al., 2011]. The strong topographically enhanced flow during SBJ events also potentially impacts aerosol transport in the CA Central Valley [Ewell et al., 1989].

[3] Because of its importance to local climate and weather conditions, numerous studies have documented SBJ structure. However, most previous studies examined relatively few SBJ events [e.g., Reeves et al., 2008; Kim and Kang, 2007; Parish, 1982] or used spatially confined data [e.g., Neiman et al., 2010; Smutz, 1986]. Reanalysis products have the potential to capture more events over a larger area. However, given the horizontal spatial extent of the northern Sierra Nevada SBJ predicted by the Rossby radius of deformation, ∼200–400 km [Pierrehumbert and Wyman, 1985] (calculated using a typical Central Valley Brunt-Väisälä frequency range of 0.01–0.02 s−1, barrier height of 2 km, and Coriolis parameter for 35 N), reanalyses' ability to represent the SBJ – a topographically generated feature – is questionable due to their coarse resolution and topography.

[4] Dynamically downscaled atmospheric data is becoming a prevalent tool in the effort to understand climate and weather processes on scales applicable to humans and ecosystems [Laprise, 2008; Kanamitsu and Kanamaru, 2007; Hughes et al., 2007; Leung et al., 2006]. Dynamical downscaling is preferable to statistical downscaling for the study of dynamical effects in complex terrain because it can capture fine-scale processes (e.g., dynamically generated orographic blocking and formation of a barrier jet from large-scale flow imposed on high-resolution terrain) otherwise absent in coarser-scale gridded data sets, and therefore, processes that would not be represented by a statistical model [van der Kamp et al., 2012; Gutmann et al., 2012; Hughes et al., 2009]. Generally, representation of surface conditions is closer to observations in dynamically downscaled products than in coarser-scale products [e.g.,Kanamitsu and Kanamaru, 2007]. However, most skill evaluations of dynamically downscaled data use only surface data or temporally coarse rawinsonde data, and the veracity of these dynamically downscaled upper-air conditions remains uncertain. Furthermore, topographically induced dynamical features like the SBJ have the potential to impact meteorological conditions aloft (i.e., up to at least 4 km MSL), where most reanalysis products are generally thought of as accurate given the distance from the topography.

[5] This study investigates whether downscaling of modern reanalysis products is necessary to adequately represent the SBJ – a CalWater science objective [Ault et al., 2011](http://www.esrl.noaa.gov/psd/calwater/)–and is possible due to a long time series of hourly boundary layer wind profiler observations at multiple locations [Neiman et al., 2010] (provided largely by NOAA's Hydrometeorological Testbed). We examine the SBJ at CCO in four data-assimilation-constrained numerical models: the NCEP/NCAR reanalysis product (NNRP; 2.5 degree grid spacing) [Kalnay et al., 1996], the North American Regional Reanalysis (NARR; 32 km grid spacing) [Mesinger et al., 2006], the California reanalysis downscaling (CaRD10; 10 km grid spacing) [Kanamitsu and Kanamaru, 2007], and NARR downscaled to 6 km horizontal resolution with the Weather Research and Forecast (WRF) model [WRF Reanalysis Downscaling: WRF-RD]. The examination of NNRP and CaRD10 data is limited to CCO, because of the former's extremely low horizontal resolution and the latter's low vertical resolution, and is included only for completeness. NARR's 32 km grid spacing and effective resolution of ∼60–120 km [Ainslie and Jackson, 2010] are potentially adequate to resolve the SBJ given its expected width of 200–400 km, therefore we more extensively compare NARR's SBJ with WRF-RD's, across the Central Valley. Since WRF-RD's climatological SBJ most closely matches that observed and looks more physical than NARR's across CA, we present a more extensive assessment of WRF-RD's kinematics and thermodynamics to determine its utility for future study of SBJ dynamics.

2. Description of Data Used and Model Configuration

2.1. Observations

[6] We use time series data spanning one or more cool seasons from eleven 915-MHz wind profilers [e.g.,Carter et al., 1995] and collocated surface meteorological instruments across California (see Figure 1 and Table 1) for validation purposes. The site with the longest continuous time series, 11 years, is at Chico (CCO) in the northern Sacramento Valley. Due to its long time series, detailed comparisons between CCO observations and the reanalysis data sets and downscalings represent the backbone of this validation study. The Sacramento profiler is owned and operated by the Sacramento Metropolitan Air Quality Management District; the other 10 profilers were deployed and operated by the National Oceanic and Atmospheric Administration's Earth System Research Laboratory (NOAA/ESRL) as part of NOAA's Hydrometeorological Testbed (HMT)[Ralph et al., 2005] and related, earlier field campaigns.

Figure 1.

Domains of WRF-RD. Color contours show terrain elevation, black line shows coastline, and red box shows boundary of 6 km domain. Site locations of observations are shown with pink circles (915-MHz wind profiler data) and black squares (surface meteorological data). Instruments are described in detail insection 2.1.

Table 1. Site Information for 915-MHz Wind Profilers and Collocated Surface Measurements (a and b, Respectively) Used in This Studya
Three-Letter AbbreviationSite NameLatitudeLongitudeElevationTemporal CoverageData Available
  • a

    Site locations can be seen in Figure 1. Temporal coverage is listed as the year during which the available cool season ends (e.g., Oct. 2000–Mar. 2001 is listed as 2001).

BBYBodega Bay38.32−123.07122001, 2004–2011a, b
CCLChowchilla37.11−120.24762004–2011a, b
CCOChico39.69−121.91412000–2011a, b
CCRConcord37.99−122.06142010–2011a, b
CFCColfax39.08−120.946442009–2011a, b
GVYGrass Valley39.17−121.116892000–2005a, b
LHSLost Hills35.62−119.69802010a
PPBPoint Piedras Blancas35.66−121.29112010–2011a, b
SACSacramento38.30−121.4262010a
SHSSloughhouse38.50−121.21502006–2011a, b
TRKTruckee39.32−120.1417962010a

[7] All wind profilers provided hourly averaged vertical profiles of horizontal wind velocity from ∼0.1 to 4–6 km above ground with ∼100 m vertical resolution and ∼1 m s−1accuracy in all weather conditions. The profiler winds were objectively edited using the vertical-temporal continuity method ofWeber et al. [1993] and then visually inspected to flag remaining outliers. In precipitating conditions, the melting level height was determined hourly from the wind profilers using the objective radar brightband height (BBH) detection method of White et al. [2002]. At each NOAA/ESRL site, 2-min observations of temperature were collected from a 10 m meteorological tower.

