1. Introduction
Over the last several decades, the large-eddy simulation (LES) technique has become one of the major tools in atmospheric boundary layer (BL) research and prominently in investigations of processes in convective and transitional boundary layers (see, e.g., recent reviews by Stoll et al. (2020) and Honnert et al. (2020) and references therein). Since detailed observations of three-dimensional fields in the atmosphere covering several decades of spatial scales are not practical at present, many studies rely on LES to provide a proxy for physically consistent distributions of dynamical and thermodynamical variables, as well as their variances and covariances. This approach has been prevalent in studies on convective boundary layers over land, including dry (e.g., Jonker et al. 1999; Moeng and Wyngaard 1984; Shin and Dudhia 2016) and moist (e.g., Lock et al. 2000; Neggers et al. 2019) convective regimes, as well as land–atmosphere interactions (e.g., Patton et al. 2005; Kang and Ryu 2016; Lee et al. 2019; Chen et al. 2020). One of the primary benefits of LES is that the approach provides not only information about the mean profiles of BL properties but also a physically consistent description of spatial variability of important atmospheric variables over a range of relevant scales. Characterizing contributions of various scales to the total variance contained in a given domain is critically important for development and evaluation of parameterizations for mesoscale and global models because much of that range falls under the horizontal grid spacing of these models (e.g., Wyngaard 2004; Honnert et al. 2011; de Roode et al. 2019).
Although details of LES setups differ widely among various convective BL studies, two broad groups of model configurations emerge. One is aimed at achieving simulation output that matches most closely a well-documented observed case, often from a targeted field campaign (e.g., Heinze et al. 2017b; Fast et al. 2019a; Schemann et al. 2020). Such simulations tend to employ realistic representations of the underlying surface, including topography, and account for proper synoptic environment through nesting LES into a mesoscale model or specifying reanalysis-based boundary conditions (e.g., Barthlott and Hoose 2015; Bauer et al. 2020; Heinze et al. 2017a; Xu et al. 2018). With current computational capabilities, LES domains in these simulations may cover large areas on the order of 100 × 100 km2, thus allowing these models to develop their own internal mesoscale circulations, although often at the expense of employing a horizontal grid spacing coarser than 100 m. Another group of model configurations involves a number of idealizations in external forcing and bottom and lateral boundary conditions. These LES are also expected to develop physically consistent distributions of atmospheric variables, which, however, might be significantly affected by the imposed regularity of surface properties and large-scale tendencies, as well as, by commonly used periodic lateral boundary conditions. In return, such semi-idealized configurations allow one to explore a wider variety of BL cases and conditions and more clearly isolate the effects and roles of specific parameters and processes. In addition, the more computationally economical setup helps to generalize findings from a specific case study using a particular LES model by applying multiple models to the same case, as in model intercomparison studies (e.g., Brown et al. 2002), and by applying the same LES model to a multitude of cases (e.g., Heinze et al. 2017a; Gustafson et al. 2020; Neggers et al. 2012).
With many efforts to improve understanding of the convective boundary layer structure and its diurnal evolution relying heavily on model-predicted fields (e.g., de Roode et al. 2004; Pergaud et al. 2009; Shin and Hong 2013), it is imperative to understand their realism, as well, as potential limitations of various configurations of LES. Implications of various aspects of LES setup are difficult to access using simulations of different cases from different models. There are, however, several models capable of running under both configuration types described above. Recently Schemann et al. (2020) compared the ability of ICON LES to reproduce local cloud observations and found that driving the model with a heterogeneous surface and lateral boundary conditions constrained by the numerical weather prediction models outperformed idealized LES as long as the model domain in the “realistic” setup was not too large.
In this study, we address how commonly employed features of LES setup affect modeled BL properties and, in particular, the distribution of the variance of vertical velocity, potential temperature, and moisture across scales. This is accomplished through a comparison of two ensembles of simulations using semi-idealized and nested larger-domain configurations of the same Weather Research and Forecasting (WRF) Model applied to the same case of the convective boundary layer observed over north-central Oklahoma on 30 August 2016 during the Holistic Interactions of Shallow Clouds, Aerosols, and Land-Ecosystems (HI-SCALE) field campaign (Fast et al. 2019b). Grid spacing is one of the key parameters of any atmospheric model, and in this study horizontal grid spacings of 100 and 300 m are employed. Following the terminology of a number of studies relying on comparable model configurations, here we refer to all analyzed simulations as LES, even though runs with the 100-m and especially 300-m grid spacings should be more properly classified as “near–gray zone” simulations (Efstathiou et al. 2018; Honnert et al. 2020) or very-large-eddy simulations (VLES) (Orszag and Staroselsky 2000; Smolarkiewicz and Prusa 2002). These grid spacings may not reproduce a well-defined inertial subrange or satisfy more formal resolution convergence metrics (e.g., Sullivan and Patton 2011), but when the convective BL is well developed, the resolved fraction of the turbulent kinetic energy remains high enough for these setups to be classified as “coarse LES” (Honnert et al. 2020; Wurps et al. 2020).
It will be shown that, in simulations using a larger domain with heterogeneous surface properties, the development of mesoscale patterns significantly affects boundary layer structure inside a smaller subdomain, comparable in size to the domain used in semi-idealized LES with periodic lateral boundary conditions, flat surface, and homogeneous surface heat fluxes. The impacted boundary layer properties include inversion height and variances of temperature, moisture, and vertical velocity, as well as distributions of these variances across spatial scales. Despite systematic differences between the two ensembles of simulations, all model configurations reproduce progressive increase in characteristic scales from vertical velocity to temperature and moisture structures.
The rest of the paper is organized as follows. Section 2 describes model configurations and an ensemble of analyzed simulations, as well as a method for determining characteristic scales of variances. The main results of the analysis are presented in section 3, and section 4 provides a summary and further discussion of key findings.
2. Approach
a. Model configuration and simulations
Simulations used in this study and listed in Table 1 are comprised of two ensembles of runs with different setups: one using a larger-domain nested LES implementation, hereafter collectively referred to as WRF-LES, and the other using more idealized periodic boundary condition LES. The latter ensemble is composed of simulations taken from the LES Atmospheric Radiation Measurement (ARM) Symbiotic Simulation and Observation (LASSO) library of data bundles, which serves as a comprehensive resource for studying continental shallow convection by merging observations with LES for an ensemble of days with observed shallow convection over the ARM Southern Great Plain (SGP) site (Gustafson et al. 2019, 2020). LASSO and similar multiday simulations by other groups (e.g., Neggers et al. 2012) have begun to be analyzed to help interpret observations and to guide parameterization development for turbulent and cloud statistics (e.g., Angevine et al. 2018; Neggers et al. 2019; Gristey et al. 2020; van Laar et al. 2019). The setup differences between the two groups of simulations are illustrated in Fig. 1. Although we refer to the second simulation ensemble collectively as LASSO in this study, this setup is representative of a much broader group of semi-idealized LES.
A schematic representation of setups for the (a) high-resolution WRF-LES and (b) LASSO runs (see text for details). Two bottom surfaces represent horizontal distributions of the latent heat flux (LH) and ground elevation above the mean sea level. Large-scale (LS) external forcing on the simulated domain, illustrated using water vapor specific humidity qυ, is applied through lateral boundary conditions for WRF-LES and as a prescribed tendency for the domain-mean moisture profile
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
Analyzed simulations.
WRF-LES is employed in several different configurations (Table 1). As a reference (simulation W1), we use a thoroughly evaluated simulation performed on a 120 × 120 km2 domain nested inside a 297‐km‐wide outer domain that encompasses much of central Oklahoma and southern Kansas (Fast et al. 2019a). 100‐ and 300‐m horizontal grid spacings are used for the inner and outer domains, respectively. Both domains extend up to 16.2 km above mean sea level (MSL) in the vertical and use 304 grid levels with a near constant 24‐m grid spacing below ∼6 km and a stretched grid spacing above. The model uses a realistic topography and employs the Noah land surface model (Chen and Dudhia 2001) with realistic initial surface moisture, vegetation, and soil types to represent land–atmosphere interactions. The initial soil moisture is obtained by interpolating in situ observations from the ARM Soil Temperature and Moisture Profile System (Cook 2018) and Oklahoma Mesonet (McPherson et al. 2007) and the land cover is based on the National Land Cover Database (Yang et al. 2018) following Fast et al. (2019a) (see Sakaguchi et al. 2022, for further details on the land surface treatment in WRF-LES). This simulation uses the 1.5-order turbulent kinetic energy (TKE) closure for the subgrid-scale (SGS) fluxes (Deardorff 1980), the RRTMG shortwave and longwave radiation (Iacono et al. 2008), and the Thompson microphysics (Thompson et al. 2008) parameterizations (see Fast et al. 2019a, for further details). Initial and boundary conditions for the outer domain are driven by the National Centers for Environmental Prediction Final (FNL) operational model global tropospheric analyses (NCEP FNL Operational Model Global Tropospheric Analyses 2000). The setup for the high-resolution inner domain is illustrated in Fig. 1a. This computationally intensive simulation has been shown to reproduce many observed features of the daytime evolution of boundary layer and convective cloud population (Fast et al. 2019a).
