Schomburg et al. (2010) describes this downscaling scheme in detail. This page will only point out some relevant aspects.

Three steps

The Schomburg scheme uses 3 steps to derive a fine active field from coarse and external data:

1. A bi-quadratic spline interpolation
2. A Schomburg rule if available to produce anomalies
3. Noise generation

The 1st step instates the coarse pixel’s value as the average value of all the fine pixels it maps to. The letter steps leave this average untouched. Some variables are not submitted to these 2 steps at all.

Training

The Schomburg rules have been devised with training data from COSMO-DE with a grid spacing of 400 m. To produce predictors, the field was upscaled to a grid spacing of 2800 m.
In the training, all data was assumed to be taken at the surface or at the airborne lowest layer, which for COSMO-DE has its center at 10 m above ground. Not coincidentally, the training area is a subset of this project’s coarse and fine data.

Implementation

This project relies on an implementation into OASIS3. It does not provide for all variables the Community Land Model requires. Model1 admits the required variables under a wrong name to OASIS3, so that they can be processed at all.

What steps of the Schomburg scheme apply to which variables?

The 2nd and 3rd step of the Schomburg scheme are implemented in source:oasis3/lib/downscal/src/src_downscal2.F90 . This code shows that neither of these steps are implemented for:

• UWIND
• VWIND
• DIRSW
• DIFSW
• LONGW
• CVPRE (fn[7])
• GSPRE

Both are implemented for:

• TEMP (fn[8])
• PRESKE1

SPWAT, described in a comment as ‘Sp. Humidity Field at Lowest Model Level’, experiences only the 3rd step, but namcouple does not call upon it. Instead, it calls SPWATKE1, which is[9] undefined in src_downscal2.F90.

Footnotes