An Improved Surface Boundary Condition for Large-Eddy Simulations Based on Monin–Obukhov Similarity Theory: Evaluation and Consequences for Grid Convergence in Neutral and Stable Conditions
Abstract
Monin–Obukhov similarity theory is used in large-eddy simulation (LES) models as a surface boundary condition to predict the surface shear stress and scalar fluxes based on the gradients between the surface and the first grid level above the surface. We outline deficiencies of this methodology, such as the systematical underestimation of the surface shear stress, and propose a modified boundary condition to correct for this issue. The proposed boundary condition is applied to a set of LES for both neutral and stable boundary layers with successively decreasing grid spacing. The results indicate that the proposed boundary condition reliably corrects the surface shear stress and the sensible heat flux, and improves grid convergence of these quantities. The LES data indicate improved grid convergence for the surface shear stress, more so than for the surface heat flux. This is either due to a limited performance of the Monin–Obukhov similarity functions or due to problems in the LES model in representing stable conditions. Furthermore, we find that the correction achieved using the proposed boundary condition does not lead to improved grid convergence of the wind-speed and temperature profiles. From this we conclude that the sensitivity of the wind-speed and temperature profiles in the LES model to the grid spacing is more likely related to under-resolved near-surface gradients and turbulent mixing at the boundary-layer top, to the SGS model formulation, and/or to numerical issues, and not to deficiencies due to the use of improper surface boundary conditions.
Keywords
Grid convergence Large-eddy simulation Logarithmic layer mismatch Monin–Obukhov similarity theory Stable boundary layer1 Introduction
One persisting problem in large-eddy simulation (LES) of the atmospheric boundary layer is the so-called logarithmic layer mismatch, the fact that the simulated wind-speed profile does not match the predicted logarithmic relation, a direct result of the inherent inability of LES models to resolve locally the dominant (small) eddies close to the surface. In this region, the subgrid-scale (SGS) turbulence parametrization dominates the flow. Also, the ability to resolve near-surface vertical gradients depends strongly on the grid spacing involved. It is commonly found that the mean wind shear is thus overestimated near the surface (Sullivan et al. 1994; Khanna and Brasseur 1997; Brasseur and Wei 2010).
The structure of the atmospheric surface layer can be essentially described through the turbulent exchange of momentum, heat, and moisture with the surface. Monin–Obukhov similarity theory (MOST, Monin and Obukhov 1954) provides a solid mathematical framework to describe this exchange in terms of turbulent fluxes and atmospheric stability. MOST includes the logarithmic law-of-the-wall in neutral conditions under the assumption of a constant-flux layer. In reality, however, the turbulent fluxes usually change with height from the surface value to zero at the top of the boundary layer. There is evidence from field and numerical experiments, though, that MOST provides reliable estimates of the surface fluxes and has been used for more than 50 years (Foken 2006). The vertical extent of the surface layer is commonly defined to be that region in which the turbulent fluxes do not vary more than \(\approx 10\%\) of their surface values. Due to the linear decrease with height (observed in steady-state conditions) the depth of the surface layer can be loosely estimated to be \(\approx 10\%\) of the boundary-layer depth.
To the authors’ knowledge, MOST is nowadays used in most state-of-the-art LES models for atmospheric boundary-layer flows as a surface boundary condition to calculate the surface Reynolds stress and the turbulent surface sensible and latent heat fluxes (e.g., Heus et al. 2010; Maronga et al. 2015; van Heerwaaren et al. 2017). LES models typically use grid spacings at a metre scale, since in recent decades, due to continuously increasing computational power of state-of-the-art supercomputers, the grid spacing has decreased from about 100 m (Deardorff 1980) down to 1 m (e.g., Maronga and Bosveld 2017) or even less (e.g., Sullivan et al. 2016). Especially under stable conditions, the dominant eddies are often smaller than 10 m and thus demand very fine grids (Beare et al. 2006; Sullivan et al. 2016; Maronga and Bosveld 2017).
The effect of grid resolution on simulation results has been studied by various authors (Sullivan et al. 1994; Khanna and Brasseur 1997; Beare et al. 2006; Brasseur and Wei 2010; Maronga and Bosveld 2017), and might be linked to the shortcomings outlined above (Brasseur and Wei 2010). In particular, it is often observed that grid convergence for simulations of the stable boundary layer is lacking, see Beare et al. (2006) and Sullivan et al. (2016). The latter used fine grid spacings down to 0.36 m (pseudo-spectral code) and still reported a sensitivity of their results to the grid spacing. Until now, a convincing explanation for this behaviour has been lacking, creating a limitation for the application of LES models for simulating the stable boundary layer. From our own experience, and in line with previous research, we found a non-convergence of the surface shear stress and heat flux for both neutral and stable boundary layers, suggesting that the non-convergence in LES models in the stable boundary layer might be related to the outlined issues of the surface boundary condition. The motivation for the present study thus was to, (a) develop a reliable methodology (i.e., a surface boundary condition for LES models) to correct the surface fluxes due to excessive wind-speed and temperature gradients close to the surface, and (b) assess the chosen surface boundary condition as a possible reason for the lack of grid convergence in neutral and particularly in stable conditions.
It is known that the application of MOST as a boundary condition imposes further limitations and problems. First, when using grid spacings in the order of a few metres, the application of MOST can become problematic. Several researchers pointed out that MOST can only be considered valid within the inertial sublayer, the upper part of the surface layer, but fails in the roughness sublayer below, in which direct effects of single surface roughness elements are present (Garratt 1980; Raupach 1992; Physick and Garratt 1995; Harman and Finnigan 2007; Basu and Lacser 2017). Garratt (1980) found, for high vegetation and neutral and unstable conditions, that the lower boundary of the inertial sublayer (\(z_*\)), i.e., the top of the roughness sublayer, depends on the average horizontal distance between trees (\(\delta \)), and estimated \(z_*\approx 3\delta \), which corresponded to \(z_*= 35z_0\), where \(z_0\) is the roughness length. Physick and Garratt (1995) noted that \(z_*\) might be much smaller for stable conditions. For a similar experimental site reported by Physick and Garratt (1995), \(z_*\) was found to be around \(50z_0\) instead. Note that these values are site and stratification specific and are taken as some first proxy at this point. Garratt (1980) and Harman and Finnigan (2007) suggested corrections for the MOST relationships below \(z_*\) in order to include the roughness sublayer. This correction, however, was derived for a specific forest canopy and for neutral atmospheric stratification only. Referring to this previous work, Basu and Lacser (2017) suggested considering \(z_*= 50z_0\) as a general rule for LES models. As an example, following this suggestion and for typical roughness lengths for surfaces covered with low vegetation (heights in the order of 0.1 m), this imposes a minimum vertical grid spacing of 5 m for LES models with a common MOST boundary condition. Basu and Lacser (2017) reported correctly that, in practice, MOST also is applied for much smaller grid spacings, despite this violation of MOST assumptions and in the lack of an alternative (e.g., Beare et al. 2006; Basu et al. 2011; Maronga 2014; Sullivan et al. 2016; Udina et al. 2016). This is particularly true for LES of the stable boundary layer, where grid spacings \(< 5~\hbox {m}\) are required to resolve the small-scale turbulence (Beare and MacVean 2004). The violation of MOST assumptions for very fine grid spacings can thus be considered one possible reason for the insufficient grid convergence observed in LES of the stable boundary layer. Contemporary atmospheric LES codes nevertheless apply MOST as a boundary condition between the surface and the first computational grid level above the surface, providing a further motivation for reviewing the possible implications of this practice.
