Aurora: regional scale

Dear @wesselb,

Using the examples available at https://microsoft.github.io/aurora/example_hres_t0.html as a
starting point, we are adapting the Aurora model inputs to a regional scale, focusing on areas
of approximately 20°x40° or smaller. So far, we have concentrated on comparing the
temperature values predicted by the global and regional versions of the model. We have also
compared these results with HRES T0 data. For example, the following figure shows the
surface temperature obtained from the global and regional Aurora and its difference for our
study region.

A similar result was obtained when we compare the Aurora and HRES T0 results.

We initially expected the largest temperature differences between the global and regional
Aurora predictions to occur at the edge of the selected area, while the central region would
show very similar values in both simulations. However, the largest differences observed over
land surfaces, specifically over the northwest African coast and the Canary Islands, whereas
there is strong agreement over the oceanic region.
Could you help us to understand why aurora predicts with so many differences the temperature values over the ocean and the land?

We understand the need for fine-tunning specific to our study region but before this task,
what other factors would you consider important as input parameters for the model? For
example, window_size, decoder_depths, etc. We believe that some of these parameters
should modify to better adapt Aurora to a regional scale, but we do not know which ones
might be more important in this respect.

Thank you in advance for your help

Regards,
Sergio León

Start with the linked Aurora high-resolution example and compare the reported global and regional temperature results against the HRES T0 data described in the issue. The issue names no repository files, tests, or entry points, and it does not define a concrete change or acceptance criteria; clarification is needed before implementation can begin.