The GEWEX Atmospheric Boundary Layer Studies (GABLS) 1, 2 and 3 are used to develop a methodology for the design and testing of Reynolds-averaged Navier–Stokes (RANS) atmospheric boundary layer (ABL) models for wind energy applications. The first two GABLS cases are based on idealized boundary conditions and are suitable for verification purposes by comparing with results from higher-fidelity models based on large-eddy simulation. Results from three single-column RANS models, of 1st, 1.5th and 2nd turbulence closure order, show high consistency in predicting the mean flow. The third GABLS case is suitable for the study of these ABL models under realistic forcing such that validation versus observations from the Cabauw meteorological tower are possible. The case consists on a diurnal cycle that leads to a nocturnal low-level jet and addresses fundamental questions related to the definition of the large-scale forcing, the interaction of the ABL with the surface and the evaluation of model results with observations. The simulations are evaluated in terms of surface-layer fluxes and wind energy quantities of interest: rotor equivalent wind speed, hub-height wind direction, wind speed shear and wind direction veer. The characterization of mesoscale forcing is based on spatially and temporally averaged momentum budget terms from Weather Research and Forecasting (WRF) simulations. These mesoscale tendencies are used to drive single-column models, which were verified previously in the first two GABLS cases, to first demonstrate that they can produce similar wind profile characteristics to the WRF simulations even though the physics are more simplified. The added value of incorporating different forcing mechanisms into microscale models is quantified by systematically removing forcing terms in the momentum and heat equations. This mesoscale-to-microscale modeling approach is affected, to a large extent, by the input uncertainties of the mesoscale tendencies. Deviations from the profile observations are reduced by introducing observational nudging based on measurements that are typically available from wind energy campaigns. This allows the discussion of the added value of using remote sensing instruments versus tower measurements in the assessment of wind profiles for tall wind turbines reaching heights of 200 m.

Wind energy flow models are progressively incorporating more realistic atmospheric physics in order to improve the simulation capacity of wind turbine and wind farm design tools. Wind resource assessment and wind turbine site suitability tools, dealing with the microscale flow around and within a wind farm, have been traditionally based on site measurements and microscale flow models relying on Monin–Obukhov similarity theory (MOST; Monin and Obukhov, 1954) that assume steady-state and are typically applied in neutral atmospheric conditions. At larger scales (than microscale), the long-term wind climatology is typically determined from a combination of historical measurements and simulations from mesoscale meteorological models at a horizontal resolution of a few kilometers. The transition from mesoscale to microscale to come up with a unified model chain is the main challenge at stake for the next generation of wind assessment tools. In order to make this possible, microscale models have to extend their range to simulate the entire atmospheric boundary layer (ABL) and include relevant physics like Coriolis as well as realistic large-scale forcing and appropriate turbulent scaling, dependent on thermal stratification, from the surface layer to the free atmosphere. The dynamics of these forcings determine the interplay between the wind climatology, relevant for the assessment of the wind resource, and the wind conditions relevant for wind turbine siting. Sanz Rodrigo et al. (2016) reviews the state-of-the-art wind farm flow modeling, methodologies and challenges for mesoscale–microscale coupling.

The design of ABL models for wind energy requires a systematic approach to verification and validation in order to demonstrate consistency of the computational code with the conceptual physical model and to quantifying deviations with respect to the real world (Sanz Rodrigo et al., 2016). The verification process is carried out using idealized test cases where the solution is known from theory or from a higher-fidelity model (code-to-code comparison). Sensitivity analysis in idealized conditions also helps determine which are the main drivers of the model, which directly affect the quantities of interest, and anticipate their main sources of uncertainty. Validation, however, deals with code-to-observation comparison to quantify the accuracy of the model at representing the real world in terms of the application of interest. From the wind energy perspective, the quantities of interest are the wind conditions that are directly related to the production of energy and the design characteristics of wind turbines.