The starting hypothesis for the present paper is that the lack of grid convergence observed in LES of neutral and stable boundary layers might be attributed to the outlined problems when using MOST as the surface boundary condition. Consequently, we extend the work of Kawai and Larsson (2012) and Maronga and Bosveld (2017) through a proper description and a thorough evaluation of an improved surface boundary condition for LES models for neutral and stable stability regimes based on MOST. The improved boundary condition is designed in such a way that the MOST assumptions are not violated and the surface fluxes are corrected to fit to the resolved profiles of wind speed and temperature. For this purpose we employ the LES model PALM (Maronga et al. 2015) and conduct a set of idealized LES of neutral and stable boundary layers.
The paper is organized as follows: Sect. 2 gives an overview of the current and the proposed surface boundary conditions for LES models. Section 3 describes the LES set-up, while results are presented in Sect. 4. Finally, a summary and outlook is presented in Sect. 5.
2 Methodology
In the following we first outline the state-of-the-art methods applied in LES models to calculate the surface fluxes of heat and momentum (Reynolds stress), and confine ourselves to the case of a dry atmosphere without the presence of humidity. The methodology can, however, be easily extended to incorporate humid air and hence the surface latent heat flux. In a second step, we describe an improved method that accounts for the excessive wind shear and temperature gradient near the surface, while simultaneously conserving the correlation between local surface fluxes and the flow adjacent to the surface.
2.1 Traditional Surface Boundary Condition
One common method that avoids this overestimation was developed by Schumann (1975) and was improved by Grötzbach (1987) (the so-called SG method, named after Schumann and Grötzbach, see Hultmark et al. 2013). This method is based on solving the averaged form of the equations above and imposing a local variation based on the ratio \(u_\mathrm{h}(z_1){/}{<}\,u_\mathrm{h}\,{>}(z_1)\). For a detailed description of the SG method and more elaborate methods, see Piomelli et al. (1989), Marusic et al. (2001), Stoll and Porte-Agel (2006), and Hultmark et al. (2013). While these methods might improve the boundary condition for purely neutral flows, they do not consider the effect of atmospheric stability. Also, their application is limited to very idealized cases as the surface is required (for most methods) to be entirely homogeneous. Many LES models, including the model PALM used in the present study, employ the IL method, which is purely local.
2.2 Proposed Surface Boundary Condition
2.2.1 Requirements for the Evaluation Height
- 1.This level must be within the atmospheric surface layer (also referred to as constant-flux layer). The surface-layer height H can be estimated to an extent over the lowest 10% of the boundary-layer height \(z_\mathrm{i}\),As several definitions exist for defining \(z_\mathrm{i}\), which also differ for different stability regimes, this remains a rather loose criterion. The chosen height should thus be as close to the surface as possible.$$\begin{aligned} z_{\mathrm{sl}} \le 0.1 z_\mathrm{i} = H\;. \end{aligned}$$(16)
- 2.The level must be within the inertial sublayer and thus above the roughness sublayer. Here, we follow (Basu and Lacser 2017) and demand thatNote that the coefficient of 50 here is the recommendation of Basu and Lacser (2017), but that it might depend on stability and the characteristics of the roughness elements (see Garratt 1980; Physick and Garratt 1995). Also, note that this requirement could be eliminated by adding a correction term for the roughness sub-layer as discussed, e.g., by Harman and Finnigan (2007).$$\begin{aligned} z_{\mathrm{sl}} \ge z_*\approx 50 z_0\;. \end{aligned}$$(17)
- 3.The flow at height \(z_{\mathrm{sl}}\) must be well-resolved by the model, a requirement that is not a physical constraint, but a numerical requirement. The height \(z_{\mathrm{sl}}\) is thus highly dependent on the model (including its SGS scheme). For the PALM model (static Deardorff SGS model), previous studies have found that the mean profiles follow MOST at height levels starting from the seventh grid level above the surface (Maronga 2014; Maronga and Reuder 2017). As the PALM model uses a Cartesian staggered grid with wind speeds and scalars defined at the vertical centre of the grid boxes, the requirement for the PALM model iswhere \(\varDelta z\) is the grid spacing in the z-direction. Note that this height is model-specific and will also depend on the numerical schemes involved. The value reported here is valid for PALM’s 5th-order advection scheme (Wicker and Skamarock 2002).$$\begin{aligned} z_{\mathrm{sl}} \ge 6.5 \varDelta z\;, \end{aligned}$$(18)
3 Experimental Set-Up
3.1 Set-Up of Neutral Boundary Layer
First, an idealized neutral boundary layer with a geostrophic wind speed of \(5~\hbox {m s}^{-1}\) and \(z_0 = 0.1~\hbox {m}\) was chosen to test the new boundary condition without taking into account effects of static stability. The horizontal domain size for all runs was about \(2 000~\hbox {m} \times 2 000~\hbox {m}\), which is in the range of domain sizes found to be sufficient for simulating neutral boundary layers (see Andren et al. 1994; Drobinski and Foster 2003; Foster et al. 2006; Ludwig et al. 2009; Knigge and Raasch 2016). The Coriolis parameter was set to a latitude of \(73^\circ \hbox {N}\) (to be consistent with the runs for stable conditions, see below). The wind speed was initialized by its geostrophic value and was constant with height. Initial perturbations were imposed on the velocity fields to trigger turbulence. Overall, 12 LES were performed, four using the IL (cases \(\hbox {IL}\_\hbox {d}X\)) and four the ESG method (cases \(\hbox {ESG}\_\hbox {d}X\), where X indicates the grid spacing used). Additionally, we conducted four runs with \(z_{\mathrm{sl}} = z_1\) (i.e., SG method) in order to separate the effect of taking an elevated level and the outlined general differences between IL and SG methods (cases \(\hbox {SG}\_\hbox {d}X\)). For each method, the equidistant grid spacing (\(\varDelta = \varDelta _x = \varDelta _y = \varDelta _z\)) was varied from 16 to 8 m, 4 m, and 2 m. For the IL method, \(z_{\mathrm{sl}}\) was defined by the numerical grid and thus set to \(0.5 \varDelta _z = z_1\), while for the ESG method \(z_{\mathrm{sl}}\) was chosen to be the height of the first grid level greater or equal 50 m. As \(z_\mathrm{i}\) varied between the runs from 880–1100 m, this was well within the inertial sublayer (equal to \(0.05 z_\mathrm{i}\)).
