Keepbits varies widely with different simulated streamflow gages?

[rsignell-usgs]

I tried computing keepbits for 99% real information for 100 simulated streamflow gages (output from the US National Water Model).

The dimensions of the Xarray dataset are gage_id and time.

So first I tried computing the whole collection along the time dimension and got a reasonable 8 bits.

Then I thought I'd try each gage individually, just to see what it produced. The results are all over the map:

Here is the full reproducible notebook.

Does this mean I'm doing something wrong, using the method for data it shouldn't be used for, or that the real information varies from station to station as determined?

[rsignell-usgs]

I'm going to take a stab at answering my own question here, as I think I've learned from the @milankl answer about snow depth here that this is basically the same issue -- some rivers have low entropy because they are dry or controlled discharge during certain parts of the year, so if I want to define keepbits for the entire variable (all gages), I should use compute the keepbits value for each gage and use the largest one.

[rsignell-usgs]

From #128 (reply in thread):

Hence in a situation where you really care to preserve information in one region, I'd suggest to analyse the information in the region with the highest information content, and then use it to compress the entire field.

I view this as confirmation that my answer to my streamflow question was correct! 😄 👨‍🎓