In addition to the reference simulation, several WRF-LES coarser-resolution sensitivity simulations employing only a larger domain with 300-m horizontal grid spacing and without a nested domain are analyzed: W3 represents the reference simulation on the coarser-resolution (outer) domain, W3_SM is the same as W3, but initialized with a smoothed soil moisture (see “variable-1” simulation in Fast et al. 2019a), and then both the W3 and W3_SM simulations are repeated with a mean horizontal wind set to zero, thereby removing large-scale advection effects (W3_NA and W3_SM_NA simulations). In addition to eliminating the mean horizontal wind, the “no-advection” simulations are initialized with horizontally uniform temperature and moisture fields and also have horizontally uniform temperature and moisture profiles on the outer boundaries of the nested domain (see Chen et al. 2020, for details). Because the mean boundary layer wind is relatively weak in the reference case (cf. Fig. S2 in Chen et al. 2020), the near-surface turbulence is due primarily to the buoyancy production. As will be shown in the following section, removing the mean wind in this case leads only to minor changes in the mean sensible and latent heat fluxes in WRF-LES with identical surface properties. The more traditional and economical LES setup for boundary layer research is represented in this study by LASSO simulations of the same HI-SCALE case. A schematic of the LASSO setup using a horizontal domain of 14.4 × 14.4 km2 is shown in Fig. 1b. LASSO output, or data bundle, is described in detail by Gustafson et al. (2019, 2020) and accessible through the ARM archive (Gustafson et al. 2017). The setup includes features common to many LES studies of the boundary layer and clouds over land, including periodic lateral boundary conditions, prescribed time-dependent but horizontally uniform surface sensible and latent heat fluxes, a flat surface for the lower boundary, and prescribed horizontally uniform time-varying tendencies for temperature and moisture fields representing large-scale advection. This study uses output from three simulations from the data bundles compiled for the 30 August 2016 case, which differ only in the large-scale forcing (i.e., prescribed convergence/divergence of large-scale wind and time tendencies for temperature and moisture) applied to drive the model. In the default run, these tendencies are derived from European Centre for Medium-Range Weather Forecasts (ECMWF) forecasts and correspond to a horizontal scale of 413 km around SGP. Another simulation is also driven by the ECMWF-derived forcing but corresponding to a smaller spatial scale of 114 km. The third simulation is forced by the tendencies obtained from convection-permitting WRF simulations constrained using the Multiscale Data Assimilation (MSDA) methodology (Li et al. 2015a,b). In all three setups, surface fluxes are prescribed based on measurements at the ARM SGP Facility. All the LASSO simulations use the Thompson microphysics scheme, as do the WRF-LES runs described earlier. Finally, to explore the effects of the domain size and horizontal grid spacing (Δx) on the studied boundary layer scales, we conducted additional simulations with the default LASSO setup but using a horizontal domain of 25 km on a side and with a grid spacing of Δx = 100 m (simulation L25_E413) and Δx = 300 m (L25_300).
The large-scale advective forcing used in LASSO is derived from a variety of sources and for a range of spatial scales but in all cases characterizes mean tendency over an area much larger than the LASSO domain and more comparable to the WRF-LES domains (Gustafson et al. 2020). The location of the LASSO domain within the area for which the forcing is derived is not defined. Instead, these simulations are supposed to generate a representative state of an atmospheric column anywhere within the forcing area (see also discussion on LASSO forcing in Gustafson et al. 2020). Therefore, in this study, we also examine how the comparative variability scale analysis for LASSO and WRF-LES is affected by the position of the analyzed subdomain within the WRF-LES computational domain. The simulation period is from 0600 to 1800 CST 30 August (local time), or 1200 UTC 30 August to 0000 UTC 31 August, while the scale analysis focuses primarily on the early afternoon hours (1200–1600 CST) when the convective boundary layer is well developed. Most of the scale analysis is performed for a central, square subdomain 9.6 km on a side, so that the same area can be fully covered by non-overlapping tiles of 6 doubling sizes (9.6 = 0.3 × 25), from 0.3 km (Δx of a coarse-resolution simulation) to 9.6 km (the domain size), which are used in computing variability scales described in section 3b. For the higher-resolution simulations, tiles of 100 and 200 m are also analyzed. While this subdomain is admittedly small, using LES domain sizes on the order of 10 km is still quite common, particularly when clouds are included in consideration. The sensitivity of the results to the subdomain size will be discussed below.
b. Characterizing scale dependency of variance
Various techniques have been employed for scale analysis of horizontally inhomogeneous atmospheric fields, including those based on wavelet transforms and one- or two-dimensional Fourier power spectra (e.g., Deardorff 1974; Roy and Avissar 2000; de Roode et al. 2004). In this study, the scale dependency of variances of dynamical and thermodynamic variables is analyzed by progressively downgrading the model output to larger tile sizes (coarsened horizontal grids) and examining how the partitioning of these variances between “resolved” and sub-tile contributions changes with the tile size (Honnert et al. 2011). Note that analogous information can be extracted from a cumulative power density spectrum, or ogive (de Roode et al. 2004; Kang 2020). At the native LES resolution (horizontal grid size Δx), all variability is assumed to be resolved, and therefore the total variance can be computed directly from grid variables. In this case, the contribution of unresolved scales to the total variance is zero (Fig. 2). At the opposite extreme, when the scale, or tile size, is equal to the size of the analyzed domain, all variability becomes “unresolved” and the contribution of the resolved scales to the total variance vanishes. Between these limits, lay the scales at which contributions to the total variance come from both “resolved” and “unresolved” structures. A scale range where these contributions are of the same order in magnitude is often termed “terra incognita” or the turbulence “gray zone” (e.g., Dorrestijn et al. 2013; Honnert et al. 2011; Wyngaard 2004). Note that although LES includes a parameterized treatment of SGS variability, its contribution to the total variances and fluxes and their partitioning is not included in the above approach. Even at the relatively coarse horizontal grid sizes of 100 and 300 m used in this study, explicitly resolved scales dominate total variances throughout the well-developed 1.5–2-km-deep convective boundary layer, which is the focus of this study, except very near the surface.
Schematic representation of the scale (tile size) dependency of resolved and unresolved normalized (fractional) variance components. L50 and L95 denote scales at which the unresolved (subtile) variance fraction is 0.50 and 0.95. Δx and NxΔx are the horizontal grid spacing and the domain size, respectively.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
For a quantitative comparison of various fields, it is convenient to identify specific scales characterizing spatial structures of the considered field. Here, we define L50 and L95 as scales (tile sizes) at which smaller (subtile) features contribute 50% and 95% of the total variance, respectively (e.g., Honnert et al. 2011). It is worth noting that because these characteristic scales are computed from fractional variance distributions (i.e., distribution of variance normalized by the total variance), they depend only on the shape of the energy spectrum, not the total variance or the amplitude of the spectrum.
In discretized models, grid spacing Δx inherently limits the size of the smallest features that can be resolved, and, in this study, it represents the smallest analyzed scale with no variability below it. The size of the smallest tile containing any “unresolved” variance is, therefore, 2Δx. It is worth keeping in mind, however, that multiple grid points are needed to actually resolve any spatial structure and, in general, four or more grid points per wavelength are needed to represent a wave pattern.
3. Results
a. Boundary layer evolution
Figure 3 shows the time evolution of planetary boundary layer height (zi) and latent and sensible heat fluxes for the analyzed simulations. All variables are horizontally averaged over the centrally located 9.6 × 9.6 km2 subdomain. In this study, zi is diagnosed based on a vertical gradient of the potential temperature profile following an algorithm by Liu and Liang (2010). Using the hourly instantaneous output of potential temperature fields, mean zi for each run is computed using values determined for each LES column in the analyzed subdomain. Systematic differences in boundary layer height evolution are seen between LASSO and WRF-LES (Fig. 3a), with LASSO’s boundary layer being on average over 400 m shallower during midday and early afternoon hours.
