1. Introduction
Numerical weather prediction (NWP) models are now routinely run at horizontal grid spacings of a few kilometers, which permits an explicit representation of deep convection. However, while larger storm structures are sufficiently realistic at these resolutions (Weisman et al. 1997; Clark et al. 2016), the initiation of convection is often tightly linked to unresolved boundary layer processes. Turbulent eddies, for instance, are essential for overcoming the convective inhibition layer. Because boundary layer turbulence parameterizations typically only represent the mean mixing caused by subgrid eddies in a deterministic way, models can exhibit biases caused by an incorrect coupling between subgrid processes and resolved convection. In particular, this has been observed in the Consortium for Small-Scale Modeling (COSMO) model, which we are using in this study (Keil et al. 2014; Baldauf et al. 2011; Craig et al. 2012).
This inconsistency becomes particularly relevant when synoptic forcing is weak and local mechanisms are the main driver for overcoming convection inhibition (Keil et al. 2014). In these situations insufficient convective initiation has been commonly observed for kilometer-scale models (see e.g., Clark et al. 2016). In operational NWP systems such biases are often compensated by tuning other parameters, such as the turbulent length scale in the boundary layer parameterization (Hanley et al. 2015). Smaller mixing lengths allow the lowest model levels to heat up more, resulting in earlier and more deep convection (Hanley et al. 2015; Baldauf et al. 2011). For example, for the operational COSMO version, a decision was made to decrease the turbulent length parameter tur_len in order to reduce a lack of summertime convection (Baldauf et al. 2011). This, however, comes at the expense of temperature biases, which are widely observed in different studies (Baldauf et al. 2011; Schraff et al. 2016; Leutwyler et al. 2017; Necker et al. 2018). Baldauf et al. (2011) argue, however, that “the improved forecasts of deep convection outweigh those disadvantages” (p. 3902, l. 8, 9) and also climate simulations with a kilometer-scale COSMO version exploit this aspect (see e.g., Leutwyler et al. 2017). Such biases, however, can pose a severe problem for data assimilation systems (Schraff et al. 2016; Necker et al. 2018) and an improvement of deep convection without the loss of temperature skill is desirable.
One mathematically well-founded method to improve the coupling between subgrid and resolved processes are stochastic parameterizations [see Berner et al. (2017), for a review]. While theoretical studies rigorously derive stochastic representations of model uncertainty for idealized models (e.g., Chapron et al. 2018; Vissio and Lucarini 2018; Demaeyer and Vannitsem 2017, 2018), more pragmatic approaches dominate applications in complex NWP models today. Holistic approaches as in Hirt et al. (2018) or the Stochastically Perturbed Parameterization Tendencies (SPPT) (Buizza et al. 1999) and Stochastic Kinetic Energy Backscatter (SKEBS) (Shutts 2005) approaches provide a bulk representation for many types of model error. In contrast, process-level stochastic perturbations directly represent model uncertainties at its known sources: Ollinaho et al. (2017) and Jankov et al. (2017, 2019) propose stochastic perturbations of uncertain model parameters (SPP); Christensen et al. (2017) update the SPPT scheme to treat different processes independently; Craig and Cohen (2006) introduce a physically motivated approach based on statistical mechanics. Such process-level stochastic parameterizations have shown to better fit observations compared to bulk approaches (Christensen et al. 2017).
In this paper, we focus on process-level stochastic perturbations to improve the representation of convective initiation in kilometer-scale models. In the first part of the paper we revise the physically based stochastic perturbation (PSP) scheme for boundary layer turbulence originally developed by Kober and Craig (2016); in the second part we introduce a stochastic perturbation scheme for subgrid-scale orography (SSOSP). The developments introduced here will improve the consistency of the PSP scheme, and add additional physical processes. The aim of the present paper is to document these changes and additions, and to provide a preliminary evaluation of their impacts on simulations of convection to determine if the anticipated physical effects are present. Based on these results, a revised version of the PSP scheme will be defined that has the potential to improve forecast skill for convective precipitation. To demonstrate improved skill will require a systematic evaluation of ensemble forecasts over a long test period. This will be the subject of a future study.
The PSP scheme (Kober and Craig 2016) aims to better couple subgrid turbulence to convective initiation by reintroducing missing subgrid variance to the resolved scales. This is done by adding stochastic perturbations to the model tendencies of temperature, humidity, and vertical velocity. The perturbation structure is given by a horizontally correlated random field that is regenerated every 10 min, a time span that represents the average lifetime of a boundary layer eddy. Since such eddies span the entire depth of the convective boundary layer, the perturbations are kept vertically coherent by applying the same random field at all model levels. Most importantly, the amplitudes of the perturbations are scaled by physical information on the subgrid variance computed in the turbulence parameterization. This links the perturbation strength to the physical processes that cause small-scale turbulence. Furthermore, this enables the scheme to adapt to different atmospheric situations.
The PSP scheme showed promising results in a range of case studies (Kober and Craig 2016). Under weak synoptic forcing, the perturbations trigger more convective cells and consequently produce more precipitation. This improved the diurnal cycle and amplitude of precipitation. Further, Rasp et al. (2018) have successfully exploited its capability to create different realizations of the convective cloud field for the same synoptic situation by generating different random fields for each ensemble member. The magnitude of the ensemble spread created by different PSP realizations was also found to be comparable to operational ensembles (Keil et al. 2019). And in a convective-scale data assimilation framework, the PSP scheme produced a beneficial impact in ensemble spread and precipitation forecasts (Zeng et al. 2019, manuscript submitted to Mon. Wea. Rev.).
However, the original version of the PSP scheme also leads to undesirable side effects. In the first part of this paper, we investigate these issues and propose modifications to alleviate them. In particular, four topics will be covered:
In its current formulation, the evolution of the random field is temporally discontinuous, being regenerated every 10 min. To avoid these discontinuities, we introduce a continuously evolving random field using an autoregressive (AR) process.
Unrealistically high nighttime precipitation has been observed using the PSP scheme (Rasp et al. 2018). As we will show, this is related to perturbations in the free troposphere. These do not fit with the rationale of the PSP scheme, which is designed to represent vertically coherent structures in convective boundary layers. Hence to remove these artifacts we restrict the perturbations to the boundary layer.
Furthermore, organized convective structures were found to be broken apart by the perturbations. To prevent this, we investigate the use of a mask to switch off the perturbations in regions where precipitation is already occurring.
Finally, the vertical velocity perturbations were found to have an insignificant effect compared to the temperature and humidity perturbations (Kober and Craig 2016). As we will discuss in this paper, this is associated with the generation of acoustic modes that lead to rapid dissipation of the imposed vertical velocity perturbations. By introducing balanced, 3D wind perturbations, we obtain lasting and effective wind perturbations.
After evaluating the modifications separately, we propose a revised version of the scheme, which we will call PSP2.