The GEWEX Atmospheric Boundary Layer Studies (GABLS) were developed by the atmospheric boundary layer community to benchmark single-column models, used by meteorological models to parameterize the ABL (Holtslag et al., 2013). While the cases are all based on observations of the ABL in relatively stationary and horizontally homogeneous conditions, it is notoriously difficult to define validation cases due to the interplay of a large number of physical processes that can modify these relatively simple conditions. Hence, the first two GABLS benchmarks used idealized conditions in order to analyze the turbulent structure of the ABL without the influence of the variability of the external large-scale forcing. GABLS1 simulated a quasi-steady stable boundary layer resulting from 9 h of uniform surface cooling (Cuxart et al., 2006). GABLS2 simulated a diurnal cycle, still with uniform geostrophic forcing, by simplifying measurements from the CASES-99 experiment in Kansas (Svensson et al., 2011). Under this idealized forcing, large-eddy simulation (LES) models have shown high consistency at predicting the ABL behavior (Beare et al., 2006). Therefore, they have been used to verify reduced-order models based on Reynolds-averaged Navier–Stokes (RANS) turbulence modeling. Hence, GABLS 1 and 2 are suitable verification cases for demonstrating the simulation capacity of ABL models for incorporating thermal stratification into turbulence modeling under uniform large-scale forcing and using prescribed surface boundary conditions.

GABLS1 showed that many boundary layer parameterizations tend to overestimate the turbulent mixing in stable conditions, leading to a too-deep boundary layer compared to LES simulations (Cuxart et al., 2006). GABLS2 showed the difficulties of comparing observations with simulations under idealized forcing and prescribed surface temperature. Holtslag et al. (2007) showed that during stable conditions there is strong coupling between the geostrophic wind speed and the surface temperature. Hence, prescribing the surface temperature inhibits the interaction of the boundary layer with the surface, which, for instance, results in large differences in the 2 m temperature predicted by the models.

The challenges of the first two GABLS exercises inspired the setup of GABLS3, which deals with a real diurnal case with a strong nocturnal low-level jet (LLJ) at the Cabauw meteorological tower in the Netherlands (Baas et al., 2009; Holtslag, 2014; Basu et al., 2011; Bosveld et al., 2014a). Here, large-scale forcing is not constant throughout the diurnal cycle but depends on time and height. Instead of prescribing the surface temperature, models are allowed to make use of their land surface schemes in order to include the dependencies between the ABL and the land surface models. The large-scale forcing is prescribed based on piece-wise linear approximations of the real forcing, derived from simulations with the Regional Atmospheric Climate Model (RACMO) mesoscale model and adjusted to match the observed surface geostrophic wind and the wind speed at 200 m. These approximations are introduced to limit the impact of the uncertainties associated with mesoscale geostrophic and advection forcing.

Based on the GABLS benchmark series, the challenges of stable boundary layers and diurnal cycles are reviewed by Holtslag et al. (2013): notably, the relation between enhanced mixing in operational weather models performance, the role of land surface heterogeneity in the coupling with the atmosphere, the development of LES models with interactive land surface schemes, the characterization of a climatology of boundary layer parameters (stability classes, boundary layer depth and surface fluxes), and the development of parameterizations for the very stable boundary layer when turbulence is not the dominant driver. These challenges are also shared by wind energy applications. Therefore, it is relevant to study the GABLS3 case within the wind energy context as a validation case with focus on rotor-based quantities of interest.

Revisiting GABLS3 for wind energy also means adopting a more pragmatic approach when it comes to adding physical complexity. In the context of developing a mesoscale-to-microscale model, it is important to identify which are the first-order physics that need to be incorporated to improve performance compared to current practices in the wind industry. For instance, adding thermal effects on turbulence modeling is important compared to the traditional hypothesis of neutral stratification, while the effects of humidity may be initially neglected.

