1. Introduction
Thermally driven topographic flows are multiscale dynamical phenomena that are frequently observed during fair-weather conditions. The involved airmass exchange between mountainous regions and adjacent plains results from flows at the scale of a whole mountain range, the scale of single valleys, and the scale of single slopes (Wagner 1932). For instance, at the meso-α scale [200–2000 km; Orlanski (1975)] plateaulike topography involves large-scale circulations between the adjacent plains and the elevated terrain [e.g., Rocky Mountains; Reiter and Tang (1984)]. At the meso-β scale (20–200 km) thermally driven plain–mountain flows have been observed, for example, in the European Alps (Weissmann et al. 2005; Lugauer and Winkler 2005; Bica et al. 2007) and in the Rocky Mountains (Banta and Cotton 1981; Bossert and Cotton 1994). These flows are forced by hydrostatic pressure gradients that emerge from differential heating (Egger 1990).
At smaller scales—for example, the scale of a single valley—the differential heating is explained by volumetric considerations. During daytime, the same amount of energy provided to a plain and a valley atmosphere involves a stronger heating of the smaller valley volume (Steinacker 1984; Whiteman 1990). During the last 60 years, theoretical concepts (Defant 1949; Vergeiner and Dreiseitl 1987; Egger 1990), observational evidence (e.g., Nickus and Vergeiner 1984; Neininger and Liechti 1984; Broder and Gygax 1985; Hennemuth and Schmidt 1985; Whiteman 1990; Henne et al. 2004; Rotach and Zardi 2007), and numerical simulations (e.g., Chow et al. 2006; Schmidli and Rotunno 2010) allowed for significant advances in the understanding of the valley wind dynamics.
Several previous studies argued that thermally driven flows are important for the formation of cumulus clouds and for the initiation of precipitating convection. Mass convergence induced by the thermally driven flows has been reported to be critical for the triggering of deep convection near mountain crests (e.g., Cotton et al. 1983; Banta 1984; Banta and Schaaf 1987). Moreover, Banta (1984) speculated that convergence at the scale of an entire mountain range may be of importance for convective initiation. This multiscale aspect has also been addressed by Reiter and Tang (1984), who concluded that the frequent occurrence of thunderstorms over the Rockies is linked to the flow convergence induced by the large-scale topographic heat low. A similar link to thermally driven mesoscale flows was proposed to explain the regional accumulation of convective and lightning activity found for some parts of the European Alps (Linder et al. 1999; Finke and Hauf 1996; Lugauer and Winkler 2005). Moreover, thermally driven flows have been reported to supply moisture from low-lying source regions (e.g., basins or adjacent plains) that in turn was found to facilitate the initiation of deep convection (Kuwagata 1997; Kuwagata and Kimura 1997; Gantner et al. 2003; Barthlott et al. 2006; Yin et al. 2011).
Other studies underlined the complexity of the causal link between the flow convergence over mountains and the formation of cumulus clouds and deep convection. For a mountain O(30 km) in size, observations presented by Geerts et al. (2008) and Demko et al. (2009) indicate that the mountain-scale flow convergence is weaker on days with deep convection than on days without it. This result has been confirmed by Demko and Geerts (2010a), who argued that thermally driven flows reduce the convective available potential energy (CAPE) due to advection of colder air.
The diversity of involved processes points to the complexity of the mutual interaction between thermally driven flows and the formation of deep orographic convection. Capturing the different scales of the governing processes appears to be challenging for the numerical representation of such flows. Indeed, subkilometer-scale horizontal grid spacings have commonly been applied to simulate valley winds and boundary layer characteristics within relatively small computational domains (Gantner et al. 2003; Zhong and Fast 2003; Zängl 2004; De Wekker et al. 2005; Chow et al. 2006). As demonstrated by Weissmann et al. (2005), however, a grid spacing of 6.6 km appears to be sufficient to resolve the plain–mountain flow toward the Alps. Even-coarser grid spacings appear to be appropriate to capture the circulations over large-scale plateaus (e.g., the Tibetan Plateau; Zängl et al. 2001).
To simulate convective fluxes and the related surface precipitation, parameterizations are required for grid spacings of larger than a few kilometers. One of the main deficits of such convection-parameterizing models (CPM) is a too-early onset of deep convection (Dai et al. 1999; Bechtold et al. 2004; Brockhaus et al. 2008). Parameterizations of deep convection are not required in cloud-resolving models (CRM; Δx ≃ 1 km) because deep convection is explicitly simulated on the numerical grid. CRMs yield reasonably good agreement with the observed evolution of cloud formation and surface precipitation (e.g., Hohenegger et al. 2008; Demko and Geerts 2010a,b) as long as the dynamical forcing for convective initiation is captured (Richard et al. 2011). Although the subgrid-scale flux contributions still need to be parameterized (e.g., Moeng et al. 2010), the convergence of bulk flow properties of kilometer-scale CRMs was demonstrated in a previous study (Langhans et al. 2012a).
As illustrated above, a multitude of topographic scales are present in complex terrain. Thus, continuous improvement can be expected for simulations of topographic flows with increasing numerical resolution. The performance of CPMs in simulating deep convection could be expected to increase equivalently with increasing resolution. Although some assumptions (e.g., convection in equilibrium with large-scale forcing or a small convective cloud fraction within a grid box) that are made in traditional parameterizations may break down at scales below ~10 km (e.g., Molinari and Dudek 1992; Arakawa 2011), an improved representation of deep convection can nevertheless be expected at such resolutions, since for increasing numerical resolution the convection schemes will respond to increasingly better resolved topographic flows. As argued by Arakawa (2011), in the high-resolution limit convection schemes should ideally become inactive and CPMs should converge toward the solution provided by CRMs.
In this paper, we will test this hypothesis by comparing month-long simulations with a CPM with relatively high numerical resolution (Δx ≃ 6.6 km) with both CRMs and previous CPM simulations that used resolutions of ≃20–50 km (e.g., Bechtold et al. 2004; Brockhaus et al. 2008; Hohenegger et al. 2008). Moreover, this study aims at quantifying the benefits related to a doubling of the horizontal resolution in a CRM: Are thermally driven flows and moist convection better represented on a 1-km grid as compared with simulations on a 2-km grid? This study also seeks a better understanding of the interaction between thermally driven flows and deep orographic convection. To address these research items, a large dataset of observations will be used in a systematic validation of the simulated diurnal cycles of thermally driven flows, their forcing, convective cloud formation, surface precipitation, and the mutual interaction between topographic flows and deep convection.
The paper is organized as follows. The numerical model and the conducted simulations are described in section 2. The observational datasets are introduced in section 3. We present our results in two sections. Section 4 contains results on the evaluation of thermally driven flows. Results on the simulated evolution of cloud formation and convective precipitation and on the link with thermally driven flows are presented in section 5. Section 6 provides a summary and conclusions.
2. Numerical model
a. Model description
The nonhydrostatic Consortium for Small-Scale Modeling model (COSMO), designed for weather and climate forecasts at resolutions ranging from the meso-β to the meso-γ scale (Steppeler et al. 2003; Doms and Förstner 2004), is utilized in this study. The model solves the fully compressible governing equations by applying a split-explicit third-order Runge–Kutta discretization in time (Klemp and Wilhelmson 1978; Wicker and Skamarock 2002). A more detailed presentation of COSMO in convection-permitting mode is given by Baldauf et al. (2011).
The package of physical parameterizations used in this study includes a radiative transfer scheme that is based on the δ-two-stream approach (Ritter and Geleyn 1992), a multilayer land surface scheme (Heise et al. 2003), a single-moment bulk cloud-microphysics parameterization (Reinhardt and Seifert 2006), and a turbulent kinetic energy–based surface transfer and PBL parameterization (Mellor and Yamada 1982; Raschendorfer 2001). Slope and shadowing effects on radiation are considered in our cloud-resolving simulations and are computed by following the method of Müller and Scherer (2005) and Buzzi (2008). A large model domain has been chosen (1000 × 1000 km2) to fully cover the European Alps (see Fig. 1a). In the vertical direction, a pressure-based hybrid coordinate is used, with 46 stretched model levels ranging from the surface to the model’s top at 20 hPa. The centers (i.e., mass points) of the lowest grid boxes are located at about 30 m AGL, and Rayleigh damping is applied in the uppermost levels.