In the second part of this manuscript, we target the influence of subgrid-scale orography (SSO) on triggering convection as a complementary process to turbulence. While the importance of resolved orography on convective initiation is unquestioned (Houze 2012), the contribution of subgrid-scale orography is unclear. For example, Kirshbaum et al. (2007b) find that scales smaller than few kilometers are not dominant for convective initiation when a range of orographic scales exist. On the other hand, Kirshbaum et al. (2007a,b) also show that small, single-scale orographic features can dictate the spacing of orographic bands, and Langhans et al. (2011) conclude that small-scale orography contributes significantly to convective initiation. Since orography can act as a source of predictability for convective scales (Bachmann et al. 2019a,b, manuscript submitted to Mon. Wea. Rev.; Anthes 1986), the accurate representation of its effects will be important for weather prediction.
Most current NWP models include a parameterization for SSO. However, these parameterizations primarily account for a deceleration of the flow by unresolved orography, either by modifying the roughness length or by including an orographic drag term. Here, we aim to account for the mechanical lifting caused by SSO and its effect on convective initiation with a newly developed stochastic perturbation scheme, called SSOSP. The new scheme closely follows the formulation of the PSP scheme: wind tendencies are randomly perturbed with an amplitude that scales with theoretical gravity waves excited by SSO. Investigating the impact of this scheme will allow us to estimate the importance of SSO on convective initiation and therefore contribute to the discussion mentioned in the previous paragraph.
In section 2, the formulation of the PSP scheme is explained in detail, as well as the four modifications mentioned above. Further, the SSOSP scheme is introduced. In section 3 we describe our experimental setup, which is used for testing both PSP and SSOSP. The results are presented in section 4 for the PSP scheme and in section 5 for the SSOSP, followed by a summary and discussion in section 6.
2. Conception of the stochastic perturbations
In this section we summarize the original perturbation scheme for boundary layer turbulence (PSP) by Kober and Craig (2016) and then describe our modifications to improve the physical consistency of the scheme. At the end of this section, we introduce the stochastic perturbations to account for mechanical lifting by subgrid-scale orography, the SSOSP scheme.
a. Physically based stochastic perturbations for subgrid turbulence (PSP)
The PSP scheme for boundary layer turbulence was first described in Kober and Craig (2016). Here we present an updated formulation according to Rasp et al. (2018).
Boundary layer turbulence is characterized by eddies with scales ranging from millimeters to around 1 km, the boundary layer height. Traditional parameterizations that account for the impact of turbulence on the resolved grid, were developed for grid sizes that are large relative to the turbulent eddies. The parameterization predicts the mean effects of a large ensemble of eddies, and the variability associated with the individual eddies is averaged out. But the kilometer-scale region represented by a grid box in a convective-scale model will contain only a small number of the largest eddies and their variability will have a substantial effect on the gridbox mean. As a result, the variability on the smallest resolved scales in the model is underrepresented, which can lead to the difficulties in overcoming convective inhibition that were described in the introduction.
b. Modifications to the PSP scheme
1) Autoregressive process: PSP-AR
2) Adaptive boundary layer height restriction: PSP-HPBLcut
It will be shown in section 4a(2) that the original PSP showed unrealistic precipitation at night, which can be related to considerable perturbations in the free troposphere present also during the night. However, the PSP was developed with the convective boundary layer in mind and many of its assumptions (notably vertically correlated eddies) are not valid for free tropospheric turbulence. In order to improve the physical integrity of the scheme and simultaneously reduce unrealistic precipitation, we limit the perturbations to the planetary boundary layer height HPBL. We determine HPBL as the height at which the bulk Richardson number reaches the critical Richardson number as implemented in the COSMO model. This value is 0.33 under stable conditions (Wetzel 1982) and 0.22 under convective conditions (Vogelezang and Holtslag 1996). Below HPBL the perturbations are applied as usual. Above, however, the perturbations are decreased linearly to zero at a height of 500 m above HPBL. This modification, applied to the PSP-AR version of the scheme, will be termed PSP-HPBLcut.
3) Masking of already precipitating areas: PSP-mask
We observed that in certain situations the original PSP scheme appeared to break up organized convection [see section 4a(3)]. We hypothesize that this is partly caused by the perturbations disturbing already existing updrafts. As a simple ad hoc fix, we test turning off the perturbations in grid columns where the vertically integrated precipitating hydrometeor content exceeds 10−3 kg m−2. This adjustment to the PSP-AR version will be labeled as PSP-mask.
4) 3D nondivergent perturbations: PSP-3D
Kober and Craig (2016) observed that the impact of the original PSP scheme was caused by the temperature and humidity perturbations while the vertical velocity perturbations had essentially no effect. Further analysis shows that the w perturbations decay rapidly in a few time steps. This can be related to pressure perturbations: Chagnon and Bannon (2005), for example, show that an injection of vertical momentum efficiently induces high-frequency acoustic waves causing a rapid removal of the imposed vertical updraft. Following their approach we extend the scheme to add perturbations to the horizontal velocity field that approximate three-dimensional nondivergence. The goal is to suppress the excitation of spurious acoustic waves so that the vertical velocity perturbations persist (see e.g., Fiedler 2002; Chagnon and Bannon 2005; Edson and Bannon 2008).
We define the horizontal perturbations by the velocity potential χ: vh = ∇χ. Consequently, the horizontal perturbations are nonrotational, which is in accordance with the incentive of the perturbations.
To avoid large vertical velocity gradients ∂w/∂z near the lower boundary, we linearly increase w perturbations from 0 at the surface to the original value at z0. We chose z0 = 500 m to include a sufficient number of vertical model levels while ensuring that the full perturbation amplitude is active at the height of the daytime inversion, where we expect the perturbations to be most effective.
Applying these balanced wind perturbations to PSP-AR will be denoted as PSP-3D.
c. Stochastic perturbations to account for mechanical lifting by subgrid-scale orography (SSOSP)
As described in the introduction, orography plays an important role in triggering convection. The impact of orography that is smaller than model resolution [i.e., subgrid-scale orography (SSO)], needs to be parameterized. In most models, this is done by an orographic drag parameterization (Lott and Miller 1997) and a modification of the roughness length (Doms et al. 2011). The former accounts for deceleration of flow by orographically induced gravity waves and blocking and the latter approximates effects within the boundary layer. However, the vertical motion responsible for triggering of convection by SSO is not addressed in these parameterizations.
For this purpose we propose stochastic perturbations to account for convective initiation by subgrid-scale orography, which we will name the SSOSP scheme. Here we focus on the effects of mechanical lifting and the resulting gravity waves. The flow over a mountain causes an upward displacement of the air parcels (i.e., a vertical velocity perturbation), which may decay with height, or propagate away from the source, depending on the stratification of the atmosphere. The upward displacement may cause the parcels to reach their level of free convection (LFC). As the relative importance of single mechanisms for convective initiation is not well understood, we neglect processes such as thermally induced upslope winds, upward and leeside convergence and flow blocking by mountains (see e.g., Geerts et al. 2008; Schmidli 2013; Langhans et al. 2013; Houze 2012; Hassanzadeh et al. 2016; Panosetti et al. 2016).