Reducing model-chain uncertainties by using on-site observations is also particularly appealing for wind energy since it is standard practice to have profile measurements at the site. Since these measurements are typically affected by site effects, we propose introducing corrections at the microscale level based on profile nudging. Hence, contrary to the original GABLS3 setup, for the sake of a more generalized mesoscale-to-microscale methodology, we propose using the large-scale tendencies computed by a mesoscale model as driving forces at microscale without introducing any correction based on measurements. Then, at microscale, the simulation can be dynamically relaxed to the profile observations to correct the hour-to-hour bias. This is also a more natural way of dealing with the wind energy model chain using an asynchronous coupling methodology where (1) a database of input forcings is generated offline by a mesoscale model (in the context of a regional wind atlas for instance), (2) site effects are simulated by a microscale ABL model forced by these mesoscale inputs and introducing a high-resolution topographic model, and (3) deviations of the model with respect to a reference observational site are corrected to remove the bias generated throughout the downscaling process. It is important to note that strict validation shall not include site observations to be able to quantify the impact of the limited knowledge of the model. The final bias-correction step allows the calibration of the model to reduce the bias and provide a more accurate wind assessment in the application context. Quantifying the correction introduced by the nudging terms in the modeling equations and their relative weight with respect to the other terms can also be used to assess the limitations of the model.

The methodology used by Bosveld et al. (2014a) to characterize large-scale forcing from mesoscale simulations will be adopted here using simulations from the Weather Research and Forecasting (WRF) model. At microscale, we use a single-column model with three RANS turbulence closure schemes of 1st, 1.5th and 2nd order. This model chain was also used by Baas et al. (2009) to design the GABLS3 case and perform a sensitivity analysis of various single column model (SCM) settings. Following a similar philosophy, we evaluate the impact of different mesoscale forcing terms and bias-correction strategies on wind energy quantities of interest.

We follow the same modeling approach used by Baas et al. (2010) to define a microscale atmospheric boundary layer model driven by realistic mesoscale forcing. This one-way meso–micro methodology allows the coupling of the models offline, facilitating the generalization of the downscaling methodology to any combination of mesoscale and microscale models working asynchronously.

The RANS equations in natural Cartesian coordinates (

Similar to the momentum equations, the energy equation in the absence of
radiative and phase-change heat transfer effects relates the tendency of
potential temperature with the mesoscale advective temperature
(

The diffusion terms in Eqs. (1), (3) and (5) are simulated assuming an
isotropic eddy viscosity that relates turbulent fluxes with the gradients of
mean flow quantities:

The

Then, the eddy viscosity is defined as

Sogachev et al. (2012) define a modified

In neutral conditions, a relationship amongst

GABLS1

Surface boundary conditions are defined based on MOST using the simulated
surface-layer friction velocity

The GABLS1 case setup is described in Cuxart et al. (2006), based on LES
simulations presented by Kosovic and Curry (2000), where the boundary layer
is driven by a prescribed uniform geostrophic wind and surface cooling rate
over a horizontally homogeneous ice surface. The following initial and
boundary conditions are used:

Single-column model simulations are run for 9 h using a 1 km high
log-linear grid of 301 points and a time step of 1 s (Sanz Rodrigo and
Anderson, 2013). Figure 1 (left) shows surface fluxes and boundary layer
height, based on shear stress, for the three turbulence models and compared
with the

GABLS1 quasi-steady vertical profiles of horizontal wind speed

A sensitivity analysis of quasi-steady ABL profiles is shown in Fig. 3,
following the same simulation approach as GABLS1 and varying the surface
cooling rate CR and the geostrophic wind

Sensitivity analysis of quasi-steady profiles at different cooling rates
CR and geostrophic wind speed

It is important to note that the quasi-steady profiles resulting from the sensitivity analysis almost never happen in real conditions. They are canonical cases that help us parameterize the ABL without dynamical effects so that we can more easily study the relationship between the main drivers of the ABL. In real conditions, the ABL is a transient phenomena that not only depends on the actual boundary conditions but also on the hours to days of history leading to them.

While the second GABLS exercise was more strongly based on observations from the CASES-99 experiment in Kansas, from the ABL forcing perspective it can still be regarded as idealized. The case corresponds to 2 consecutive clear and dry days with a strong diurnal cycle. Since the focus of the study was the intercomparison of boundary layer schemes, the forcing conditions were simplified to facilitate the comparison among the various turbulent closures rather than an assessment of their accuracy against the actual observations.

The case setup and model intercomparison is described in Svensson et
al. (2011). The boundary conditions are prescribed in terms of a uniform
geostrophic wind of

GABLS2 surface temperature profile (Svensson et al., 2011) and alternative 48 h periodic cycle used to obtain a diurnal cycle independent of initial conditions.