(a) The topography in the modeling domain for CR-1, and representations of the topography in (b) CR-2 and (c) CP-7. (d) Enlargement of the dashed box shown in (a) to indicate the location of the surface observational sites. Bullets and triangles represent measurement sites for standard meteorological parameters and surface radiative fluxes (ASRB), respectively. Topography is shaded every 200 m. Countries and mountain ranges mentioned in the text and the location of Milan (M) are indicated in (a). The dashed line in (c) indicates the region used for an Alpine-scale evaluation.
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
Weak fourth-order quasi-horizontal numerical diffusion is applied here only on velocity variables to ensure numerical stability and to avoid artificial vertical transport of heat and moisture. If applied to potential temperature, mixing along terrain-following levels was found to heat valleys during the night and thereby hampered the modeling of strong pressure gradients and related downvalley flows. Further description and justification of this setup are provided by Langhans et al. (2012b).
In COSMO, the surface roughness lengths arise from two contributions. One part relates to the subgrid-scale variance of the topography, and a second part depends on the type of land cover in a grid box. The latter is provided by the Global Landcover 2000 Database (GLC2000; http://www-gvm.jrc.it/glc2000). Plant characteristics (e.g., leaf area index and root depth) are derived from the dominant land cover within a grid box. Soil types are determined using the Digital Soil Map of the World (http://www.fao.org/nr/land/soils/en/).
Synthetically generated satellite brightness temperatures are diagnosed online from the forward radiative transfer model known as Radiative Transfer for Television and Infrared Observation Satellite Operational Vertical Sounder (RTTOV), which has been coupled to COSMO (Keil et al. 2006). RTTOV was originally developed at the European Centre for Medium-Range Weather Forecasts (ECMWF; Eyre 1991) and has undergone several modifications since then (e.g., Saunders et al. 1999; Chevallier et al. 2001; Matricardi et al. 2001; Chevallier and Kelly 2002). Both clear-sky and cloudy-sky brightness temperatures are computed such that cloudy grid points can easily be identified.
b. Numerical experiments
A month-long time period with pronounced thermally driven orographic flows and deep convection has been selected. July of 2006 has been the subject of several previous studies (MeteoSchweiz 2012; Hohenegger et al. 2008, 2009; Schlemmer et al. 2011; Langhans et al. 2012a,b) because intense deep orographic convection occurred repeatedly over the Alps for several days with strong irradiance and weak synoptic forcing. Only during the beginning (5–8) and end (27–30) of the month is precipitation related to synoptic disturbances. All simulations are initialized at 0000 UTC 1 July 2006 and end at 1800 UTC 31 July 2006 (UTC = Central European summer time − 2 h).
Initial and 6-hourly boundary conditions are provided by the operational analysis of the ECMWF at a horizontal resolution of ~25 km. To ensure a short spinup period for soil moisture, all simulations are initialized with well-balanced soil moisture distributions that were obtained from long-term climate runs with COSMO-CLM [see Jäger et al. 2008; here, CLM indicates the climate version of the Lokalmodell (the predecessor of COSMO), or Limited-Area Model]. Only the period dominated by diurnal convection between 0000 UTC 9 July and 0000 UTC 27 July 2006 will be evaluated in this paper. All mean diurnal cycles presented below relate to this period.
An overview and specifications of the numerical simulations are presented in Table 1. Two cloud-resolving simulations, CR-1 and CR-2, using horizontal resolutions of ~1.1 and ~2.2 km (0.01° and 0.02° on the rotated latitude–longitude grid), respectively, are performed. The parameterization of shallow convection had no noticeable impact on the results. It was thus switched off in CR-1 and CR-2. A convection-parameterizing simulation, CP-7, is conducted with a horizontal resolution of ~6.6 km (0.06°). The corresponding long time steps of the three simulations are 6, 15, and 60 s, respectively. CP-7’s resolution is still better than that for current regional climate models [e.g., the European Commission Ensembles-Based Predictions of Climate Changes and Their Impacts project (“ENSEMBLES”) at ~25 km (Christensen et al. 2010) and the Coordinated Regional Downscaling Experiment (CORDEX) at ~10 km (Giorgi et al. 2009)] but is probably too coarse to resolve deep convection explicitly. Although more complex parameterizations are available, we decided to use the frequently applied Tiedtke mass-flux scheme (Tiedtke 1989) in CP-7, because the applied mass-flux approach still entails the fundamental concept of many of the more recent developments (e.g., Bechtold et al. 2001, 2004). Many other currently (at ~10 km) used parameterizations (e.g., Arakawa and Schubert 1974; Bougeault 1985; Kain and Fritsch 1990) apply similar assumptions (e.g., small fractional coverage of convection in a grid box) and typically reveal similar deficits in simulations of continental summer convection (Bechtold et al. 2004; Guichard et al. 2004; Brockhaus et al. 2008). Convection is triggered if test parcels reach their level of free convection. The mass flux at cloud base is determined by the moisture convergence in the subcloud layer. Although this scheme was not designed for scales that are smaller than ~10 km, experience indicates that a 6.6-km run may still improve the errors commonly found in GCMs and lower-resolution regional climate models. Similar setups are thus applied in operational NWP applications, such as mesoscale ensemble prediction (e.g., Marsigli et al. 2005; Eckel and Mass 2005).
Overview and specifications of the numerical experiments in this paper. The applied horizontal grid spacings (km), time steps (s), and topographic filter cutoffs (km) are listed. The application of topographic correction of radiative fluxes is also indicated.
The Shuttle Radar Topography Mission (http://srtm.csi.cgiar.org) provides a high-resolution topographic dataset that has been truncated to each model grid. To prevent numerical instabilities related to steep terrain, the topography is low-pass filtered with a cutoff at about 4Δx in all three simulations (see Figs. 1a–c). All CR-1 parameters related to soil type, land use, and plant cover were obtained by nearest-neighbor interpolation from CR-2. The roughness lengths utilized in our simulations decrease with increasing numerical resolution, because the variance of the unresolved topography decreases. The effects of terrain on radiative fluxes are considered only in CR-1 and CR-2.
Two additional sensitivity simulations, CR-2-T7 and CR-7, are presented in this paper. CR-2-T7 applies the same setup as CR-2 but uses the topography of CP-7 together with the interpolated coarser-resolution roughness lengths and land-use characteristics of CP-7. The comparison with CR-2 and CP-7 will help to differentiate the effects of topographic and numerical resolution. CR-7 applies the same setup as CP-7 but without using any parameterization for moist convection.
3. Observations
a. Data description
Automatic measurements of standard atmospheric parameters are provided by the national weather services of Austria (Zentralanstalt für Meteorologie und Geodynamik: ZAMG), Germany (Deutscher Wetterdienst: DWD), and Switzerland (MeteoSwiss). Hourly measurements of 10-m winds, 2-m temperature, 2-m humidity, and pressure are obtained from stations located in or near a selection of major Alpine valleys. A list of the primary station sites and the corresponding valleys is given in Table 2, and their locations are indicated in Fig. 1d. Valleys have been selected such that observations are available from at least one station in the valley and from at least one station that is located close to the valley exit or the adjacent foreland. Data from plain–valley pairs will be used to compute the horizontal pressure gradient (see section 3b). Observations from one additional station are evaluated for the Inn, Reuss, Rhein, and Rhone valleys, respectively.
Overview of stations used for the valley wind and pressure gradient analysis. For each valley the last-mentioned station corresponds to the “plain” station or the station at the valley exit ( subscript P). Stations located within valleys (subscript V) are used for evaluating the 10-m valley wind. The pressure gradients have been computed between the plain stations and the corresponding valley station(s). Abbreviations are introduced for station names, and the observed upvalley wind direction α is given for all valley stations.
The model grid points associated with individual stations are determined by minimizing an optimal distance to the station. The latter is defined as a weighted average of the horizontal and vertical distances, with stronger weight given to the vertical distance (Kaufmann 2008). For each valley station, the most frequently observed upvalley and downvalley flow directions have been determined for the periods 1000–1800 UTC (see α in Table 2) and 1800–1000 UTC, respectively. The up- and downvalley wind directions were used to compute the along-valley wind component at each valley station.