Inspired by the PSP scheme, we introduce stochastic vertical velocity perturbations that scale with the amplitude of the SSO lifting and the induced gravity waves. The horizontal wind is perturbed to approximate nondivergence as described in section 2b(4). Since the errors in the estimation of the nondivergent wind perturbations are likely to be largest over strong orography where the grid is most distorted, there will be an additional source of noise in these regions. However, we were unable to identify any obvious effects in the simulation results.
3. Model simulations, observations, and simulation period
To evaluate the impact of the PSP variants and the SSOSP we compute several simulations with different setups. We investigate each modification of PSP individually on one day with weak synoptic forcing of convection, 6 June 2016. For comparison, we use 30 May 2016 as an example of a strong forcing day with synoptically driven, coherent precipitation. Based on these results, we decide on a suitable configuration for a revised PSP version, PSP2, and compute 10 subsequent days with total simulation time of 24 h to confirm our findings. For the SSOSP scheme we focus on 6 June 2016. See Table 1 for an overview of the experiments.
Overview of the considered experiments and corresponding model settings.
a. Simulation period
We select a 10-day period from 29 May to 7 June 2016 for our experiments, which was characterized by heavy precipitation over Germany (Piper et al. 2016). The first half of the period was dominated by an upper-level trough and associated low pressure systems over Germany. The resulting synoptic lifting combined with southeasterly advection of moist air caused heavy rainfalls, particularly on 30 May (Fig. 1). In the second half of the period, the trough made way for a persistent ridge structure reminiscent of an omega-block. Then the synoptic situation was calm, allowing large instability to build up each day followed by strong convection. This behavior was particularly dominant on 6 June. This period provides a good testing ground for the stochastic schemes, offering both strongly and weakly forced convection.
Synoptic charts at (a) 0000 UTC 30 May and (b) 0000 UTC 5 Jun 2016. White lines represent mean sea level pressure (hPa). Colors represent 500-hPa geopotential (dam). Figures created from the operational analyses of the ECMWF IFS.
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
b. Model and simulation setup
We use the COSMO model (Baldauf et al. 2011), version 5.4 g, in a convection-permitting setup with Δx = 0.025°, roughly 2.8 km, and a time step of 25 s. The setup uses 50 vertically stretched model layers ranging from 10 m above ground to 22 km above mean sea level. Shallow convection is parameterized using the Tiedtke scheme. Details on parameterizations can be found in Doms et al. (2011). The setup follows the operational COSMO-DE setup with 461 by 421 grid points centered over Germany at 50°N, 10°E.
The only major deviation from the current operational setup is a change of the tuning parameter tur_len in the boundary layer scheme to 500 m. As discussed in the introduction, a smaller value was applied as an ad hoc correction in the operational model to increase the rate of convective initiation. In this way we will examine whether the PSP scheme is able to achieve realistic convective initiation without this correction.
Initial conditions are taken from the COSMO-KENDA ensemble data assimilation system (Schraff et al. 2016), which is based on a local ensemble transform Kalman filter to directly assimilate conventional observations. Additionally, radar observations are assimilated through latent heat nudging. In our experiments we only use the deterministic analysis. Using initial conditions provided by a convective-scale data assimilation system reduces the model spinup compared to downscaled initial conditions. Boundary conditions are provided by global ICON forecasts. Our simulations are started at 0000 UTC on each of the simulation days and extend for 24 h.
A simulation without stochastic perturbations, also with tur_len = 500 m, is denoted as Reference.
c. Observations
We compare our simulations to radar-derived precipitation fields. For this we use the Radar Online Aneichung (RADOLAN) quality-controlled radar observations (EY product) provided by the German Weather Service, which is based on European radar reflectivities and provides radar coverage for most of our domain (DWD 2018a,b). For all analyses of our simulations we restrict the geographical region to grid points where radar data is available.
For this study we focus on the systematic impact of the PSP scheme on the behavior of the model. The comparison to radar observation only serves as a guideline. Trying to improve the forecasts scores would require much longer simulation periods and tuning of several model parameters (e.g., tur_len). We anticipate, however, that the systematic impacts of the PSP scheme are independent of the specific model configuration.
4. Results
PSP modifications
1) Autoregressive Process in PSP-AR
To test the influence of the AR random field we focus on 6 June, a weakly forced day on which the perturbations are expected to have a large influence on the precipitation. The original PSP scheme (PSP), in which the random field is regenerated every 10 min, leads to an increase and earlier onset of precipitation (Fig. 2a) compared to the unperturbed reference simulation. We note that the diurnal cycle and amplitude of precipitation from radar observations (Fig. 2a) are already well matched by the Reference simulation. The PSP simulations consequently produce too much precipitation. This behavior is, however, not necessarily typical of longer periods [see section 4a(5)]. As mentioned above, due to the tuning problem of NWP, we concentrate on the systematic impacts of the PSP schemes, not their performance compared to observations. Switching to a continuously evolving random field (PSP-AR) further increases the precipitation amount. Example precipitation fields in Fig. 2b confirm that this increase in precipitation is a consequence of more convective cells and thus, more triggering.
(a) Hourly accumulated domain-mean precipitation for 6 Jun 2016. The domain-mean is computed only for the region of available radar data. (b) Snapshots of hourly precipitation accumulation at 1500 UTC 6 Jun 2016. Gray shading indicates missing radar data. The different experiments are listed in Table 1 (the perturbed simulations shown were computed using α0 = 7.2).
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
We explain this behavior by the different meaning of τeddy in the PSP-AR setup: For the AR process τ represents the average decay scale while for single grid points and time ranges longer correlation times are possible. These might be more effective in triggering precipitation. The PSP-AR perturbation can be scaled by tuning αtuning, which we will do after considering also the other modifications. Note also that these results are robust for different random seeds (not shown). In the remainder of this paper we will always use the AR random field, unless otherwise specified.
2) Adaptive boundary layer height restriction in PSP-HPBLcut
Next, we investigate the impact of constraining the perturbations to the height of the planetary boundary layer (PSP-HPBLcut). Figure 3 shows the horizontal standard deviations of the perturbations. Without limiting the perturbation height (PSP-AR) the perturbations are largest in the boundary layer but significant perturbations in the midtroposphere are present as well. As discussed above this is conceptually inconsistent with the rationale of the PSP scheme. Limiting the height in HPBLcut removes this inconsistency. This modification leads to a decrease in the total precipitation amount and, most importantly, removes the precipitation peak at night (Fig. 2a).
Standard deviation of tendency perturbations over the horizontal domain for each vertical model level and hourly timesteps: (top) temperature, (middle) specific humidity, and (bottom) vertical velocity. Perturbations for the (left) PSP-AR and (right) PSP-HPBLcut simulations.