GABLS2 time–height contour plots of wind velocity

Figure 1 (right) shows the surface fluxes and stability parameter of the
three turbulence models compared with the SCM results of the GABLS2 model
intercomparison of Svensson et al. (2011) and the LES results of Kumar et
al. (2010). The three models are within the scatter of the SCM reference
results and close to the LES results. Compared to the LES simulations, the

The GABLS3 setup is described in Bosveld et al. (2014a). The case analyzes
the period from 12:00 UTC 1 July to 12:00 UTC 2 July 2006 at the Cabauw
Experimental Site for Atmospheric Research (CESAR), located in the
Netherlands (51.971

Roughness map for a 30

Time–height contour plots of the longitudinal wind component

The CESAR measurements are carried out at a 200 m tower, free of obstacles up to a few hundred meters in all directions. The measurements include 10 min averaged vertical profiles of wind speed, wind direction, temperature and humidity at heights 10, 20, 40, 80, 140 and 200 m, as well as surface radiation and energy budgets. Turbulence fluxes are also monitored at four heights: 3, 60, 100 and 180 m. A RASS profiler measures wind speed, wind direction and virtual temperature above 200 m.

The selection criteria for GABLS3 consisted of the following filters applied
to a database of 6 years (2001–2006): stationary synoptic conditions, clear
skies (net long-wave cooling > 30 W m

Mesoscale forcing is derived from simulations with the Advanced Research Weather Forecasting model (WRF), version 3.8 (Skamarock et al., 2008). Kleczek et al. (2014) made a sensitivity study of WRF for different grid setups, boundary layer schemes, boundary conditions and spin-up time. Reasonably good results of the vertical wind profile in stable conditions (at midnight) are obtained, although the dependency on the PBL scheme and grid setup is important.

Mesoscale simulations are reproduced here using the same domain setup used as
reference by Kleczek et al. based on three concentric square domains centered
at the Cabauw site. The model is driven by 6-hourly ERA Interim reanalysis
data from ECMWF (European Centre for Medium-Range Weather Forecasts), which
come at a resolution of approximately 80 km. Three domains, all with
183

To derive mesoscale forcing, the momentum budget components (also called
tendencies) are directly extracted from WRF since they are computed by the
solver (Lehner, 2012). Curvature, due to the curvilinear coordinate system in
WRF, and horizontal diffusion tendencies were neglected since they are
comparatively small with respect to the other terms of the momentum budget.
Figure 7 shows contour plots of the longitudinal wind component and the
momentum budget terms of Eq. (2). These quantities have been spatially and
temporally averaged to filter out microscale fluctuations. The spatial filter
is based on 4

Magnitude

Figure 8 shows the effect of

The derived mesoscale forcing is consistent with that obtained by Bosveld et al. (2014a), based on simulations with the RACMO model at a horizontal resolution of 18 km. Advection tendencies show narrower peaks compared to those from Bosveld et al. (2014a). It is difficult to say where these differences come from since we used different input data and horizontal and temporal resolutions. In order to facilitate the implementation and interpretation of the mesoscale forcing in the GABLS3 SCM intercomparison, simplified mesoscale forcing was defined by adjusting piecewise linear approximations of the RACMO tendencies to obtain a reasonable agreement of the wind speed at 200 m.

Despite the filtering process, the resulting smooth fields in Fig. 7
still show large mesoscale disturbances in the advective tendencies,
especially during nighttime conditions at greater heights where vertical
diffusion is low. The geostrophic wind is more uniform, showing some decrease
in intensity with height (baroclinicity). At rotor level (Fig. 8) the
pressure gradient force is quite stationary throughout the whole cycle, with a
sudden change of 50

The dynamical origin of the nocturnal low-level jet was originally described
by Blackadar (1957) as an inertial oscillation that develops in flat terrain
due to rapid stabilization of the ABL during the evening transition under
relatively dry and cloud-free conditions (see also Baas et al., 2011; van de
Wiel et al., 2010). The daytime equilibrium of pressure gradient, Coriolis
and frictional forces is followed by a sudden decrease in vertical mixing due
to radiative cooling during the evening transition. This results in an
imbalance of forces. The residual mixed layer in the upper part of the ABL is
decoupled from the surface and the Coriolis force induces an oscillation in
the wind vector around the geostrophic wind, producing an acceleration of the
upper air that is manifested as a low-level jet at relatively low heights.
At Cabauw this happens 20 % of the nights, with jet heights between 140
and 260 m and jet speeds of 6–10 m s

