It's day light savings isn't it!?

[thodson-usgs]

@gzt5142

[thodson-usgs]

This can be a teaching example for one of the earlier notebooks.
I think the simplest solution is to convert to UTC

[gzt5142]

Are you suggesting that the -5h offset is appropriate for half of the year, but not the other half? so only half the data matches... Hmm.

I would have thought that time stamps were UTC. I'll have to look at it tomorrow.

[gzt5142]

I think that the original notebook succeeded because of these lines:

# make sure the indices match
obs.index = obs.index.to_period('D')
mod.index = mod.index.to_period('D')
## note: mod and obs are pandas series

My hypothesis is that it re-indexed such that the time information was dropped. That particular pair of commands is specific to dataframes, which is why they worked on the pd.Series of that original notebook. Just need to find the equivalent action for an xarray dataset/dataarray, and I think that will solve the issue.

[gzt5142]

I think this will do it...

x = obs.sel(gage_id="USGS-03339000")
x_hat = model.sel(gage_id="USGS-03339000").resample(time='1D', offset='5h').mean()
x['time'] = x['time'].dt.floor('D')
x_hat['time'] = x_hat['time'].dt.floor('D')
(x.time == x_hat.time).count()

I get 15310 from that count.

The dt.floor('D') is the xarray replacement for the pandas to_period capability. Time data in xarray seems to have a special method associated with a datetime data type... .dt can do lots of stuff I've never considered before. Manipulates datetime values in useful ways.

[thodson-usgs]

@gzt5142, thanks! Yes, I was aware of dt but I probably assumed floor did something else.