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
3) Masking of already precipitating areas in PSP-mask
In the original and the PSP-AR scheme, we observed a breaking up of larger-scale precipitation structures. An example for this can be seen on 30 May (Fig. 4c), a strongly forced convective day. A coherent, synoptically driven precipitation structure is present over western Germany and Belgium, which is well reproduced by the reference simulation. Adding the PSP-AR perturbations breaks up this structure into many smaller convective cells. Turning off the perturbations in precipitating columns PSP-mask corrects this problem but also reduces the impact on weak forcing days (see Fig. 2). A too small threshold for precipitating hydrometeors can be responsible for switching off the perturbations before deep convection has developed.
Impact of precipitation mask: time evolution for 30 May 2016 for (a) domain averaged precipitation and (b) structure component of SAL. (c) Example precipitation fields for 0800 UTC 30 May 2016.
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
A quantitative measure for the structure of the precipitation field is the S(tructure) component of the SAL score (Wernli et al. 2008) displayed in Fig. 4b. Negative S values imply that the simulated precipitation cells are too small and peaked compared to the radar observations while S = 0 suggests a perfect match in terms of structure. For details on the computation, see appendix B. During the morning hours on 30 June, when the synoptic structure is most salient, the PSP-AR simulations show a clear reduction of the S score, indicating smaller cells. The precipitation mask reverts the structure back to its original value.
The precipitation mask solves the problem of the original scheme, where organized convection was broken up, but as noted above, also reduces the triggering effect in weak forcing situations.
4) 3D nondivergent perturbations in PSP-3D
As a last modification to the PSP scheme, we investigate the impact of adding nondivergent wind perturbations. For these experiments we reduce the perturbation amplitude αtuning to α1 = 2 because larger horizontal wind perturbations resulted in numerical instabilities. The domain averaged precipitation time series are displayed in Fig. 5a. As expected the precipitation increase is lower for the new perturbation amplitude α1 compared to the original perturbation amplitude α0 (see Table 1). We first investigate the impact of the original w perturbations without horizontal wind perturbations by comparing PSP-AR, α1 to PSP-AR Tqυ, α1, where only temperature and humidity are perturbed. Consistent with the results of Kober and Craig (2016), w has a negligible effect in this experiment.
(a) As Fig. 2, but considering different PSP setups for analyzing the 3D modification. Note that α1 = 0.28α0. (b) Structure component S of SAL for selected simulations as indicated by colors and line style. The RMS between 1100 and 2200 UTC, when most precipitation occurs, is given in the legend as well. Values close to zero indicate similar structure as radar observations. Positive/negative S imply that the simulated precipitation cells are too small/widespread when compared to radar observations. (c) As Fig. 2b, but for different PSP setups.
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
Adding the u and υ components to enforce nondivergent wind perturbations results in a marked increase in precipitation even without the T and qυ perturbations (PSP-3D w, α1). Including perturbations to T and qυ again (PSP-3D all, α1) further increases the effect. This suggests that the effect of nondivergent wind perturbations is now of same order of magnitude as the T and q perturbations. We emphasize that the relative amplitudes of the imposed wind and thermodynamic perturbations are determined from the subgrid variances derived from the turbulence scheme.
We further investigate the impact of the nondivergent wind perturbations on the precipitation structure (Figs. 5b,c). PSP-3D all, α1 has fewer but larger convective cells compared to PSP, α0 at a comparable total precipitation amount. Again we look at the S score to quantify the convective structure. In general, the model simulations show negative S values. This confirms the visual impression that the cells are too small and intense and lack the stratiform precipitation regions in the observations. During the daytime the PSP-AR simulation has slightly improved scores compared to the unperturbed reference simulations. The nondivergent wind perturbations result in a further improvement, particularly during the time of maximum convection in the afternoon. Hence, balanced wind perturbations lead to larger cell structures.
In general, the nondivergent wind perturbations are very effective in triggering convection and also appear to induce mesoscale circulations that cause larger precipitation cells.
5) Combining the modifications for PSP2
Finally, we choose a most suitable combination of modifications to PSP for further investigation. This includes the AR random field, HPBLcut and the nondivergent wind perturbations and will be denoted as PSP2. Since the 3D wind perturbations already improve the precipitation structure (see Fig. 4c, PSP2), we do not include the precipitation mask, which has the undesirable side effect of reducing convective initiation. Because of the many changes to the original scheme we start by determining a reasonable perturbation amplitude αtuning. Corresponding domain averaged precipitation time series are displayed in Fig. 6 for αtuning = 1, 1.5, 2. We select αfinal = 1.5, which produces perturbation amplitudes comparable to the original PSP (see blue line in Fig. 2) and use this value henceforth.
Domain averaged, hourly accumulated precipitation for 6 Jun considering different amplitudes for the new PSP2 setup.
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
We now simulate the entire high-impact weather period. For the analysis we divide it into a strong forcing period (29 May–2 June) and a weak forcing period (3–7 June). For both synoptic situations the modified PSP2 scheme produces an earlier peak of precipitation (Fig. 7a). This effect is desirable, since it addresses the known problem of late initiation of convection in kilometer-scale models. However, it should be noted that for this particular model and this weather period, the change does not result in an overall improvement in the precipitation forecast in comparison to radar observations.
(a) Daily cycle of hourly accumulated, domain averaged precipitation separated into a weakly and a strongly forced period. (b) Daily cycle of S(tructure) component of SAL for the weakly forced period and (c) FSS for the whole period (dotted), strong (dashed), and weak (solid) forcing. The FSS was computed using a threshold of 0.1 mm h−1 and a scale of 227 km.
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
For the weak forcing days the precipitation amplitude is also increased, while for the strong forcing days the amplitude is decreased slightly. This reduction could be caused by a reduction of CAPE in the evening due to earlier triggering or the 3D wind perturbations.
We again look at the S component of the SAL score to investigate the impact on precipitation structure (Fig. 7b). For the weak forcing period the PSP2 scheme shows a shift toward more coherent precipitation. This is a change in the right direction but compared to observations the convective cells are still too small scale.
Finally, in Fig. 7c we look at a common metric for assessing precipitation forecast skill, the fraction skill score (FSS). Displayed is the FSS for a scale of 227 km with a precipitation threshold of 0.1 mm h−1. Results have also been computed for other scales and thresholds (not shown). As expected the skill decreases for smaller scales and higher thresholds, however, the relative performance of the reference and PSP2 forecasts is similar for all values. In particular, the PSP2 perturbations lead to a slight improvement in skill for the weak forcing days, particularly during the time of maximum convective activity. For the strong forcing days, there is a skill increase during the time of convective initiation but a decrease of skill toward the evening hours. This is likely connected with the too early decrease in precipitation caused by the perturbations as discussed above. To evaluate the significance of these differences, longer periods need to be investigated.
To summarize, the new version, PSP2, includes the AR random field, the restriction of the perturbations to the boundary layer and the balanced wind perturbations. As a preliminary estimate, we retune the perturbation amplitude αtuning to 1.5 to match the precipitation amplitude of the original PSP scheme. For operational use, this choice would have to be revisited to produce the best overall precipitation forecast over a long test period. Although a clear overall improvement in forecast skill scores was not demonstrated in the preliminary evaluation here, the specific impacts of the changes to the scheme were as expected. In particular, PSP2 maintains the desired effect of the original PSP scheme, namely an earlier peak and increased amplitude of convection precipitation in weak forcing situations, while several drawbacks have been eliminated. In addition, it improves the cell structure and the spatial forecast skill compared to the unperturbed simulation.