Longwave and shortwave radiative fluxes are obtained from three stations of the Alpine Surface Radiation Budget network (ASRB; Marty et al. 2002). The locations of the Cimetta (CIM; 1670 m), Davos (DAV; 1610 m), and SLF-Versuchsfeld (SLF; 2540 m) stations are indicated in Fig. 1d.
A comprehensive dataset for hourly surface precipitation has recently been developed for Switzerland by Wüest et al. (2010). It is based on a daily gridded precipitation dataset that is constructed from gauge measurements with a target resolution of about 2 km. Hourly information obtained from the Swiss radar composite (Germann et al. 2006) is then used to disaggregate the gridded daily analysis.
Brightness temperatures are provided by the Spinning and Enhanced Visible and Infrared Imager (SEVIRI) installed on Meteosat-8 (Schmetz et al. 2002). SEVIRI delivers data with a high spatiotemporal resolution (15 min; 3 × 3 km2 at nadir). Cloudy pixels are identified from the observed temperatures using a modified version of the cloud-detection algorithm described by Khlopenkov and Trishchenko (2007). The modified version has been provided by R. Stöckli.1
In this study, the observed temperatures of the 10.8-μm (T10.8) channel are compared with synthetically generated temperatures from COSMO (see section 2). Three cloud categories are differentiated: low clouds (T10.8 > 0°C), midlevel clouds (−20° ≤ T10.8 ≤ 0°C), and high clouds (T10.8 < −20°C). Observed temperature soundings revealed corresponding altitude ranges at roughly z < 4 km, 4 ≤ z ≤ 7 km, and z > 7 km MSL, respectively. Prior to the analysis, the observed T10.8 temperatures were interpolated to CR-2’s grid.
Examples of observed brightness temperature distributions are briefly described for two common situations: shallow convection at 1000 UTC and deep convection at 1600 UTC (see Figs. 2a,b). Also shown is the difference between visible reflectance and visible background reflectance (Figs. 2c,d). At 1000 UTC (Figs. 2a,c) relatively warm shallow orographic cumuli have formed over the western parts of the Alps (e.g., France, Switzerland, and western Austria); four mature convective clouds with cold cloud-top temperatures are already present in the Black Forest region of Germany and the Vosges region in eastern France (see Fig. 1a). Convective cells also rise into a preexisting upper-level cloud over the eastern Alps. At 1600 UTC (Figs. 2b,d) some shallow cumuli have further evolved to mature convective storms with widespread and cold cloud tops. Particularly in the southwestern Alpine region, convective plumes intrude even into the lower stratosphere (see green pixels in Fig. 2b).
Horizontal distributions of Meteosat-8 (a),(b) brightness temperatures T10.8 (K) and (c),(d) difference between visible reflectance and its background for (left) 1000 UTC and (right) 1600 UTC 12 Jul 2006. The subregion shown is indicated by the solid box in Fig. 1a. Topography is contoured at 1000 m. Green pixels in (b) show locations of overshooting tops detected from large brightness temperature differences (>2 K) between the 6.7-μm water vapor channel and the 10.8-μm infrared window channel (Schmetz et al. 1997). Green pixels in (c) and (d) indicate missing values. Black arrows in (b) and (d) point at some of the high clouds appearing with relatively warm brightness temperatures T10.8.
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
The examples given also illustrate that optically thin clouds may appear with relatively warm cloud tops at some pixels (see black arrows in Figs. 2b and 2d). For transparent clouds the measured infrared emission is influenced by the emission from lower-lying clouds and/or the warm surface. Thus, some pixels of the observed and the modeled brightness temperatures will be classified as low- or midlevel clouds, although they are actually located at higher altitudes. Such situations may occur, for example, during the dissipative stage of a thunderstorm or at the margin of a spreading cirrus anvil.
b. Observed horizontal pressure gradient
Plain–valley station pairs are used to evaluate the observed horizontal pressure gradient forcing. The computation of such gradients for stations at different elevations is nontrivial. In addition to the frequently applied method of reducing the pressure hydrostatically to a common level prior to evaluating the dynamically relevant horizontal gradients (e.g., Reiter and Tang 1984; Pauley 1998; Lugauer and Winkler 2005), alternative methods have been proposed.
Geerts et al. (2008) proposed to split the station pressure into a 24-h running mean and a perturbation 
Another alternative has been proposed by Nickus and Vergeiner (1984, hereinafter NV84), who used smooth analytical virtual temperature profiles to reduce the pressure measured at two stations to a common level. Their fictitious boundary layer temperature profiles mimic a smooth stability-dependent transition from a stably stratified surface layer during night or a superadiabatic surface layer during day toward a prescribed upper-level temperature gradient.
Relative to the “traditional” pressure reduction using the assumption of a fixed lapse rate and relative to Geerts et al.’s method, an evaluation of radiosonde soundings and modeled thermodynamic profiles (both not shown) revealed a considerably improved computation of the horizontal pressure gradient if NV84’s profiles were used in combination with information about the PBL’s state retrieved from CR-1. The advantage of this method originates from considering surface-layer lapse rates, which can considerably influence the layer-mean temperature. A detailed description of the applied method and a comparison with the other two methods is provided in Langhans (2012). The station pairs for which pressure gradients have been evaluated are indicated in Table 2. For four of the eight plain stations that were considered, the gradient was computed with respect to two different valley stations, such that in total the horizontal pressure gradient was evaluated for 12 plain–valley station pairs.
4. Thermally driven flows
Station measurements are explored in this section to validate thermally driven flows and the pressure-gradient forcing as simulated by CR-1, CR-2, and CP-7. First, the diurnal evolution and frequency distributions of valley winds are studied. Second, the pressure-gradient force is evaluated from station pairs and by analyzing the simulated pressure distributions. Third, the radiation budget at the surface, responsible for most of the warming or cooling of the boundary layer, is validated.
a. Evolution of the along-valley wind
The 10-m along-valley flow as simulated by the three simulations is compared with observations by analyzing mean diurnal cycles for the valley stations that are listed in Table 3. Figure 3 shows the observed and simulated along-valley wind component as an average over all valley stations. Also shown is the absolute wind speed. The observations show that the averaged upvalley wind sets in at 0800 UTC, peaks at 4 m s−1 around 1300 UTC, and turns to a downvalley wind at 1900 UTC.
Overview of the daytime (1000–1800 UTC) wind statistics for the three simulations and for each valley station. RMS errors and biases are computed from the mean diurnal cycles during the evaluation period 0000 UTC 9 Jul and 0000 UTC 27 Jul 2006 for the three parameters (upvalley wind component, wind speed, and wind direction). The last row shows wind speed statistics obtained from the station-averaged diurnal cycle (see Fig. 3).
Mean diurnal cycle of along-valley wind component (solid) and absolute wind speed (dashed) from observations (black) and simulations (gray) (a) CR-1, (b) CR-2, and (c) CP-7. Shown are averages over all valley stations listed in Table 3. The along-valley velocity component has been determined as the most frequent valley-wind direction observed during up- and downvalley wind regimes. Arrows on the x axis indicate the times of sunrise (SR) and sunset (SS).
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
All simulated station-averaged mean diurnal cycles reveal a daytime upvalley flow and a reversal to nighttime downvalley flow (Figs. 3a–c). Differences with the observations and among simulations exist, however. First, all three simulations exhibit an upvalley wind that is too weak. An improvement to stronger upvalley winds is obtained with increasing numerical resolution, with CR-1 outperforming CR-2 and, in particular, CP-7. Second, the onset and ending of upvalley flow are captured very well by the CRMs, but CP-7 delays the ending by about 2 h. Third, the downvalley winds are slightly too weak—in particular, in CP-7.
For the 12 stations considered, the RMSE and the bias of the upvalley wind decrease with increasing resolution at seven and five stations, respectively. The improvements with resolution are also reflected in the RMSE and the bias of the station-averaged diurnal cycle (see Table 3). The improvements result both from increased wind speeds (e.g., at PIO, CHU, GAR, MAG, and SIO) and from improved wind directions (e.g., at OBE, MAG, and GAR). This behavior is exemplified in Figs. 4a and 4b, which show the valley wind at CHU and SIO, respectively. Overall, the RMSE and bias of the CRMs at these stations are comparable to the results obtained in previous numerical studies of valley winds (e.g., Chow et al. 2006).
