Sensitivity of the WRF Model to Boundary-Layer Forcing: a Case Study from the COPS Experiment

Sensitivity of a high-resolution numerical model to boundary-layer forcing: a case study from the COPS experiment.(Need better title)

Burton, Gadian, Blyth, Lock, Mobbs, etc. and anyone else!

1. Introduction.

In this study, it will be shown that the choice of boundary-layer and land-surface schemes is crucial when attempting to simulate an observed deep convective cloud. The results of sensitivity tests are presented, for a case from the COPS campaign: the ability of the model to correctly represent boundary-layer processes during the deep convective cloud event of the 15th July 2007 will be investigated. With an isolated cloud forming over complex terrain, during a period of high pressure, this event represents a particularly challenging test case and constitutes an ideal scenario for model assessment. The term assessment is used purposefully here, instead of the more common alternative, verification – since there is no direct attempt here to compare model results with observations. This seeming deficiency is intentional, as only one of the simulations presented here produced any deep convective cloud at all – and this particular simulation is described as part of a separate model intercomparison in this issue (Barthlott et al. 2010, hereafter B10).

The use of the WRF (Weather Research and Forecasting) model is now widespread: its use in meteorological modelling (both operational forecasting and analysis of test cases) is now extensive. However, the ability of the WRF boundary layer schemes to properly represent the atmosphere above complex orography is not well understood: this has important consequences for modelling work involving the use of WRF in such cases. Additionally, there is a large degree of freedom (as in any numerical model) in the choice of physical schemes and packages; it is not easy to determine a priori which combination of physical schemes will suit a chosen case study. It should be noted that, while the following study is derived from WRF model results, the relevance to other numerical models is significant, in showing that the appropriate choice of physical packages is highly important - and can mean the difference between no convective cloud and deep, intense convection; additionally, and more importantly, the work presented here will hopefully illuminate the important physical processes which need to be represented well in any numerical model, if the realistic simulation of deep convection is to be achieved. The ability of other numerical models to reproduce this case of deep convection is described elsewhere in this issue (see B10).

The large number of physical schemes available to the user of WRF (or any sophisticated NWP model) represents a difficulty in itself: the combination of the various physical modules and parametrisations is often a result of inspired guesswork; and to produce the optimal combination of these (or a combination which yields results closest to those which we expect, or observe) requires a certain amount of sensitivity testing. The ease with which various physical parametrisations can be switched in WRF, however, causes this to be a relatively straightforward task. Indeed, the large degree of freedom in choosing physical options in the model leads to a vast number of possible model configurations, too many for a single study. In this paper we shall concentrate upon the effect of (i) varying the different boundary-layer schemes and (ii) changing the way in which the land surface is represented.From the groups of simulations, the potential problems in the available boundary layer schemes (such as boundary layers being too well-mixed) will be demonstrated.

In Section 2, the COPS experiment is briefly described; Section 3 contains details of the WRF model and its representations of the boundary layer and land surface models; results of the sensitivity tests are presented in Section 4, and this study concludes with recommendations regarding the choice of physical schemes(Section 5).

2. The COPS experiment.

Short text here. No need to repeat what will be contained in other parts of this issue.

2.1. Description of 15th July convective cloud.

Not much here – it is covered in (Barthlott et el. 2009, i.e. this issue) etc. Text to that effect.

3. Description of WRF.

The Weather and Research Forecasting model (WRF, Ref.) is a state-of-the art numerical model originally developed by NCAR and other US institutes and agencies. Highly portable and designed to run under various (easily changed) configurations, the model includes a full suite of physical packages. More text about WRF. The version of WRF used in this study is version 3.1 (released April 2009); however, this study has relevance to earlier (and presumably, subsequent) releases of the model which include the boundary-layer and surface-scheme representations described below.

3.1. Description of the general model configuration.

In the simulations described in this paper, WRF was set up with the physical schemes as displayed in Tables I and II. Table I shows the physical schemes that are common to all the runs described in this paper; Table II describes the differences between model runs.

Figure X. shows the three nests used in the model simulations. The ratio of grid sizes in parent:child nests is a constant 1:3, and the timestepping in the model integrations follows the same rule. The feedback between nests is a two-way process, with information being fed from the smaller scales to the larger, in addition to the usual downscaling of energy. The model was initialized with NCEP GFS analyses, freely available on the internet (REF). The analyses used were at 1 degree by 1 degree resolution and with 27 vertical levels; these analyses were employed to both initialize the model simulation, and to provide boundary conditions (on the outer domain) every 6 hours, until the model forecast limit of 24 hours. On the inner domain, the boundary conditions are derived from the outer domain.

TableI. Properties of the simulations that are common to all runs. For further details of the physical schemes, see the WRF Technical Note (Ref.)

