0

I tried to implement a moving percentile on climatic data using xarray. I've a dataset with daily min and max temperature, something like this:

<xarray.Dataset>
Dimensions:    (latitude: 60, longitude: 20, time: 14456)
Coordinates:
  * latitude   (latitude) float64 39.05 39.05 39.15 39.15 ... 41.85 41.95 41.95
  * longitude  (longitude) float64 8.05 8.15 8.25 8.35 ... 9.65 9.75 9.85 9.95
  * time       (time) datetime64[ns] 1980-01-01 1980-01-02 ... 2019-07-30
Data variables:
    tn         (time, latitude, longitude) float32 nan nan nan ... nan nan nan
    tx         (time, latitude, longitude) float32 nan nan nan ... nan nan nan
Attributes:
    E-OBS_version:  20.0e
    Conventions:    CF-1.4
    References:     http://surfobs.climate.copernicus.eu/dataaccess/access_eo...
    history:        Wed May 13 17:26:55 2020: ncrcat tn_ens_mean_0.1deg_reg_1...
    NCO:            netCDF Operators version 4.8.1 (Homepage = http://nco.sf....

For each day of the year I want to compute a moving quantile taking a moving window of 5 days. For better understanding see this pic, the window take 5 days of each year and compute the quantile (the green, red and blue boxes represent the moving quantile filter).

I didn't found the correct combination of rolling and groupby operators to achieve my purpose. So I solved it using a (slow) for loop. I created a list of dates and I use sel operator to select and compute the quantile, something like:

# all_days_lists is a list of lists
for days_list in all_days_lists:
   # select days of interest
   tmp_ds = xdataset.sel(time=days_list)
   # compute 90th percentile
   p90 = tmp_ds.quantile(0.9,dim='time')

Any ideas to solve this problem? Thanks a lot.

francesco
  • 1

1 Answers1

0

I think this will work for daily data:

da.rolling(time=5).reduce(np.percentile, q=90)

You can use reduce to apply most numpy functions to a rolling window of your data.

Charles
  • 97
  • 5