Time evolution of (a) PBL height and surface fluxes of (b) latent and (c) sensible heat for all simulations. All variables are averaged over the central 9.6 × 9.6 km2 subdomain.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
Within each group of simulations, differences in zi can reach 200 m. LASSO simulations driven by large-scale advective tendencies derived from the same large-scale model (ECMWF) produce nearly identical zi, regardless of whether the forcing is derived for area of 400 km (L14_E413) or 100 km (L14_E114) on a side. Using forcing from a different large-scale model (L14_M300) leads to a shallower boundary layer in the afternoon. This change in zi is attributed to differences in the advective tendencies and subsidence because prescribed time-dependent LH and SH fluxes are identical in all LASSO simulations presented here by design (Figs. 3b,c). Prescribed large-scale vertical motions are significantly stronger in the MSDA than in the ECMWF forcing. Around local noon, for example, MSDA-prescribed subsidence near the top of the boundary layer is on the order of 1 cm s−1, while in the ECMWF the vertical motion is much weaker (∼0.1 cm s−1) and in the opposite direction. The described stronger effect of the forcing source on zi compared to the forcing scale may, in general, be case specific but is likely to be applicable to other LASSO shallow convection cases. Several factors affect differences between ECMWF and MSDA forcings derived for comparable horizontal scales, including different large-scale models and different data assimilation procedures, while the scale dependency of any forcing is tied primarily to synoptic conditions, as manifested in mesoscale gradients in various fields on a given day. For LASSO cases of shallow convection, however, large-scale conditions around the SGP are often quite uniform, especially in the morning hours, so that the effect of the scale in the forcing is diminished while biases inherent in the source models are carried over into the forcings.
A comparison of L14_E413 and L25_E413 illustrates that increasing the horizontal extent of the periodic domain from 14.6 to 25 km has a negligible effect on zi evolution. Coarsening the grid from 100 to 300 m increases zi in the afternoon (cf. L25_E413 and L25_300), which is consistent with previous resolution sensitivity studies (e.g., Beare 2014; Wurps et al. 2020). Bopape et al. (2020) also found increasing zi with grid spacing in LES and coarse LES of a convective BL and attributed it to the overly penetrative thermals in the lower-resolution simulations altering mean temperature profile at the BL top and weakening the inversion. It must be noted, however, that in our simulations the resolution-related increase in zi is small relative to the difference between LASSO and WRF-LES ensembles.
The convective boundary layer entrainment fluxes, and therefore changes in zi, are controlled by a number of physical processes, including the engulfment of the free-tropospheric air, instabilities induced by wind shear-driven waves, penetrating and recoiling convective eddies, and propagation and excitation of gravity waves and their interaction with the turbulent eddies (e.g., Wulfmeyer et al. 2016). The intensities of the above processes are related to the strength of vertical motions inside and around the entrainment layer, which can be characterized by the vertical velocity variance,
Profiles of variance for (a) vertical velocity, (b) potential temperature, (c) water vapor mixing ratio, and (d) vertical flux of the virtual potential temperature at 14 h. The height is normalized by the boundary layer depth zi. Horizontal dashed line indicates the middle of the boundary layer (z = 0.5 zi). Logarithmic x axes are used for (b) and (c) to distinguish
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
Among the WRF-LES runs, W1 has the shallowest BL during much of the day (Fig. 3a). This is the only simulation in the WRF-LES group that uses the same Δx = 100 m as the majority of the LASSO runs. Compared to the reference W1 simulation, the deepening of the boundary layer in the morning is accelerated in W3 runs (Fig. 3a). The effect of coarsening the grid, however, is different between LASSO and WRF-LES setups, even though in both configurations BL is deeper in simulations on a coarser grid (cf. W3 vs W1 and L25_300 vs L25_E413). In WRF-LES, using Δx = 300 m leads to larger variances for w, θ, and qυ, as well as stronger entrainment (Fig. 4). In LASSO, however, there is no increase in
Another notable difference between LASSO and WRF-LES is the sharp reduction in zi later in the afternoon in some simulations from the latter group. These instances of BL collapse appear to be related to cold pools propagating from deeper convection outside the central subdomain, which occurs in WRF-LES runs but not in LASSO. We return to this aspect in section 3d.
Stronger turbulence and a deeper convective boundary layer can develop from a larger sensible heat flux from the surface. To some degree, this is true for the presented WRF-LES results. However, results from WRF-LES do not show one-to-one correspondence between SH and zi (Figs. 3a,c), as simulations with nearly identical LH and SH can exhibit zi differences of 200 m or more (e.g., compare W3 and W3_NA).
Unlike LASSO, where the surface heat fluxes are prescribed from observations at the SGP facility, WRF-LES computes these fluxes interactively and initial soil moisture is one of the main factors controlling LH and SH evolution in these simulations. In the reference simulation W1, as well as in its coarse-resolution analog W3, both SH and LH fluxes match well observation-based fluxes used in LASSO. This indirectly supports that surface and atmospheric properties over the SGP Central Facility region in WRF-LES are realistic. In the smoothed soil moisture simulations (W3_SM and W3_SM_NA), the top layer of soil around the SGP is initially drier than in the reference simulation, such that the mean soil moisture over the analyzed 9.6 × 9.6 km2 subdomain is reduced by nearly a third (not shown). This leads to smaller latent heat fluxes in the W3_SM and W3_SM_NA simulations (Fig. 3b). This reduction in LH is, at least partially, compensated by an increase in the sensible heat flux in these two runs (Fig. 3c), which contributes to deepening of the boundary layer (Fig. 3a).
Overall, in the considered case, the largest and most systematic differences in zi and profiles of
b. Characteristic variability scales in a developed convective boundary layer
A visual illustration of relevant scales for horizontal distributions of vertical velocity and water vapor mixing ratio from the reference simulation is provided by Fig. 5. The difference in the sizes of the features is clearly noticeable between the two panels, with characteristic scales for moisture nearly double those for vertical velocity. Although higher (lower) values of water vapor mixing ratio are generally associated with updrafts (downdrafts), the correlation between the two is not perfect, and a region of the lowest qυ (near x = 0.5 km and y = 7 km) is associated with quiescent vertical motions. This is because, even though the vertical advection exerts a major influence on qυ, instantaneous moisture content at any level depends on the vertical velocity profile and its time history, in addition to the local w values.
Horizontal cross sections of (a) vertical velocity w and (b) water vapor mixing ration qυ through the analyzed subdomain of the W1 simulation at 14 h at the middle of the boundary layer (z = 0.95 km). Also shown are scales for the 50th (L50) and 95th (L95) percentiles of variance and the boundary layer depth (zi).
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
The scale dependency of the unresolved fraction of the total variance of vertical velocity, potential temperature, and water vapor mixing ratio for all simulations from Table 1 is shown in Figs. 6a–c. Variances are calculated for the horizontal cross sections taken through the middle of the boundary layer (z = 0.5 zi) at time t = 14 h when the convective boundary layer is well developed. Note that fractions of resolved variance, as illustrated in the conceptual Fig. 2, represent mirror images of the unresolved fractions about the 0.5 line and are omitted here for clarity.
Scale dependency of variance fractions for (a),(d) vertical velocity; (b),(e) potential temperature; and (c),(f) water vapor mixing ratio for (a)–(c) the complete scale range and (d)–(f) variance contained only in structures larger than 300 m. All variances are computed for horizontal cross sections through the analyzed subdomain in the middle of the boundary layer (z = 0.5 zi) at 14 h. Horizontal dashed lines mark SGS variance fractions of 0.50 and 0.95. Vertical dashed and dotted lines indicate L50 and L95 scales, respectively, for the reference simulation W1 (see text for details). Representation of simulations with 300-m grid spacing is identical in the left and right columns.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
For vertical velocity, the scale partition of the variance is controlled primarily by the native resolution of the model. In simulations with Δx = 100 m, features smaller than 300 m contain 25%–40% of the resolved variance (Fig. 6a), while simulations using Δx = 300 m obviously have no resolved variability at and below 300-m scale. Among the higher-resolution runs, LASSO simulations have a higher fractional variance at smaller scales than the W1. This is not because smaller eddies in LASSO are more intense, but rather because the variability at larger scales is suppressed, possibly because of a doubly periodic LASSO domain or flat terrain. Hence, even if the intensity of eddies of a certain size were identical, smaller total variance in LASSO would lead to higher fractional variance. The total vertical velocity variance (for the same 9.6 × 9.6 km2 subdomain) is indeed larger in W1 than in any of the LASSO simulations by at least a third (Fig. 4). Expectedly, in all simulations, nearly all variability in the vertical velocity fields is contained at scales smaller than the boundary layer depth, which is near 1.5 km for LASSO and 2.0 km for WRF-LES in the early afternoon (Fig. 6a).