A known issue for simulations of the neutral boundary layer are inertial oscillations in the wind profile. By carefully looking at the time series of \(<u_*>\), we found that these oscillations are sufficiently damped throughout the boundary layer after 60 h of simulation time in all cases. In particular, we found that these oscillations are very weak in the lower boundary layer and thus they did not affect the results within the surface layer. The simulations all ran for 84 h and the data were averaged and analyzed during the last 24 h (14,400 samples). Note that the overbar symbol is used in the following to denote time-averaged quantities.
Table 1 gives an overview of the simulations performed for neutral conditions and whether the individual runs fulfil the requirements defined in Sect. 2.2.1. Note that all cases with IL and SG methods by implication fail to fulfil these requirements, while for the ESG method only case ESG_d16, i.e. with the coarsest grid, does not fulfil all requirements. The height of the surface layer was estimated to be \(H = 0.1 z_\mathrm{i}\), where \(z_\mathrm{i}\) was calculated as in Beare et al. (2006), as the height at which the mean shear stress fell to 5% of its surface value.
Summary of LES runs performed for neutral stratification
Case | \(\varDelta \) (m) | \(z_{\mathrm{sl}}\) (m) | H (m) | \(z_{\mathrm{sl}} \le 0.1 z_\mathrm{i}\) | \(z_{\mathrm{sl}} \ge 50 z_0\) | \(z_{\mathrm{sl}} \ge 6.5 \varDelta z\) | CPU (h) |
---|---|---|---|---|---|---|---|
\(\hbox {IL}\_\hbox {d}16^*\) | 16.0 | 8.0 | 110 | Yes | Yes | No | 650 |
\(\hbox {IL}\_\hbox {d}8^*\) | 8.0 | 4.0 | 86 | Yes | yes | No | 2350 |
\(\hbox {IL}\_\hbox {d}4^*\) | 4.0 | 2.0 | 79 | Yes | No | No | 49,050 |
\(\hbox {IL}\_\hbox {d}2^*\) | 2.0 | 1.0 | 88 | Yes | No | No | 310,900 |
\(\hbox {SG}\_\hbox {d}16^*\) | 16.0 | 8.0 | 101 | Yes | Yes | No | 600 |
\(\hbox {SG}\_\hbox {d}8^*\) | 8.0 | 4.0 | 87 | Yes | Yes | No | 2400 |
\(\hbox {SG}\_\hbox {d}4^*\) | 4.0 | 2.0 | 81 | Yes | No | No | 47,800 |
\(\hbox {SG}\_\hbox {d}2^*\) | 2.0 | 1.0 | 90 | Yes | No | No | 209,350 |
\(\hbox {ESG}\_\hbox {d}16^*\) | 16.0 | 56.0 | 99 | Yes | Yes | No | 450 |
\(\hbox {ESG}\_\hbox {d}8\) | 8.0 | 52.0 | 86 | Yes | Yes | Yes | 2450 |
\(\hbox {ESG}\_\hbox {d}4\) | 4.0 | 50.0 | 77 | Yes | Yes | Yes | 45,150 |
\(\hbox {ESG}\_\hbox {d}2\) | 2.0 | 51.0 | 99 | Yes | Yes | Yes | 127,431 |
3.2 Set-Up of Stable Boundary Layer
A weakly-stable boundary layer was used to evaluate the performance of the proposed boundary condition, with the set-up similar to that used in the GABLS1 model inter-comparison outlined in Beare et al. (2006). The initial potential temperature was set constant with height throughout the model domain to 265 K in our simulations (i.e., a stable boundary layer (SBL) developed in a neutrally-stratified environment). Note that in GABLS1, a capping inversion starting at a height of 100 m was used instead. A model domain of about \(500~\hbox {m} \times 500~\hbox {m} \times 500~\hbox {m}\) was used, with equidistant grid spacing, and which is sufficient for a SBL that reaches up to \(\approx 200~\hbox {m}\). Geographical latitude was set to \(73^\circ \hbox {N}\) as in GABLS1. In order to stimulate turbulence, a random potential temperature perturbation of amplitude 0.1 K was applied for height levels of up to 50 m. The vertical velocity was set to zero at the surface and top boundary of the domain, with top boundary conditions set to free-slip conditions. A continuous surface cooling rate of \(0.25~\hbox {K h}^{-1}\) was applied to create and maintain a stably-stratified flow. The geostrophic wind speed was set to \(8~\hbox {m s}^{-1}\) and the roughness was set as in the neutral simulations to \(z_0 = z_{0h} = 0.1~\hbox {m}\). As the simulated SBL was significantly shallower than for the neutral cases, i.e. \(z_\mathrm{i} = 150{-}170~\hbox {m}\), we followed a more rigid concept and fixed \(z_{\mathrm{sl}}\) to exactly 13 m. Unlike for neutral conditions, where a height in the middle of the surface layer was employed, we decided to use a height value very close to the top of the surface layer. This was done in order to use as coarse grid spacings as possible while simultaneously fulfilling all requirements for the ESG method. The grid spacings were then varied in such a way that a prognostic level was exactly located at a height of 13 m. Five cases were hence performed, each using the IL and the ESG method with grid spacings of 5.2 m, 2.8 m, 1.37 m, and 1.04 m. Note that the choice of \(z_{\mathrm{sl}}\) is always case-specific and must be adjusted depending on simulation set-up.
The simulations ran for 18 h, which is somewhat longer than in the GABLS1 case. As for neutral conditions, we observed inertial oscillations in the neutral layer above the boundary layer, which did not penetrate into the boundary layer itself. As the surface was continuously cooled, no true steady temperature state could be achieved, but we observed convergence of time series (e.g., friction velocity) after about 16 h. As direct consequence, a shorter time-averaging of 1 h was applied as in Beare et al. (2006) from hours 17 to 18 (600 samples). We also tested other averaging intervals and earlier averaging periods, but found no effects on the results.
