Speeding up calculations

[JannisHoch]

On my Mac with 16GB RAM and M1 chip, processing roughly half a year of daily at 0.1 degree resolution and global scale** took me 30+ min*** for the Oudin method, accoring to % time in jupyter notebook. It used around 50 % of my CPU and 20 % of my memory, according to htop.

For methods requiring more input data, the processing time will likely increase.

Maybe worthwile looking into ways to speed up the calculations, such as numba, if that's compatible with xarray?

** so yes, using gridded data. And admittedly, it also contained the ocean surface. In total, 1800*3600 cells had to be processed per time step.

*** I stopped it then.

[mvremec]

Hi Jannis,

this is indeed a good point! Thanks!
I will look into this and see if we can use our unvectorized functions or if we need to vectorize them.
xarray.apply_ufunc might help.

Cheers, Matevz

To do:

• [check whether we need to vectorize our function to use xarray.apply_ufunc or Numba]
• [check xarray.apply_ufunc]