[8] During the HMT winters of 2005–2006 and 2006–2007, rawinsondes were released episodically at Sloughhouse, California (SHS, 92 and 43 total sondes for each respective winter), during the landfall of extratropical cyclones. These sondes recorded temperature, relative humidity, and horizontal wind velocity at closely spaced pressure intervals. Data from 49 of these sondes released during 10 SBJ events was used to assess the mean ambient static stability in the northern Central Valley (as represented by SHS static stability) and to provide validation against the mean static stability portrayed in WRF-RD for those same events.

2.2. WRF-RD Simulation

[9] The high resolution dynamical downscaling was generated by the Weather Research and Forecast model, Version 3.0.1.1 (WRF) [Skamarock et al., 2008]. The downscaling simulation contained an 18 km horizontal resolution domain covering California and extending west over the Pacific Ocean, and a 6 km domain covering only California two-way nested in the 18 km domain (Figure 1). Both resolutions are available hourly. At 6 km horizontal resolution, all the major mountain complexes in California are resolved, including the major ridgelines transverse to the primary axis of the Sierra Nevada (Figure 2). Each domain contained 27 vertical levels, with the vertical grid stretched to place the highest resolution in the lower troposphere. In the 18 km domain, the Kain-Fritsch cumulus parameterization was used [Kain, 2004]; in the 6 km domain only explicitly resolved convection could occur. Both domains used the Yonsei University (YSU) boundary layer scheme [Hong et al., 2006], the Morrison 2-moment microphysics scheme [Morrison et al., 2009], the rapid radiative transfer model longwave radiation scheme [Mlawer et al., 1997], the Dudhia shortwave radiation scheme [Dudhia, 1989], and the Noah land surface model with 4 ground layers [Chen and Dudhia, 2001].

Figure 2.

Elevation map of California as represented by (a) NNRP, (b) NARR, (c) WRF-RD, and (d) USGS 30-s digital elevation map (http://eros.usgs.gov/). Colors show elevation, as noted in the colorbar. Thick black line shows real coastline. White cross shows location of CCO, orange ‘m’ shows location of nearest grid point to CCO in the model products (second closest grid point in NARR – see text for details).

[10] WRF-RD data was generated for 11 cool seasons (water years 2001–2011). Throughout this period, WRF was reinitialized three hours prior to the first of each month and every 5 days, 3 h, thereafter. The first three hours of each run were discarded for model spin-up, resulting in a temporally continuous run, with slight meteorological discontinuities between full model initializations. Interior and lateral boundary conditions are provided by NARR (section 2.3). This dynamical downscaling technique captures the variability in meteorological conditions over an 11-year period (1995–2006) in Southern California when generated with WRF's predecessor model [Hughes and Hall, 2010; Hughes et al., 2009].

2.3. Other Reanalysis Products

[11] NNRP [Kalnay et al., 1996] is a global reanalysis data set generated with the NCEP global operational model frozen at its 1995 version, with data input through NCEP's climate data assimilation system. The gridded version has 2.5 degree horizontal resolution and is available on 17 standard meteorological pressure levels every 6 h.

[12] NARR is a 32 km/45 layer product created by ingesting surface and upper air observations over the continental U.S. with the Regional Data Assimilation System into the NCEP Eta Model to produce a spatially and temporally consistent and well-represented meteorological record [Mesinger et al., 2006]. NARR's lateral boundary conditions are provided by NNRP, and NARR data is available every 3 h for 1979-present.

[13] CaRD10 [Kanamitsu and Kanamaru, 2007] uses the Regional Spectral Model to dynamically downscale NNRP to 10 km horizontal resolution over California. CaRD10 data was available hourly from 1948 to 2008; data from Jan. 1, 2001 to Dec. 31, 2008 was used. CaRD10 has the same vertical resolution as NNRP (standard meteorological pressure levels), although the data was generated at higher-vertical-resolution sigma layers.

3. Identification of the SBJ in Reanalysis Products, and Comparison With Observations

3.1. SBJs at CCO in Four Model Products of Varying Resolution

[14] To determine SBJ conditions within the observed and modeled data sets, this study uses the following criteria identified in Neiman et al. [2010]. First, the profile had to occur during the cool season and had to have a relative maximum in the Sierra-parallel component of the flow (hereafter, V; directed from 160° to 340°) of greater than 12 m s−1 below 3 km MSL (i.e., below crest level). If more than one relative maximum was observed, the maximum with greatest V was taken as SBJ amplitude (hereafter, Vmax). Also, V had to decrease by more than 2 m s−1with increasing height between Vmax altitude of and 3 km MSL. Furthermore, Vmax must occur at or above 200 m and the points above and below Vmax must contain data. These criteria were designated to ensure that only those profiles exhibiting an unambiguous SBJ signature are included in the subsequent analyses. Finally, SBJ “cases” were defined as consecutive profiles fulfilling the above criteria for 8 h or longer: 8 consecutive profiles in the observations, WRF-RD, and CaRD10, three consecutive profiles in NARR, or two consecutive profiles in NNRP.

[15] Using this algorithm, hours with SBJ case conditions over the 11-year period were identified in the observations, and composite Sierra-parallel (V) and Sierra-perpendicular (hereafter, U; directed from 250° to 70°) wind profiles were constructed from those observations and at the grid point closest to CCO in WRF-RD, CaRD10, and NNRP for the SBJ times. Prior to compositing, all data was put on a common, high-resolution z-grid via linear interpolation (to handle altitude-varying vertical coordinates); however, data set native vertical resolutions are shown inFigure 3. In NARR, the grid point closest to CCO had a surface elevation of 487 m MSL, so the second closest grid point (63 m elevation) was used to more closely match the site altitude (41 m) at CCO. The locations of these grid points (from NNRP, NARR, and WRF-RD) relative to CCO are shown inFigures 2a–2c. Figures 3a and 3b show U and V component composites for all four models, which all have a reasonable representation of U (Figure 3a), although for NARR, WRF-RD, and CaRD10 U is roughly 3–5 m s−1smaller than observed throughout the depth of the profile. Unsurprisingly given its low-resolution topography (Figure 2), the SBJ is completely absent in the NNRP V composite during observed SBJ cases (Figure 3b). All three of the other models, however, have some semblance of an SBJ in their V composites, albeit 1–3 m s−1 (10–25%) weaker than observed, and with Vmax in NARR and CaRD10 roughly two times above the observed altitude of 800 m.