Table 2 summarizes the cases performed for stable conditions and their key parameters.
Summary of LES runs performed for stable stratification
Case | \(\varDelta \) (m) | \(z_{\mathrm{sl}}\) (m) | H (m) | L (m) | \(z_{\mathrm{sl}} \le 0.1 z_\mathrm{i}\) | \(z_{\mathrm{sl}} \ge 50 z_0\) | \(z_{\mathrm{sl}} \ge 6.5 \varDelta z\) | CPU (h) |
---|---|---|---|---|---|---|---|---|
\(\hbox {IL}\_\hbox {d}5\_\hbox {sbl}^*\) | 5.2 | 2.6 | 16 | 98 | Yes | No | No | 39 |
\(\hbox {IL}\_\hbox {d}3\_\hbox {sbl}^*\) | 2.8 | 1.4 | 17 | 101 | Yes | No | No | 562 |
\(\hbox {IL}\_\hbox {d}2\_\hbox {sbl}^*\) | 2.0 | 1.0 | 16 | 101 | Yes | No | No | 3046 |
\(\hbox {IL}\_\hbox {d}15\_\hbox {sbl}^*\) | 1.37 | 0.685 | 15 | 97 | Yes | No | No | 9537 |
\(\hbox {IL}\_\hbox {d}1\_\hbox {sbl}^*\) | 1.04 | 0.52 | 14 | 95 | Yes | No | No | 42,768 |
\(\hbox {ESG}\_\hbox {d}5\_\hbox {sbl}^*\) | 5.2 | 13.0 | 17 | 151 | Yes | Yes | No | 38 |
\(\hbox {ESG}\_\hbox {d}3\_\hbox {sbl}^*\) | 2.8 | 13.0 | 17 | 168 | Yes | Yes | No | 594 |
\(\hbox {ESG}\_\hbox {d}2\_\hbox {sbl}\) | 2.0 | 13.0 | 16 | 172 | Yes | Yes | Yes | 3007 |
\(\hbox {ESG}\_\hbox {d}15\_\hbox {sbl}\) | 1.37 | 13.0 | 16 | 176 | Yes | Yes | Yes | 10,560 |
\(\hbox {ESG}\_\hbox {d}1\_\hbox {sbl}\) | 1.04 | 13.0 | 15 | 182 | Yes | Yes | Yes | 43,960 |
\(\hbox {DYN}\_\hbox {d}5\_\hbox {sbl}^*\) | 5.2 | 2.6 | 18 | 105 | Yes | No | No | 40 |
\(\hbox {DYN}\_\hbox {d}3\_\hbox {sbl}^*\) | 2.8 | 1.4 | 17 | 103 | Yes | No | No | 550 |
\(\hbox {DYN}\_\hbox {d}2\_\hbox {sbl}^*\) | 2.0 | 1.0 | 16 | 100 | Yes | No | No | 3100 |
\(\hbox {DYN}\_\hbox {d}15\_\hbox {sbl}^*\) | 1.37 | 0.685 | 15 | 96 | Yes | No | No | 9850 |
\(\hbox {DYN}\_\hbox {d}1\_\hbox {sbl}^*\) | 1.04 | 0.52 | 14 | 98 | Yes | No | No | 42,500 |
4 Results
In order to evaluate the two different applied boundary conditions, we first compare the results of IL and ESG methods for purely neutral conditions and thus neglect the possible complications imposed by stratification. In the second part of this section, we evaluate the performance and implications for stable conditions.
4.1 Neutral Boundary Layer
In neutral conditions, the boundary condition is reduced to the prediction of the surface shear stress and thus provides an ideal prerequisite for analyzing the behaviour of LES using different methods to determine this stress.
4.1.1 Logarithmic-Layer Mismatch
In order to evaluate the effect of using the horizontally-averaged wind speed instead of the local value (IL method) separately, we additionally plotted the SG method data in Fig. 2 (red lines). As expected, the behaviour is similar to the IL method close to the surface (first grid point). At higher levels, the SG method is closer to the theoretical MOST profile than the IL method, which is indicating higher \(u_*\) values compared to the IL method runs. We might ascribe this to the fact that the horizontal mean gradient is used for calculating the surface shear stress instead of the local gradient in the IL method so that the excessive wind shear near the surface is smaller. However, note that the overestimation is still well pronounced (see Fig. 2), so that the SG method alone does not resolve the logarithmic layer mismatch.
Figure 2 (blue lines) shows the same data for the ESG method. Here we immediately recognize that the wind speed profiles agree remarkably well to the predicted ones and are also much better than for the SG method. This result is in agreement with Kawai and Larsson (2012), who showed similar plots in their study of a neutral boundary layer configuration. The main reason for this is the higher value of \(u_*\) (and thus \(<\overline{u_*}>\)) which was calculated to match the wind speed at a height around 50 m instead of the first grid level above surface. Furthermore, it is noteworthy that for coarser grids (cases ESG_d16 and ESG_d8) we observe the expected overestimation of the wind speed near the surface, being a direct implication of the correction involved in the ESG method, which provokes a shift of the dimensionless wind-speed profiles towards smaller values (as the scaling parameter \(<\overline{u_*}>\) is higher). Rather surprising is the fact that this overestimation is not visible for finer grid spacings (cases ESG_d4 and ESG_d2). Here, the ESG method apparently is able to also correct the excessive wind shear near the surface. Also, it should be noted that the ESG method also seems to provide a good correction in cases when the surface layer is not well-resolved and the criteria for the application of MOST as a boundary condition are not all met (i.e., case ESG_d16).
4.1.2 Effect on Grid Convergence
The SG method, though showing consistently slightly larger values than the IL method, apparently does not improve grid convergence of \(<u_*>\). For the cases using the ESG method, however, we note that \(<u_*>\) is generally higher than for the IL and SG methods, an expected consequence of the excessive wind shear leading to smaller values in the IL method. The data also reveal a weak decrease in the mean value with decreasing grid spacing from \(0.26~\hbox {m s}^{-1}\) in case ESG_d16 to \(0.25~\hbox {m s}^{-1}\) in case ESG_d4. This corresponds to a decrease by 2.3%, which is thus only half of the decrease observed for the IL method, indicating better convergence for the ESG method. However, we also observe a slight increase in the friction velocity from case ESG_d4 to ESG_d2, which might be suggestive to approach grid convergence. Note that the minimum and maximum values have converged, though. We thus ascribe the small increase in \(<\overline{u_*}>\) from case ESG_d4 to ESG_d2 to dynamic instabilities in the flow that temporarily alter the mean state of the boundary layer (streaks or roll-like structure, see above), but which are probably not resolved at coarser grid spacings. In general, we can conclude from Fig. 3 that the grid convergence of the surface shear stress is improved when using the ESG method.