Scale dependencies of variances of temperature and moisture show a more complicated picture than that of vertical velocity, although qualitatively similar sensitivities to the model horizontal grid spacing and open versus periodic lateral boundary condition remain (Figs. 6b,c). Compared to the vertical velocity, there is a general trend for shifting variance toward larger scales for temperature and moisture, consistent with earlier findings (e.g., Honnert et al. 2011; de Roode et al. 2019). This is clearly demonstrated by the L50 and L95 scales for the reference simulation W1 increasing from the top (vertical velocity) to the bottom (water vapor) panels in Fig. 6.
Part of the differences seen in Fig. 6 is attributable to the boundary layer depth variations among the simulations (Fig. 3a) because zi exerts a major control on the size of structures in the boundary layer (e.g., de Roode et al. 2004). Hence, when the variability scales are normalized by zi, the spread of scale dependencies from the analyzed simulations is reduced (Figs. 7a–c). In particular, the separation of the fractional variance distributions between simulations conducted with two grid spacings becomes clearer, as W1 and all LASSO runs collapse to a nearly identical scale dependency. An ensemble of coarser-resolution simulations (all W3 runs and L25_300) retains significant spread in scale dependencies of variances, particularly for θ and qυ (Figs. 7b,c), highlighting the important role of processes generating inhomogeneities in these fields outside the analyzed subdomain and on scales larger than zi.
As in Fig. 6, but with the scale L normalized by zi.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
To more directly compare distributions of variability across larger scales from simulations with different native resolutions, we repeat the scale analysis for the 100-m simulations downgraded to Δx = 300 m, essentially truncating power spectra from the higher-resolution simulations at 300-m scale. These results are shown in Figs. 6d–f and 7d–f. Distributions of variance across scales larger than 300 m (Figs. 6d–f) show that the differences between LASSO and WRF-LES setups are reduced compared to when the full scale range is considered (Figs. 6a–c), with most significant changes seen at scales smaller than 2 km. LASSO and W1 simulations, however, still put a larger variance fraction in smaller scales. For vertical velocity, for example, LASSO and W1 contain half of the variance in 300–600-m range and a third in the 600–1200-m-scale range, while for the coarser-resolution WRF-LES runs partitioning of w variance between these two scale ranges is reversed (Fig. 6d).
c. Profiles of characteristic variability scales
For a more systematic and quantitative comparison of different simulations, we now turn to the profiles of L50 and L95 (Fig. 8). The presented profiles are averaged in time between 12 and 16 h using hourly output. Since deeper boundary layers tend to support larger structures, both scales and altitude here are normalized by zi to eliminate parts of the variability in characteristic scales among simulations that can be explained by the differences in boundary layer depths (Fig. 3a).
Profiles of characteristic horizontal scales of variances for (a),(e) vertical velocity w; (b),(f) potential temperature θ;(c),(g) water vapor mixing ratio qυ; and (d),(h) cloud water mixing ratio qc. Spatial structures with scales smaller than L50 (solid lines) and L95 (dashed lines) contain 50% and 95% of resolved variance in the considered subdomain, respectively. The L50 and L95 scales are computed for (a)–(d) the total variance and (e)–(h) variance in the scales larger than 300 m only. Representation of simulations with 300-m grid spacing is identical in the left and right columns. LASSO cases are shown in red, the reference WRF-LES case is in black, and coarser-resolution WRF-LES cases are in blue. See legend for symbols and color shades marking specific simulations from Table 1. All profiles are averaged in time between 1200 and 1600 local time. Variability scales (L) and vertical coordinate (z) are normalized by the boundary layer depth zi. The profile in (d) also shows L50 (filled symbols) and L95 (open symbols) for surface sensible (SH) and latent (LH) heat fluxes for WRF-LES runs, as denoted by symbols in the legend.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
For vertical velocity in the boundary layer, L50,w/zi collapses nicely into two distinct parabolic profiles corresponding to two horizontal grid spacings (Fig. 8a). The shape of the profile results from generally smaller eddies prevalent near the ground and around zi with larger eddies developing in the middle of the boundary layer. The normalized median scale (L50,w/zi) for the reference simulation W1 almost exactly matches L50,w/zi for LASSO simulations using the same 100-m grid spacing (Fig. 8a), suggesting that majority of the variability in that field comes from eddies whose sizes are governed by the local boundary layer depth. Increasing the LASSO domain to 25 km (L25_E413 simulation) does not have any discernible impact on variability scales and neither do the changes in the LASSO large-scale forcing. Changing grid spacing, however, has a clear and nearly identical effect on L50,w/zi in both LASSO and WRF-LES setups. In the boundary layer, L50,w/zi is larger for 300-m runs than for 100-m runs by a factor of 1.5–1.7. Note that in the higher-resolution runs 50% of w variance is contained in features smaller than about 500 m, which simulations on a 300-m grid are not capable of representing (Fig. 6a). If we only consider variance of structures larger than 300 m, then L50,w/zi from all simulations agree well throughout the BL depth (Fig. 6e).
For the largest eddies of the L95,w/zi size, the direct effect of grid spacing is less pronounced and L95,w/zi for W1 lines up more closely with L95,w/zi for other WRF-LES, even though the latter use a coarser 300-m horizontal grid. Although the deeper boundary layers in WRF-LES correspond to larger eddies than are seen in LASSO, normalization by zi reduces these differences. Vertical velocity structures wider than zi are likely driven by either processes outside the analyzed subdomain or effects of topography, which are neglected in the LASSO setup. The range of the surface heights within the analyzed subdomain is only a few tens of meters (or less than two vertical level spacings), but over 70% of surface height variance is contained in scales larger than 5 km (Fig. 9). Hence, any effect of topography will certainly tend to increase the variability scales for all impacted fields.
Scale dependency of variance fractions for surface elevation for the central subdomain (thick lines) and for four subdomains described in section 3d (thin lines). Because nearly all of the variance is contained in scales larger than 1 km, the differences between W1 (Δx = 100 m) and W3 (Δx = 300 m) are small.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
Above the boundary layer, both L50,w/zi and L95,w/zi profiles vary significantly more among different runs, indicating the importance of factors other than boundary layer depth. Here, the characteristic variability scales for the reference simulation (W1) correspond closer to other WRF-LES runs than to LASSO (Fig. 8a). Gravity waves and mesoscale circulations of the scales larger than O(10) km caused by topography and/or convection outside the analyzed subdomain likely contribute to enhanced L50,w and L95,w above the boundary layer in WRF-LES.
The L50/zi and L95/zi profiles for potential temperature (Fig. 8b) and water vapor mixing ratio (Figs. 8c) show some of the same signatures as those for vertical velocity, most notably strong dependency on grid spacing: tripling horizontal grid spacing increases L50,θ/zi by a factor of 2.5 (Fig. 8b). There are also several noteworthy distinctions. The separation of characteristic scales between LASSO and WRF-LES groups of runs in the boundary layer is significantly larger for potential temperature than for vertical velocity or moisture. In the middle of the boundary layer, in particular, there is a strong local maximum in L50,θ/zi and L95,θ/zi in most WRF-LES runs that is not observed in any of the LASSO simulations. The maxima are less pronounced for smoothed soil moisture runs (W3_SM and W3_SM_NA), in which sensible heat fluxes over the analyzed subdomain are larger (Fig. 3b). However, significant spread among other simulations with comparable fluxes suggests that the primary reason for the wider temperature structures in the middle of the boundary layer in WRF-LES is neither grid spacing nor the mean surface flux values. It is likely that these larger θ structures are related to processes outside the considered subdomain, perhaps to propagation of outflow from deeper convection that begins to develop in the afternoon elsewhere in the WRF-LES domain (see also section 3d). This is indirectly supported by the W1 profiles of L50,θ/zi and
Compared to L50/zi and L95/zi for vertical velocity and potential temperature, profiles of these scales for water vapor shift further to larger values. The shape of the
Removing variance in the 100–300-m range has a smaller effect on L50 and L95 for θ and qυ (Figs. 8f,g) than for w (Fig. 8e) because that scale range contains a smaller fraction of
Although our focus here is on the properties of a well-developed convective boundary layer, and a comprehensive analysis of overlaying clouds is outside the scope of the present study, Fig. 8d illustrates the characteristic scale profiles for the cloud water mixing ratio qc for reference. The L50 and L95 for qc at cloud base near the top of the boundary layer are determined by the vertical velocity structures at that level. In the free troposphere, the qc scales remain nearly constant, or even slightly decrease with altitude, while the characteristic scales of w, θ, and qυ structures grow larger due to diminished small-scale turbulence there.