Mean diurnal cycle of the valley-wind component (m s−1) computed for the 18-day evaluation period at (a) CHU, (b) SIO, and (c) ALT (see Table 2 for station identifiers). Observations are in black, and simulations are in color (see legend).
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
At other stations, the RMSEs or biases do not decrease continuously with increasing resolution. For instance, a positive wind speed bias decreases with coarser resolution at AIG and KUF, such that CR-1 does not yield the best agreement with observations there. Conversely, CR-1 generates a less pronounced upvalley flow than do CR-2 and CP-7 at ALT and thus results in less agreement with the observations (see Table 3 and Fig. 4c). Closer analysis of the simulated flow field around ALT revealed large horizontal gradients in wind speed that complicate a successful validation at ALT as compared with other stations. In addition, the proximity of buildings to ALT casts some doubt on the representativeness of the observations. In general, little information on the stations’ surroundings was available, and therefore the representativeness of those point measurements remains uncertain. CR-1’s inability to capture stronger winds at ALT is nevertheless surprising.
b. Frequency distributions
The simulated along-valley 10-m wind at single stations is further analyzed in this section to better understand the influence of numerical resolution on the model skill described in the previous section. To this end, frequency distributions of wind direction and wind speed (see Fig. 5) are computed for the upvalley wind period (1000–1800 UTC).
Matrix of observed and simulated frequency distributions of daytime near-surface wind at valley stations for the evaluation period between 1000 and 1800 UTC. Bin width is 22.5°, and distributions of wind speed (m s−1) are given for each direction (see color bar). An average speed is indicated for frequent (≥10%) wind directions. The frequency scale is indicated for INT simulated by CR-1.
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
Figure 5 affirms the improved model skill with increasing resolution as found for many stations (see Table 3). The differences between CR-2 and CR-1 are not explained by one common deficit in CR-2, but may result from improvements in wind speed (e.g., at CHU, INT, and PIO), from improvements in wind direction (e.g., at OBE), or often from improvements in both (e.g., at GAR, MAG, and SIO). CP-7 not only underestimates the wind speed but also results in a considerable offset in wind direction. In summary, individual deficits can be identified for each simulation and for each station, but an improvement in model skill with increasing resolution is found at many stations.
c. Evolution of the horizontal pressure gradient
Although the numerical resolution clearly affects the representation of flows at the scale of single valleys, additional analysis showed that the plain–mountain flow (Alpine pumping; see Lugauer and Winkler 2005; Weissmann et al. 2005) is only slightly weaker in CP-7 than in the CRMs. Thus, at small scales the underlying pressure-gradient forces appear to differ considerably, as valley winds are much stronger in the CRMs. The large-scale forcing of the plain–mountain flow, however, appears to be very similar among the simulations. To evaluate the differences in the dynamical forcing, first the observed horizontal pressure gradients are computed from plain–valley station pairs as described in section 3b. The simulated gradients between plains and valleys are obtained by interpolating the three-dimensional pressure field to the level of the grid point associated with the higher-elevation station. Figure 6 shows the mean diurnal cycle of the obtained horizontal pressure gradient as an average over all 12 plain–valley pairs (see Table 2). The gradients have been decomposed into gradients of 
Mean diurnal cycle of the horizontal pressure gradient averaged over all 12 station pairs listed in Table 2. Shown are gradients of perturbation pressure p′ (solid) and mean pressure 
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
The observed evolution of the perturbation gradient reveals the pressure wave associated with a stronger daytime heating and nighttime cooling of the valley atmosphere. It reaches −2.2 hPa (100 km)−1 at 1300 UTC concurrent with the strongest valley winds. This observed peak is comparable to the 5-yr climatological value of pressure gradients presented by Lugauer and Winkler (2005). The perturbation gradient forcing starts at 0800 UTC and ends at 1900 UTC and thus explains the simultaneous onset and ending of the observed mean upvalley wind (see Fig. 3).
The perturbation gradients generated by the models reveal an evolution that is qualitatively similar to that observed. The models, however, exhibit amplitudes that are too small during both the day- and nighttime. Of interest is that the perturbation gradient generated by CP-7 is only slightly smaller than for the CRMs and explains the relatively small difference in the magnitude of the simulated plain–mountain flows, as mentioned previously. Note that the simulated gradients in 
In summary, this analysis shows that the plain–valley pressure-gradient forcing generated by the CRMs appears to be only slightly enhanced relative to CP-7. Further analysis of the dynamical forcing is provided in the next section.
d. Simulated horizontal pressure distributions
Horizontal pressure distributions (hPa) of (a),(d) 
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
In Switzerland and western Austria, the horizontal gradients in 
e. Surface energy budget
The dynamical forcing described above arises primarily from solar energy input at the ground, which in turn provides sensible heat to the lower atmosphere. To prevent linking different dynamical forcings to possible differences in the energy supply at the surface, a logical way to complete the evaluation of the valley wind forcing is to also validate the surface energy balance of the models. Although the surface radiation budget has been analyzed at three stations, results from only one station will be presented here because the emerging conclusions are independent of the station. Observed and modeled radiative fluxes are presented for CIM (Fig. 8). The net radiative energy supply Q* at the surface results from the downward- and upward-directed shortwave and longwave fluxes (i.e., SW↓, SW↑, LW↓, and LW↑).
Mean diurnal cycles of the surface energy budget components. Shown are observed and modeled (a) surface radiative fluxes for CIM and (b) net surface radiation at CIM and simulated SW↓ averaged over the Alpine region (indicated in Fig. 1c). Fluxes directed toward the surface are defined as being positive.
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
Figure 8a demonstrates the models’ capability to capture the diurnal evolution of the radiative flux components. Small differences are found among the CRMs, and CP-7 results in smaller SW↓ and slightly larger LW↓ during the day. Relative to observations, all simulations underestimate SW↓, the reflected SW↑, and LW↑. Net radiation is captured well in the CRMs but is underestimated by CP-7 (Fig. 8b). A decreased incoming radiation in CP-7 is detected also from Alpine-scale averages of SW↓ (Fig. 8b). Thus, the slightly weaker plain–mountain pressure-gradient forcing found in CP-7 (Fig. 6) might result not only from the coarser topography but could also result from a reduced energy supply. Further analysis of this aspect will be provided in section 5c.
5. Cloud formation and precipitation
The described initiation of dry mesoscale flows within the boundary layer is commonly regarded as a prerequisite for the subsequent formation of cumulus clouds and surface precipitation. This section presents the observed and simulated evolution of clouds and precipitation. It is also of interest to enhance our knowledge of the observed and modeled interaction between boundary layer convergence and surface precipitation.
a. Cloud formation
The cloud cover is presented for a larger Alpine region to study the evolution from shallow to deep convective orographic clouds in more detail. We begin with an analysis of the simulated cloud patterns on one particular day. Figure 9 shows synthetically generated infrared temperatures from the simulations at 1000 and 1600 UTC 12 July 2006. The corresponding observed temperatures are shown in Fig. 2 and have been described in section 3.
As in Figs. 2a and 2b, but from COSMO simulations (a),(b) CR-1, (c),(d) CR-2, and (e),(f) CP-7 at (left) 1000 and (right) 1600 UTC 12 Jul 2006. Synthetic brightness temperatures of COSMO have been computed online as described in section 2. As in Fig. 2b, green pixels indicate locations of overshooting tops.
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
In agreement with observations, the CRMs produce several growing cumulus clouds at 1000 UTC over the western Alps (Figs. 9a,c). Qualitatively similar to observations, the CRMs even predict isolated mature convective storms with cold cloud tops over the Black Forest. In contrast, CP-7 results in several very broad mature convective clouds over both the Black Forest and the southwestern Alps, and as a consequence the high cloud cover is considerably overestimated (Fig. 9e). All runs result in cloud-top temperatures for the cloud system over the eastern Alps that are too cold. This issue also appeared frequently for high clouds during the night.
Later, at 1600 UTC, cumulus clouds have evolved to deep mature cumulonimbus clouds in all simulations (Figs. 9b,d,f). High cirrus clouds have already spread radially from isolated deep convective updrafts. The latter frequently penetrate through the tropopause in the CRMs (see green pixels in Figs. 9b,d). A comparison with Fig. 2b shows good qualitative agreement with the observed temperatures. Again, CP-7 produces a fraction of high clouds that is too large.