Statistics of the simulated friction velocity regarding their horizontal variability after 84 h of simulation time
Case | Mean (m s\(^{-1}\)) | \(\sigma _{u_*}\,(\mathrm{m\,s}^{-1})\) | Skewness | Kurtosis |
---|---|---|---|---|
IL_d16 | 0.243 | 0.032 | 0.50 | 2.72 |
IL_d8 | 0.236 | 0.039 | 0.39 | 2.68 |
IL_d4 | 0.230 | 0.042 | 0.26 | 3.02 |
IL_d2 | 0.232 | 0.053 | 0.17 | 3.23 |
ESG_d16 | 0.259 | 0.018 | \(-\) 0.05 | 2.44 |
ESG_d8 | 0.257 | 0.022 | 0.07 | 2.63 |
ESG_d4 | 0.253 | 0.026 | 0.12 | 2.72 |
ESG_d2 | 0.255 | 0.032 | 0.11 | 2.83 |
To summarize the results for neutral conditions, it was shown that the ESG method corrects the calculated surface shear stress efficiently for the exceeding wind shear near the surface. For the cases studied, however, grid convergence of the wind-speed profiles was not improved, which suggests that the finest grid spacing of 2 m was possibly not sufficient to resolve the near-surface wind-speed profile good enough. Also, deficiencies of the SGS model might play an important role here. Nevertheless, the overall differences in \(<\overline{u_*}>\) between the runs with different grid spacings were not found to be larger than 2.3%, which is smaller than the 4.5% observed for the IL method, and which appears tolerable for LES applications.
4.2 Stable Boundary Layer
The SBL provides a much more complex test bed for the evaluation of the ESG method, mainly because the Obukhov length now is an additional scaling parameter that accounts for the combined effect of shear production and buoyancy. The Obukhov length is used in the similarity functions to correct the logarithmic profile. Furthermore, not only is a boundary condition required for \(u_*\), but one is also required for \(\theta _*\). In the next sections we will generally follow the same strategy as for the neutral boundary layer set-up and first evaluate the IL against the ESG method for linking \(u_*\) and \(\theta _*\) to the surface layer profiles of wind speed and temperature. In a second step we will then discuss the issue of grid convergence.
4.2.1 Logarithmic-Layer Mismatch
Figure 8 (blue lines), however, reveals slightly different results than for neutral conditions. For stable conditions, we first note that the ESG method is (as for neutral conditions) effective in closing the gap between theoretical and LES profiles (see differences between solid and dashed lines in Fig. 8) and thus provides a good correction for \(u_*\). However, here we also note that there is a continuous tendency in the wind-speed profiles. For cases ESG_d5_sbl and ESG_d3_sbl, the LES profiles suggest lower wind speeds than predicted by MOST, while for cases ESG_d15_sbl and ESG_d1_sbl, the LES profiles more and more overestimates the theoretical profile. Before we try to find an explanation for this sensitivity to the grid spacing, we first analyze the temperature profiles.
4.2.2 Effect on Grid Convergence
In the previous section, we identified two sensitivities when using the ESG method. First, the wind-speed profiles showed an increasing overestimation of the theoretical profile (or in other words, a too high \(u_*\)) with decreasing grid spacing. Second, the temperature profiles showed the opposite tendency, underestimating the theoretical ones with decreasing grid spacing (i.e., too low \(\theta _*\)). The respective statistics for \(u_*\) and \(\theta _*\) are shown in Fig. 10a, b. As for neutral stratification, we note that \(u_*\) is about 23% higher for the ESG method when compared to the IL method, an implication from the correction of the exceeding wind speed near the surface. Besides, it is evident that there is much better grid convergence for the ESG method compared to the IL method. In contrast, \(\theta _*\) reveals that for both methods grid convergence is lacking and \(\theta _*\) is monotonically decreasing with decreasing grid spacing, showing an almost linear trend for grid spacings of 2 m and less. For the ESG method we observe generally lower values of \(\theta _*\) than for the IL method, which is consistent with the findings from our discussion of Fig. 9 (see above). Mathematically, a smaller \(\theta _*\) results in larger values of L as can be inferred from Eq. 8 when entered in Eq. 2. Figure 10c (see also Table 2) shows the statistics of L for both methods. Indeed, we observe a continuous increase in L for the ESG method. However, the opposite trend is observed for the IL method. As L is defined through both \(u_*\) and \(\theta _*\), the trend of decreasing surface shear stress with decreasing grid spacing also comes into play. As \(u_*\) has a higher weight in the calculation of L (as its number enters squared), the resulting L reflects basically the trend of \(u_*\) for the IL method. In contrast, \(u_*\) for the ESG method was found to be rather constant (i.e., converged) so that the trend in \(\theta _*\) then dominates the behaviour of L, leading to different trends of L for IL and ESG methods.
The finding that L increases with decreasing grid spacing for the ESG method while it remains rather constant or slightly decreasing for the IL method would indicate a different effect on the wind-speed and temperature profiles. For the ESG method we might expect stronger and larger eddies with increasing values of L (by increasing wind shear and/or smaller surface heat flux) and hence possibly a greater depth of the boundary layer, while we would suspect that the opposite is the case for the IL method. Interestingly, Fig. 11 reveals that for both methods, the boundary-layer depth decreases to the same amount (and that apparently there is neither a correlation of the boundary-layer depth changes to L nor to \(u_*\)). This rather surprising finding leads us to the conclusion that the lack of grid convergence here is not linked to the surface fluxes and stresses. The variance profiles shown in Fig. 12 indicate a monotonically decreasing mixing with decreasing grid spacing throughout the boundary layer, giving further support to the fact that no grid convergence is reached (the same is found for the heat-flux profiles, not shown). Our findings are in line with Sullivan et al. (2016) who also investigated the GABLS1 case with grid spacings down to 0.39 m and showed similar results for mean profiles (thus suggesting a similar decrease in the turbulent mixing). We thus ascribe this to generally under-resolved turbulence in the LES runs. It remains unclear, however, why the variance profiles (or the total TKE, not shown) decrease with decreasing grid spacing, as more and more turbulence should be resolved. It goes beyond the scope of this paper to further investigate this. While Fig. 11 suggests that differences are largest in the upper boundary layer, these still might be the result of inadequate resolution of the turbulence near the surface. This might lead to an erroneous surface friction and result in different boundary-layer depths, which in turn are most prominently visible at the top of the boundary layer. However, we can also suspect that mixing processes near the top of the boundary layer might contribute to the grid dependence. Moreover, numerical dissipation of the advection scheme which is a known feature for large wavenumbers might contribute to the grid dependence. Furthermore, the SGS model formulation might play an important role as suggested by previous studies (e.g., Porte-Agel 2000; Porte-Agel et al. 2004; Basu and Porte-Agel 2006; Lu and Porte-Agel 2013). Previous studies that simulated the GABLS1 case could achieve much better grid convergence when using dynamic SGS models (Beare et al. 2006; Basu and Porte-Agel 2006; Lu and Porte-Agel 2013). Note, however, that the GABLS1 case differs from the present set-up by a capping inversion, which might inhibit the growth of the boundary layer and might lead to under-resolved turbulence in the entrainment zone. The capping inversion thus potentially has an effect on grid convergence. A more detailed analysis of these processes goes beyond the scope of the present study. For a more rigorous discussion on under-resolved turbulence in stable conditions, see Sullivan et al. (2016). In summary, however, we must note that the observed grid convergence effects are likely LES model dependent.
