xarray：将数据变量与新的连续维度中的离散观察值相结合

Question

I am working with a crop calendar that records the day of the year (doy) at which a given phenological state occurs - here the mean planting ( plant ) and harvest ( harvest ) seasons (note that the nan printed below are pixels on oceans, the other values contain int ):

<xarray.Dataset>
Dimensions:  (y: 2160, x: 4320)
Coordinates:
  * x        (x) float64 -180.0 -179.9 -179.8 -179.7 ... 179.7 179.8 179.9 180.0
  * y        (y) float64 89.96 89.88 89.79 89.71 ... -89.71 -89.79 -89.88 -89.96
Data variables:
    plant    (y, x) float32 nan nan nan nan nan nan ... nan nan nan nan nan nan
    harvest  (y, x) float32 nan nan nan nan nan nan ... nan nan nan nan nan nan

I need to combine the two variables in a dataarray of dimension (doy: 365, y: 2160, x: 4320) in order to track, for each pixel, the phenological state as a function of the doy. Conceptually, the steps I identified so far are:

1. assigne a numerical value for each state, eg, off=0 , plant=1 , harvest=2
2. use the doy as an index to the corresponding day in the doy dimension of the new dataarray and assign the numerical value corresponding to the state
3. complete the values in between using something similar to pandas.DataFrame.fillna with method='ffill'

I went through the Reshaping and reorganizing data and the Combining Data pages, but with my current understanding of xarray I honestly don't know where to start.

Can anyone point me in a direction? Is what I am trying to do even achievable using only matrix operations or do I have to introduce loops?

PS: Apologies for the confusing formulation of the question itself. I guess that only reflects something fundamental that I am still missing.

Answer 1

You can exploit xarray's automatic broadcasting rules to create a boolean mask of all dates above/below an array of dates indexed by x/y:

# I'm assuming your "day of year" values are 1-indexed, and you're
# using a 365-day calendar. I'll leave leap year handling to you :)
days_of_year = xr.DataArray(
    np.arange(1, 366), dims=["day_of_year"], coords=[np.arange(1, 366)],
)

# broadcast against eachother with (x, y) <= (day_of_year, )
planted = ds.plant <= days_of_year
harvested = ds.harvest <= days_of_year

state = planted + harvested

The result will be an array with dimensions (x, y, day_of_year) containing the codes you described.

A note on growing season data:

I've worked with this type of data before, and one thing to watch out for is areas globally where the growing season spans Jan 1, which would break your above method. Alternatively, you could just create a "is_growing_season" mask which flexibly handles areas with growing seasons spanning calendar years:

is_growing_season = xr.where(
    ds.harvest >= ds.plant,
    ((days_of_year >= ds.plant) & (days_of_year <= ds.harvest)),
    ((days_of_year >= ds.plant) | (days_of_year <= ds.harvest)),
)