Match oceanographic profile data to calibrated echo data products

[ctuguinay]

From conversation between @leewujung @LOCEANlloydizard and some of the NOAA NWFSC folks -

It would be pretty helpful to have a function to match oceanographic profile data (e.g., from floats, CTDs, oceanographic models etc.,) to calibrated echo data products (i.e., Sv raw resolution, Sv interpolated, MVBS). The user should provide a dataset with (time, depth) or (latitude, longitude, depth) dimensions, and these should be enforced.

In the case when time exists in profile data: Nearest-neighbor match calibrated echo data ping_time.

In the case when latitude and longitude exists in profile data: Match based on calibrated echo data latitude and longitude. This is potentially similar to what is done in this function by @eliascapriles-NOAA - https://github.com/nmfs-ost/AA-SI_Oceanography/blob/0787d0ed3af5217bb59722d88332779a24461ecf/src/aaoceanography/utils/geo_location.py#L63.

[leewujung]

I think this is not just to MVBS, but also for Sv - either raw resolution or interpolated (#1597) ones - since it is common to want to keep everything at the highest resolution possible.

@ctuguinay could you change the issue title and description to reflect this? Thanks!

[ctuguinay]

@leewujung When matching on raw resolution Sv, are you then thinking to match it based on range sample too? Depth values would be matched on the 2D ping time and range sample grid. Anyways, I think this probably works out just fine.

[leewujung]

@ctuguinay : oh you're right I hadn't thought about this - I agree it should be matched to the depth variable alined with the range_sample dimension. But now that you brought it up, I wonder if we should restrict this function to only be applied on cases when the depth are uniform across channels, otherwise say for N channels with different range resolution, we will need to create N different matched oceanographic slices??

also adding @johnepohl-noaa @ElizabethMPhillips to the thread.

[ctuguinay]

@leewujung For the N difference matched oceanographic slices are you thinking of something like this?

(channel 1, time1) = depth array [0, 10, 20]
(channel 1, time2) = depth array [0, 5, 15]
(channel 2, time1) = depth array [5, 10, NaN]

where the depth array is indexed by range sample [0,1,2].

First match based on the (channel, time) tuple first and then match data across range sample.

[ctuguinay]

But yeah I agree, that might be pretty complicated. What could also happen is we create a function that does the matching in the uniform case, and then wrap it with xarray.apply_ufunc or something that could apply it across (channel, time) tuples efficiently in the non-uniform case 😆

[ElizabethMPhillips]

I'll have to think on this a little more, but it seems to me like having a feature that can handle non-uniform values (depth) across channels would be ideal. If I'm thinking of this correctly, that would allow a user to bring in raw oceanographic data (e.g., unique casts that stopped at different depths) - if they wanted to - and matched just to areas where the time/lat/lon/depth align, as well as smoothed ocean data frames that can then be matched to gridded acoustic data...But I suspect there will almost always be some sort of mismatch in depth/range that should have a way to get handled in echopype

[johnepohl-noaa]

Hi! Thanks for adding me into this conversation, looking forward to it. I think I understand the general discussion here. The only thing I'd add, is that its important to recognize there are definitely places where sharp horizontal changes in hydrography can occur rapidly. So users of a feature that allows them to 'link' their pings to the nearest CTD, need to be aware of artifacts across their general survey region that might make them want to match to a different CTD. For example, in the CCE, a survey line running along the Mendocino Escarpment can potentially get different water mass characteristics, depending on if they link to a CTD north or south of the escarpment. That is a bathymetric effect that will stand apart of a simple Lat, Lon, or Depth match. Policing this seems outside the scope of EchoPype etc, but perhaps it can be mentioned in some kind of Users Guide.