Extracting the lon and lat values only from indexes where the value isn't NaN

[ytmben]

I have the following Data array and I want to export a list of lons and lats to a .csv or .txt, but only for lonlat pairs have have a value ~NaN.
Here is my array:

<xarray.DataArray 'deptho' (latitude: 72, longitude: 83)> Size: 24kB
array([[      nan,       nan,       nan, ...,       nan,       nan,
              nan],
       [      nan,       nan,       nan, ...,       nan,       nan,
              nan],
       [6.672306 , 6.672306 , 7.5111613, ...,       nan,       nan,
              nan],
       ...,
       [      nan,       nan,       nan, ...,       nan,       nan,
              nan],
       [      nan,       nan,       nan, ...,       nan,       nan,
              nan],
       [      nan,       nan,       nan, ...,       nan,       nan,
              nan]], shape=(72, 83), dtype=float32)
Coordinates:
  * latitude   (latitude) float32 288B 31.02 31.06 31.1 ... 33.9 33.94 33.98
  * longitude  (longitude) float32 332B 32.54 32.58 32.62 ... 35.88 35.92 35.96
Attributes:
    long_name:      Bathymetry
    standard_name:  sea_floor_depth_below_geoid
    units:          m

Any suggestions?

[tayfuryldz]

A convenient way is to turn the two spatial dimensions into a single point dimension, remove the points where the data value is NaN, then convert the surviving coordinates to a table:

points = (
    da.stack(point=("latitude", "longitude"))
      .dropna("point")
)

out = points.to_dataframe(name="deptho").reset_index()
out[["longitude", "latitude"]].to_csv("lonlat.csv", index=False)

For your 2-D deptho(latitude, longitude), stack() creates one entry for each latitude/longitude pair. dropna("point") then removes exactly the entries whose deptho value is NaN, while the stacked index preserves the corresponding latitude and longitude labels.

If you also want the depth value in the output, just write:

out[["longitude", "latitude", "deptho"]].to_csv(
    "lonlat_depth.csv", index=False
)

An alternative is da.where(da.notnull(), drop=True), but note that drop=True drops coordinate labels for which all values are false; it does not by itself flatten away individual NaN cells inside the remaining 2-D rectangle. Stacking and then dropping NaNs is therefore the direct operation for a list of valid coordinate pairs.

I also checked this on a small 2-D DataArray with the same dimension layout before posting; the resulting rows contain only the lon/lat pairs corresponding to non-NaN values.