Handling of missing data in verification

[yt87]

I have 15 years of sea ice forecast data, 9 months of daily forecasts issued once per month. I also have analysis for the same period, with about 20% of data missing. I am struggling with combining the two datasets together. Later on, I group the data and compute verification scores, like this one: xarray-contrib/xskillscore#396.
Here is an example of what I came with, a toy data:

import numpy as np
import pandas as pd

import xarray as xr

refix = pd.Index(np.arange(0, 21, 10), name="reference")
leadix = pd.Index(np.arange(1, 5), name="lead")

fcst = xr.DataArray(
    np.arange(12).reshape(4, 3), coords=(leadix, refix), dims=("lead", "reference"), name="fcst"
)
# simulate missing data
timeix = pd.Index([3, 4, 5, 7, 8, 10, 11, 12, 15, 17, 20, 21, 24], name="time")
obs = xr.DataArray(np.arange(timeix.size), coords=(timeix,), dims=("time"), name="obs")

# method='nearest' is necessary for .sel to work
temp = xr.concat(
    [obs.sel(time=rt + fcst["lead"], method="nearest") for rt in fcst["reference"]], "reference"
).transpose()
ds = xr.merge([temp, fcst])
# nullify the effect of method='nearest'
ds["obs"] = ds["obs"].where(ds.time==ds.reference + ds.lead)
ds

This appears to work, but I am not happy about the code. I have to use method='nearest' to select appropriate slice for each forecast reference time, but at the end those computed values are discarded (there are too many gaps to to allow for interpolation). Is there a better approach?

[yt87]

DataArray.reindex is the answer. Sorry for the noise.