5 Summary and Outlook
In the present study we introduced an improved boundary condition based on MOST for use in LES models. The main concept behind this boundary condition is the use of an elevated level for evaluating the MOST relationships in order to calculate the surface shear stress and scalar fluxes, instead of being limited to the first grid level above the surface. This concept is based on the previous work for neutral conditions by Kawai and Larsson (2012) and was here extended to non-neutral conditions. By taking into account three criteria for the choice of this elevated level, violation of the assumptions of MOST as outlined by Basu and Lacser (2017) is consistently avoided. Also, the improved method ensures the correlation between the near-surface turbulent structures and the surface fluxes, and thus does not corrupt the flow features near the surface. For evaluation of the improved boundary condition, we conducted a comprehensive set of LES for both a neutral and a weakly stable boundary-layer configuration using both conventional boundary conditions based on evaluating MOST at the first grid level and with the proposed new boundary condition at an elevated level.
The results of the performed LES indicate that the improved boundary condition is a solid method to avoid discrepancies between the surface fluxes (shear stress and surface sensible heat) and the resolved surface-layer wind-speed and temperature profiles by the LES model under neutral and stable conditions. For neutral conditions, it could be shown that the surface shear stress is not only corrected towards larger values that are in line with the resolved wind-speed profile, but it also effectively corrects the excessive wind shear observed near the surface. We also found that the correction implied by the improved boundary condition performed better for the surface shear stress than for the surface sensible heat flux in stable conditions. We could ascribe this to the fact that under stable conditions, both \(u_*\) and \(\theta _*\) must be calculated. These are linked via a single parameter, L. Here, we found that the calculated L represented a compromise to achieve best agreement with both the simulated wind-speed and temperature profiles, but has an inherent bias to better represent the wind-speed profile. Moreover, we could show that the simulated boundary layers with the proposed boundary condition are deeper (under stable conditions by 6%), which is an expected result as the correction of the surface shear stress involves higher values of \(u_*\) and thus increased turbulent mixing.
The starting hypothesis for the present study was that the well-known lack of grid convergence in LES of the SBL might be attributed to the lack of grid convergence of the surface shear stress and the surface sensible heat flux. Our results gave evidence that this is not the case and that there is no link between the differences in the mean profiles and differences in the surface fluxes, particularly for stable conditions. Instead we suppose that LES models with current resolution are still not able to capture strong curvature very close to the surface, even at grid spacings in the order of 1 m, which might explain the observed tendency of shallower boundary layers for finer grid spacings. Previous studies also suggested that the formulation of the SGS model is a key aspect for grid convergence. Furthermore, numerical aspects like numerical dissipation, the reduction of higher-order advection schemes near the surface, or effects at the top of the boundary layer might be possible causes for the observed lack of grid convergence. LES runs using very sophisticated SGS models and a vertical nesting approach with very fine grid spacings near the surface would be a logical continuation to investigate at what grid spacing true grid convergence can be achieved and what the responsible processes are. We are currently working on the implementation of such a nesting approach in the PALM model and plan to tackle this issue in a follow-up study.
Our results clearly demonstrate that the proposed boundary condition eliminates key issues in the application of MOST as a boundary condition in LES models and can thus be considered as an addition or alternative to advanced SGS schemes in order to avoid the excessive near-surface wind shear and generate more realistic boundary-layer heights in LES models. Also, the MOST-based boundary conditions proposed by Marusic et al. (2001) and Hultmark et al. (2013) can probably be improved based on our findings. However, our proposed method also has some limitations. First, as horizontally-averaged values are used to calculate the surface fluxes, the method is not applicable in the presence of surface heterogeneity, imposed for instance by heterogeneous land use or roughness as in such cases the atmospheric state in an elevated level might no longer be representative for the underlying surface (e.g., due to an internal boundary layer). Furthermore, the method is difficult to apply in complex terrain or in the presence of buildings, so usage might be restricted to academic idealized studies. Second, Eq. 13 in its current form is only valid for non-zero mean wind speeds and needs to be modified to be used under free convective conditions, i.e, by falling back to the IL method in such cases to preserve a minimum friction velocity (see also Schumann 1987). Furthermore, note that we did not vary the parameters \(z_{\mathrm{sl}}\) and \(z_0\) in our simulated cases. Hultmark et al. (2013) and Stoll and Porte-Agel (2006), however, showed that there might be sensitivities on the results. More rigorous testing of the robustness of the methodology and the dependence of the results on surface roughness are desirable. Due to the computational demands involved, particularly for the finest grid spacings used, we could only do a few test runs where we varied \(z_{\mathrm{sl}}\) and therein we did not see a strong sensitivity of our general results and findings. Note, however, that a height-dependence of the results is to be expected as the constant-flux layer is a theoretical construct and, in reality (or the model), fluxes do vary with height within the surface layer.
Finally, in order to relax the strict requirements for the choice of the grid spacing in LES models, we plan to incorporate the roughness sublayer in the boundary condition formulated as outlined by Harman and Finnigan (2007).
