Misrepresentation of the number of precipitation events

[TanguiP]

Hello,

I am using xesmf to interpolate (conservative interpolation) precipitation products from their native grid to a common one for comparison purpose. However, I have recently found something odd. I am comparing precipitation frequency at each grid point (I.e. the number of non-null precipitation) and the interpolation failed to recreate this characteristic of the precipitation.
For a dataset and a variable called ds_1 and precipitation_3h, the frequency is given by the command: ds_1.precipitation_3h.where(ds_1.precipitation_3h>0).count(dim="time")
The following graphic show the mean precipitation (left) and the frequency (right) of the same precipitation estimate before (down) and after (up) interpolation.

Input grid (resolution 0.1°)
<xarray.Dataset>
Dimensions: (lat: 50, lon: 50, time: 248)
Coordinates:

• lon (lon) float32 290.0 290.1 290.2 ... 294.8 294.9 295.0
• lat (lat) float32 30.05 30.15 30.25 ... 34.75 34.85 34.95
• time (time) datetime64[ns] 2015-01-01 ... 2015-01-31T21:00:00
  Data variables:
  precipitation_3h (lat, lon, time) float32 0.03335 0.03451 ... 0.0 0.5153

Output grid (resolution of 0.25°)
<xarray.Dataset>
Dimensions: (lat: 19, lat_b: 20, lon: 19, lon_b: 20)
Coordinates:
k float64 0.1265

• lon (lon) float32 290.2 290.5 290.8 291.0 ... 294.0 294.2 294.5 294.8 (resolution of 0.25)
• lon_b (lon_b) float64 290.1 290.4 290.6 290.9 ... 294.1 294.4 294.6 294.9
• lat (lat) float32 30.25 30.5 30.75 31.0 31.25 ... 34.0 34.25 34.5 34.75
• lat_b (lat_b) float64 30.12 30.38 30.62 30.88 ... 34.12 34.38 34.62 34.88

I’ve tried to use other interpolation methods without success. Can you tell me if it is my fault, or this is a limit of the interpolation method?
Regards,
Tangui

[raphaeldussin]

This version of the package is no longer maintained. Would you mind moving your issue to https://github.com/pangeo-data/xESMF?