| DATE | DATA SOURCE | DURATION |
| Spring 1994 | MSRC-SUNYSB | 17 Days |
| Summer 1994 | ARM Web Site | 16 Days |
| Fall 1994 | CSU Web Site | 19 Days |
| Spring 1995 | ARM Web Site | 18 Days |
| Spring 1995 | CSU Web Site | 17 Days |
| Fall 1995 | CSU Web Site | 24 Days |
Relaxation Techniques
After running our SCM for each of the six IOPs, we often found large differences
between model temperature and humidity and observations.
The temperature errors can be as high as 20K towards the end of a model
run. Other SCM modeling groups participating in ARM have also found similar
errors (see Notes from the LLNL SCM Workshop, April 1996). The introduction of
a simple relaxation correction (time constant = 24 hrs)
on the model temperature and humidity profiles drastically reduces these errors.
Model temperature results from the Summer 1995 IOP are shown in
Figure 1 (no relaxation) and
Figure 2 (with relaxation).
Fall 1994 IOP SCM temperature results are shown in
Figure 3 and
Figure 4. The temperature error in these two model
runs were reduced between 50-80%.
Analysis of the temporal mean temperature correction for each of the six IOPs tested (Figure 5) revealed some interesting patterns. Namely, below about 800 mb, the SCM was producing temperatures higher than observed in all six IOPs. Between 800 and 300 mb, there seemed to be no trend, while above 300 mb the SCM again produced temperatures that were too high compared to the sounding measurements. These areas where the model is consistently producing excessively warm temperatures may indicate either a deficiency in the model physics or problems in the production of the forcing terms. We suggest that an intermodel comparison between the SCM groups be conducted to help determine the source of these errors.
Prognostic Cloud Liquid Water
The model precipitation results from each of the six IOPs in Table 1 are shown in
Figure 6a and
Figure 6b.
These model runs were performed using a relaxation correction (Tau=24 hours) to
the model temperature and humidity profiles.
The SCM precipitation
compares very well with surface measurements from the Oklahoma Mesonet
in 5 of the 6 IOPs. These encouraging results allowed us to begin a
preliminary investigation of whether the inclusion of cloud liquid water
as a prognostic variable improves the model results when compared to
ARM measurements. We tested 5 different model configurations (shown in Table 2) which differed
only in the specification of cloud liquid water, cloud optical thickness
and effective droplet radius.
| Cloud Liquid Water | Cloud Optical Thickness | Effective Droplet Radius | |
| CCM2 | None | Specified=f(T,P) | Not Applicable |
| CW | Explicit | Specified=f(T,P) | Not Applicable |
| CWRF | Explicit | Calculated | Fixed (10 microns) |
| CWRV | Explicit | Calculated | Varying (warm clouds only) |
| CWRI | Explicit | Calculated | Varying (all clouds) |
Our results show that total cloud fraction (Figure 7), downwelling surface shortwave (Figure 8) and outgoing longwave (Figure 9) are much better approximated (compared to surface and satellite measurements) using model configurations that include cloud liquid water as a prognostic variable.
Model Cloud Height Results
A detailed analysis of the model cloud fractions from the Spring 1994
IOP indicate that the SCM is underestimating the amount of low clouds (surface
to 700 mb) and overestimating the amount of high clouds (above 400 mb)
compared to satellite estimates (see Figure 10).
The observed low cloud amount reaches a maximum on April 22 and is
concurrent with a reduction in the observed surface downwelling solar
radiation (Figure 11). However, the observed
reduction in downwelling solar on April 22 is relatively modest, suggesting
that the low clouds may be optically and geometrically thin clouds.
Reruning the model with a much finer
vertical resolution (53 layers compared to 16 layers) resulted in model
cloud fields closer to the satellite estimates
(Figure 12). The model downwelling solar from
the high resolution run is also closer to surface observations
(Figure 13).
A major implication
from this result is that the forcing data should be supplied
on a vertical grid fine enough to support these "high" resolution model
runs.
Sensitivity to Surface Latent Heat Flux
Several SCM experiments were conducted to investigate the sensitivity of
model precipitation to changes in the magnitude of the surface latent
heat flux. A sensitivity experiment was performed for each of the six IOP data sets
listed in Table 1.
A flow chart illustrating the structure of each sensitivity experiment is
shown in Figure 14. Each sensitivity experiment
consisted of three individual SCM runs. The SCM is first run (Control) using 24-hour
relaxation with the correction terms being saved. In the next two runs (Experiment 1 and 2)
the correction terms from the control run are used in place of relaxation. The latent
heat flux in Experiment 1 is 90% of that used in the Control run, while the latent
heat flux in Experiment 2 is 110% of that used in the Control run.
To help analyze the results of these experiments we have defined the factor A_r, where
A_r = LN [ ( {Prec_110} - {Prec_090} ) / ( {FQ_110} - {FQ_090} ) ].
| IOP | A_r |
| Spring 1994 | +0.608 |
| Summer 1994 | -0.350 |
| Fall 1994 | +0.506 |
| Spring 1995 | +0.347 |
| Summer 1995 | +0.195 |
| Fall 1995 | -0.649 |
The results from the six sets of experiments varied considerably. In some experiments the change in precipitation exceeded the changed in the surface latent heat flux, while in other cases the opposite held true. Figure 15 shows the temporal mean vertical velocity profile from each of the six IOPs. The mean vertical velocity is predominantly upwards in 5 of the 6 IOPs. However, there is considerable variability between the 6 profiles. We found that the term A_r is closely related to the vertical velocity in the lower troposphere. Figure 16 shows the term A_r vs. the mean vertical velocity in the region 900-600 mb for each of the six IOPs. These experiments illustrate, that under certain conditions, an error in the surface latent heat flux = D, may result in an error in the model precipitation of up to 2D.
