Run "Filtering on CPUs versus GPUs" with data from pangeo catalog

[leticiafabre]

I am running this example but loading the NCAR data from pangeo catalog. Insted of filtering SST I filtered SSH just to test something different. That is what I got until now:

I think the next step is to select the area before filtering to save computional effort(?)

[paigem]

This looks amazing @leticiafabre!

Regarding your question - yes, in general it is best to select the region first before computing. But it depends how you wrote your code. If you wrote the code in xarray and only do the computation (or plotting) at the end, then you are already essentially slicing your region before computing.

[NoraLoose]

Yay, nice work @leticiafabre! 🎉

Have you played around with the filter scale: making it smaller and larger? You could also plot the difference SSH - filtered SSH, which will give you the small scales only (while the filtered SSH gives you the large scales).

[leticiafabre]

Thanks @paigem and @NoraLoose!

I will try change the filter scale and plot the difference SSH - filtered SSH.
Meanwhile, @anaclarabmariani is selecting a different area in the ocean to work in this project

[leticiafabre]

I run with 30km scale and that is the result:

Is the minimum scale for the filter the size of the grid?

[NoraLoose]

@leticiafabre look at those pretty plots - yay! 🎉

Another thing you could try is to label the x- and y-axis with longitude and latitude values (rather than nlon and nlat, which are just grid indices).

Is the minimum scale for the filter the size of the grid?

Correct, the minimum scale is the grid scale.

[paigem]

The POP model doesn't make it super straightforward to plot with lat/lon rather that nlat/nlon. Feel free to give it a try, but we can also make a note to discuss this more during our Zoom call tomorrow! 😊

[anaclarabmariani]

Here is the figure that I showed you today on the call. I tryed to select the North Tropical Atlantic Region and filtered the 60km scale.

[NoraLoose]

Looks awesome @anaclarabmariani 🎉

I have two small comments:

• If you'd like to allow for more space between the 3 subpanels you can use

plt.subplots_adjust(hspace=0.4) 

before you save the figure, and play around with the value for hspace, see here for more information.

• I just looked at your notebook, and I think you are using not a filter with a 60km scale but a filter with a factor of 60 (I know it is confusing!). Since the input data is on a 0.1 degree grid, a factor of 60 means that you filter to a scale of 0.1 * 60 = 6 degree. Does this make sense?

[anaclarabmariani]

• Cool! Thank you, Nora! I'll try to do that!

• About the filter scale: yes, that makes much more sense! Thank you, I'll work on that too.