Figure 3.

(a) Terrain perpendicular and (b) terrain parallel composite wind speed versus altitude at CCO (or grid point closest to CCO) in observations (red line), WRF-RD (blue), CaRD10 (green), NARR (black), and NNRP (purple), on hours identified as occurring within SBJ events identified in the observations. (c) Terrain perpendicular and (d) terrain parallel composite wind speed versus altitude at CCO (or grid point closest to CCO) in observations, WRF-RD, CaRD10, NARR, and NNRP (colors same as in Figures 3a and 3b), on hours identified as falling during SBJ events within observations, WRF-RD, CaRD10, NARR, and NNRP, respectively. Circles on each line show product-native vertical resolution.

[16] To eliminate the effect of timing errors on the SBJ composites, we then applied the Neiman et al. [2010]SBJ objective identification algorithm to each data set, producing an individual SBJ case-list for each data set. Composite SBJ U- and V- profiles using these model-specific cases are shown inFigures 3c and 3d. The number of SBJ cases is 204, 141 (194 when weighted by time series' length), 159, and 9 in WRF-RD, CaRD10, NARR, and NNRP, respectively; 256 cases are tagged in the observations (Table 2). This reduction of SBJ cases as grid spacing increases is likely due to the varying widths of SBJ events: As grid spacing increases, there is likely an overall tendency for SBJ cases with smaller Rossby radius (i.e., cases with smaller Brunt-Väisälä frequency) to be unresolvable. The U composites (Figure 3c) are similar to those composited on observed SBJ cases (Figure 3a), with differences as large as 5 m s−1below 1 km MSL. Surprisingly, NNRP does manage to produce “jets,” but the composite of these 9 cases is kinematically unlike the observed SBJ composite given its deep, high maximum along-barrier flow. NNRP and CaRD10 both overestimate Vmax's amplitude and its altitude, which may occur because cases with larger Rossby radius also would be stronger, due to their larger Brunt-Väisälä frequency [Olson and Colle, 2009]. NARR's Vmax altitude overestimation possibly arises from the high elevation of the adjacent downwind grid point (see Figure 2b), whereas CaRD10s SBJ Vmax altitude is clearly controlled by the vertical resolution of the retained data. WRF-RD has the most accurate representation of the SBJ at CCO (Table 2); WRF-RD's Vmax altitude (∼900 m) is only slightly higher than that observed (∼800 m), although it slightly (∼5%) underestimates the amplitude of Vmax.

Table 2. A Summary of SBJ Statistics at CCO From Observations and Model Data Sets
Data SetHorizontal Grid SpacingCCO/Grid Point Altitude (m, MSL)Number of SBJ CasesVmax Altitude (m, MSL)Vmax Amplitude (m s−1)
  • a

    To account for CaRD10s shorter time period (8 years), we have estimated the 11 year total by multiplying the 8-year total by 11/8. The 8 year total is shown first, estimated 11 year total follows in parenthesis.

915 MHz wind profilerN/A4125680015.39
WRF-RD6 km3820490014.63
CaRD1010 km11141 (194)a145016.73
NARR32 km63159145016.62
NNRP2.5 degrees7339155016.51

[17] In sections that follow, we compare SBJ composites in the observations, WRF-RD, and NARR with one another. To ensure the composites draw on the same SBJ events, we construct matched subsets of the cases identified in observations, WRF-RD, and NARR individually for which at least one hour in the case of one data set falls within 12 h of an hour in the case of the other data set. When a single case in one data set overlaps multiple cases in another, the cases in the latter are combined. For the WRF-RD/NARR/observation comparisons (section 3.2), we use 117 triply matched cases – where cases have been temporally matched and combined across WRF-RD, NARR, and the observations. For the WRF-RD/observation comparisons (section 4.3), we doubly matched 150 cases within WRF-RD and the observations.

3.2. Comparison of SBJs in WRF-RD and NARR

[18] Although WRF-RD's composite SBJ profile is closer to that observed than NARR's (Figure 3d), and WRF-RD contains a more accurate number of SBJ events, NARR's composite SBJ profile is closer than expected to observations and NARR contains a significant percentage of observed SBJ events. Is the effort required to dynamically downscale to a resolution finer than the 32-km resolution of NARR necessary?

[19] To address this question, we first compare plan views of composite wind velocities in WRF-RD and NARR for the 117 triply matched SBJ cases (Figure 4). At 1000 m MSL (Figures 4a and 4b), the strong north- to north-northwestward directed core of the SBJ is visible in both products in the northeastern Central Valley with composite speeds greater than 15 m s−1. However, the grid point to grid point variability is smaller in WRF-RD than in NARR, which has large changes in wind speed between adjacent grid points, and the SBJ eastward extent is tighter against the Sierras in WRF-RD than in NARR, because of the former's more resolved topography. The high wind speeds of the SBJ core also extend farther westward in WRF-RD than in NARR. One of the largest differences between WRF-RD and NARR is west of the coastline from the San Francisco Bay area northwest to Eureka: WRF-RD has strong (∼15 m s−1) southwesterly winds at the coast that intensify out to 125 W, whereas NARR's winds over and west of this section of coast are much weaker (∼10 m s−1). At 2000 m MSL, both products show southwesterly flow in the northern half of the domain that strengthens over land and then turns north-northeastward over the northern Sierra. WRF-RD contains considerably stronger lee-side downsloping flow in the northeastern quarter and along the northern edge of the domain, implying a stronger (and perhaps more realistic) representation of lee-side mountain wave activity.

Figure 4.