5. SSOSP scheme
We now turn our attention to the SSO perturbations. First, we investigate the impact on the precipitation amount for different values of the correlation time scale τ and perturbation amplitude αref where αref = (αtuning × 2 h)/τ. This rescales the final perturbation amplitude for different values of τ with respect to a reference time scale τref = 2 h. To see the effects of the orography more clearly, we analyze orographic and flat grid points separately by defining orographic/flat grid points as grid points where the SSO standard deviation is larger/lower than 20 m.1,2 Corresponding domain averaged precipitation is displayed in Fig. 8. For orographic grid points increasing αref from 0.18 to 0.42 leads to more precipitation but, counterintuitively, a further increase of αref causes a reduction in the precipitation amount. Interestingly, flat grid points only experience a reduction with increased αref. One hypothesis for this decrease is that the perturbations enhance mixing of boundary layer and free tropospheric air, reducing instability, and leading to a reduction in precipitation. Increasing τ monotonically leads to more precipitation. More slowly evolving perturbations allow stronger updrafts to develop. These stronger updrafts cause an increased probability of convective initiation.
Domain averaged, hourly accumulated precipitation for a several different SSOSP settings. Precipitation averaged over (a) all grid points, that have a SSO-std ≥20 m, and (b) grid points with SSO-std <20 m.
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
For maximum impact we chose αref = 0.42 and τ = 5 h for further ensemble experiments with 10 members. The domain averaged precipitation time series in Fig. 9 shows reasonable spread between ensemble members for both orographic and flat grid points. This spread is comparable to spread generated by the original PSP scheme [see Kober and Craig (2016), Fig. 7d]. While spread over orographic regions is desired, further insight is necessary to understand the behavior over flat grid points. Moreover, the behavior of individual ensemble members suggests a significant increase in precipitation over orographic grid points and decrease over flat grid points. This is further illustrated in Fig. 9 where ensemble precipitation fields are displayed for two time steps: precipitation within the SSOSP scheme mainly occurs in the alpine region while most parts of Germany are precipitation free.
(a),(b) Domain averaged, hourly accumulated precipitation for an ensemble simulation with τ = 5 h, α = 1.05 (αref = 0.42). The bold red line gives the ensemble mean, while red shading indicates the standard deviation. Single ensemble members are displayed in thin red lines. In (a) precipitation is averaged over all grid points, that have a SSO-std ≥20 m, while (b) summarizes grid points with SSO-std <20 m. (c),(d) Example precipitation contours at (c) 1200 and (d) 1600 UTC for a 10-member ensemble. Shown are 0.1 mm precipitation contours for radar in gray (filled), reference in black, and SSOSP member in red.
Citation: Monthly Weather Review 147, 11; 10.1175/MWR-D-19-0060.1
Overall, the SSOSP scheme shows a nonmonotonic dependence on perturbation amplitude αref, while increasing the correlation time scale τ appears to give increasing precipitation over orographic regions. For flat regions however, the scheme causes an undesired reduction of precipitation.
6. Summary and discussion
a. Stochastic perturbations for subgrid turbulence (PSP)
In the first part of this paper, four concerns of the original PSP scheme were addressed and corrected by modifying different aspects of the original PSP scheme:
The autoregressive process in PSP-AR generates a physically more realistic, continuous evolution of the random field and hence refines the concept of the scheme. The performance is qualitatively comparable to the original PSP.
In PSP-HPBLcut we constrain vertically correlated perturbations to the boundary layer in accordance with the vertical extent of boundary layer eddies. This removes perturbations in the free troposphere that resulted in spurious nighttime precipitation.
By excluding perturbations in precipitating regions in PSP-mask we are able to reestablish organized convective structures. However, we decided not to include the PSP-mask adjustments in the revised PSP2 version for two reasons: first, the other modifications improved the breakup of organized structures already; second, this ad hoc fix has the undesirable side effect of reducing convective initiation overall.
In PSP-3D we included 3D nondivergent wind perturbations to prevent rapid dissipation of vertical velocity perturbations. Interestingly, these consistent wind perturbations are at least as effective as the T and qυ perturbations in triggering convection. The relative importance of buoyant and mechanical lifting in initiating convective updrafts in nature is still unknown. Both processes were found by Torri et al. (2015) to play a role in secondary initiation by cold pools. It also appears that the cell structures in the simulations with nondivergent wind perturbations were more comparable to radar observations. The relationship between triggering processes and the resulting convective structures is an important topic for future research.
We have combined the first, second and fourth modification to define the PSP2 scheme and selected a tuning parameter of αtuning = 1.5 to match the precipitation amplitude of the original PSP. That the optimal value of the tuning parameter is of order one is an encouraging sign that our physical reasoning is appropriate. By evaluating PSP2 on a longer time period, we find its behavior well in accordance with the rationale of the original scheme: when synoptic forcing is weak we observe a small shift in the diurnal cycle and an increase in total precipitation amplitude; this could counteract long-standing biases in current NWP systems (Clark et al. 2016; Baldauf et al. 2011). During strongly forced episodes or at night the scheme has only a small impact.
In addition to the systematic impact of the PSP scheme on convective initiation, studies have also demonstrated its ability to increase ensemble spread (Rasp et al. 2018; Keil et al. 2019.). This has potential to help reduce the underdispersion in many current convective-scale ensemble forecasting and data assimilation systems (see e.g., Berner et al. 2017; Ollinaho et al. 2017). Direct radar assimilation might benefit especially from such a perturbation scheme where sufficient spread in the first guess ensemble is particularly necessary. One remaining, undesired effect of the PSP2 scheme is a decrease of precipitation toward the evening. We note that this has been identified also for the original PSP by Rasp et al. (2018) who related this to a lack of convective organization. Organization of convection is caused by processes other than the ones addressed with the PSP2 scheme, most notably cold pools. A better representation of these features and their ability to trigger new convection might be necessary to capture organized evening convection. We aim to address this in a follow-up study.
Overall, we conclude the PSP revisions by recommending PSP2 instead of the original PSP: PSP2 contains comparable advantages as the PSP but without most of its drawbacks.
The question whether the PSP/PSP2 scheme improves the forecasts is more challenging to answer. Dealing with a highly complex model with dozens of adjustable parameters inevitably conjures up the curse of model tuning (Palmer and Weisheimer 2011). Over many years the parameters in an operational model are modified to achieve the best forecast scores. This most likely invoked a multitude of compensating errors. Introducing a more physical parameterization, therefore, demands a careful retuning of the parameters, which is a considerable computational effort. Furthermore, simulations using a full ensemble prediction system over a sufficiently long time period are necessary to thoroughly validate the impact of the PSP2 scheme. We intend to address these challenges in future work to allow for a more conclusive evaluation of forecast skill. In this paper we have focused on understanding the systematic changes caused by the perturbation schemes rather than on absolute forecast scores. The acquired physical understanding will help to guide development of a potential operational forecasting system.
b. Stochastic perturbations for subgrid orography (SSOSP)
In addition to the stochastic perturbations for turbulence, we have developed SSOSP, a scheme to account for subgrid orography and its impact on convective initiation. Doing so we consider mechanical lifting by SSO taking into account the stratification of the atmosphere. Again, reasonable values of the tuning parameter αtuning are of order one, supporting our physical reasoning.