Table II. The four runs considered in this paper, and their boundary-layer and land-surface schemes. The names in column 1 constitute the descriptive labels used in the remainder of this paper.

With reference to Table II, there are four different simulations presented in this paper; only one of which managed to produce the observed deep convective cloud. The four different runs, which are combinations of two different boundary-layer (BL) schemes and two different land-surface models (LSM), were otherwise identical. The BL and LSM schemes used in this study are described below.

3.2. Description of the boundary-and surface-layer schemes.

In WRF the boundary layer (BL) schemes can be broadly classified according to the way in which the turbulence (especially the representation of larger scale eddies) is treated. Any sophisticated BL scheme should possess the property that even when the vertical gradient of the mean value of θ is zero, the turbulent flux of heatcan be non-zero (this then allows large-scale eddies to develop over large depths in the atmosphere: see, e.g. Stull X). There are two principal means of ensuring this.

3.2.1. BL: “Countergradient” schemes.

The motivation behind this type of BL scheme is that turbulent fluxes are allowed to depend upon local mean gradients with the addition of an extra, or countergradient, term (see, for example, Stensrud X):

=

Where C is a prognostic field such as u, v or θ.

(The terminology “counter” applied since the corrective term acts in the opposite direction to the mean gradient of the field.). This formal statement of the fluxes ensures that the desirable property noted above is satisfied (if γ is non-zero). The countergradient term, denoted by ,can then be derived as a function of (for example) the field flux at the surface and a suitable velocity scale, with the inevitable introduction of a number of constants. The full description of this BL scheme is outside the scope of this note: suitable treatments can be found in (REFS).

3.2.2. BL: TKE-based schemes.

In this representation of the BL scheme, an extra equation is incorporated into the numerical model. This extra equation determines the production and dissipation of turbulent kinetic energy (TKE). There are numerous means by which the equations of TKE can be presented (see e.g. Stull) but all have the required property that turbulent fluxes do not vanish in a region of zero mean gradient. TKE-based BL schemes are often thought of as being more sophisticated, since the TKE equations are derived from a direct treatment of the equations of motion (see Stull REFS). In spite of this, the TKE scheme is still subject to some assumptions and depends upon certain empirically-determined constants (the equation set itself is still not closed) .

Note that for both the countergradient and the TKE-based schemes, the numerical implementation is one-dimensional in the vertical. In the simulations described below, the YSU scheme is of the countergradient type, and the MYJ is of the TKE type.

Given that we are seeking to simulate a case with known deep convection, it would seem to be imperative that the surface is represented accurately in the model: if the surface heat and moisture budgets, for example, are not adequately represented, then there may be an inadequate representation of convective processes. The part of the WRF model that deals with these surface fluxes, and feeds the latter into the lower levels of the BL scheme, are the land surface models, and these are described in the next section.

3.2.3. Thermal Diffusion model.

The following is a summary and paraphrasing of information derived from A Description of the Advanced Research WRF Version 3 (Refs), which constitutes a technical companion to the WRF user guide (REF) and includes a comprehensive list of references.

In the thermal diffusion treatment of the surface heat and moisture fluxes, there is no explicit vegetation included. Soil temperatures are prescribed at five discrete levels, plus a deep-layer average; soil moisture is derived from seasonal averages. This model is one-dimensional.(REFS)

3.2.4. NOAH LSM.

This parametrisation of the surface is more sophisticated than the thermal diffusion scheme described in 3.2.3. The possibility of, and allowance for, evapotranspiration and vegetative effects, in addition to drainage and runoff effects, is a major feature of the model. Soil temperature and soil water and ice are kept at four model levels. As for the (simpler) thermal diffusion approach, the NOAH model is effectively a column model. (REFS)

4. Results of the sensitivity tests.

4.1 The mechanism of cloud formation on the 15th July.

In all four runs, a clear signal is derived from a convergence line which formed over much of the Black Forest on the afternoon of the 15th July. In the north of the COPS region, this convergence line was observed in surface and upper-air instrumentation (Kottmeier et al. 2009) and, in the absence of any indications of instability in observed soundings at this time, is thought to be the driving mechanism behind cloud formation in that region. Due to the lack of any observing stations in the region of the observed deep cloud considered here, however, the connection (or rather extension) of the convergence line to the south cannot be confirmed experimentally (although radar evidence in B10 seems to suggest that it is); the numerical simulations presented below do suggest, however, that a convergence line stretched from the north to the south of the Black Forest. This can be most clear seen in the THERM-YSU simulation. (Figs 1).