: Time–height contour plots of wind velocity

Revisiting the GABLS3 in wind energy terms means evaluating the performance
of the models with application-specific quantities of interest. These
quantities are evaluated across a reference rotor span of 160 m, between 40
and 200 m, characteristic of an 8 MW large wind turbine. Aside from hub-height
wind speed

The rotor equivalent wind speed is specially suitable for accounting for wind
shear in wind turbine power performance tests (Wagner et al., 2014). The
REWS is the wind speed corresponding to the kinetic energy flux through the
swept rotor area, when accounting for the vertical shear:

Wind shear is defined by fitting a power-law curve across the rotor wind
speed points

Validation results can be quantified based on the mean absolute error MAE
metric:

It is important to note that the errors computed here are particular for this
diurnal cycle test case and cannot be associated with the general accuracy of
the SCM in other situations. It is more important to discuss the results in
relative terms to explain, for instance, the impact of adding modeling
complexity as we go from idealized to more realistic forcing. Then, if a
simulation is used as a reference to quantify this relative improvement, it
is convenient to use a normalized MAE (NMAE) by dividing with respect to the MAE
of the reference simulation:

Table 1 shows a list of the simulations performed with the single-column
model using different settings in terms of surface boundary conditions and
mesoscale forcing. The SCM simulations have been run with the same grid setup
of GABLS2, i.e., 4 km long log-linear grid with 301 levels and a time step
of 1 s. The simulations are grouped according to different model evaluation
objectives as described in the last column of Table 1. Table 2 shows the
MAE and normalized MAE, with respect to the reference

List of simulations and objectives for the sensitivity analysis of single-column models.

MAE and normalized MAE with respect to the reference

Time–height contour plots of wind velocity

Figure 9 shows time–height contour plots of wind velocity, wind direction
and potential temperature for the WRF simulation, the reference SCM simulation
without nudging (

At the rotor area, the peak of the REWS is well predicted by both the
mesoscale and the

By introducing profile nudging, these deviations are corrected to a large
extent in the lower part of the ABL. Since the weighting function of the
nudging terms

The

The third objective in the model evaluation strategy of Table 1 is to
determine if there is a choice of boundary condition for the energy equation
that is more adequate in the prediction of quantities of interest. Basu et
al. (2008) demonstrated using MOST arguments that using a prescribed surface
heat flux as a boundary condition in stable conditions should be avoided. MOST
is imposed at the surface by prescribing the mesoscale 2 m temperature
(

Interestingly enough, prescribing the observed 2 m temperature instead of the mesoscale 2 m temperature results in a 23 % increase in REWS MAE. This is due to a mismatch between the surface (observed) and top (simulated) boundary conditions, which leads to a less accurate prediction of potential temperature gradients throughout the ABL. In effect, despite the large bias in the prediction of the potential temperature, the mesoscale simulation still does a good job of simulating the diurnal evolution of vertical potential temperature gradients, which are ultimately the main feedback in the simulation of the wind speed fields via the buoyancy term in the turbulence equations. Then, using the mesoscale 2 m temperature as indirect surface boundary condition seems to be the most appropriate choice. This is a standard output in meteorological models and surface stations; therefore, it makes sense to use it for practical reasons and as a standard in wind energy campaigns and flow models.