To provide a more quantitative and a more systematic insight into the observed and modeled cloud evolution, we computed frequency distributions for a large Alpine region (see Fig. 1c) and analyzed the resulting mean diurnal cycles for our 18-day evaluation period. The fraction covered by each cloud category (see section 3) is visualized together with the mean diurnal cycle of surface precipitation in Fig. 10. The observations (Fig. 10a) show a highly correlated diurnal evolution of high cloud cover and surface precipitation. Hardly any high clouds are observed during the morning, a peak of 27% is reached during the afternoon, and afterward the high cloud cover decreases rapidly. Low, midlevel, and high clouds start to form around 0800, 0900, and 1100 UTC, respectively (see symbols in Fig. 10a). The total cloud cover evolves from 20% in the morning to 65% in the afternoon and decays again during the night.
Mean diurnal cycles of cloud cover and surface precipitation obtained from (a) Meteosat-8, (b) CR-1, (c) CP-7, and (d) CR-2. The cloud fraction has been computed for the Alpine region indicated in Fig. 1c. High clouds (light gray), midlevel clouds (medium gray), and low-level clouds (dark gray) are defined by brightness temperature ranges in section 3. Thin (thick) solid black lines show modeled (observed) precipitation for Switzerland, and dashed lines represent modeled precipitation for the Alpine region. Stars indicate the time of initial growth for each of the three cloud layers, defined here as the first occurrence of a fractional increase of larger than 1% (30 min)−1.
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
The mean diurnal cycles of both the CRMs (Figs. 10b,d) and CP-7 (Fig. 10c) reveal the same qualitative features as are observed. A deficit that affects all runs—in particular, CP-7—is a too-large percentage of high cloud cover. The total cloudiness is overestimated—in particular, during the night when few high clouds are observed. Closer analysis revealed significant amounts of ice and snow content that persisted in the upper troposphere throughout the night. Idealized simulations using COSMO in a very similar setup and with the same bulk microphysics parameterization confirm a likely overestimated lifetime of ice clouds (Schlemmer 2011).
The evolution of total cloud cover and partial cloud cover differs little between CR-1 and CR-2. The latter show a too-short transition period from shallow to deep convective clouds, as low clouds form around 0930 UTC and midlevel and high clouds form around 1000 UTC (Figs. 10b,d). In CP-7, all three cloud types begin to form at ~0930 UTC (Fig. 10c) and the peak cloudiness is reached too early at 1600 UTC.
These findings explain two unresolved aspects from the previous analysis. First, the too-early and too-excessive peak of the cloud fraction of CP-7 explains its decreased shortwave incoming radiation around noon (see Fig. 8). Second, a comparison with radiosoundings (not shown) revealed too-large humidities in the upper troposphere for all runs. This result can be attributed to the overestimated fraction of high clouds. An explanation for the too-long cloud lifetime is still missing but could be related to the sedimentation of ice crystals, which is a process that is not considered in the applied bulk microphysics scheme.
b. Surface precipitation
Figure 11a allows for a quantitative comparison of the mean diurnal cycles of observed and simulated precipitation in Switzerland. The observations reveal an onset of surface precipitation around 0900 UTC and a peak in rainfall at 1800 UTC.
Mean diurnal cycles of surface precipitation averaged over (a) Switzerland and (b),(c) the Alpine region (indicated in Fig. 1c). Dashed curves in (b) and (c) show the simulated mean diurnal cycle of convergence CON (s−1) for the Alpine region. The black curve in (a) shows observed rain rates. Sensitivity experiments are presented in (c).
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
The CRMs result in good overall agreement with the observations: the onset, the timing and magnitude of the peak value, and the nighttime rain rates are captured well. The differences between CR-1 and CR-2 are small, and the validation against observations is excellent, in particular when considering that the observations systematically underestimate precipitation. The observations neglect systematic rain gauge biases, and a station network is used that lacks appropriate coverage of high-altitude regions. In contrast, the precipitation peak in CP-7 occurs 4 h too early. This deficiency has been reported frequently in the literature (e.g., Bechtold et al. 2004; Brockhaus et al. 2008). Moreover, CP-7 significantly overestimates precipitation events during the night. The same aspects and discrepancies are reflected in precipitation at a larger Alpine scale (Figs. 10b–d and 11b).
The observed and simulated spatial distributions of accumulated precipitation in Switzerland are shown in Fig. 12. Most of the observed precipitation is focused over elevated regions while little precipitation is measured over the flat Swiss plateau (between the Jura and the main Alps; see Fig. 1a). The CRMs capture this distribution well, with most of the precipitation over elevated terrain and little over the flat regions. CP-7 also produces widespread precipitation over flat regions but shows less overestimation of precipitation than the CRMs do for elevated regions. The RMSE, biases, and mean absolute errors are shown in Table 4. CR-1 performs best but has a wet bias since the daytime precipitation over terrain is overestimated (see also Fig. 11a). This aspect worsens in CR-2, which thus shows a larger bias, RMSE, and mean absolute error than does CP-7. As mentioned above, however, the observed rain rates are also underestimating precipitation over terrain and are available only to a very limited region of the modeling domain. This uncertainty likely limits this quantitative comparison.
Accumulated precipitation (mm) in Switzerland during the 18-day evaluation period from (a) observations, (b) CR-1, (c) CP-7, and (d) CR-2. Dotted areas indicate regions with missing observational data.
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
RMSEs, MAEs, and biases of simulated distributions of accumulated precipitation in Switzerland for simulations CR-1, CR-2, and CP-7 for the 18-day evaluation period between 0000 UTC 9 Jul and 0000 UTC 27 Jul 2006.
c. The link between thermally driven flows and deep convection
The mean diurnal evolution of CON is depicted for all three simulations in Fig. 11b. All runs yield a peak of the flow convergence at approximately midday, considerably before the peak rain rates are observed (see Fig. 11a). Differences among the CRMs are very small and, in line with the aforementioned similarity in the magnitude of the plain–mountain flows, CP-7 generates an only slightly smaller peak convergence.
To identify the influence of the numerical resolution and to isolate the role of the convection parameterization, we conducted one additional cloud-resolving experiment (CR-2-T7) that uses the topography of CP-7 (as described in section 2). The resulting flow convergence is surprisingly similar to that of CR-2 (Fig. 11c). It is apparent that the Alpine-scale convergence is not governed by the small-scale topography, and neither is the convective precipitation in the Alpine region (Fig. 11c). Thus, rather than unresolved topography, the convection scheme itself is responsible for the too-early precipitation peak in CP-7. Indeed, the underestimation of the flow convergence in CP-7 is partly explained by a too-large fractional cloud cover (Figs. 8b and 10b).
These findings are supported by simulation CR-7, which results in an evolution of the mountain-scale mass convergence that is very similar to that of CR-2-T7 and CR-2, despite the switched-off convection scheme (see Fig. 11c). Although deep convection is underresolved in CR-7, the smaller nighttime precipitation rates and the timing of the daytime precipitation agree much better with CR-2 and the observations than with CP-7.
To analyze the relation between convective rainfall and thermally driven flows on a daily time scale, we further calculated the correlation of daily precipitation and the along-valley wind component. To estimate the forced lifting induced by the upvalley flow, we evaluated the time-integrated lateral inflow into the valleys for the average valley wind of all Swiss stations. If incompressibility and zero mass flux at valley endings are assumed, these integrals provide an estimate for the total PBL lifting provided during the upvalley-flow period. The provided lifting (only phases of upvalley flow are considered) has been evaluated for each day and for three distinct periods: the full day, before 1600 UTC, and before 1200 UTC. Then, the corresponding correlation coefficients with daily accumulated precipitation r24, r16, and r12 have been evaluated.
Figure 13 shows observed and simulated time series of rain rates and station-averaged along-valley winds in Switzerland. Weak upvalley flow is observed on days with large rain rates, and enhanced upvalley flow occurs on days with little precipitation (Fig. 13a). Indeed, the observations reveal a moderate anticorrelation r24 of 0.67 (see also Fig. 13a). Note that relatively short periods of upvalley flow are observed on days with significant rainfall and thus could partly explain the negative correlation. The correlations of the integrated lifting provided before 1200 and before 1600 UTC are also negative (r12 = −0.37 and r16 = −0.61), however. Thus, even the upvalley flow preceding convective precipitation is relatively weak on days with large rain rates. A closer analysis of atmospheric soundings during this period (not shown) pointed out that heavy rain rates are tied to days with enhanced atmospheric instability (i.e., larger CAPE).