At 10Z, zones of convergence can be seen forming on the western slopes of the Black Forest, which grow and nearly join by 11Z; to the east, a line of convergence (oriented in a roughly west-east direction) is apparent at 10Z. By 13Z, the line of convergence appears to be a coherent structure, covering nearly the whole of the region from south to north; and the previously existing, separate line to the east has dissipated. As the afternoon progresses, the line is seen to move to the east; while preserving its rather sinuous structure. By 18Z the line has fragmented and lost all coherency. Although these plots are at a level of 1.5km above sea level, the signature of the convective line can be seen at lower levels. Thus, Fig. 2 shows the lowest-level vertical velocity and divergence field for the NOAH-MYJ run. Although the line of convergence appears more complex (due to feedbacks from the evolving cloud fields), the line is still clearly defined. This convergence line appears in all the runs in this study.

While the YSU-THERMAL combination of physical schemes is, prima facie, the simplest simulation (in terms of the complexity of the physical treatment of the underlying processes), it nevertheless offers insights into the dynamical situation. Primarily because no deep clouds were observed in this run, it can function as a type of idealised simulation, the dynamical processes appearing more clearly, as they are not disturbed by the effects of clouds (such as downdrafts). Thus, in this simulation the convergence line (shown in Fig. 1) is seen clearly, whereas in that of the NOAH-MYJ case, the structure of the line (and some of the apparent coherency) is made more complex due to the presence of cloud-related severe updrafts and accompanying downdrafts (see Fig. 2). Thus, in analyzing the simulations, the YSU-THERMAL case allows the determining physical processes to become apparent, the details (and hopefully more realistic properties) of which are illustrated by the NOAH-MYJ simulation (the only simulation to produce deep cloud).

In the following, it is instructive to examine a north-south cross-section (defined in Figs. 1 and 2 and subsequent figures) and investigate the boundary-layer properties along this transect as the convergence line moves across it. The southeast-northwestern oriented convergence line noted above moves eastward across the cross-section (as shown in Figs. 1.) Due to this orientation, the region of upward motion moves southwards across the cross-section (see Fig. 3 which shows the cross-section for the NOAH-MYJ run) giving the impression that clouds are moving to the south; this is not the case, and is simply a consequence of the orientation of the convergence line as it passes through the cross-section. Between 14 and 15Z, the convergence line is stationary near the location of the deep convective updraft; this stationarity allows the potential for severe updrafts to develop in the NOAH-MYJ simulation.In the other simulations (Fig 4), the same progression of southward moving updrafts can be seen along the cross-section. However in these cases the vertical velocities are not as significant, and updrafts do not penetrate beyond the LFC. By 16Z and beyond, the convergence line has, in all simulations, weakened and fragmented, and the deep cloud rapidly dissipates.

The cloud obtained in the NOAH-MYJ run (Fig. 5) shows a quite realistic structure, with a large anvil spreading to the north (as observed in reality); patterns of updrafts and downdrafts (including a rear inflow) that would be associated with such a deep convective cloud; and high values of w associated with the large value of CAPE for the nearby sounding (see below).

4.2 Boundary-layer response.

Skew-t plots are shown in Fig. 6 for a point near the site (48.XW, 8.XE) of severe convective activity (see B10). For all the measure of convective potential (such as LI, SWEAT, TT) the NOAH-MYJ run displays the highest values, and thus the potential for the severest convective activity, at least as measure by these metrics; and does, indeed, produce the most severe response to the passing of the convergence line.

For the CCL in each of the soundings, it is apparent that the CCL is lowest for the NOAH-MYJ run; the two highest (in altitude) values of the CCL are THERM-YSU and NOAH-YSU, suggesting that the use of this BL scheme is causing the PBL to be too dry; the use of the NOAH scheme mitigates this to some extent. This is further confirmed by an examination of the cross-section of cloud water mixing-ratio (Fig. 7). The YSU scheme redistributes surface moisture too readily to upper levels of the PBL, where it can be detrained into the free atmosphere. The counter-gradient term γ (see Eqn. 1) is seen here to be responsible for transferring moisture too readily away from the lower levels in the simulations where the YSU scheme is used. In the latter, as was noted in Sect. X., turbulent transfer is allowed between levels which are not adjacent (in order to simulate large-scale eddies); in this case this has the deleterious effect of transporting moisture away to upper levels. In the MYJ cases, turbulent mixing is on adjacent levels, and the moisture is kept to a large extent within the lower levels, where it can act as a reservoir of energy. This is illustrated in Fig. 7 which shows the difference in qv between the simulations NOAH-MYJ and NOAH-YSU. The only significant negative areas in this plot are above 2km, suggesting that not only is the YSU boundary-layer of a larger vertical extent, it has also redistributed the moisture too evenly.It is possible that some tuning of the countergradient term may improve the resulting simulations when using the YSU boundary-layer scheme; this is, however, beyond the scope of this brief study.