Time series of surface-layer characteristics using different surface
boundary conditions for potential temperature with the

GABLS3 time series of rotor-based quantities of interest from top
to bottom: rotor equivalent wind speed REWS, hub-height wind direction
WD

Adding mesoscale tendencies to microscale ABL simulations requires the
generation of tendencies from a mesoscale model. The question is how
important these tendencies are in the assessment of quantities of interest.
This is the fourth objective in the model evaluation strategy of Table 1. The
modulation of the LLJ evolution by the mesoscale tendencies in the GABLS3
episode is discussed by Baas et al. (2010) and Bosveld et al. (2014a). They
use a SCM to switch on and off different forcing mechanisms and show their
relative impact on the evolution of the LLJ. Figure 10 shows time–height
plots of different SCM simulations: with all mesoscale tendencies included
(

In the first 100 m above the ground, where turbulence diffusion is important, advection tendencies are relatively small and using surface geostrophic forcing provides a realistic evolution of the diurnal cycle. Above 100 m advective tendencies become a dominant force in the modulation of the equilibrium between Coriolis and pressure gradient forces. If only surface geostrophic forcing is applied at greater heights, the wind speed and direction are way off. In terms of the REWS NMAE, removing potential temperature tendencies does not have a significant impact, while additionally removing momentum tendencies results in a 24 % increase in error. Using just the surface geostrophic wind as forcing increases the error by an additional 100 %. Hence, realistic forcing requires the characterization of the horizontal pressure gradient variations with time and height as the main drivers. Then, even though advection tendencies come with high uncertainty, introducing mesoscale momentum advection still results in significant improvement. Potential temperature advection in this case shows some improvement in the wind direction and wind shear, but this is compensated for with a deterioration of wind speed and wind veer; therefore, the overall impact on REWS is not significant.

In homogeneous terrain conditions, such as those of the GABLS3 case, we
should not expect improvements when using the offline meso–micro simulations
with a RANS model with respect to online mesoscale simulations with a
boundary layer scheme since the surface conditions have not changed and the
turbulence models are similar. Instead, by using the same surface conditions,
we demonstrated that using mesoscale tendencies was an effective solution to
drive a microscale ABL model offline without introducing significant
additional uncertainties due to the coupling between the models. It is also
not surprising to find large errors in the WRF model hour to hour, sometimes
even larger than in the reanalysis input data, since the higher resolution of
the model brings additional variability that is physically realistic but
is not necessarily well represented by the models (Baas et al., 2010; Bosveld
et al., 2014). In aggregated terms, it has been demonstrated that adding
mesoscale-generated advection tendencies was beneficial for the SCM
simulations, even though their hourly contribution was not obvious due
to phase errors for instance. A way of improving the transient behavior
of the microscale model is to introduce bias correction through nudging.
Here, we explore the profile nudging method of Eq. (4), which depends on the
timescale

Two scenarios of nudging are considered in Table 1, making use of the Cabauw
instrumentation as a proxy for typical setups that could be used in the wind
energy context. The first scenario corresponds to mast-based instrumentation
where we can routinely measure and assimilate in the model wind speed and
temperature. By convention, temperature measurements start at 2 m and wind
speed measurements at 10 m. Then, the mast height is varied from 80 m
(

The second scenario corresponds to a lidar setup whose range typically
starts from 40 m and goes up to 200–400 m. Here, only wind speed profiles
are assimilated. Again, considering a default nudging timescale of 1 h, an improvement of 53 and 58 %
is observed when assimilating data up to
200 and 400 m, respectively. Measuring above the rotor range has
little benefit in this case. Comparing the two scenarios, mast or lidar, for a nudging
range up to 200 m, it is observed that the main advantage of assimilating
potential temperature profiles is in improving the wind shear and veer
predictions. This is also observed at shorter nudging timescales,
particularly during the morning transition (Fig. 12). Figure 8 shows the
magnitude and direction of the nudging correction, vertically averaged over
the rotor range and compared to the other forcing terms. Using a nudging
timescale of 60 min results in corrections of less than 1 m s

Vertical profiles of horizontal wind speed

Figure 13 shows the vertical wind profiles of horizontal wind speed and wind direction at midnight and during the morning transition. At midnight, the WRF model performs reasonably well at developing the nocturnal LLJ and the nudging corrections mainly affect the wind direction profile. In contrast, the morning transition is not well captured by the model and large nudging corrections are needed in both wind speed and direction. In both cases, the transition at 400 m between the corrected and uncorrected parts of the profile is apparent. Using a linear decaying weight of the nudging correction above 200 m produces a reasonably smooth transition.

