A downscaling algorithm needs something like a fine data file to determine the coordinate system and attributes of its product. If available a priori, it can also learn a lot from fine fields, obviously. downscale expects the geometrical surface height as its only fine predictor.

Test data

There is test data in the code for fine and coarse data. Its format is always NetCDF.

downscale/data_oasis3/ann

The fine data is all in one file, lfff00000000c400m.nc. It covers one layer of 3.10 *10^10 m^2 = 31 000 km^2 with 176 400 points, which corresponds to a resolution of 419 m.

lfff00100000_fin.nc

There was one curious file in there, lfff00100000_fin.nc. It contains fine-scale temperature data, but it is never used. Furthermore, it shows weird stripe artifacts in the plot.

Complete data

This first source of data is again the CLM (Climate Limited-Area Modeling) community like with the coarse data. Another possibility is the much finer SRTM data, which will have to be upscaled, i.e. interpolated to the appropriate grid.