Least squares polynomial fit with Dask about esmlab

[bradyrx]

One thing we've come across at climpred is that polyfit and other functions return zeros or NaNs when they encounter missing data in a time series. This isn't a problem to worry about if you only anticipate esmlab to be used in climate model output, but observational data frequently has missing data.

We deal with this by temporarily doing a linear interpolation to fill in the NaNs, then force those back to NaNs on the returned output (in the case of detrending; wouldn't need to replace the NaNs if just getting the slope): https://github.com/bradyrx/climpred/blob/8c1700e94d124424f5f2adf5250a22f2ff8bb3dc/climpred/stats.py#L118

We also have some snippets in there to handle Datasets in addition to DataArrays.

[klindsay28]

@kmpaul , I'm curious why this issue was closed with no explanatory comment.
It doesn't look like the enhancement has been implemented.

[andersy005]

@klindsay28, I just re-opened it

[andersy005]

I am planning on working on this as part of pydata/xarray#3349