Also shown in Fig. 8d are scales for the horizontally inhomogeneous sensible and latent heat fluxes (SH and LH) for WRF-LES runs. In the LASSO setup, the surface fluxes are prescribed to be horizontally uniform (Fig. 1b) and, therefore, the variability scales of the fluxes are not defined for these simulations. The L50 and L95 for SH and LH are similar to the corresponding scales for θ and qυ at the lowest levels, resulting in the characteristic scales for LH being larger than for SH. Simulations with smoothed initial soil moisture have smaller L95 for SH and LH than other WRF-LES runs with the same grid spacing (compare, e.g., W3 vs W3_SM and W3_NA vs W3_SM_NA), but this is not the case for L50,SH and L50,LH, or L95,θ and
d. Spatial heterogeneity, scale of the forcing, and representativeness of a small domain
In the preceding sections, our analysis of the WRF-LES runs focused on a single 9.6 × 9.6 km2 subdomain. While this subdomain size is representative of a LASSO setup, it is much smaller than the actual WRF-LES computational domain. The central location of the analyzed subdomain corresponds to the location of the ARM SGP Central Facility and to the center of the area for which large-scale forcing driving LASSO simulations is derived. The LASSO forcing itself, however, represents temperature and moisture tendencies averaged over a much larger area that is more comparable to the WRF-LES domain (Li et al. 2015a,b; Gustafson et al. 2020), and the position of the LASSO domain within that larger area is not defined. In other words, LASSO simulation results assume spatial homogeneity within the forcing region O(100 × 100) km2 and should arguably be representative of this larger area, not just the 10-km box around the Central Facility. It is therefore instructive to explore how the location of the considered subdomain inside the WRF-LES domain can affect the comparison between LASSO and WRF-LES.
To do that we repeat our analysis for four subdomains within the W1 and W3 simulation domains. These subdomains are shifted by ±40 km from the center in south–north and east–west directions, as illustrated in Fig. 10. The subdomains differ not only in variability of surface elevation (Fig. 9), but also in land surface type, which affects latent and sensible heat fluxes. The influence of the processes outside these subdomain, e.g., through advection, can also be different in these regions, especially later in the simulations when mesoscale patterns in the larger domain become more prominent.
Horizontal distributions of (a) the top-level soil moisture, (b) water vapor mixing ratio in the middle of the boundary layer, (c) surface latent heat flux, and (d) surface height above the mean sea level. All fields correspond to time =14 h. Squares in the middle of each panel mark the analyzed subdomains corresponding to the Central Facility location. Four corner subdomains labeled northwest (NW), northeast (NE), southeast (SE), and southwest (SW) are picked to represent spatial variability of local statistics in simulations W1 and W3 (see text for details).
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
A comparison of Figs. 3 and 11 shows that the zi, LH, and SH differ more among different subdomains from the same simulation than among different WRF-LES runs analyzed over the same centrally located subdomain. The daily maximum zi for the four W1 and W3 subdomains ranges from 1.7 km in the northeast to 2.3 km in the northwest (Fig. 11a), eclipsing the range of zi from central subdomains in WRF-LES runs from Table 1 (Fig. 3a). The subdomain location has an even more pronounced effect on LH, with a factor-of-3 difference between the lowest fluxes in the west and the highest in the northeast (Fig. 11b). The range of daily maximum SH values also increases from 60 to 100 W m−2 (cf. Figs. 3c and 11c).
Time evolution of (a) PBL height, (b) latent heat flux, and (c) sensible heat flux for various subdomains from the W1 and W3 simulations. Results for the L14_E413 LASSO simulation are also shown for reference.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
Interestingly, while the LH for the central subdomain is in the middle of the LH range for the corner subdomains (Fig. 11b), SH in the central subdomain is on the lower boundary of the SH range for the corner subdomains (Fig. 11c). This likely contributes to the slowest deepening of the boundary layer for the central subdomain in the morning hours (Fig. 11a). The physical conditions leading to these differences arise from a combination of differing vegetation types as well as soil moisture gradients within WRF-LES (Fig. 10).
Mean latent and sensible heat fluxes for each subdomain agree well between W1 and W3 simulations throughout the day (Figs. 11b,c), emphasizing the dominant control of the local surface properties. In the morning hours, this agreement in surface fluxes translates into similar behavior of zi among subdomains in these simulations. For example, before noon in both W1 and W3 largest zi is found in the NW subdomain, followed by the SW, while the NE has the shallowest BL of the four corner subdomains. At this time, the grid-spacing bias is fairly persistent, and zi values from the coarser-resolution W3 simulation are systematically larger than from W1 (Fig. 11a). Later in the simulations, however, factors other than the surface fluxes come into play in determining zi as the development of mesoscale circulations inside larger WRF-LES domain starts to affect the local BL properties. For example, the boundary layer collapse around t = 17 h seen in Figs. 3 and 11 for nearly all simulations and subdomains is contributed by the propagation of cold pools inside the analyzed subdomains from neighboring precipitating areas, as illustrated for the reference W1 simulation in Fig. 12. In this case, the only W1 subdomain without nearby precipitation is located in the southeast (Fig. 12), and this is the only location where zi changes little in the afternoon and does not change significantly between 16 and 18 h (Fig. 11a).
Horizontal distributions of the potential temperature from the W1 simulation at z = 400 m above ground (corresponding to 0.5 zi over the central subdomain) at t = 18 h. Gray shading marks areas with accumulated precipitation larger than 0.01 mm. Indicated subdomains are as in Fig. 10.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
The distribution of variance across scales within various subdomains, shown in Fig. 13, qualitatively follows the patterns discussed in section 3b and illustrated in Fig. 7. Despite large differences in surface moisture (Fig. 10a), vegetation types (see Fast et al. 2019a) and, correspondingly, LH and SH fluxes among subdomains, for each subdomain in W1 or W3 simulation variability shifts to larger scales for potential temperature and moisture compared to vertical velocity. Model grid spacing, once again, plays a dominant role in determining the fraction of variance contained in structures smaller than 1–2 km. For vertical velocity, clusters of lines corresponding to W1 (100-m grid) and W3 (300-m grid) are clearly separated, with only the most humid NE subdomain showing propensity toward significantly larger structures than are seen in the central subdomain (Fig. 13a). Topography also likely plays a role in the increased contribution of structures near zi and larger to the variance of w in the NE subdomain because the elevation distribution in this region is dominated by the scales of 2–4 km (Fig. 9).
As in Fig. 7, but for scale dependency of the fractional variance for the central and four corner subdomains from W1 and W3 simulations. Results for the LASSO simulation L14_E413 are also shown for reference.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
As we saw in Fig. 7, here as well, the range of scales over which the fractional variance is changing is broader for temperature and moisture than for vertical velocity. Still, the effects of model grid spacing and relationships among variability scales for different variables are consistent for all subdomains.
Distributions of the variances across scales larger than 300 m brings fractional variances from W1 and W3 in different subdomains closer together for smaller scales (i.e., less than 0.5 zi) but has little affect at scales on the order of zi and larger (Figs. 13d–f).
The same conclusion can be reached for vertical profiles of normalized L50 and L95 scales for the ensemble of subdomains shown in Fig. 14 (cf. Fig. 8). Despite significant differences in surface fluxes, boundary layer depth, and other properties among various locations of analyzed subdomains, L50,w/zi in the boundary layer and
As in Figs. 8a–d, but for various subdomains from W1 and W3 simulations. Solid lines and filled symbols correspond to the L50 scale and dashed lines and open symbols correspond to the L95 scale. Yellow and blue shading indicate ranges of scales among four different corner sites in W1 and W3 simulations, respectively. Profiles for the default LASSO simulation L14_E413 are also shown for reference.
Citation: Monthly Weather Review 150, 6; 10.1175/MWR-D-21-0244.1
It is notable that the range of values among the corner subdomains is relatively large (note the logarithmic scale of the x axis) and comparable to the spread among various WRF-LES runs plotted for the same central subdomain and shown in Fig. 8. LASSO simulations, lacking surface heterogeneity and topography effects, as well as advection of any spatially coherent features into the subdomain, result in notably smaller L95/zi than WRF-LES at t = 14 h, particularly for θ and qυ. While this may not have a strong impact on the size of the largest clouds this early in the afternoon, since at this time
Overall, the results in this section show that the location of the subdomain has a strong effect on variability scales, particularly those of temperature and moisture. At the same time, the relationships among variability scales for different variables, as well as for simulations using different horizontal grid spacings, described in the previous section, appear to hold.