The impact of the SSOSP on precipitation is not clear-cut. While we observe a precipitation increase over orographic regions the scheme also leads to a reduction of precipitation over flat terrain. Furthermore, there is a nonmonotonic relationship between the perturbation amplitude and the precipitation amount that may be related to a combination of mechanical triggering and a mixing-induced reduction of CAPE. An ensemble simulation with SSOSP showed that spread over orographic terrain is generated, but also over flat regions which is conceptually not desired.
These behaviors of the SSOSP make it hard to employ. Furthermore, the impacts are not clearly complementary to those of PSP2: the precipitation impact of SSOSP is either too weak or coupled with an undesired impact on flat regions, and the generation of ensemble spread by SSOSP can also be achieved with PSP2. For these reasons, we refrained from including the SSOSP perturbations in the PSP2 configuration. Conceptually, however, schemes for different subgrid processes could easily be combined.
In the introduction we referred to the open question of whether small-scale orography matters for convective initiation. Here we tested this by modeling subgrid mechanical lifting on the smallest resolved scale. Our results suggest that the subgrid orographic triggering effect most likely is not that relevant. On the one hand, to see the desired effect in mountainous regions we had to increase the perturbation strength to a point where unwanted effects occurred over flat terrain. On the other hand, the perturbations introduced by our scheme are active on the smallest resolved scales of the kilometer-scale model. Actual small-scale orography perturbations would be even smaller, and therefore likely even less influential. In general, our results confirm in a realistic setup what has been observed in other studies (Langhans et al. 2013; Schneider et al. 2018; Tucker and Crook 2005; Kirshbaum et al. 2007a): conceptually, SSO may be relevant in regions of insignificant resolved orography; yet due to the fractal, scale-invariant structure of orography (Turcotte 1987) regions of high subgrid orographic variations are generally also regions of high grid-scale orography. This larger scale, resolved orography may dominate the triggering of convection so that SSO is virtually irrelevant for convective initiation.
Acknowledgments
The research leading to these results has been done within the subproject “Representing forecast uncertainty using stochastic physical parameterizations” of the Transregional Collaborative Research Center SFB/TRR 165 (www.wavestoweather.de) funded by the German Research Foundation (DFG). We further acknowledge K. Kober and F. Brundke for developing a previous version of the SSOSP scheme that was used as baseline for the final SSOSP version.
APPENDIX A
SSOSP: Detailed Description
Details on the SSOSP scheme are given in the following two sections. First the formulation for orographically induced gravity waves is given in section a. Then, in section b, we approximate the gravity wave amplitude within the framework of the subgrid-scale orography for the parameterization.
a. Scaling orographically induced vertical velocity
b. Parameterization
The terms w0 and m are then used in Eqs. (7)–(9) to determine the amplitude w′ of the perturbations. Note that only grid points where μsso > 10 m are perturbed. The general framework for w′ is then given by Eqs. (7)–(9).
APPENDIX B
Computational Setup and Details
a. Computational details
Analysis and visualization was done using Python. The code and data are available from the authors upon request. For reading and processing the COSMO model output files the ens_tools package, which is based on the xarray library (Hoyer and Hamman 2017), was used which is being developed within the Waves to Weather project and will be available publicly soon.
b. SAL computation
We compute the SAL score using the threshold R* for identifying precipitation objects by using radar observations as reference; R* is computed based on the suggestion by Wernli et al. (2009): R* = (1/15)R95, where R95 corresponds to the 95th percentile of precipitation of grid points exceeding 0.1 mm h−1; R95 is computed separately for radar and forecast and for each time step. Different threshold choices, such as fixed precipitation amounts, did not change the outcome qualitatively.
REFERENCES
Anthes, R. A., 1986: The general question of predictability. Mesoscale Meteorology and Forecasting, P. Ray, Ed., Springer, 636–656.
Bachmann, K., C. Keil, and M. Weissmann, 2019a: Impact of radar data assimilation and orography on predictability of deep convection. Quart. J. Roy. Meteor. Soc., 145, 117–130, https://doi.org/10.1002/qj.3412.
Bachmann, K., C. Keil, G. Craig, M. Weissmann, and C. Welzbacher, 2019b: Predictability of deep convection in idealized and operational forecasts: Effects of radar data assimilation, orography and synoptic weather regime. Mon. Wea. Rev., https://doi.org/10.1175/MWR-D-19-0045.1, in press.
Baines, P., and T. Palmer, 1990: Rationale for a new physically-based parametrization of subgrid-scale orographic effects. ECMWF Tech. Memo. 169, https://doi.org/10.21957/h4h36b3u.
Baldauf, M., A. Seifert, J. Förstner, D. Majewski, M. Raschendorfer, and T. Reinhardt, 2011: Operational convective-scale numerical weather prediction with the COSMO model: Description and sensitivities. Mon. Wea. Rev., 139, 3887–3905, https://doi.org/10.1175/MWR-D-10-05013.1.
Berner, J., and Coauthors, 2017: Stochastic parameterization: Toward a new view of weather and climate models. Bull. Amer. Meteor. Soc., 98, 565–588, https://doi.org/10.1175/BAMS-D-15-00268.1.
Bierdel, L., P. Friederichs, and S. Bentzien, 2012: Spatial kinetic energy spectra in the convection-permitting limited-area NWP model COSMO-DE. Meteor. Z., 21, 245–258, https://doi.org/10.1127/0941-2948/2012/0319.
Buizza, R., M. Milleer, and T. Palmer, 1999: Stochastic representation of model uncertainties in the ECMWF ensemble prediction system. Quart. J. Roy. Meteor. Soc., 125, 2887–2908, https://doi.org/10.1002/qj.49712556006.
Chagnon, J. M., and P. R. Bannon, 2005: Adjustment to Injections of mass, momentum, and heat in a compressible atmosphere. J. Atmos. Sci., 62, 2749–2769, https://doi.org/10.1175/JAS3503.1.
Chapron, B., P. Dérian, E. Mémin, and V. Resseguier, 2018: Large-scale flows under location uncertainty: A consistent stochastic framework. Quart. J. Roy. Meteor. Soc., 144, 251–260, https://doi.org/10.1002/qj.3198.
Christensen, H., S.-J. Lock, I. Moroz, and T. Palmer, 2017: Introducing independent patterns into the stochastically perturbed parametrization tendencies (SPPT) scheme. Quart. J. Roy. Meteor. Soc., 143, 2168–2181, https://doi.org/10.1002/qj.3075.