Composite winds for triply matched SBJ conditions in (a and c) NARR and (b and d) WRF-RD at 1000 m (Figures 4a and 4b) and 2000 m (Figures 4c and 4d) above mean sea level. Arrows show wind speed and direction and colors show speed (m s−1). Thin black contours show model-resolution terrain, with contours every 800 m starting at 100 m. Thick black line shows actual coastline. Heavy black lines AB and CD in Figure 4b show location of cross sections shown inFigure 5, with circles at locations used to identify NARR grid points.

[20] Differences between the WRF-RD and NARR composite SBJs are comparably striking in Sierra-perpendicular and -parallel cross sections (Figure 5). Given their respective representations of the topography, the Sierra-perpendicular cross sections have similarities: Both WRF-RD and NARR have a ‘terrain-following’ SBJ over the east side of the Central Valley and above the western foothills of the Sierra (Figure 5a), with a jet maximum approximately 1 km above their respective topographies. The jet peaks in strength just west of the western slope of the Sierra Nevada, and the enhanced north-northwestward momentum is absent east of about 121 W. Both data sets share an envelope of weak cross-barrier flow that follows the topography, with strong cross-barrier flow (>15 m s−1) aloft (Figure 5b). However, V in the lowest 500 m of the Central Valley within NARR is much weaker, and NARR's inadequately resolved topography hints at the too-high SBJ core over CCO (compareFigure 5a with Figure 3d). Furthermore, the coastal topography height is underrepresented in NARR, leading to a complete lack of ‘coastal jet’ west of the coastline compared with the ∼10 m s−1V in WRF-RD that extends out from the coastline for about 1/2-degree longitude. The along-Sierra cross section also reveals differences between the two products: WRF-RD (Figure 5d) has a smoothly increasing V that maximizes just upwind (south) of the northern Central Valley topography at approximately 40 N, whereas NARR (Figure 5c) has two local maxima in V at 39 N and 40 N that undulate in altitude.

Figure 5.

Composite Sierra-parallel (filled contours) and Sierra-perpendicular (line-contours) wind for triply matched SBJ conditions in (a and c) NARR and (b and d) WRF-RD across a Sierra-perpendicular (Figures 5a and 5b) and Sierra-parallel (Figures 5c and 5d) cross section. Locations of cross sections are shown inFigure 4. Colorbar at bottom shows magnitude of each wind component (both filled and contours); negative Sierra-perpendicular winds are shown with black dashed lines. Contour interval for line-contours is 2.5 m s−1.

[21] How do these differences in wind speed and direction impact water vapor flux, a variable to first order directly proportional to orographic precipitation amount [Neiman et al., 2009; Hughes et al., 2009; Smith and Barstad, 2004]? To address this, we first show the column-integrated water vapor transport (IVT,Figure 6), calculated from water vapor mixing ratios (q), and zonal and meridional wind components (u and v) interpolated to altitudes from the surface to 4 km MSL. The zonal and meridional IVT components are calculated by inline image* ū* inline image* dz and inline image* inline image* inline image* dz, respectively, where overbars denote layer averages, ρ is air density, and dzis the distance between layers. Both WRF-RD and NARR have greater than 200 kg m−1 s−1southwesterly IVT offshore that penetrates the gap between the coastal mountains, turning northward as it approaches the Sierra Nevada. IVT in WRF-RD is approximately 50% larger throughout much of the region. Vertical integration and inclusion of water vapor reduce grid point-to-grid point discontinuities visible in NARR's wind fields (Figure 4), though some remain in IVT. Finally, over and in the lee of the southern Sierra Nevada are drier in WRF-RD than in NARR, likely because in WRF a larger proportion of the moisture is transported northwest by the SBJ.

Figure 6.

Composite integrated water vapor flux for triply matched SBJ conditions in (a) NARR and (b) WRF-RD. Integration is from the surface to 4 km MSL. Arrows show integrated water vapor flux magnitude and direction and colors show flux magnitude (kg m−1 s−1). Thin black contours show model-resolution terrain, with contours every 800 m starting at 100 m. Thick black line shows actual coastline. Heavy black lines AB and CD in Figure 6b show location of cross sections shown inFigure 7, with circles at locations used to identify NARR grid points.

[22] Figure 7shows Sierra-perpendicular and -parallel water vapor flux in the cross sections ofFigure 5. Here, water vapor flux has been computed by q * U * ρ and q * V * ρfor the Sierra-perpendicular and Sierra-parallel components, respectively, withq and ρ defined as for IVT and U and V defined as winds in the SBJ rotated coordinate system (i.e., as in Figure 5). The cross sections of water vapor flux in WRF-RD and NARR are qualitatively similar to the corresponding cross sections of wind. Along-barrier fluxes are largest near the surface within the Central Valley and west of the coastal mountains, just below the altitudes with the largest along-barrier winds. Since water vapor mixing ratios are largest at the surface (not shown) and across-barrier winds generally increase with height (Figure 5), across-barrier water vapor fluxes peak ∼700 m above the surface over the ocean and at a similar altitude AGL east of the Sierra crest. In contrast, over the Central Valley the large across-barrier fluxes are 2–2.5 km AGL, suggesting the SBJ acts as a ‘virtual barrier’, displacing the cross-barrier vapor fluxes vertically. Water vapor fluxes and winds have similar differences between WRF-RD and NARR. However, the maximum ‘terrain’-following across-barrier fluxes are much larger in WRF-RD, as are the maximum along-barrier fluxes west of the coastal mountains.

Figure 7.

Composite Sierra-parallel (filled contours) and Sierra-perpendicular (line-contours) water vapor flux for triply matched SBJ conditions in (a and c) NARR and (b and d) WRF-RD across a Sierra-perpendicular (Figures 7a and 7b) and Sierra-parallel (Figures 7c and 7d) cross section. Locations of cross sections are shown inFigure 6b. Colorbar at bottom shows magnitude of each water vapor flux component (both filled and contours); negative Sierra-perpendicular fluxes are shown with black dashed lines. Contours interval for line-contours is 0.01 kg m−2 s−1. Large purple arrows on Figures 7a and 7b show locations used for ocean and near CCO transport calculations of Figure 8.