4. Summary and conclusions
The presented study examines the dependency of characteristics of a simulated continental convective boundary layer on various aspects of LES setup. Two groups of simulations of the case observed on 30 August 2016 during the HI-SCALE field campaign (Fast et al. 2019b) are conducted using the same model (WRF) and are different primarily in horizontal domain size and lateral and bottom boundary conditions (Table 1). The first group includes LASSO simulations (Gustafson et al. 2017, 2020) on a smaller O(10 × 10) km2 doubly periodic horizontal domain that are driven by prescribed horizontally uniform surface sensible and latent heat fluxes and profiles of temperature and moisture tendencies representing large-scale advection effects. WRF-LES runs in the second group are conducted on a larger O(100 × 100) km2 horizontal domain and use reanalysis-based lateral boundary conditions and an interactive land surface model with realistic topography and land cover types. In both configurations, grid spacings of 100 and 300 m are tested and additional sensitivity runs explore effects of domain size, large-scale advective forcing, background horizontal wind, and inhomogeneity in initial fields. Comparative analysis performed for a common subdomain focuses on the horizontally averaged boundary layer height, sensible and latent heat fluxes, variance profiles, and variance distribution across spatial (horizontal) scales. For a quantitative comparison of characteristic scales of horizontal variability in vertical velocity, potential temperature, and water vapor mixing ratio, we use the median variability scale L50, defined such that scales smaller and larger than that contribute equally to the total variance, and L95 scale, for which 95% of the variance is contained in features smaller than that scale.
The key findings include the following:
-
Simulations using the two setups form two distinct groups in terms of BL depth evolution, with zi being on average nearly 500 m higher in WRF-LES than in LASSO from late morning to midafternoon. The key features that separate these configurations are domain size (significantly larger in WRF-LES) and bottom boundary condition (inhomogeneous in WRF-LES and homogeneous in LASSO). Both these features favor the development of significant mesoscale variability in WRF-LES. Other setup details affect zi evolution to a lesser extent. Coarsening grid spacing from 100 to 300 m, for example, increases zi by less than 100 m. Differences in local, i.e., within the analyzed subdomain, sensible and latent surface heat fluxes among WRF-LES runs also have a relatively small effect on zi, although simulations with higher SH flux tend to develop deeper boundary layers.
-
Simulations using a 100-m horizontal grid spacing show that in a well-developed convective boundary layer L50,w/zi ≤ 0.3, L50,θ/zi ≤ 0.5, and
-
In simulations using a coarser horizontal grid spacing of 300 m, the L50 scales are overestimated for all considered variables. This is true, although to a smaller degree, even when variability in the range between 100 and 300 m is excluded from consideration in higher-resolution simulations (Figs. 6d–f). This is not an unexpected finding given that the actually resolved structures are often considered to be on the order of six grid spacings, which for a 300-m grid brings it up to the boundary layer depth (≈2 km). In the considered case, W3 simulations begin to correctly resolve vertical velocity features at 4Δx since their fractional variance of w for scales larger than 1.2 km converges to that of the reference simulation (Fig. 6d). The sensitivity of the variance distribution across scales to model grid spacing is important to keep in mind since using grid spacings larger than 100 m is not uncommon, especially in studies focused on cloudy boundary layers and parameterization development (e.g., Bretherton and Blossey 2017; de Roode et al. 2019; Schalkwijk et al. 2015; Shin and Hong 2015; Xu et al. 2018).
-
Nearly quadrupling the area of the LASSO horizontal domain to 25 × 25 km2 has only a minor and nonsystematic effect on scales of variability and, therefore, on characteristic horizontal size of simulated features for all considered variables (cf. simulations L14_E413 and L25_E413). LASSO has switched to using this larger domain for its routine modeling of shallow convection. Presented results suggest that while this change may improve simulation statistics and reduce noise in time series of domain mean quantities (Gustafson et al. 2020), such a moderate increase in the domain size does not appear to lead to the formation of significantly larger structures in the analyzed variables, at least in the considered case of a continental boundary layer with shallow convection. This behavior of LASSO runs is likely a consequence of applying horizontally uniform forcing for surface fluxes and large-scale vertical motions and advective tendencies.
-
In WRF-LES runs using a larger domain, heterogeneous surface properties, and interactive land surface model, the development of internal circulations and mesoscale patterns on scales of tens and hundreds of kilometers affect the mean state and variability inside an analyzed subdomain. It is worth noting that even in the simulations referred to as “no advection,” the horizontal velocity components are only set to zero at the boundaries of the 297 × 297 km2 WRF-LES domain. Freely evolving internal circulations inside this domain are not precluded and can produce “large-scale” advection tendencies for any given locale, even when the advective tendencies averaged over the whole domain zero out. These internal circulations in WRF-LES are likely driven by the spatial inhomogeneity of surface properties, such as topography, soil moisture, and land use type (Chen et al. 2020), but secondary flow patterns can also develop from more random features, such as deep convective clouds and cold pools over parts of the domain (Fig. 12).
The presented comparative analysis provides several insights into the interpretation of simulation results from a LASSO-type setup. Surface forcing (fluxes) observed at a particular location can be representative of a traditional LES scale domain O(10 × 10) km2 around that site but can deviate significantly for comparable domains at other locations few tens of kilometers away. At the same time, mean temperature and moisture tendencies derived for a region O(100 × 100) km2 do not necessarily represent the actual atmospheric forcing for a central, or any other, smaller O(10 × 10) km2 subregion inside. Consequently, combining such surface and atmospheric forcing may represent neither the local nor mean condition very well. This may complicate a comparison between small domain simulations and local observations because the realism of simulations will depend strongly on how well the mean advective tendencies represent condition over that location. This study finds that significant mesoscale variability is present across the forcing area even for a relatively uniform continental conditions over the U.S. Southern Great Plains. Potentially stronger effects may be imposed by greater variability of surface properties (e.g., Talbot et al. 2012) and steeper terrain (e.g., Chow et al. 2006).
The presented results focus on the properties and characteristic scales of a well-developed convective boundary layer, and a detailed analysis of overlaying clouds is outside the scope of the current study. For details about observed and modeled cloud properties for the considered case the readers are referred to Fast et al. (2019a), Chen et al. (2020), and Sakaguchi et al. (2022). It must be acknowledged, however, that clouds with bases located around the bottom of the inversion layer may affect the convective BL structure (e.g., Lareau et al. 2018). Using an ensemble of semi-idealized, or LASSO-like, LES, Brown et al. (2002) found that well-established convective boundary layer scalings are largely applicable to the subcloud layer in a case similar to the one considered here, which lends credence to the presented characteristic scale analysis for the early afternoon hours. That said, the effects of the LES setup options on various interactions and feedbacks between clouds and a subcloud layer, as discussed, for example, by van Stratum et al. (2014), remain to be investigated. Furthermore, later in the afternoon, although analyzed subdomains remain essentially precipitation free, deeper precipitating clouds appear in other parts of the larger model domain and begin exerting significant influence on the considered boundary layer properties in the WRF-LES, including through cold pool propagation. This highlights the importance of considering processes occurring outside the immediate neighborhood of a location of interest under such conditions and underscores the difficulty of accounting for mesoscale and synoptic effects in a semi-idealized model setup.
By running ensemble simulations for each case using different large-scale advective tendencies from a variety of sources, LASSO provides a measure of the sensitivity of the results to the forcing. For the case of 30 August 2016, differences among LASSO runs with three considered forcing options appear to be significantly smaller than differences between LASSO and WRF-LES ensembles or the variability among differently positioned subdomains in the WRF-LES. While these considerations are gleaned primarily from the analysis of boundary layer height and variability scales, they are likely applicable to other aspects of simulations as well. Schemann et al. (2020) explored an additional model configuration in which a small-domain LES is constrained by the fields from a mesoscale model at the lateral boundaries, thereby foregoing periodic boundary conditions. That setup demonstrated an improvement in simulated local cloud statistics, and in the future it will be interesting to apply a scale analysis to such simulations to establish possible propagation of larger-scale features into the LES domain.
Acknowledgments.
This study was supported by the U.S. Department of Energy Office of Science Biological and Environmental Research as part of the Atmospheric Systems Research (ASR) Program. The Pacific Northwest National Laboratory is operated by Battelle for the U.S. Department of Energy under Contract DE-AC05-76RL01830. Computing resources for the simulations were provided by the National Energy Research Scientific Computing Center (NERSC), a DOE Office of Science user facility supported under Contract DE-AC02-05CH11231; the ARM Data Center Computing Facility; and the Environmental Molecular Science Laboratory’s (EMSL) on its computational cluster Cascade. We thank three anonymous reviewers for their comments that helped improve presentation of the results.
Data availability statement.
The WRF community model data are available from the National Center for Atmospheric Research (NCAR) at http://www2.mmm.ucar.edu/wrf/users. Forcing data are available from the ARM facility archive, https://www.arm.gov, sponsored by the DOE Office of Science. Output from LASSO simulations is available at https://doi.org/10.5439/1342961.