Clark, P., N. Roberts, H. Lean, S. P. Ballard, and C. Charlton-Perez, 2016: Convection-permitting models: A step-change in rainfall forecasting. Meteor. Appl., 23, 165–181, https://doi.org/10.1002/met.1538.
Craig, G. C., and B. G. Cohen, 2006: Fluctuations in an equilibrium convective ensemble. Part I: Theoretical formulation. J. Atmos. Sci., 63, 1996–2004, https://doi.org/10.1175/JAS3709.1.
Craig, G. C., and T. Selz, 2018: Mesoscale dynamical regimes in the midlatitudes. Geophys. Res. Lett., 45, 410–417, https://doi.org/10.1002/2017GL076174.
Craig, G. C., C. Keil, and D. Leuenberger, 2012: Constraints on the impact of radar rainfall data assimilation on forecasts of cumulus convection. Quart. J. Roy. Meteor. Soc., 138, 340–352, https://doi.org/10.1002/qj.929.
Demaeyer, J., and S. Vannitsem, 2017: Stochastic parametrization of subgrid-scale processes in coupled ocean–atmosphere systems: Benefits and limitations of response theory. Quart. J. Roy. Meteor. Soc.,143, 881–896, https://doi.org/10.1002/qj.2973.
Demaeyer, J., and S. Vannitsem, 2018: Stochastic parameterization of subgrid-scale processes: A review of recent physically based approaches. Advances in Nonlinear Geosciences, A. A. Tsonis, Ed., Springer, 55–85.
Doms, G., and Coauthors, 2011: A description of the nonhydrostatic regional COSMO model. Part II: Physical parameterization. Deutscher Wetterdienst, Offenbach, Germany, 161 pp., http://www.cosmo-model.org/content/model/documentation/core/cosmoPhysParamtr.pdf.
DWD, 2018a: Hoch aufgelöste Niederschlagsanalyse und vorhersage auf der Basis quantitativer Radar und Ombrometerdaten für grenzüberschreitende Fluss-Einzugsgebiete von Deutsch land im Echtzeitbetrieb—Beschreibung des Kompositformats version 2.4.4. (High resolution precipitation analysis and prediction based on quantitative radar and ombrometer data for cross-border river catchment for Germany in real-time operation - Description of the composite format version 2.4.4). DWD, accessed 13 May 2019, https://www.dwd.de/DE/leistungen/radolan/radolan_info/radolan_radvor_op_komposit_format_pdf.pdf?_blob=publicationFile&v=6.
DWD, 2018b: RADOLAN/RADVOR Produktübersicht (RADOLAN/RADVOR product overview). DWD, accessed 13 May 2019, https://www.dwd.de/DE/leistungen/radolan/produktuebersicht/radolan_produktuebersicht_pdf.pdf?_blob=publicationFile&v=6.
Edson, A. R., and P. R. Bannon, 2008: Nonlinear atmospheric adjustment to momentum forcing. J. Atmos. Sci., 65, 953–969, https://doi.org/10.1175/2007JAS2278.1.
Fiedler, B. H., 2002: A wind transform for acoustic adjustment in compressible models. Mon. Wea. Rev., 130, 741–746, https://doi.org/10.1175/1520-0493(2002)130<0741:AWTFAA>2.0.CO;2.
Geerts, B., Q. Miao, and J. C. Demko, 2008: Pressure perturbations and upslope flow over a heated, isolated mountain. Mon. Wea. Rev., 136, 4272–4288, https://doi.org/10.1175/2008MWR2546.1.
Gill, A. E., 1982: Atmosphere-Ocean Dynamics. Academic Press, 662 pp.
Hanley, K. E., R. S. Plant, T. H. Stein, R. J. Hogan, J. C. Nicol, H. W. Lean, C. Halliwell, and P. A. Clark, 2015: Mixing-length controls on high-resolution simulations of convective storms. Quart. J. Roy. Meteor. Soc., 141, 272–284, https://doi.org/10.1002/qj.2356.
Hassanzadeh, H., J. Schmidli, W. Langhans, L. Schlemmer, and C. Schär, 2016: Impact of topography on the diurnal cycle of summertime moist convection in idealized simulations. Meteor. Z., 25, 181–194, https://doi.org/10.1127/metz/2015/0653.
Hirt, M., L. Schielicke, A. Müller, and P. Névir, 2018: Statistics and dynamics of blockings with a point vortex model. Tellus, 70A, 1–20, https://doi.org/10.1080/16000870.2018.1458565.
Houze, R. A., 2012: Orographic effects on precipitating clouds. Rev. Geophys., 50, RG1001, https://doi.org/10.1029/2011RG000365.
Hoyer, S., and J. Hamman, 2017: xarray: N-D labeled arrays and datasets in Python. J. Open Res. Software, 5 (1), http://doi.org/10.5334/jors.148.
Jankov, I., and Coauthors, 2017: A performance comparison between multiphysics and stochastic approaches within a North American RAP ensemble. Mon. Wea. Rev., 145, 1161–1179, https://doi.org/10.1175/MWR-D-16-0160.1.
Jankov, I., J. Beck, J. Wolff, M. Harrold, J. B. Olson, T. Smirnova, C. Alexander, and J. Berner, 2019: Stochastically perturbed parameterizations in an HRRR-based ensemble. Mon. Wea. Rev., 147, 153–173, https://doi.org/10.1175/MWR-D-18-0092.1.
Keil, C., F. Heinlein, and G. C. Craig, 2014: The convective adjustment time-scale as indicator of predictability of convective precipitation. Quart. J. Roy. Meteor. Soc., 140, 480–490, https://doi.org/10.1002/qj.2143.
Keil, C., F. Baur, K. Bachmann, S. Rasp, L. Schneider, and C. Barthlott, 2019: Relative contribution of soil moisture, boundary layer and microphysical perturbations on convective predictability in different weather regimes. Quart. J. Roy. Meteor. Soc., https://doi.org/10.1002/qj.3607, in press.
Kirshbaum, D. J., G. H. Bryan, R. Rotunno, and D. R. Durran, 2007a: The triggering of orographic rainbands by small-scale topography. J. Atmos. Sci., 64, 1530–1549, https://doi.org/10.1175/JAS3924.1.
Kirshbaum, D. J., R. Rotunno, and G. H. Bryan, 2007b: The spacing of orographic rainbands triggered by small-scale topography. J. Atmos. Sci., 64, 4222–4245, https://doi.org/10.1175/2007JAS2335.1.
Kober, K., and G. C. Craig, 2016: Physically based stochastic perturbations (PSP) in the boundary layer to represent uncertainty in convective initiation. J. Atmos. Sci., 73, 2893–2911, https://doi.org/10.1175/JAS-D-15-0144.1.
Langhans, W., A. Gohm, and G. Zängl, 2011: The orographic impact on patterns of embedded convection during the August 2005 alpine flood. Quart. J. Roy. Meteor. Soc., 137, 2092–2105, https://doi.org/10.1002/qj.879.