Time series of surface precipitation in Switzerland (black) and mean along-valley wind in Switzerland (gray) for (a) observations, (b) CR-1, (c) CR-2, and (d) CP-7. The near-surface along-valley wind has been averaged over all Swiss valley stations. The upvalley wind periods are shaded in gray. The correlation between daily accumulated precipitation (0000 UTC–0000 UTC next day) and the integrated mass flux for three different periods (0000–0000 next day, 0000–1600, and 0000–1200 UTC) is indicated (see text for further explanation).
Citation: Journal of Applied Meteorology and Climatology 52, 6; 10.1175/JAMC-D-12-0167.1
The correlations produced by the CRMs (Figs. 13b,c) reveal excellent agreement with the observations. The abrupt onset of downvalley flows following heavy rainfall is captured well by both CR-1 and CR-2 (e.g., on 13 and 22 July 2006) but not by CP-7. The latter seemingly fails in simulating the dynamical response to convective precipitation, as upvalley winds persist during periods of convective precipitation and the anticorrelation r24 is too small (Fig. 13d). The anticorrelations r12 and r16 are too large in CP-7 because of the pronounced overestimation of cloud cover on convective days (see Fig. 10b) and the corresponding reduction of incoming radiation (see Fig. 8).
6. Summary and conclusions
Long-term simulations of cloud-resolving simulations (Δx ≃ 2.2 and 1.1 km) and convection-parameterizing simulations (Δx ≃ 6.6) have been systematically validated against observations. This study focused on thermally driven flows, their pressure gradient forcing, cumulus formation, and deep orographic convection over the European Alps. To this end, an 18-day high pressure summer episode with weak synoptic forcing, intense solar heating, and a frequent initiation of deep orographic convection has been selected.
The characteristic diurnal sequence of processes can be summarized as follows. A pronounced diurnal cycle of the pressure gradient between the valleys and the plains forces a mesoscale plain–mountain flow and upvalley winds in the major Alpine valleys. On average, the upvalley flow sets in at 0800 UTC, peaks at 1300 UTC, and turns to a downvalley flow at 1900 UTC. First, shallow cumuli form over the Alps at 0800 UTC, followed by the formation of midlevel and high clouds at 0900 and 1100 UTC, respectively. Convective precipitation sets in at 0900 UTC, peaks late in the afternoon at 1800 UTC, and decays rapidly until 2300 UTC.
The evaluation of the cloud-resolving simulations using grid spacings of 1.1 and 2.2 km, respectively, showed the following results:
The diurnal evolution of the thermally driven flows, cumulus formation, and surface precipitation is very well captured. Overall, doubling the horizontal resolution had relatively little influence with the exception of improvements related to the representation of valley winds.
Shortcomings are related to too-weak along-valley pressure gradients, too-weak upvalley flows, and a too-short transition period from shallow to deep convective clouds.
The convection-parameterizing simulation using a grid spacing of 6.6 km and the Tiedtke mass-flux scheme showed substantially worse agreement with observations than did the CRMs. This simulation and further sensitivity experiments revealed the following results:
The activation of the convection scheme results in a too-early precipitation peak relative to observations and the CRMs. This discrepancy can be attributed to the behavior of the convection parameterization rather than to the underresolved small-scale topography. The latter was found to have little influence on the timing and magnitude of precipitation rates. Indeed, a 6.6-km run without a convection scheme improved the timing of the precipitation considerably.
The mesoscale plain–mountain flow can in principle be represented at such “coarse” resolution. Best results were obtained when the convection scheme was turned off. If the convection scheme is applied, the cloud cover was too early and too dense, which was found to shield the incoming shortwave radiation and to weaken the Alpine-scale flow convergence.
Our findings demonstrate very good agreement between CRMs and observations and an insensitivity of CRMs to horizontal grid spacing. We thus encourage long-term simulations at CRM resolutions. CPMs with a resolution below 10 km could potentially capture the Alpine-scale mass convergence to a very good degree. The applied convection scheme results in considerable disagreement with observations and CRMs in terms of cloud cover and precipitation, however. Despite considerably increased numerical resolution, the CPM thus suffers from deficits that are similar to those that are well known from coarser-resolution regional- and global-scale simulations.
For the link between mesoscale flow convergence and moist convection, both observational data and simulations suggest that intense convective rainfall is not directly related to enhanced mountain-scale mass convergence but rather is related to a stronger atmospheric instability. Nevertheless, it remains an open question as to what degree the mesoscale flows influence the atmospheric stability over the mountain range and to what degree deep convection is controlled by the large-scale moisture supply. An idealized modeling framework could be helpful to address this question and other factors such as the role of mountain geometry, moisture supply, and synoptic conditions.
Acknowledgments
The Center for Climate Systems Modeling (C2SM) at ETH Zurich is acknowledged for providing financial, technical, and scientific support. This work was partially funded by the Swiss National Science Foundation (Grant 200021_132614). Observational data were provided by the following national weather services: DWD, MeteoSwiss, and ZAMG. We thank Evelyne Richard and Reto Stöckli for constructive advice and the Swiss National Supercomputing Centre (CSCS) for providing access to its high-performance compute clusters; access to the COSMO model was kindly provided by the COSMO consortium. We also acknowledge constructive comments by two anonymous referees.
REFERENCES
Arakawa, A., 2011: Multiscale modeling of the moist-convective atmosphere—A review. Atmos. Res., 102, 263–285.
Arakawa, A., and W. H. Schubert, 1974: Interaction of a cumulus cloud ensemble with the large-scale environment. J. Atmos. Sci., 31, 674–701.
Baldauf, M., A. Seifert, J. Förstner, D. Majewski, M. Raschendorfer, and T. Reinhardt, 2011: Operational convective-scale numerical weather prediction with the COSMO model: Description and sensitivities. Mon. Wea. Rev., 139, 3887–3905.
Banta, R. M., 1984: Daytime boundary-layer evolution over mountainous terrain. Part I: Observations of the dry circulations. Mon. Wea. Rev., 112, 340–356.
Banta, R. M., and W. R. Cotton, 1981: An analysis of the structure of local wind systems in a broad mountain basin. J. Appl. Meteor., 20, 1255–1266.
Banta, R. M., and C. B. Schaaf, 1987: Thunderstorm genesis zones in the Colorado Rocky Mountains as determined by traceback of geosynchronous satellite images. Mon. Wea. Rev., 115, 463–476.
Barthlott, C., U. Corsmeier, C. Meißner, F. Braun, and C. Kottmeier, 2006: The influence of mesoscale circulation systems on triggering convective cells over complex terrain. Atmos. Res., 81, 150–175.
Bechtold, P., E. Bazile, F. Guichard, P. Mascart, and E. Richard, 2001: A mass-flux convection scheme for regional and global models. Quart. J. Roy. Meteor. Soc., 127, 869–886.
Bechtold, P., J.-P. Chaboureau, A. Beljaars, A. K. Betts, M. Köhler, M. Miller, and J.-L. Redelsperger, 2004: The simulation of the diurnal cycle of convective precipitation over land in a global model. Quart. J. Roy. Meteor. Soc., 130, 3119–3137.
Bica, B., and Coauthors, 2007: Thermally and dynamically induced pressure features over complex terrain from high-resolution analyses. J. Appl. Meteor. Climatol., 46, 50–65.
Bossert, J. E., and W. R. Cotton, 1994: Regional-scale flows in mountainous terrain. Part I: A numerical and observational comparison. Mon. Wea. Rev., 122, 1449–1471.
Bougeault, P., 1985: A simple parameterization of the large-scale effects of cumulus convection. Mon. Wea. Rev., 113, 2108–2121.
Brockhaus, P., D. Lüthi, and C. Schär, 2008: Aspects of the diurnal cycle in a regional climate model. Meteor. Z., 17, 433–443.
Broder, B., and H. A. Gygax, 1985: Terrain-induced effects on the ozone, temperature, and water vapor daily variation in the upper part of the PBL over hilly terrain. Tellus, 37B, 259–271.
