Experiences with the Hirlam including the CBR scheme

Wim de Rooy

Introduction

In this note verification results are shown of the Hirlam including the CBR (Cuxart, Bougeault, Redelsperger, 2000) scheme for vertical diffusion. In the first part of this note wind speed at the two lowest model levels and 10m is compared with measurements made in a 200m high meteorological tower. In the second part the results of Hirlam including CBR are compared with the Hirlam including the Holtslag scheme during the "Danish" 3 december storm. Finally some plans for the future are presented.

Verification of Hirlam CBR wind speed against tower observations

For the period of one month (July 1996) we compared wind speed at 10m and the lowest two model levels with observations made in the 200m high meteorological tower in Cabauw (near the centre of the Netherlands). In the first figure the typical behaviour, at least for this month, of the model wind speed at 140 (second lowest model level) and 10m is shown. During strong wind, near neutral, conditions, the wind speed at 140m is largely underestimated. See e.g. Fig. 1 (also in postscript) at hour 528 where the wind speed at 140m is underestimated with 5.5 m/s . At the same time the 10m wind speed is about correct or even a bit too high.

Figure 2 shows that this suggests too much vertical mixing of momentum.

In Figure 3 (also in postscript ) the overall results for the complete month are shown. The strong negative bias in the wind speed at 140m is the result of the underestimation's during near neutral, strong wind conditions. The 30m wind speed has a positive bias. With Figure 2 in mind this clearly indicates the too strong momentum mixing. The almost zero bias of the 10m wind speed can be explained by the difference in roughness length between the model and the local conditions in Cabauw (this Representativeness Problem is discussed in De Rooy, 1999). Cabauw is situated in a relatively open terrain in comparison with the corresponding average Hirlam grid box area. Hence the actual (direction dependent) roughness length in Cabauw is significantly smaller than the roughness length used in the model to transform the wind speed at 30m downwards to 10m. Here we have a nice illustration of a compensating error and the fact that the wind speed at 10m should not, or very carefully, be used for validation purposes.

Comparison between Hirlam with CBR and Hirlam with Holtslag during a storm

As we have seen from the last section, the CBR scheme seems to result in too much vertical mixing during strong wind situations. We therefore decided to investigate the performance during a storm and to compare the results with the output of Hirlam including the Holtslag scheme for vertical diffusion. For the comparison the "Danish" 3 December 1999 storm is chosen. During this storm a station near the coast of Denmark reported 85 kts, just before it crashed. So near neutral conditions in this area can be guaranteed. All verification results in this section refer to 3 december 18:00 GMT when the storm was at its maximum. Further we only present +12h forecasts (other forecast lengths show the same behaviour). At the KNMI, Hirlam version 4.3, thus including the Holtslag scheme for vertical diffusion, is still the operational version. Therefore data produced by the operational model could be used for this comparison study. To obtain the results for the CBR version, Hirlam version 4.6.2 including the CBR scheme, was started at 1 December 0:00 GMT in order to let the model spin up and simulate the normal assimilation cycle. The area on which both model versions run is the same. However to save disk space, the output fields of the CBR version cover a smaller area as can be seen in Figs. 4, 5 and 6.

Figure 4 Pressure at mean sea level

Figure 4, where the pressure at mean sea level is plotted, shows that the depression core is situated at about the same location for both Hirlam versions. However, the depression core is approximately 5 hPa less deep in the CBR version (CBR=965 hPa, Holtslag=960 hPa, observed=958 hPa). This might be explained by the stronger vertical mixing of momentum in the CBR version which will lead to a stronger ageostrophic wind and as a consequence an increased filling up of the depression core.

Fig. 5 Wind speed at 10m

Fig. 5 with the wind speed at 10m shows the larger area with high wind speeds in the CBR version. Looking e.g. at the centre of the Netherlands, the wind speed is 20 m/s in the CBR and 12.5 m/s in the Holtslag version. As mentioned before the wind speed at 10m is not a good validation measure but the Holtslag values are near the observed wind speeds whereas the CBR winds are almost twice as high. Again the results are consistent with the too-much-mixing-hypothesis of the last section (see Fig. 2).

Fig. 6 Wind speed at 925 hPa

Finally the wind speed at 925 hPa is plotted in Fig. 6. This figure clearly reveals the much higher wind speeds at 925 hPa with the Holtslag scheme. This implies the relatively small difference in wind speed near the surface and at higher levels in the CBR scheme. In summary, it can be stated that this storm case confirms the behaviour of the CBR scheme described in the last section, namely too much vertical mixing of momentum during near neutral conditions.

Plans for the future

The above mentioned results were send to Joan Cuxart, who is the "C" in CBR and also the person who made the CBR scheme available for Hirlam. Three points in his reply are mentioned here.

Also in Spain, results with the CBR scheme show too high 10m wind speeds (although other parameters generally slightly improved)
A constant in the momentum flux formulation was originally 0.067 but mainly due to results with the Ayotte case this constant was set to 0.4 (therewith increasing the vertical mixing with a factor 6). On basis of the results mentioned in this paper this choice should be revisited. Accidentally we already did some experiments with this constant set to 0.06 at the KNMI. The results against Cabauw tower data showed that the vertical mixing was much more realistic but still somewhat too large.
Experiments with a new formulation for the mixing length, proposed by Redelsperger, has been tested within Meso-Nh in a 1D framework (which uses 0.067 for the above mentioned constant). These experiments show a further reduction in the vertical mixing and the 10m wind speed.

So a combination of the last two points might solve the problem and will result in realistic vertical mixing of momentum in the CBR scheme during near neutral conditions. At the moment the necessary modifications are further investigated at the INM.

Acknowledgements
I would like to thank Geert Lenderink, Sander Tijm, and Pier Siebesma for many helpful discussions and suggestions

References

Cuxart, J., P. Bougeault, and J.-L. Redelsperger, 2000: A turbulence scheme allowing for mesoscale and large-eddy simulations. Quarterly Journal of the Royal Meteorological Society, Vol. 126 pp. 1-30

De Rooy, W.C., 1999: Hirlam near-surface ouput. Investigation of error sources and possible improvements in the Hirlam 2m-temperature and 10m-wind speed, Hirlam newsletter No. 33