REFERENCES
Angevine, W. M., J. Olson, J. Kenyon, W. I. Gustafson, S. Endo, K. Suselj, and D. D. Turner, 2018: Shallow cumulus in WRF parameterizations evaluated against LASSO large-eddy simulations. Mon. Wea. Rev., 146, 4303–4322, https://doi.org/10.1175/MWR-D-18-0115.1.
Barthlott, C., and C. Hoose, 2015: Spatial and temporal variability of clouds and precipitation over Germany: Multiscale simulations across the “gray zone.” Atmos. Chem. Phys., 15, 12 361–12 384, https://doi.org/10.5194/acp-15-12361-2015.
Bauer, H. S., S. K. Muppa, V. Wulfmeyer, A. Behrendt, K. Warrach-Sagi, and F. Spath, 2020: Multi-nested WRF simulations for studying planetary boundary layer processes on the turbulence-permitting scale in a realistic mesoscale environment. Tellus, 72A, 1–28, https://doi.org/10.1080/16000870.2020.1761740.
Beare, R. J., 2014: A length scale defining partially-resolved boundary-layer turbulence simulations. Bound.-Layer Meteor., 151, 39–55, https://doi.org/10.1007/s10546-013-9881-3.
Bopape, M. J. M., R. S. Plant, and O. Coceal, 2020: Resolution dependence of turbulent structures in convective boundary layer simulations. Atmosphere, 11, 986, https://doi.org/10.3390/atmos11090986.
Bretherton, C. S., and P. N. Blossey, 2017: Understanding mesoscale aggregation of shallow cumulus convection using large-eddy simulation. J. Adv. Model. Earth Syst., 9, 2798–2821, https://doi.org/10.1002/2017MS000981.
Brown, A. R., and Coauthors, 2002: Large-eddy simulation of the diurnal cycle of shallow cumulus convection overland. Quart. J. Roy. Meteor. Soc., 128, 1075–1093, https://doi.org/10.1256/003590002320373210.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface-hydrology model with the Penn State-NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585, https://doi.org/10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Chen, J. Y., S. Hagos, H. Xiao, J. D. Fast, and Z. Feng, 2020: Characterization of surface heterogeneity-induced convection using cluster analysis. J. Geophys. Res. Atmos., 125, e2020JD032550, https://doi.org/10.1029/2020JD032550.
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, https://doi.org/10.1175/JAM2322.1.
Cook, D. R., 2018: Soil Temperature and Moisture Profile (STAMP) system instrument handbook. Rep. DOE/SC-ARM-TR-186, 23 pp., https://www.arm.gov/publications/tech_reports/handbooks/stamp_handbook.pdf.
de Roode, S. R., P. G. Duynkerke, and H. J. J. Jonker, 2004: Large-eddy simulation: How large is large enough? J. Atmos. Sci., 61, 403–421, https://doi.org/10.1175/1520-0469(2004)061<0403:LSHLIL>2.0.CO;2.
de Roode, S. R., and Coauthors, 2019: Turbulent transport in the gray zone: A large eddy model intercomparison study of the CONSTRAIN cold air outbreak case. J. Adv. Model. Earth Syst., 11, 597–623, https://doi.org/10.1029/2018MS001443.
Deardorff, J. W., 1974: Three-dimensional numerical study of turbulence in an entraining mixed layer. Bound.-Layer Meteor., 7, 199–226, https://doi.org/10.1007/BF00227913.
Deardorff, J. W., 1980: Stratocumulus-capped mixed layers derived from a three-dimensional model. Bound.-Layer Meteor., 18, 495–527, https://doi.org/10.1007/BF00119502.
Dorrestijn, J., D. T. Crommelin, A. P. Siebesma, and H. J. J. Jonker, 2013: Stochastic parameterization of shallow cumulus convection estimated from high-resolution model data. Theor. Comput. Fluid Dyn., 27, 133–148, https://doi.org/10.1007/s00162-012-0281-y.
Efstathiou, G. A., R. S. Plant, and M. J. M. Bopape, 2018: Simulation of an evolving convective boundary layer using a scale-dependent dynamic Smagorinsky model at near-gray-zone resolutions. J. Appl. Meteor. Climatol., 57, 2197–2214, https://doi.org/10.1175/JAMC-D-17-0318.1.
Fast, J. D., L. K. Berg, Z. Feng, F. Mei, R. Newsom, K. Sakaguchi, and H. Xiao, 2019a: The impact of variable land-atmosphere coupling on convective cloud populations observed during the 2016 HI-SCALE field campaign. J. Adv. Model. Earth Syst., 11, 2629–2654, https://doi.org/10.1029/2019MS001727.
Fast, J. D., and Coauthors, 2019b: Overview of the HI-SCALE field campaign: A new perspective on shallow convective clouds. Bull. Amer. Meteor. Soc., 100, 821–840, https://doi.org/10.1175/BAMS-D-18-0030.1.
Gristey, J. J., G. Feingold, I. B. Glenn, K. S. Schmidt, and H. Chen, 2020: Surface solar irradiance in continental shallow cumulus fields: Observations and large-eddy simulation. J. Atmos. Sci., 77, 1065–1080, https://doi.org/10.1175/JAS-D-19-0261.1.
Gustafson, W. I., and Coauthors, 2017: Atmospheric Radiation Measurement (ARM) research facility. LASSO Data Bundles. 36° 36′ 18.0″N, 97° 29′ 6.0″W: Southern Great Plains Central Facility (C1). ARM Data Archive, accessed 23 February 2022, https://doi.org/10.5439/1342961.
Gustafson, W. I., and Coauthors, 2019: Description of the LASSO data bundles product. Rep. DOE/SC-ARM-TR-216, U.S. Department of Energy, Office of Science, 171 pp., https://www.osti.gov/servlets/purl/1469590.
Gustafson, W. I., and Coauthors, 2020: The Large-Eddy Simulation (LES) Atmospheric Radiation Measurement (ARM) Symbiotic Simulation and Observation (LASSO) activity for continental shallow convection. Bull. Amer. Meteor. Soc., 101, E462–E479, https://doi.org/10.1175/BAMS-D-19-0065.1.
Heinze, R., C. Moseley, L. N. Boske, S. K. Muppa, V. Maurer, S. Raasch, and B. Stevens, 2017a: Evaluation of large-eddy simulations forced with mesoscale model output for a multi-week period during a measurement campaign. Atmos. Chem. Phys., 17, 7083–7109, https://doi.org/10.5194/acp-17-7083-2017.
Heinze, R., and Coauthors, 2017b: Large-eddy simulations over Germany using ICON: A comprehensive evaluation. Quart. J. Roy. Meteor. Soc., 143, 69–100, https://doi.org/10.1002/qj.2947.
Honnert, R., V. Masson, and F. Couvreux, 2011: A diagnostic for evaluating the representation of turbulence in atmospheric models at the kilometric scale. J. Atmos. Sci., 68, 3112–3131, https://doi.org/10.1175/JAS-D-11-061.1.
Honnert, R., and Coauthors, 2020: The atmospheric boundary layer and the “gray zone” of turbulence: A critical review. J. Geophys. Res. Atmos., 125, 26, https://doi.org/10.1029/2019JD030317.
Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A. Clough, and W. D. Collins, 2008: Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models. J. Geophys. Res., 113, D13103, https://doi.org/10.1029/2008JD009944.
Jonker, H. J. J., P. G. Duynkerke, and J. W. M. Cuijpers, 1999: Mesoscale fluctuations in scalars generated by boundary layer convection. J. Atmos. Sci., 56, 801–808, https://doi.org/10.1175/1520-0469(1999)056<0801:MFISGB>2.0.CO;2.
Kang, S. L., 2020: Effects of mesoscale surface heterogeneity on the afternoon and early evening transition of the atmospheric boundary layer. Bound.-Layer Meteor., 174, 371–391, https://doi.org/10.1007/s10546-019-00493-w.
Kang, S. L., and J. H. Ryu, 2016: Response of moist convection to multi-scale surface flux heterogeneity. Quart. J. Roy. Meteor. Soc., 142, 2180–2193, https://doi.org/10.1002/qj.2811.
Lareau, N. P., Y. Y. Zhang, and S. A. Klein, 2018: Observed boundary layer controls on shallow cumulus at the ARM Southern Great Plains site. J. Atmos. Sci., 75, 2235–2255, https://doi.org/10.1175/JAS-D-17-0244.1.
Lee, J. M., Y. Y. Zhang, and S. A. Klein, 2019: The effect of land surface heterogeneity and background wind on shallow cumulus clouds and the transition to deeper convection. J. Atmos. Sci., 76, 401–419, https://doi.org/10.1175/JAS-D-18-0196.1.