[23] To summarize the impact these differences in water vapor flux at each grid point have on total water vapor transport through the region, we calculate area averaged water vapor transport over two subsections of the across-barrier cross section (i.e.,Figure 6b, AB), near CCO (between 121.6 W and 122.2 W; right set of arrows in Figures 7a and 7b) and over the ocean (west of 124 W; left set of arrows in Figures 7a and 7b). Grid point values of water vapor flux (calculated for Figure 7), are integrated through two depths of atmosphere (the surface to 2 km MSL and 2 km to 4 km MSL), integrated horizontally across the ‘near CCO’ and ‘ocean’ subsections, and then the resultant transport (in kg s−1) is divided over the total amount of atmosphere in each data set for each subsection to account for different underlying topography. These calculations yield area-averaged barrier-parallel water vapor flux (Uf, kg m−2 s−1), barrier-perpendicular flux (Vf, kg m−2 s−1), total water vapor flux (tot, kg m−2 s−1), and the ratio of parallel to perpendicular fluxes (Uf/Vf, dimensionless), shown in Figure 8.

Figure 8.

Composite area-averaged integrated water vapor transport for triply matched SBJ conditions in (cool colors) WRF-RD and (warm colors) NARR. Integration is from (group 1 and 2) the surface to 2 km and (group 3 and 4) 2 km to 4 km MSL. Flux has been horizontally and vertically averaged over subsections shown at the bottom ofFigures 7a and 7b: (group 1 and 3) near CCO between 122.2 W and 121.6 W, and (group 2 and 4) over the ocean west of 124 W. Water vapor fluxes have been normalized by the atmospheric area of the subsection considered to account for any differences in the underlying terrain of the two data sets, yielding units of kg m−2 s−1. Water vapor transport magnitudes are shown on the left axis; ratio magnitudes are shown on the right axis.

[24] Comparing the water vapor transport between WRF-RD and NARR reveals some broad similarities and striking differences. In both WRF-RD and NARR, in the surface-2 km layer near CCO Vf is larger than Uf: The SBJ diverts water vapor that would be across-barrier to the along-barrier direction. However, this diversion is more pronounced in WRF-RD than in NARR, since the Uf/Vf ratio is smaller and less than one in WRF-RD, and is also more vertically confined in WRF-RD than in NARR, which has a larger and greater than one 2–4 km Uf/Vf ratio. In addition, the area-averaged total water vapor transported in WRF-RD is nearly 50% greater than that transported in NARR over the ocean subsection and in the 2–4 km layer near CCO. Since NARR drives WRF-RD at the edge of WRF's 18 km domain (Figure 1), this suggests the better-resolved mesoscale processes in WRF-RD concentrate water vapor transport as the extratropical cyclones associated with these barrier jets make landfall.

[25] Although NARR's representation of the climatological SBJ at CCO is fairly realistic, plan views and cross sections reveal that its representation of topography introduce suspicious-looking factors: large grid point-to-grid point discrepancies with seemingly isolated maximum wind speeds, a complete lack of coastal barrier jet (we expect a coastal barrier jet during SBJ cases [Neiman et al., 2006]), and substantial differences in magnitude and direction of water vapor transport. These factors severely limit NARR's applicability for an investigation of SBJ dynamics, and require higher resolution, as provided by the 6 km resolution WRF-RD.

4. Validation of WRF-RD for Investigation of SBJ Dynamics

[26] Its physically reasonable representation of the composite SBJ (Figure 3d) suggests the WRF downscaling has potential for accurately capturing SBJ dynamics. However, prior to using it to investigate the SBJ, a fuller kinematic and thermodynamic assessment of WRF-RD is necessary.

4.1. WRF-RD's Representation of Snow Level and Surface Temperature

[27] To give more confidence in WRF-RD's representation of dynamic processes throughout California during SBJ events, we assess its representation of thermodynamic variability across all hours (including hours without SBJ events) at locations across the region. For this assessment, we examine two variables: approximate snow level altitude and temperature 2 m above the ground surface (T2m). Snow level altitude [represented in WRF-RD by altitude of the zero-degree Celsius isotherm (hereafter 0°C altitude) and in the observations by BBH (described insection 2.1)] is critical for hydrologic processes [Neiman et al., 2011; Lundquist et al., 2010; White et al., 2002] since it determines what percentage of precipitation will fall as snow in a given watershed. Furthermore, snow level altitudes during atmospheric river events – key precipitation producers in California [Dettinger et al., 2011; Ralph and Dettinger, 2011] – are often extraordinarily high [Neiman et al., 2008], motivating an assessment of this variable in WRF-RD. T2m, although diagnostic in WRF-RD, gives some representation of the temperature within the boundary layer, critical for SBJ representation (for additional discussion seesection 4.2).

[28] A comparison of observed BBH and WRF-RD altitude closest to 0°C at CCO is shown inFigure 9. Because of melt processes, the 0°C isotherm elevation will be 100–300 m above the BBH [Gourley and Calvert, 2003; White et al., 2002], and WRF-RD's 0°C altitude is slightly above observed BBH for most hours as expected. Some hours with very high observed BBH have WRF-RD 0°C slightly lower, indicating a slight cold bias during very warm storm conditions (which may be atmospheric river storms). The red-circled points along the left axis ofFigure 9, where the WRF-RD 0°C altitude is much greater than BBH, have been manually inspected and are incorrectly objectively tagged BBHs. This general good agreement between WRF-RD and observed snow level is highlighted by a high correlation coefficient (0.89), and an additive bias of 155.8 m within the expected range of agreement given melt processes.

Figure 9.

Observed radar brightband height (BBH) as tagged by objective tagging algorithm (abscissa) and WRF-RD altitude closest to zero degrees Celsius (interpolated from sigma layers to 50 m increments) at grid point closest to CCO (ordinate). Hours circled in red have been identified as suspect objective algorithm tags by inspection of their virtual temperature. Black dashed line shows 0°C = BBH. Correlation coefficients that include and exclude the red points are shown in the lower right in red and blue text, respectively.