Notes
Acknowledgements
We would like to thank Sukanta Basu and the three anonymous reviewers for their very detailed and constructive comments that helped to improve the manuscript significantly. The present paper is part of the ISOBAR project funded by the Research Council of Norway (RCN) under the FRINATEK scheme (Project Number: 251042/F20). The first author would like to thank Bert Holtslag at Wageningen University, Netherlands, and Joachim Reuder at University of Bergen, Norway, for fruitful discussions on the topic. All simulations were performed on the Cray XC40 at The North-German Supercomputing Alliance (HLRN), Berlin. NCL (The NCAR Command Language (Version 6.1.2) [Software]. (2013). Boulder, Colorado: UCAR/NCAR/CISL/VETS. https://doi.org/10.5065/D6WD3XH5) was used for data processing and visualization.
References
- Andren A, Brown AR, Graf J, Mason PJ, Moeng CH, Nieuwstadt FTM, Schumann U (1994) Large-eddy simulation of a neutrally stratified boundary layer: a comparison of four computer codes. Q J R Meteorol Soc 120:1457–1484CrossRefGoogle Scholar
- Basu S, Lacser A (2017) A cautionary note on the use of Monin–Obukhov similarity theory in very high-resolution large-eddy simulations. Boundary-Layer Meteorol 163:351–355. https://doi.org/10.1007/s10546-016-0225-y CrossRefGoogle Scholar
- Basu S, Porte-Agel F (2006) Large-eddy simulation of stably stratified atmospheric boundary layer turbulence: a scale-dependent dynamic modeling approach. J Atmos Sci 63:2074–2091CrossRefGoogle Scholar
- Basu S, Holtslag A, Bosveld FC (2011) GABLS3 LES intercomparison study. In: ECMWF/GABLS workshop on “Diurnal cycles and the stable atmospheric boundary layer”, ECMWF, pp 75–82Google Scholar
- Beare RJ, MacVean MK (2004) Resolution sensitivity and scaling of large-eddy simulations of the stable boundary layer. Boundary-Layer Meteorol 112:257–281CrossRefGoogle Scholar
- Beare RJ, Cortes MAJ, Cuxart J, Esau I, Golaz C, Holtslag AAM, Khairoutdinov M, Kosovic B, Lewellen D, Lund T, Lundquist J, Mccabe A, Macvean MK, Moene A, Noh Y, Poulos G, Raasch S, Sullivan P (2006) An intercomparison of large-eddy simulations of the stable boundary layer. Boundary-Layer Meteorol 118:247–272CrossRefGoogle Scholar
- Bou-Zeid E, Meneveau C, Parlange MB (2005) A scale-dependent lagrangian dynamic model for large eddy simulation of complex turbulent flows. Phys Fluids 17:025,105. https://doi.org/10.1063/1.1839152 CrossRefGoogle Scholar
- Brasseur JG, Wei T (2010) Designing large-eddy simulation of the turbulent boundary layer to capture law-of-the-wall scaling. Phys Fluids 22:021303CrossRefGoogle Scholar
- Deardorff JW (1980) Stratocumulus-capped mixed layers derived from a three-dimensional model. Boundary-Layer Meteorol 18:495–527CrossRefGoogle Scholar
- Drobinski P, Foster RC (2003) On the origin of near-surface streaks in the neutrally-stratified planetary boundary layer. Boundary-Layer Meteorol 108:247–256CrossRefGoogle Scholar
- Foken T (2006) 50 years of the Monin-Obukhov similarity theory. Boundary-Layer Meteorol 119:431–447. https://doi.org/10.1007/s10546-006-9048-6 CrossRefGoogle Scholar
- Foster RC, Vianey F, Drobinski P, Carlotti P (2006) Near-surface coherent structures and the vertical momentum flux in a large-eddy simulation of the neutrally-stratified boundary layer. Boundary-Layer Meteorol 120:229–255CrossRefGoogle Scholar
- Garratt JR (1980) Surface influence upon vertical profiles in the atmospheric near-surface layer. Q J R Meteorol Soc 106:803–819CrossRefGoogle Scholar
- Grachev AA, Andreas EL, Fairall CW, Guest PS, Persson POG (2007) SHEBA flux-profile relationships in the stable atmospheric boundary layer. Boundary-Layer Meteorol 124:315–333. https://doi.org/10.1007/s10546-007-9177-6 CrossRefGoogle Scholar
- Grötzbach G (1987) complex flow phenomena and modeling, encyclopedia of fluid mechanics, vol 6, Gulf Publishing, chap Direct numerical and large eddy simulations of turbulent channel flows, pp 1337–1391Google Scholar
- Harman IN, Finnigan JJ (2007) A simmple unified theory for flow in the canopy and roughness sublayer. Boundary-Layer Meteorol 123:339–363. https://doi.org/10.1007/s10546-006-9145-6 CrossRefGoogle Scholar
- Heinz S (2008) Realizability of dynamic subgrid-scale stress models via stochastic analysis. Monte Carlo Methods Appl 14(4):311–329CrossRefGoogle Scholar
- Heus T, Van Heerwaarden CC, Jonker HJJ, Siebesma AP, Axelsen S, Dries K, Geoffroy O, Moene AF, Pino D, De Roode SR, Vilà-Guerau de Arellano J (2010) Formulation of the dutch atmospheric large-eddy simulation (DALES) and overview of its applications. Geosci Model Dev 3:415–444CrossRefGoogle Scholar
- Hultmark M, Calaf M, Parlange MB (2013) A new wall shear stress model for atmospheric boundary layer simulations. J Atmos Sci 70:3460–3470. https://doi.org/10.1175/JAS-D-12-0257.1 CrossRefGoogle Scholar
- Kawai S, Larsson J (2012) Wall-modeling in large eddy simulation: length scales, grid resolution, and accuracy. Phys Fluids 24:1–11. https://doi.org/10.1063/1.3678331 CrossRefGoogle Scholar
- Khanna S, Brasseur JG (1997) Analysis of Monin-Obukhov similarity from large-eddy simulations. J Fluid Mech 345:251–286CrossRefGoogle Scholar
- Knigge C, Raasch S (2016) Improvement and development of one- and two-dimensional discrete gust models using a large-eddy simulation model. J Wind Engin Indust Aerodyn 153:46–59. https://doi.org/10.1016/j.jweia.2016.03.004 CrossRefGoogle Scholar
- Knigge C, Auerswald T, Raasch S, Bange J (2015) Comparison of two different methods that provide highly resolved atmospheric turbulence data for simulations of stall effects. Comput Fluids 108:57–66. https://doi.org/10.1016/j.compfluid.2014.11.005 CrossRefGoogle Scholar