Buzzi, M., 2008: Challenges in operational numerical weather prediction at high resolution in complex terrain. Ph.D. thesis, ETH Zurich, 187 pp.
Cassano, J. J., and T. R. Parish, 2000: An analysis of the nonhydrostatic dynamics in numerically simulated Antarctic katabatic flows. J. Atmos. Sci., 57, 891–898.
Chevallier, F., and G. Kelly, 2002: Model clouds as seen from space: Comparison with geostationary imagery in the 11-μm window channel. Mon. Wea. Rev., 130, 712–722.
Chevallier, F., P. Bauer, G. Kelly, C. Jakob, and T. McNally, 2001: Model clouds over oceans as seen from space: Comparison with HIRS/2 and MSU radiances. J. Climatol., 14, 4216–4229.
Chow, F. K., A. P. Weigel, R. L. Street, M. W. Rotach, and M. Xue, 2006: High-resolution large-eddy simulations of flow in a steep alpine valley. Part I: Methodology, verification, and sensitivity experiments. J. Appl. Meteor. Climatol., 45, 63–86.
Christensen, J. H., E. Kjellström, F. Giorgi, G. Lenderink, and M. Rummukainen, 2010: Weight assignments in regional climate models. Climate Res., 44, 179–194.
Cotton, W. R., L. G. Raymond, P. J. Wetzel, and R. L. McAnelly, 1983: A long-lived mesoscale convective complex. Part I: The mountain-generated component. Mon. Wea. Rev., 111, 1893–1918.
Cotton, W. R., G. H. Bryan, and S. C. van den Heever, 2011: Storm and Cloud Dynamics. Academic Press, 809 pp.
Dai, A., F. Giorgi, and K. E. Trenberth, 1999: Observed and model-simulated diurnal cycles of precipitation over the contiguous United States. J. Geophys. Res., 104, 6377–6402.
Dalu, G. A., M. Baldi, R. A. Pielke, and G. Leoncini, 2003: Mesoscale nonhydrostatic and hydrostatic pressure gradient forces—Theory. J. Atmos. Sci., 60, 2249–2266.
Defant, F., 1949: Zur Theorie der Hangwinde, nebst Bemerkungen zur Theorie der Berg- und Talwinde (On the theory of slope winds, together with observations on the theory of mountain and valley winds). Arch. Meteor. Geophys. Bioklimatol., 1, 421–450.
Demko, J. C., and B. Geerts, 2010a: A numerical study of the evolving convective boundary layer and orographic circulation around the Santa Catalina Mountains in Arizona. Part I: Circulation without deep convection. Mon. Wea. Rev., 138, 1902–1922.
Demko, J. C., and B. Geerts, 2010b: A numerical study of the evolving convective boundary layer and orographic circulation around the Santa Catalina Mountains in Arizona. Part II: Interaction with deep convection. Mon. Wea. Rev., 138, 3603–3622.
Demko, J. C., B. Geerts, Q. Miao, and J. A. Zehnder, 2009: Boundary layer energy transport and cumulus development over a heated mountain: An observational study. Mon. Wea. Rev., 137, 447–468.
De Wekker, S., D. G. Steyn, J. D. Fast, M. W. Rotach, and S. Zhong, 2005: The performance of RAMS in representing the convective boundary layer structure in a very steep valley. Environ. Fluid Mech., 5, 35–62.
Doms, G., and J. Förstner, 2004: Development of a kilometer-scale NWP-system: LMK. COSMO Newsletter, No. 4, 159–167. [Available online at http://www.cosmo-model.org/content/model/documentation/newsLetters/newsLetter04/chp9-5.pdf.]
Eckel, F. A., and C. F. Mass, 2005: Aspects of effective mesoscale short-range ensemble forecasting. Wea. Forecasting, 20, 328–350.
Egger, J., 1990: Thermally forced flows: Theory. Atmospheric Processes over Complex Terrain, Meteor. Monogr., No. 45, Amer. Meteor. Soc., 43–58.
Eyre, J. R., 1991: A fast radiative transfer model for satellite sounding systems. ECMWF Tech. Memo. 176, 28 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/tm/001-300/tm176.pdf.]
Finke, U., and T. Hauf, 1996: The characteristics of lightning occurrence in southern Germany. Beitr. Phys. Atmos., 69, 361–374.
Gantner, L., M. Hornsteiner, J. Egger, and G. Hartjenstein, 2003: The diurnal circulation of Zugspitzplatt: Observations and modeling. Meteor. Z., 12, 95–102.
Geerts, B., Q. Miao, and J. K. Demko, 2008: Pressure perturbations and upslope flow over a heated, isolated mountain. Mon. Wea. Rev., 136, 4272–4288.
Germann, U., G. Galli, M. Boscacci, and M. Bolliger, 2006: Radar precipitation measurement in a mountainous region. Quart. J. Roy. Meteor. Soc., 132, 1669–1692.
Giorgi, F., C. Jones, and G. Asrar, 2009: Addressing climate information needs at the regional level: The CORDEX framework. WMO Bull., 58, 175–183. [Available online at http://www.wmo.int/pages/publications/bulletin_en/archive/58_3_en/documents/58_3_giorgi_en.pdf.]
Guichard, F., and Coauthors, 2004: Modelling the diurnal cycle of deep precipitating convection over land with cloud-resolving models and single-column models. Quart. J. Roy. Meteor. Soc., 130, 3139–3172.
Heise, E., M. Lange, B. Ritter, and R. Schrodin, 2003: Improvement and validation of the multi-layer soil model. COSMO Newsletter, No. 3, 198–203. [Available online at http://www.cosmo-model.org/content/model/documentation/newsLetters/newsLetter03/cnl3-chp9-12.pdf.]
Henne, S., and Coauthors, 2004: Quantification of topographic venting of boundary layer air to the free troposphere. Atmos. Chem. Phys., 4, 497–509.
Hennemuth, B., and H. Schmidt, 1985: Wind phenomena in the Dischma valley during DISKUS. Arch. Meteor. Geophys. Bioklimatol., 35, 361–387.
Hohenegger, C., P. Brockhaus, and C. Schär, 2008: Towards climate simulations at cloud-resolving scales. Meteor. Z., 17, 383–394.
Hohenegger, C., P. Brockhaus, C. S. Bretherton, and C. Schär, 2009: The soil moisture–precipitation feedback in simulations with explicit and parameterized convection. J. Climate, 22, 5003–5020.
Jäger, E. B., I. Anders, D. Lüthi, B. Rockel, C. Schär, and S. I. Seneviratne, 2008: Analysis of ERA40-driven CLM simulations for Europe. Meteor. Z., 17, 349–367.
Kain, J. S., and J. M. Fritsch, 1990: A one-dimensional entrainment/detrainment plume model and its application in convective parameterization. J. Atmos. Sci., 47, 2784–2802.
Kaufmann, P., 2008: Association of surface stations to NWP model grid points. COSMO Newsletter, No. 9, 54–55. [Available online at http://www.cosmo-model.org/content/model/documentation/newsLetters/newsLetter09/cnl9-10.pdf.]
Keil, C., A. Tafferner, and T. Reinhardt, 2006: Synthetic satellite imagery in the Lokal-Modell. Atmos. Res., 82, 19–25.
Khlopenkov, K. V., and A. P. Trishchenko, 2007: SPARC: New cloud, snow, and cloud shadow detection scheme for historical 1-km AVHHR data over Canada. J. Atmos. Oceanic Technol., 24, 322–343.
Klemp, J., and R. Wilhelmson, 1978: Simulation of three-dimensional convective storm dynamics. J. Atmos. Sci., 35, 1070–1096.
Kuwagata, T., 1997: An analysis of summer rain showers over central Japan and its relation with the thermally induced circulation. J. Meteor. Soc. Japan, 75, 513–527.
Kuwagata, T., and F. Kimura, 1997: Daytime boundary layer evolution in a deep valley. Part II: Numerical simulation of the cross-valley circulation. J. Appl. Meteor., 36, 883–895.
Langhans, W., 2012: Multiscale aspects of cloud-resolving simulations of moist summer convection over complex terrain. Ph.D. thesis, ETH Zurich, 207 pp.