[29] Correlations between WRF-RD's 0°C altitude and profiler derived BBHs at 10 additional sites are shown inTable 3. At 9 of the remaining 10 sites, the correlation is above 0.8 (once suspect values are removed), and at 3 sites it is at or above 0.9. Inspection of the scatterplots for each site (not shown) reveals that 0°C altitude is generally higher than BBH, with an additive bias at all sites within the 100–300 m range expected given melt processes. The site with very low correlation, TRK, is 1805 m MSL so its BBH is confined to a relatively narrow altitude range above the surface compared with other sites; additionally, this site was operational for one cool season, so there are few hours with BBH measurements. Together these factors most likely cause the low correlation.

Table 3. Correlations and Additive Bias Between Objectively Tagged BBH and Altitude in WRF-RD Interpolated to 50 m Altitude Bands Closest to 0°C, at All Wind Profiler Locationsa
StationCorrelationAdditive BiasNumber of Points
  • a

    Second correlation given for CCL and CCO is correlation with incorrectly tagged BBHs removed (see section 4a). Also shown is the total number of cool season BBH hits available at each site.

BBY0.94277.8195
CCL0.79/0.91191.2/153.71934
CCO0.78/0.90191.2/155.83483
CCR0.90200.9383
CFC0.82181.3965
GVY0.85138.71332
LHS0.88251.7168
PPB0.84201.3405
SAC0.89211.6484
SHS0.90183.31268
TRK0.18114.659

[30] A similar hour-by-hour comparison of WRF-RD's T2m with observed near-surface temperature was performed at 8 stations (seeTable 1). The correlation coefficients, mean absolute errors (MAE), and additive biases for those comparisons are shown in Table 4. At the 6 interior locations, correlations range from 0.79 to 0.93, indicating good agreement in T2m variability; the coastal locations (PPB and BBY) have lower correlations (0.47 and 0.57), which partly reflects reduced variability of coastal surface temperature. MAE at these 8 locations ranges from 1.5 to 2.45 K, and – perhaps more importantly – at all locations except GVY WRF-RD exhibits a 0.44–1.93 K warm bias at the surface; we will discuss this warm bias and its implications more insection 4.2.

Table 4. Correlation, MAE, and Additive Bias Between Observed Surface Air Temperature and WRF-RD Temperature 2 m AGL, at Locations With Surface Dataa
StationCorrelationMAE (K)Additive Bias (K)Number of Points
  • a

    Locations with surface data are shown in Figure 1. Also shown here is the total number of cool season measurements available at each site.

BBY0.571.911.437230
CCL0.872.431.8132790
CCO0.892.421.6347833
CCR0.792.451.935713
CFC0.892.221.3310358
GVY0.931.50−0.1118515
PPB0.472.131.836841
SHS0.831.850.4417879

4.2. WRF-RD's Winds

[31] The composite WRF-RD SBJ profiles (Figure 3) have reasonable representation of winds, but a fuller picture of hour-to-hour wind variability at multiple locations is necessary to strengthen our confidence in its representation of SBJ dynamics. For that reason, we compare the terrain-perpendicular (U) and terrain-parallel (V) wind components for all hours, irrespective of whether or not SBJs were observed, at 11 locations with long-term wind profiler measurements (Figure 1).

[32] At BBY – the profiler with the least terrain-perturbed (and most ‘maritime’) conditions (Figure 10) – we see good agreement in both components of the wind at all three altitudes shown (1000, 2000, and 3000 m AGL). This good agreement is reflected in high correlation coefficients (0.85 to 0.89), although there is a weak bias in the WRF-RD wind speeds that is more pronounced in V and at higher altitudes in U.

Figure 10.

(a, c, and e) Barrier-perpendicular and (b, d, and f) barrier-parallel wind components in WRF-RD (ordinate) versus those observed (abscissa) at BBY at 1000 m AGL (Figures 10a and 10b), 2000 m AGL (Figures 10c and 10d), and 3000 m AGL (Figures 10e and 10f). Blue dots show values for each hour, red line shows x = y, and green line shows linear fit. Correlation coefficient is noted in upper left of each panel.

[33] Similarly at CCO (Figure 11), we see generally good agreement between WRF-RD and observed winds at all three altitudes, as indicated by fairly high correlation values (0.62 to 0.81). The highest correlation (0.81) is in V at 1000 m AGL – close to the altitude of climatological SBJ Vmax – and the lowest correlation is at the same altitude in U; this low correlation is at least in part due to the wind variation at that altitude being primarily in V, while U exhibits comparatively little variability due to the stagnant character of the cross-barrier flow in SBJ conditions. The weak bias we see at BBY is even more pronounced at CCO, particularly during hours with large V magnitudes. Focusing on V at 1000 m AGL (Figure 11b), we notice that the scatterplot has an ‘elbow’ near 20 m s−1, above which WRF-RD underpredicts V to a larger degree. Since V at 1000 m is close to climatological Vmax altitude SBJ, this bias explains WRF-RD's weaker climatological SBJ (Figure 3d), and would contribute to its tendency to have fewer SBJ events than observed (section 3.1).

Figure 11.

(a, c, and e) Barrier-perpendicular and (b, d, and f) barrier-parallel wind components in WRF-RD (ordinate) versus those observed (abscissa) at CCO at 1000 m AGL (Figures 11a and 11b), 2000 m AGL (Figures 11c and 11d), and 3000 m AGL (Figures 11e and 11f). Blue dots show values for each hour, red dashed line shows x = y, and green dashed line shows linear fit. Correlation coefficient is noted in upper left of each panel.

[34] What is causing WRF-RD's V weak bias during SBJ events? Although elements in the simulation setup could be responsible (e.g., errors in lateral boundary conditions, too coarse horizontal and/or vertical resolution, etc.), we investigate thermodynamic conditions during SBJ events to reveal any systematic biases in the simulation consistent with reduced SBJ strength.Figure 12 shows composite soundings at SHS (146 km southeast of CCO; Figure 1) during 49 soundings launched during 10 SBJ events. WRF-RD has a slight cold, wet bias from ∼400 m–4 km and a slight warm, dry bias in the lowest ∼300 m. Inspection of squared Brunt-Väisälä frequency, N2, for the two composite profiles (Figure 12c; the dry Brunt-Väisälä frequency is used because of the profiles' relatively low relative humidities) reveals that the variation of WRF-RD's temperature bias with height leads to a large underestimation of stability in the lowest ∼700 m. A recent idealized numerical study [Olson and Colle, 2009] found barrier jet strength strongly correlates with atmospheric stability, suggesting WRF-RD's reduced SBJ strength could be due to its underestimate of stability in the Central Valley. This underestimation of stability could be due to overly efficient mixing of WRF-RD's PBL [Kim and Hong, 2009] or possibly insufficient vertical resolution.

