Wind speed and explicit aggregations

[jonasbhend]

This PR implements both derived parameters (e.g. wind speed from eastward and northward components) and explicit handling of aggregation periods. The latter allows us to evaluate hourly, 6-hourly and daily precipitation in the same evalml experiment or meteogram run.

Out of scope

Adopt the novel funtionality for plotting of animations of forecast fields. This is a somewhat larger refactor due to the different data layout and combination of high-resolution inner and global outer domain necessary for forecast maps.

Summary of changes

• Handle derived parameters in data_input
• Explicit aggregations for precipitation

Example output

TOT_PREC1	TOT_PREC6

Example dashboard: file:///M:/zue-prod/fc_development/seamless/S-RUC/evaluation/MRB-855_dashboard.html

[dnerini]

Very cool, thanks for this much needed feature! The PR looks already very mature, I just wonder whether we could already use it to integrate over plain variables too? Eg 24h min/max temperature? Or would you say this is out of scope?

[jonasbhend]

Very cool, thanks for this much needed feature! The PR looks already very mature, I just wonder whether we could already use it to integrate over plain variables too? Eg 24h min/max temperature? Or would you say this is out of scope?

Good suggestion, I will have a look. Two things though:

1. daily minimum and maximum temperatures are not the same as daily minimum and maximum of the hourly instantaneous temperatures. So in essence we should at least think about modelling these separately in the long term.
2. The current aggregation is rolling, i.e. moving window. This is a bit counter to the daily min/max application, but maybe it is not too hard to integrate, I just don't know yet.

[dnerini]

Good point! So maybe a better scope for this PR could be to stick to a rolling type of aggregation, eg 24h min/max/mean temperature?

[jonasbhend]

That we could definitely do, yes. And if you select the 00 UTC timestamps, you basically have daily min/max

[jonasbhend]

I was to quick there. No I don't think adding maximum / minimum temperature is a good idea as we do not support different aggregating functions. The current PR works for sunshine duration, precipitation, etc., but I would suggest we can discuss and implement alternative aggregation functions in a separate PR. It's a good point though, as we will need to support alternative aggregation functions for wind gusts, temperature, etc. Moving window aggregation, however, is not very efficient. Here it works because for sums, you can just use the cumsum / lagged-diff trick.

[clairemerker]

AI assisted review ;)

I added a few comments, feel free to address or not!