Langhans, W., J. Schmidli, and C. Schär, 2012a: Bulk convergence of cloud-resolving simulations of moist convection over complex terrain. J. Atmos. Sci., 69, 2207–2228.
Langhans, W., J. Schmidli, and C. Schär, 2012b: Mesoscale impacts of explicit numerical diffusion in a convection-permitting model. Mon. Wea. Rev., 140, 226–244.
Linder, W., W. Schmid, and H. H. Schiesser, 1999: Surface winds and development of thunderstorms along southwest–northeast-oriented mountain chains. Wea. Forecasting, 14, 758–770.
Lugauer, M., and P. Winkler, 2005: Thermal circulation in South Bavaria—Climatology and synoptic aspects. Meteor. Z., 14, 15–30.
Marsigli, C., F. Boccanera, A. Montani, and T. Paccagnella, 2005: The COSMO-LEPS mesoscale ensemble system: Validation of the methodology and verification. Nonlinear Processes Geophys., 12, 527–536.
Marty, C., R. Philipona, C. Fröhlich, and A. Ohmura, 2002: Altitude dependence of surface radiation fluxes and cloud forcing in the Alps: Results from the Alpine surface radiation budget network. Theor. Appl. Climatol., 72, 137–155.
Matricardi, M., F. Chevallier, and S. Tjemkes, 2001: An improved general fast radiative transfer model for the assimilation of radiance observations. ECMWF Tech. Memo. 345, 42 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/tm/301-400/tm345.pdf.]
Mellor, G. L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys. Space Phys., 20, 851–875.
MeteoSchweiz, cited 2012: Juli 2006: Klimatologisch ein extremer Monat (July 2006: A climatologically extreme month). [Available online at http://www.meteoschweiz.admin.ch/web/de/wetter/wetterereignisse/juli_2006_erste_bilanz.html.]
Moeng, C.-H., P. P. Sullivan, M. F. Khairoutdinov, and D. A. Randall, 2010: A mixed scheme for subgrid-scale fluxes in cloud-resolving models. J. Atmos. Sci., 67, 3692–3705.
Molinari, J., and M. Dudek, 1992: Parameterization of convective precipitation in mesoscale models: A critical review. Mon. Wea. Rev., 120, 326–344.
Müller, M. D., and D. Scherer, 2005: A grid- and subgrid-scale radiation parameterization of topographic effects for mesoscale weather forecast models. Mon. Wea. Rev., 133, 1431–1442.
Neininger, B., and O. Liechti, 1984: Local winds in the upper Rhone valley. GeoJournal, 8, 265–270.
Nickus, U., and I. Vergeiner, 1984: The thermal structure of the Inn valley atmosphere. Meteor. Atmos. Phys., 33, 199–215.
Orlanski, I., 1975: A rational subdivision of scales for atmospheric processes. Bull. Amer. Meteor. Soc., 56, 527–530.
Pauley, P. M., 1998: An example of the uncertainty in sea level pressure reduction. Wea. Forecasting, 13, 833–850.
Raschendorfer, M., 2001: The new turbulence parameterization of LM. COSMO Newsletter, No. 1, 89–98. [Available online at http://www.cosmo-model.org/content/model/documentation/newsLetters/newsLetter01/newsLetter_01.pdf.]
Reinhardt, T., and A. Seifert, 2006: A three-category ice-scheme for LMK. COSMO Newsletter, No. 6, 115–120. [Available online at http://www.cosmo-model.org/content/model/documentation/newsLetters/newsLetter06/cnl6_reinhardt.pdf.]
Reiter, E. R., and M. Tang, 1984: Plateau effects on diurnal circulation effects. Mon. Wea. Rev., 112, 638–651.
Richard, E., and Coauthors, 2011: Forecasting summer convection over the Black Forest: A case study from the Convective and Orographically-induced Precipitation Study (COPS) experiment. Quart. J. Roy. Meteor. Soc., 137, 101–117.
Ritter, B., and J. F. Geleyn, 1992: A comprehensive radiation scheme for numerical weather prediction models with potential applications in climate simulations. Mon. Wea. Rev., 120, 303–325.
Rotach, M. W., and D. Zardi, 2007: On the boundary-layer structure over highly complex terrain: Key findings from MAP. Quart. J. Roy. Meteor. Soc., 133, 937–948.
Saunders, R. W., M. Matricardi, and P. Brundel, 1999: An improved fast radiative transfer model for assimilation of satellite radiance observations. Quart. J. Roy. Meteor. Soc., 125, 1407–1425.
Schlemmer, L., 2011: The diurnal cycle of midlatitude, summertime moist convection over land in an idealized cloud-resolving model. Ph.D. thesis, ETH Zurich, 207 pp.
Schlemmer, L., C. Hohenegger, J. Schmidli, C. S. Bretherton, and C. Schär, 2011: An idealized cloud-resolving framework for the study of midlatitude diurnal convection over land. J. Atmos. Sci., 68, 1041–1057.
Schmetz, J., S. A. Tjemkes, M. Gube, and L. van de Berg, 1997: Monitoring deep convection and convective overshooting with Meteosat. Adv. Space Res., 19, 433–441.
Schmetz, J., P. Pili, S. Tjemkes, D. Just, J. Kerkmann, S. Rota, and A. Ratier, 2002: An introduction to Meteosat Second Generation (MSG). Bull. Amer. Meteor. Soc., 83, 977–992.
Schmidli, J., and R. Rotunno, 2010: Mechanisms of along-valley winds and heat exchange over mountainous terrain. J. Atmos. Sci., 67, 3033–3047.
Steinacker, R., 1984: Area–height distribution of a valley and its relation to the valley wind. Contrib. Atmos. Phys., 57, 64–71.
Steppeler, J., and Coauthors, 2003: Meso-gamma scale forecasts using the nonhydrostatic model LM. Meteor. Atmos. Phys., 82, 75–96.
Tiedtke, M., 1989: A comprehensive mass flux scheme for cumulus parameterization in large-scale models. Mon. Wea. Rev., 117, 1779–1800.
Vergeiner, I., and E. Dreiseitl, 1987: Valley winds and slope winds—Observations and elementary thoughts. Meteor. Atmos. Phys., 36, 264–286.
Wagner, A., 1932: Neue Theorie des Berg- und Talwindes (A new theory of mountain and valley winds). Meteor. Z., 49, 329–341.
Weissmann, M., F. J. Braun, L. Ganter, G. J. Mayr, S. Rahm, and O. Reitebuch, 2005: The Alpine mountain–plain circulation: Airborne Doppler lidar measurements and numerical simulations. Mon. Wea. Rev., 133, 3095–3109.
Whiteman, C. D., 1990: Observations of thermally developed wind systems in mountainous terrain. Atmospheric Processes over Complex Terrain, Meteor. Monogr., No. 45, Amer. Meteor. Soc., 5–42.
Wicker, L. J., and W. C. Skamarock, 2002: Time-splitting methods for elastic models using forward time schemes. Mon. Wea. Rev., 130, 2088–2097.
Wüest, M., M. Litschi, A. Altenhoff, C. Frei, M. Hagen, and C. Schär, 2010: A gridded hourly precipitation dataset for Switzerland using rain-gauge analysis and radar-based disaggregation. Int. J. Climatol., 30, 1764–1775.
Yang, X., 1991: A study of nonhydrostatic effects in idealized sea breeze systems. Bound.-Layer Meteor., 54, 183–208.
Yin, S., W. Li, D. Chen, J.-H. Jeong, and W. Guo, 2011: Diurnal variations of summer precipitation in the Beijing area and the possible effect of topography and urbanization. Adv. Atmos. Sci., 28, 725–734.
Zängl, G., 2004: A reexamination of the valley wind system in the Alpine Inn valley with numerical simulations. Meteor. Atmos. Phys., 87, 241–256.
Zängl, G., J. Egger, and V. Wirth, 2001: Diurnal winds in the Himalayan Kali Gandaki valley. Part II: Modeling. Mon. Wea. Rev., 129, 1062–1080.
Zhong, S., and J. Fast, 2003: An evaluation of the MM5, RAMS, and Meso-Eta models at subkilometer resolution using VTMX field campaign data in the Salt Lake valley. Mon. Wea. Rev., 131, 1301–1322.
The detection of low clouds using this method has been found to be sensitive to sunrise and sunset, such that interpretation of the data is avoided during these periods.