Figure 12.

(a) Potential temperature, (b) relative humidity, and (c) composite squared Brunt-Väisälä frequency during the 49 soundings that were launched at SHS during SBJ events (blue line with triangles) and the same variables from WRF-RD at the grid point closest to SHS and the hour closest to the sonde launches (red dashed line with circles). Triangles and circles show the native vertical resolution in the sondes and WRF-RD, respectively.

[35] To examine WRF-RD's representation of U and V at all 11 profilers in the region, we compute correlations between the observed and modeled wind components at each altitude (Figure 13). Nearly all profilers at nearly all altitudes have correlations greater than 0.7 in both components; in fact, most correlations are above 0.8, particularly in the V-component. Curiously, although most profilers have their lowest correlations in the bottom 1 km of the U-component, only one other profiler – CCL, another site that frequently exhibits SBJ conditions – shows a correlation dropoff comparable to that seen at CCO. Inspection of the individual scatterplots at all 11 locations (not shown) reveals that CCO is the only site with a noticeable ‘elbow’ in V near 1 km AGL. This is possibly due to few hours with high V-component speeds at the other locations, since CCO lies closest to WRF-RD's SBJ climatological maximum location (i.e., the location with speeds near 20 m s−1 in Figures 4a and 4b).

Figure 13.

Correlation coefficient between profiler (a) Sierra-perpendicular and (b) Sierra-parallel wind and WRF-RD output at the closest grid point as a function of altitude at 11 locations shown onFigure 1. Temporal coverage of each profiler can be found in Table 1.

[36] In summary, WRF-RD represents wind variability well throughout the region at most altitudes. At two locations – CCO and CCL – where most of the wind variability is in V, WRF-RD struggles with variability in U in the lower troposphere, but its representation of V is quite good. Finally, although WRF-RD's reduced Central Valley stability likely introduces a significant weak bias into WRF-RD's SBJ at CCO (and possibly throughout the SBJ areal maximum), it does not preclude use of WRF-RD for an investigation of SBJ dynamics and impact on precipitation, since it primarily impacts the relatively few hours with extreme along-barrier wind speeds and only moderately impacts the climatology (cf.Figure 11b and Figure 3d).

4.3. WRF-RD's SBJ

[37] Our final assessment of WRF-RD's SBJ representation focuses on SBJ events at CCO on an individual case basis. We begin by matching cases across WRF-RD and the observations temporally, as described insection 3.1, resulting in 150 doubly matched cases. We then construct composite profiles for each case, and use two criteria to determine whether WRF-RD's representation of that case warrant its inclusion in the SBJ catalog. First, we compute the correlation coefficient between WRF-RD and observed case-averaged V-profiles, to constrain each case's V-profile ‘shape’. Cases whose V-profile correlation is less than zero are rejected; inspection of the histogram of V-profile correlations (not shown) reveals a break in the correlation distribution at zero. Second, we calculate the altitude of Vmax in the case-averaged profiles; a comparison of this altitude in WRF-RD and observed wind profiles (Figure 14) reveals general good agreement in Vmax altitude, with a tendency for WRF-RD to underestimate SBJ Vmax altitude for “high” SBJs above 1.5 km. We then compute the absolute difference in Vmax altitude between WRF-RD and observed. Cases with Vmax height differences greater than 750 m are rejected (red circled cases inFigure 14).

Figure 14.

Observed SBJ case-average Vmax altitude (abscissa) and WRF-RD SBJ case-average Vmax altitude at grid point closest to CCO (ordinate). Cases circled in red fall outside the Vmax altitude 750 m difference criterion. Black dashed line shows WRF-RD SBJ Vmax altitude = Observed SBJ Vmax altitude.

[38] Application of these two criteria results in rejection of 23 of the 150 doubly matched cases. Figure 15 shows example profiles from one retained case (Figures 15a and 15b) and one case rejected due to negative V-profile correlation (Figures 15c and 15d). The retained case profiles agree very well, aside from WRF-RD having somewhat weaker westerlies between 3 and 4 km. Many of the retained cases have similar agreement, although some have larger differences in SBJ V-profile magnitude (note that because of WRF-RD's weak bias discussed insection 4.2, we did not base any rejection criteria on Vmax amplitude). The rejected case also shows good agreement in the U-component, but the V-component reveals a secondary Vmax not captured by WRF-RD.

Figure 15.

(a) Terrain perpendicular and (b) terrain parallel composite wind speed versus altitude at CCO (or grid point closest to CCO) in observations (red) and WRF-RD (blue) during an SBJ case that was retained. (c) Terrain perpendicular and (d) terrain parallel composite wind speed versus altitude at CCO (or grid point closest to CCO) in observations (red) and WRF-RD (blue) during an SBJ case that was rejected.

[39] Inspection of each of the 150 doubly matched case composite profiles (not shown) reveals two types of profiles that WRF-RD commonly misrepresents. First, profiles with an observed secondary Vmax are not uncommon – 39 of the 150 doubly matched observed SBJ case composites exhibit a secondary maximum – but the secondary maximum is generally not captured as a true double Vmax in WRF-RD but rather as a broad single maximum as inFigure 15d. Only one case contains a true double Vmax in WRF-RD. These double Vmax cases make up a large percentage of rejected cases: 13 of the 23 rejected cases contain double Vmax in the observations, and are a clear weakness of WRF-RD. Second, all but three of the cases with large Vmax altitude differences have Vmax altitudes higher in observations than in WRF-RD, suggesting WRF-RD has difficulty representing ‘high’ SBJs. Both of these issues possibly stem from WRF-RD's reduced vertical resolution above 1 km.