Li, Z. J., S. Feng, Y. G. Liu, W. Y. Lin, M. H. Zhang, T. Toto, A. M. Vogelmann, and S. Endo, 2015a: Development of fine-resolution analyses and expanded large-scale forcing properties: 1. Methodology and evaluation. J. Geophys. Res. Atmos., 120, 654–666, https://doi.org/10.1002/2014JD022245.
Li, Z. J., J. C. McWilliams, K. Ide, and J. D. Farrara, 2015b: A multiscale variational data assimilation scheme: Formulation and illustration. Mon. Wea. Rev., 143, 3804–3822, https://doi.org/10.1175/MWR-D-14-00384.1.
Liu, S. Y., and X. Z. Liang, 2010: Observed diurnal cycle climatology of planetary boundary layer height. J. Climate, 23, 5790–5809, https://doi.org/10.1175/2010JCLI3552.1.
Lock, A. P., A. R. Brown, M. R. Bush, G. M. Martin, and R. N. B. Smith, 2000: A new boundary layer mixing scheme. Part I: Scheme description and single-column model tests. Mon. Wea. Rev., 128, 3187–3199, https://doi.org/10.1175/1520-0493(2000)128<3187:ANBLMS>2.0.CO;2.
McPherson, R. A., and Coauthors, 2007: Statewide monitoring of the mesoscale environment: A technical update on the Oklahoma Mesonet. J. Atmos. Oceanic Technol., 24, 301–321, https://doi.org/10.1175/JTECH1976.1.
Moeng, C. H., and J. C. Wyngaard, 1984: Statistics of conservative scalars in the convective boundary-layer. J. Atmos. Sci., 41, 3161–3169, https://doi.org/10.1175/1520-0469(1984)041<3161:SOCSIT>2.0.CO;2.
NCEP FNL Operational Model Global Tropospheric Analyses, 2000: NCEP FNL Operational Model Global Tropospheric Analyses, continuing from July 1999. National Center for Atmospheric Research, Computational and Information Systems Laboratory, accessed 1 September 2021, https://doi.org/10.5065/D6M043C6.
Neggers, R. A. J., A. P. Siebesma, and T. Heus, 2012: Continuous single-column model evaluation at a permanent meteorological supersite. Bull. Amer. Meteor. Soc., 93, 1389–1400, https://doi.org/10.1175/BAMS-D-11-00162.1.
Neggers, R. A. J., P. J. Griewank, and T. Heus, 2019: Power-law scaling in the internal variability of cumulus cloud size distributions due to subsampling and spatial organization. J. Atmos. Sci., 76, 1489–1503, https://doi.org/10.1175/JAS-D-18-0194.1.
Orszag, S. A., and I. Staroselsky, 2000: CFD: Progress and problems. Comput. Phys. Commun., 127, 165–171, https://doi.org/10.1016/S0010-4655(00)00032-1.
Patton, E. G., P. P. Sullivan, and C. H. Moeng, 2005: The influence of idealized heterogeneity on wet and dry planetary boundary layers coupled to the land surface. J. Atmos. Sci., 62, 2078–2097, https://doi.org/10.1175/JAS3465.1.
Pergaud, J., V. Masson, S. Malardel, and F. Couvreux, 2009: A parameterization of dry thermals and shallow cumuli for mesoscale numerical weather prediction. Bound.-Layer Meteor., 132, 83–106, https://doi.org/10.1007/s10546-009-9388-0.
Roy, S. B., and R. Avissar, 2000: Scales of response of the convective boundary layer to land-surface heterogeneity. Geophys. Res. Lett., 27, 533–536, https://doi.org/10.1029/1999GL010971.
Sakaguchi, K., and Coauthors, 2022: Determining spatial scales of soil moisture—Cloud coupling pathways using semi‐idealized simulations. J. Geophys. Res. Atmos., 127, e2021JD035282, https://doi.org/10.1029/2021JD035282.
Schalkwijk, J., H. J. J. Jonker, A. P. Siebesma, and E. Van Meijgaard, 2015: Weather forecasting using GPU-based large-eddy simulations. Bull. Amer. Meteor. Soc., 96, 715–723, https://doi.org/10.1175/BAMS-D-14-00114.1.
Schemann, V., K. Ebell, B. Pospichal, R. Neggers, C. Moseley, and B. Stevens, 2020: Linking large-eddy simulations to local cloud observations. J. Adv. Model. Earth Syst., 12, e2020MS002209, https://doi.org/10.1029/2020MS002209.
Shin, H. H., and S. Y. Hong, 2013: Analysis of resolved and parameterized vertical transports in convective boundary layers at gray-zone resolutions. J. Atmos. Sci., 70, 3248–3261, https://doi.org/10.1175/JAS-D-12-0290.1.
Shin, H. H., and S. Y. Hong, 2015: Representation of the subgrid-scale turbulent transport in convective boundary layers at gray-zone resolutions. Mon. Wea. Rev., 143, 250–271, https://doi.org/10.1175/MWR-D-14-00116.1.
Shin, H. H., and J. Dudhia, 2016: Evaluation of PBL parameterizations in WRF at subkilometer grid spacings: Turbulence statistics in the dry convective boundary layer. Mon. Wea. Rev., 144, 1161–1177, https://doi.org/10.1175/MWR-D-15-0208.1.
Smolarkiewicz, P. K., and J. M. Prusa, 2002: VLES modelling of geophysical fluids with nonoscillatory forward-in-time schemes. Int. J. Numer. Methods Fluids, 39, 799–819, https://doi.org/10.1002/fld.330.
Stoll, R., J. A. Gibbs, S. T. Salesky, W. Anderson, and M. Calaf, 2020: Large-eddy simulation of the atmospheric boundary layer. Bound.-Layer Meteor., 177, 541–581, https://doi.org/10.1007/s10546-020-00556-3.
Sullivan, P. P., and E. G. Patton, 2011: The effect of mesh resolution on convective boundary layer statistics and structures generated by large-eddy simulation. J. Atmos. Sci., 68, 2395–2415, https://doi.org/10.1175/JAS-D-10-05010.1.
Talbot, C., E. Bou-Zeid, and J. Smith, 2012: Nested mesoscale large-eddy simulations with WRF: Performance in real test cases. J. Hydrometeor., 13, 1421–1441, https://doi.org/10.1175/JHM-D-11-048.1.
Thompson, G., P. R. Field, R. M. Rasmussen, and W. D. Hall, 2008: Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization. Mon. Wea. Rev., 136, 5095–5115, https://doi.org/10.1175/2008MWR2387.1.
van Laar, T. W., V. Schemann, and R. A. J. Neggers, 2019: Investigating the diurnal evolution of the cloud size distribution of continental cumulus convection using multiday LES. J. Atmos. Sci., 76, 729–747, https://doi.org/10.1175/JAS-D-18-0084.1.
van Stratum, B. J. H., J. Vila-Guerau de Arellano, C. C. van Heerwaarden, and H. G. Ouwersloot, 2014: Subcloud-layer feedbacks driven by the mass flux of shallow cumulus convection over land. J. Atmos. Sci., 71, 881–895, https://doi.org/10.1175/JAS-D-13-0192.1.
Wulfmeyer, V., S. K. Muppa, A. Behrendt, E. Hammann, F. Spath, Z. Sorbjan, D. D. Turner, and R. M. Hardesty, 2016: Determination of convective boundary layer entrainment fluxes, dissipation rates, and the molecular destruction of variances: Theoretical description and a strategy for its confirmation with a novel lidar system synergy. J. Atmos. Sci., 73, 667–692, https://doi.org/10.1175/JAS-D-14-0392.1.
Wurps, H., G. Steinfeld, and S. Heinz, 2020: Grid-resolution requirements for large-eddy simulations of the atmospheric boundary layer. Bound.-Layer Meteor., 175, 179–201, https://doi.org/10.1007/s10546-020-00504-1.
Wyngaard, J. C., 2004: Toward numerical modeling in the “terra incognita.” J. Atmos. Sci., 61, 1816–1826, https://doi.org/10.1175/1520-0469(2004)061<1816:TNMITT>2.0.CO;2.
Xu, H. X., Y. J. Wang, and M. Z. Wang, 2018: The performance of a scale-aware nonlocal PBL scheme for the subkilometer simulation of a deep CBL over the Taklimakan desert. Adv. Meteor., 2018, 8759594, https://doi.org/10.1155/2018/8759594.
Yang, L. M., and Coauthors, 2018: A new generation of the United States National Land Cover Database: Requirements, research priorities, design, and implementation strategies. ISPRS J. Photogramm. Remote Sens., 146, 108–123, https://doi.org/10.1016/j.isprsjprs.2018.09.006.