The series of GABLS test cases for the evaluation of ABL models have been used for the design of a single-column model that uses realistic forcing by means of mesoscale tendencies and nudging at microscale. The model includes three different turbulent closures that produce consistent results in the idealized cases GABLS cases 1 and 2. A sensitivity analysis of quasi-steady simulations following the GABLS 1 approach shows how the wind conditions at rotor heights are correlated mostly with the geostrophic wind in unstable conditions and with the local atmospheric stability in stable conditions. The main difference between the models in the GABLS 2 diurnal case resides in a larger turbulent kinetic energy as the order of the closure model is increased.

The GABLS3 diurnal cycle case has been revisited and evaluated in terms of wind energy specific metrics. Instead of using the adjusted mesoscale tendencies of the original GABLS3 setup, the mesoscale tendencies computed by WRF are directly used to force the SCM. Momentum budget analysis shows the relative importance of the different forcing terms in the momentum equations. Through spatial and temporal averaging, the high-frequency fluctuations due to microscale effects are filtered out. Compared to the WRF simulation, we use mesoscale tendencies to drive SCM simulations, resulting in consistent flow fields despite the more simplified physics of the ABL.

Using sensitivity analysis on the mesoscale tendencies, it is shown that the main driver of the ABL is the time- and height-dependent horizontal pressure gradient. Advection terms come with high uncertainties and hour to hour they can lead to large errors. Nevertheless, their impact in terms of aggregated errors in quantities of interest is positive.

The

Instead of adjusting at mesoscale, corrections are introduced at microscale through observational profile nudging to make use of the routine measurements collected in wind energy campaigns. Mast-based and lidar-based profiler setups are compared to show the added value of measuring at greater heights than the hub height, which is the main advantage of lidar systems. Sensitivity to the nudging timescale is large, especially to compensate errors introduced by the mesoscale advection forcing.

The GABLS cases show the complexity of interpreting mesoscale forcing. While the pressure gradient force is dominated by large scales and is reasonably well captured in the reanalysis data, advection tendencies depend on the physical parameterizations of the mesoscale model. Baas et al. (2010) presented an alternative case based on the ensemble averaging of nine diurnal cycles that meet the GABLS3 selection criteria. This composite case, like the presented GABLS3 case, is entirely based on forcing from a mesoscale model, and facilitates the assessment of the main features of the diurnal cycle by canceling out the mesoscale disturbances of the individual days. As a result, the composite case shows great improvement versus considering any single day separately. Hence, the assessment of mesoscale to microscale methodologies is more appropriate in a climatological rather than a deterministic sense. Otherwise, dynamical corrections like profile nudging would be required.

SCM simulations over horizontally homogeneous terrain are a convenient methodology for the design of ABL models given their simpler code implementation and interpretation of results compared to a three-dimensional setting in heterogeneous conditions. This allows testing surface boundary conditions, turbulence models and large-scale forcings more efficiently before implementing them in a three-dimensional microscale model. In a three-dimensional model, advection would be solved by the model through surface heterogeneities and velocity gradients across the lateral boundaries. Spatially averaged, height- and time-dependent mesoscale forcing from horizontal pressure gradients could be introduced as a column body force throughout the three-dimensional domain similar to how it was done in GABLS3. By spatial averaging over a larger scale than the microscale domain, we expect to filter out disturbances in the pressure gradient due to unresolved topography in the mesoscale model. These topographic effects will be modeled with a high-resolution topographic model in the three-dimensional microscale simulation. Such a model chain would still assume that the mesoscale forcing is horizontally homogeneous throughout the microscale domain but with changes in height and time through source terms in the momentum equations. Nudging local corrections would be introduced through horizontal and vertical weight functions that limit the correction to the local vicinity of the observation sites as it is done in mesoscale models (Stauffer and Seaman, 1990). This relatively simple implementation of meso–micro coupling is valid for RANS and LES models and allows easier characterization of mesoscale inputs than using three-dimensional fields.

The original GABLS3 input and validation data can be found on the KNMI GABLS
website (

The authors declare that they have no conflict of interest.

This article was produced with funding from the “MesoWake” Marie Curie International Outgoing Fellowship (FP7-PEOPLE-2013-IOF, European Commission's grant agreement number 624562). Edited by: J. Meyers Reviewed by: B. Holtslag and two anonymous referees