Mesoscale Parallel Experiment Log

Experiment Name

Eta-32 with modified treatment of surface data in the Eta 3DVAR analysis

Parallel Slot

ETAL

Control Slot

ETAV

Start date of parallel experiment

12Z 2002/10/30

End date of parallel experiment

00Z 2002/11/19

Environmental Modeling Center scientists

David Parrish, Eric Rogers

Abstract (including Motivation, Hypothesis and Method)

The 32-km EDAS/Eta parallel system is being used to test changes to the treatment of surface data in the Eta 3DVAR analysis
ANALYSIS CHANGES

The EDAS uses on the order of 10000 land surface observations of T, u, v, p and q every 3 hours. With the exception of surface pressure, none of this data is used in the GDAS. There is mounting evidence that this surface data is corrupting the eta forecasts. This is because the current background error for 3dvar has fixed vertical and horizontal correlation lengths which are only allowed to vary in the vertical. Generally data at the surface do not accurately reflect what is going on above the boundary layer. But the vertical correlation lengths used currently by 3dvar extrapolate the surface data by the same large distance everywhere, regardless of whether this is justified. The simple solution is to not use this data, as in the GDAS, but then the surface analyses are degraded. As a compromise, some changes have been made in the definition of vertical correlation length, background error variance, near surface wind-mass coupling, and observation error which attempt to minimize the impact of surface data away from a thin layer near the model surface.

1. vertical correlation length: The analysis variables for the eta 3dvar are stream and potential function, unbalanced temperature, unbalanced surface pressure, and moisture. The vertical correlation length is changed only for unbalanced temperature and moisture. These are reduced smoothly to a value of 5mb at the eta model surface. The transition starts about 30mb above the surface. This leads to significant horizontal variation in the analysis increment near the model surface, because ot the step mountains.

2. background error variance: The background error for unbalanced temperature only is increased smoothly using the same function of distance from the surface. Over the same transition layer of 30mb, the error increases so that it is 2 times larger at the surface.

3. wind-mass coupling: The statistical coupling between stream function and temperature (to get the balanced part of the temperature) is weakened using the same distance function as in 1 and 2. The coupling decreases to zero at the model surface.

4. Observation error: It was found from experiments that undesirable wind increments were produced near the surface if the vertical correlation length for stream and potential function were changed. This is why only the unbalanced temperature and moisture correlation lengths are modified. To still compensate for negative influence of surface wind data without tossing the data, the observation error for surface winds was increased by a factor of 3.

Experiment changes log

Background links

Evaluation of parallel results

Daily forecast maps

Daily forecast stats

Verification of precipitation and against rawinsondes / surface data

Conclusion

Mesoscale Parallel Home Page

Disclaimer