Langhans, W., J. Schmidli, O. Fuhrer, S. Bieri, and C. Schär, 2013: Long-term simulations of thermally driven flows and orographic convection at convection-parameterizing and cloud-resolving resolutions. J. Appl. Meteor. Climatol., 52, 1490–1510, https://doi.org/10.1175/JAMC-D-12-0167.1.
Leutwyler, D., D. Lüthi, N. Ban, O. Fuhrer, and C. Schär, 2017: Evaluation of the convection-resolving climate modeling approach on continental scales. J. Geophys. Res. Atmos., 122, 5237–5258, https://doi.org/10.1002/2016JD026013.
Lott, F., and M. J. Miller, 1997: A new subgrid-scale orographic drag parametrization: Its formulation and testing. Quart. J. Roy. Meteor. Soc., 123, 101–127, https://doi.org/10.1002/qj.49712353704.
Mellor, G. L., and T. Yamada, 1982: Development of a turbulence closure model for geophysical fluid problems. Rev. Geophys., 20, 851–875, https://doi.org/10.1029/RG020i004p00851.
Necker, T., M. Weissmann, and M. Sommer, 2018: The importance of appropriate verification metrics for the assessment of observation impact in a convection-permitting modelling system. Quart. J. Roy. Meteor. Soc., 144, 1667–1680, https://doi.org/10.1002/qj.3390.
Ollinaho, P., and Coauthors, 2017: Towards process-level representation of model uncertainties: Stochastically perturbed parametrizations in the ECMWF ensemble. Quart. J. Roy. Meteor. Soc., 143, 408–422, https://doi.org/10.1002/qj.2931.
Palmer, T., and A. Weisheimer, 2011: Diagnosing the causes of bias in climate models—Why is it so hard? Geophys. Astrophys. Fluid Dyn., 105, 351–365, https://doi.org/10.1080/03091929.2010.547194.
Panosetti, D., S. Böing, L. Schlemmer, and J. Schmidli, 2016: Idealized large-eddy and convection-resolving simulations of moist convection over mountainous terrain. J. Atmos. Sci., 73, 4021–4041, https://doi.org/10.1175/JAS-D-15-0341.1.
Piper, D., M. Kunz, F. Ehmele, S. Mohr, B. Mühr, A. Kron, and J. Daniell, 2016: Exceptional sequence of severe thunderstorms and related flash floods in May and June 2016 in Germany—Part 1: Meteorological background. Nat. Hazards Earth Syst. Sci., 16, 2835–2850, https://doi.org/10.5194/nhess-16-2835-2016.
Raschendorfer, M., 2001: The new turbulence parameterization of LM. COSMO Newsletter, No. 1, COSMO, 89–97, http://www.cosmo-model.org/.
Rasp, S., T. Selz, and G. C. Craig, 2018: Variability and clustering of midlatitude summertime convection: Testing the Craig and Cohen theory in a convection-permitting ensemble with stochastic boundary layer perturbations. J. Atmos. Sci., 75, 691–706, https://doi.org/10.1175/JAS-D-17-0258.1.
Schättler, U., G. Doms, and C. Schraff, 2015: A description of the nonhydrostatic regional COSMO-model. Part VII: User’s guide. Consortium for small-scale modelling (COSMO). Deutscher Wetterdienst, Offenbach, Germany, 227 pp., http://www.cosmo-model.org/content/model/documentation/core/cosmo_userguide_5.04.pdf.
Schmidli, J., 2013: Daytime heat transfer processes over mountainous terrain. J. Atmos. Sci., 70, 4041–4066, https://doi.org/10.1175/JAS-D-13-083.1.
Schneider, L., C. Barthlott, A. I. Barrett, and C. Hoose, 2018: The precipitation response to variable terrain forcing over low mountain ranges in different weather regimes. Quart. J. Roy. Meteor. Soc., 144, 970–989, https://doi.org/10.1002/qj.3250.
Schraff, C., H. Reich, A. Rhodin, A. Schomburg, K. Stephan, A. Periáñez, and R. Potthast, 2016: Kilometre-scale ensemble data assimilation for the COSMO model (KENDA). Quart. J. Roy. Meteor. Soc., 142, 1453–1472, https://doi.org/10.1002/qj.2748.
Shutts, G., 2005: A kinetic energy backscatter algorithm for use in ensemble prediction systems. Quart. J. Roy. Meteor. Soc., 131, 3079–3102, https://doi.org/10.1256/qj.04.106.
Torri, G., Z. Kuang, and Y. Tian, 2015: Mechanisms for convection triggering by cold pools. Geophys. Res. Lett., 42, 1943–1950, https://doi.org/10.1002/2015GL063227.
Tucker, D. F., and N. A. Crook, 2005: Flow over heated terrain. Part II: Generation of convective precipitation. Mon. Wea. Rev., 133, 2565–2582, https://doi.org/10.1175/MWR2965.1.
Turcotte, D. L., 1987: A fractal interpretation of topography and geoid spectra on the Earth, Moon, Venus, and Mars. J. Geophys. Res. Solid Earth, 92, E597–E601, https://doi.org/10.1029/JB092iB04p0E597.
Vissio, G., and V. Lucarini, 2018: A proof of concept for scale-adaptive parametrizations: The case of the Lorenz ’96 model. Quart. J. Roy. Meteor. Soc., 144, 63–75, https://doi.org/10.1002/qj.3184.
Vogelezang, D., and A. Holtslag, 1996: Evaluation and model impacts of alternative boundary-layer height formulations. Bound.-Layer Meteor., 81, 245–269, https://doi.org/10.1007/BF02430331.
Weisman, M. L., W. C. Skamarock, and J. B. Klemp, 1997: The resolution dependence of explicitly modeled convective systems. Mon. Wea. Rev., 125, 527–548, https://doi.org/10.1175/1520-0493(1997)125<0527:TRDOEM>2.0.CO;2.
Wernli, H., M. Paulat, M. Hagen, and C. Frei, 2008: SAL—A novel quality measure for the verification of quantitative precipitation forecasts. Mon. Wea. Rev., 136, 4470–4487, https://doi.org/10.1175/2008MWR2415.1.
Wernli, H., C. Hofmann, and M. Zimmer, 2009: Spatial forecast verification methods intercomparison project: Application of the SAL technique. Wea. Forecasting, 24, 1472–1484, https://doi.org/10.1175/2009WAF2222271.1.
Wetzel, P. J., 1982: Toward parameterization of the stable boundary layer. J. Appl. Meteor., 21, 7–13, https://doi.org/10.1175/1520-0450(1982)021<0007:TPOTSB>2.0.CO;2.
Wilks, D. S., 2005: Statistical Methods in the Atmospheric Sciences. Academic Press, 648 pp.
Zeng, Y., T. Janjić, A. De Lozar, S. Rasp, U. Blahak, A. Seifert, and G. C. Craig, 2019: Comparison of methods accounting for subgrid-scale model error in convective-scale data assimilation. Mon. Wea. Rev., in review.