[40] The retained SBJ case composite U and V profiles (Figure 16) closely resemble the composite profiles for all SBJ cases (Figures 3c and 3d), suggesting we have retained enough cases to adequately represent SBJ climatology. Doubly matching the cases and our rejection criteria have improved WRF-RD's agreement with observations slightly: WRF-RD's composite U-profile now nearly overlays that observed, with a difference of 2.2 m s−1or less throughout the profile depth. WRF-RD's retained SBJ case composite V is approximately 2 m s−1weaker than that observed above 1 km MSL, reflecting WRF-RD's tendency to slightly underestimate SBJ strength. However, WRF-RD's composite V-profile closely agrees with observations in terms of profile shape, with a strong maximum between 700 and 800 m that decreases monotonically by about 10 m s−1 from its maximum to 3.5 km MSL.

Figure 16.

(a) Terrain perpendicular and (b) terrain parallel wind speed versus altitude at CCO (or grid point closest to CCO) in observations (red) and WRF-RD (blue) averaged over all retained SBJ cases.

5. Summary and Discussion

[41] We investigate the climatological representation of the Sierra Barrier Jet in four numerical model products at different resolutions ranging from 2.5 degree to 6 km – two reanalysis data sets, NNRP and NARR, and two downscaled reanalysis data sets, CaRD10 and WRF-RD – and compare this representation with available wind profiler measurements at CCO over an 11 year period. Using an SBJ objective identification tool, we identify SBJ cases in the CCO wind profiler data, as well as in the four numerical model products at the grid point nearest this site (second closest grid point was used in NARR) to assess the spatial resolution necessary to correctly capture this topographically induced low-level jet. The three higher-resolution products – NARR, CaRD10, and WRF-RD – all have somewhat realistic composite SBJ V-profiles and SBJ case counts. WRF-RD's composite V-profile is closest in amplitude and shape to that observed.

[42] Since NARR contains a large proportion of observed SBJs and its composite SBJ V contains a strong – albeit too elevated – Vmax, we then compare WRF-RF and NARR composite winds and water vapor transport throughout California to reveal the additional benefit gleaned from the dynamical downscaling technique for SBJ representation. Plan views and cross sections of these composites reveal that NARR's relatively coarse representation of California's detailed topography introduces suspicious-looking factors: Large grid point-to-grid point discrepancies with seemingly isolated maximum wind speeds, a complete lack of coastal barrier jet, and substantial differences in magnitude and direction of water vapor transport. These factors severely limit NARR's applicability for an investigation of SBJ dynamics and justify our use of the 6-km resolution WRF-RD. We then undertake a comprehensive validation of WRF-RD by comparing snow levels, surface temperature, and winds at a network of 11 locations throughout the region with available wind profiler and surface data. WRF-RD's 0°C isotherm altitude compares well with observed BBH at 10 of the 11 locations, and its near-surface temperature also agrees well with that observed, albeit with a slight warm bias. The correlation between observed and modeled wind at the 11 wind profilers is greater than 0.6 at all altitudes in both components for all but U at 2 of the profilers, and it is between 0.8 and 0.9 for many altitudes and locations. These wind comparisons reveal a slight weak bias in WRF-RD winds during hours with strong winds; this bias is largest at CCO in V at 1000 m – near the composite SBJ Vmax altitude. This weak SBJ bias can be explained at least in part by WRF-RD's Central Valley static stability during SBJ events being not strong enough below ∼700 m, as revealed by a comparison of WRF-RD thermodynamic profiles with rawinsonde data.

[43] A catalog of 150 modeled SBJ events that have significant timing overlap with observed SBJ events is created and used to further assess WRF-RD's representation of the winds during SBJ events. Two criteria based on individual, temporally matched case composite V profiles determine whether each case is retained in a final tally of SBJ cases that WRF-RD accurately depicts: positive correlation between observed and WRF-RD composite V profiles, and Vmax altitude difference less than 750 m. After application of these criteria, we find WRF-RD accurately represents 56 percent of the observed SBJ cases occurring during the last 11 cool seasons, albeit with the previously noted weak bias: Of the 150 doubly matched cases, 127 are deemed ‘accurate’ based on the correlation and Vmax criteria. These 127 doubly matched cases were originally 143 cases prior to case combination, which is 56 percent of the original 256 observed cases.

[44] This study demonstrates that important differences exist between the 6 km resolution dynamically downscaled reanalysis and coarser resolution reanalysis data sets. Comparison with unique wind profiler observations from NOAA's HMT show the advantages of the dynamical downscaling at 6 km. Furthermore, the striking differences between WRF-RD and NARR's water vapor transport – much larger IVTs in WRF-RD and directional differences above and below crest height – likely impact precipitation distribution along the windward slope of the Sierra and coastal ranges and throughout the northern Sacramento Valley. Thus we recommend that future studies requiring accurate representation of the SBJ and other orographically generated processes use downscaling methods demonstrated here.

[45] Its importance to local climate and weather processes – most notably to precipitation distribution – motivate further study of the SBJ. Its representation within standard reanalysis data sets is clearly hampered by their relatively coarse representation of the topography, yet these products are often used to provide meteorological context for air quality studies. Although computationally expensive, dynamical downscaling provides a means to realistically capture meteorological conditions during SBJ events. Using the catalog of SBJ events created during this study, WRF-RD will be used to document kinematic and thermodynamic conditions leading up to and following SBJ events, and to further document the impact SBJs have on precipitation distribution. Any results further documenting the impact the SBJ has on precipitation distribution have the potential to profoundly impact water storage and management decisions for the United States' most populous state.

Acknowledgments

[46] NARR data was retrieved from the National Center for Atmospheric Research's Research Data Archive. NNRP data was provided by the NOAA Earth Systems Research Laboratory Physical Sciences Division, in Boulder, Colorado, USA, from their Web site at http://www.esrl.noaa.gov/psd/. CaRD10 data was provided by Masao Kanamitsu, Dan Cayan, and Mary Tyree at Scripps University. This work was conducted in the context of CalWater's (http://www.esrl.noaa.gov/psd/calwater/) atmospheric river – precipitation research theme hypothesis. Mimi Hughes was also supported by a National Research Council Research Application Program Postdoctoral Fellowship. Thanks go to Allen White for figure generation. We would also like to thank three anonymous reviewers for their comments, and Alice DuVivier for help in editing, both of which enhanced the manuscript.