- Lu H, Porte-Agel F (2013) A modulated gradient model for scalar transport in large-eddy simulation of the atmospheric boundary layer. Phys Fluids 25:015,110. https://doi.org/10.1063/1.4774342 CrossRefGoogle Scholar
- Ludwig FL, Chow FK, Street RL (2009) Resolved velocity structures and momentum fluxes in large-eddy simulations of neutral boundary layer flow. J Appl Meteorol Climat 48:1161–1180CrossRefGoogle Scholar
- Maronga B (2014) Monin-Obukhov similarity functions for the structure parameters of temperature and humidity in the unstable surface layer: results from high-resolution large-eddy simulations. J Atmos Sci 71:716–733CrossRefGoogle Scholar
- Maronga B, Bosveld FC (2017) Key parameters for the life cycle of nocturnal radiation fog: a comprehensive large-eddy simulation study. Q J R Meteorol Soc 143:2463–2480. https://doi.org/10.1002/qj.3100 CrossRefGoogle Scholar
- Maronga B, Reuder J (2017) On the formulation and universality of Monin-Obukhov similarity functions for mean gradients and standard deviations in the unstable surface layer: results from surface-layer resolving large-eddy simulations. J Atmos Sci 74:989–1010. https://doi.org/10.1175/JAS-D-16-0186.1 CrossRefGoogle Scholar
- Maronga B, Heinze R, Hoffmann F, Kanani-Sühring F, Keck M, Ketelsen K, Letzel MO, Sühring M, Raasch S (2015) The parallelized large-eddy simulation model (PALM) version 4.0 for atmospheric and oceanic flows: Model formulation, recent developments, and future perspectives. Geosci Model Dev 8:2515–2551. https://doi.org/10.5194/gmd-8-2515-2015 CrossRefGoogle Scholar
- Marusic I, Kunkel GJ, Porte-Agel F (2001) Experimental study of wall boundary conditions for large-eddy simulation. J Fluid Mech 446:309–320. https://doi.org/10.1017/S0022112001005924 CrossRefGoogle Scholar
- Mokhtarpoor R, Heinz S (2017) Dynamic large eddy simulation: stability via realizability. Phys Fluids 29(10):105,104CrossRefGoogle Scholar
- Monin AS, Obukhov AM (1954) Basic laws of turbulent mixing in the atmosphere near the ground. Tr Akad Nauk SSSR Geofiz Inst 24:163–187Google Scholar
- Nikitin NV, Nicoud F, Wasistho B, Squired KD, Spalart PR (2000) An approach to wall modeling in large-eddy simulations. Phys Fluids 12:1629CrossRefGoogle Scholar
- Panofsky HA (1963) Determination of stress from wind and temperature measurements. Q J R Meteorol Soc 89:85–94. https://doi.org/10.1002/qj.49708937906 CrossRefGoogle Scholar
- Panofsky HA, Dutton JA (1984) Atmospheric turbulence, models and methods for engineering applications. Wiley, New YorkGoogle Scholar
- Physick WL, Garratt JR (1995) Incorporation of a high-roughness lower boundary into a mesoscale model for studies of dry deposition over complex terrain. Boundary-Layer Meteorol 74:55–71CrossRefGoogle Scholar
- Piomelli U, Ferziger J, Moin P, Kim J (1989) New approximate boundary conditions for large eddy simulations for wall-bounded flows. Phys Fluids 1A:1061. https://doi.org/10.1063/1.857397 CrossRefGoogle Scholar
- Porte-Agel F (2000) A scale-dependent dynamic model for scalar transport in large-eddy simulations of the atmospheric boundary layer. Boundary-Layer Meteorol 112:81–105CrossRefGoogle Scholar
- Porte-Agel F, Meneveau C, Parlange MB (2004) A scale-dependent dynamic model for large-eddy simulation: application to a neutral atmospheric boundary layer. J Fluid Mech 415:261–284. https://doi.org/10.1017/S0022112000008776 CrossRefGoogle Scholar
- Raasch S, Franke T (2011) Structure and formation of dust-devil-like vortices in the atmospheric boundary layer–a high resolution numerical study. J Geophys Res. https://doi.org/10.1029/2011JD016010
- Raupach MR (1992) Drag and drag partitioning on rough surfaces. Boundary-Layer Meteorol 60:375–395CrossRefGoogle Scholar
- Saiki EM, Moeng CH, Sullivan PP (2000) Large-eddy simulation of the stably stratified planetary boundary layer. Boundary Layer Meteorol 95:1–30CrossRefGoogle Scholar
- Schumann U (1975) Subgrid scale model for finite difference simulations of turbulent flows in plane channels and annuli. J Comput Phys 18:376–404CrossRefGoogle Scholar
- Schumann U (1987) Minimum friction velocity and heat transfer in the rough surface layer of a convective boundary layer. Boundary-Layer Meteorol 44:311–326CrossRefGoogle Scholar
- Stoll R, Porte-Agel F (2006) Effect of roughness on surface boundary conditions for large-eddy simulation. Boundary-Layer Meteorol 118:169–187. https://doi.org/10.1007/s10546-005-4735-2 CrossRefGoogle Scholar
- Sullivan PP, McWilliams JC, Moeng CH (1994) A subgrid-scale model for large-eddy simulation of planetary boundary-layer flows. Boundary-Layer Meteorol 71:247–276CrossRefGoogle Scholar
- Sullivan PP, Weil JC, Patton EG, Jonker HJJ, Mironov DV (2016) Turbulent winds and temperature fronts in large-eddy simulations of the stable atmospheric boundary layer. J Atmos Sci 73:1815–1840. https://doi.org/10.1175/JAS-D-15-0339.1 CrossRefGoogle Scholar
- Udina M, Sun J, Kosovic B, Soler MR (2016) Exploring vertical turbulence structure in neutrally and stably stratified flows using the weather research and forecasting-large-eddy simulation (WRF-LES) model. Boundary-Layer Meteorol 161:355–374. https://doi.org/10.1007/s10546-016-0171-8 CrossRefGoogle Scholar
- van Heerwaaren CC, van Stratum BJH, Heus T, Gibbs JA, Fedorovich E, Mellado JP (2017) MicroHH 1.0: a computational fluid dynamics code for direct numerical simulation and large-eddy simulation of atmospheric boundary layer flows. Geosci Model Dev 10:3145–3165. https://doi.org/10.5194/gmd-10-3145-2017 CrossRefGoogle Scholar
- Wicker LJ, Skamarock WC (2002) Time-splitting methods for elastic models using forward time schemes. Mon Wea Rev 130:2088–2097CrossRefGoogle Scholar
- Williamson JH (1980) Low-storage Runge-Kutta schemes. J Comput Phys 35:48–56CrossRefGoogle Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.