Averaging of footprints should be done after merging.

[rt17603]

Original report by Luke Western (Bitbucket: lwuob, GitHub: lukewestern).

---

Summary:

Currently we average measurement data, average footprints and match these. We should instead, if feasible, match obs to footprints then average this merged dataset.

Problem is if e.g. there is only 1 measurement in a day with 24H averaging as the footprint will be averaged to 24H even though only 1H of data is used.

Possible Fixes:

Could e.g. always read in data at 1H averaging, merge with footprints and average the merged data set.

[rt17603]

Original comment by Rachel Tunnicliffe (Bitbucket: rt17603, GitHub: rt17603).

---

So the fix I have been putting in for the parallel part of OpenGHG is to use an explicit “sampling_period” for the obs data meaning we don’t just assume the observation frequency = the length between observations made. In that case if the sampling period for the one measurement was < footprint frequency (e.g. < 1H) this should just be aligned with the correct footprint rather than averaged to 24H. (unless I’m misunderstanding your scenario?)

How would your proposed fix fit alongside that do you think? You’re thinking reindex and then resample in some way?

[rt17603]

Original comment by Luke Western (Bitbucket: lwuob, GitHub: lukewestern).

---

We can perhaps discuss at the next modelling meeting as it may be a case that you have a better way of doing things, but my first thought would be to:

1. Read data always at 1H averaging (as this is the fp resolution). We can discuss how applicable this is, i.e. we care about the length of time in which the instrument samples over rather than the period between measurements.
2. Merge data with footprints
3. Resample merged dataset to desired averaging period.

‌

[rt17603]

Original comment by Luke Western (Bitbucket: lwuob, GitHub: lukewestern).

---

This also applies to filtering: filtering should be done before measurements are averaged to frequency lower than footprints.