issue_comments

For reference, it may be helpful to try your example on a smaller dataset: ds = xr.Dataset() ds['data1'] = xr.DataArray(np.arange(100), coords={'t1': np.linspace(0, 1, 100)}) ds['data1b'] = xr.DataArray(np.arange(100, 200), coords={'t1': np.linspace(0, 1, 100)}) ds['data2'] = xr.DataArray(np.arange(200, 500), coords={'t2': np.linspace(0, 1, 300)}) ds['data2b'] = xr.DataArray(np.arange(600, 900), coords={'t2': np.linspace(0, 1, 300)}) This is what the data looks like: ```

ds <xarray.Dataset> Dimensions: (t1: 100, t2: 300) Coordinates: * t1 (t1) float64 0.0 0.0101 0.0202 0.0303 0.0404 0.05051 0.06061 ... * t2 (t2) float64 0.0 0.003344 0.006689 0.01003 0.01338 0.01672 ... Data variables: data1 (t1) int64 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ... data1b (t1) int64 100 101 102 103 104 105 106 107 108 109 110 111 112 ... data2 (t2) int64 200 201 202 203 204 205 206 207 208 209 210 211 212 ... data2b (t2) int64 600 601 602 603 604 605 606 607 608 609 610 611 612 ... and here's what happens with `where`: ds.where(ds.data1 < 50, drop=True) <xarray.Dataset> Dimensions: (t1: 50, t2: 300) Coordinates: * t1 (t1) float64 0.0 0.0101 0.0202 0.0303 0.0404 0.05051 0.06061 ... * t2 (t2) float64 0.0 0.003344 0.006689 0.01003 0.01338 0.01672 ... Data variables: data1 (t1) float64 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 11.0 ... data1b (t1) float64 100.0 101.0 102.0 103.0 104.0 105.0 106.0 107.0 ... data2 (t2, t1) float64 200.0 200.0 200.0 200.0 200.0 200.0 200.0 ... data2b (t2, t1) float64 600.0 600.0 600.0 600.0 600.0 600.0 600.0 ... ```

I suspect this probably isn't really doing what you want, unless you really want two-dimensional versions of data2 and data2b. It probably makes more sense to subset out the relevant variables first.

Broadcasting producing gigantic arrays without any warning is really a NumPy issue, e.g., trynp.where(True, np.zeros((1000000, 1)), np.ones((1, 1000000))) (this will probably crash your computer!).

. So in my case extracting the data with the shared dimension using ds.drop() is appropriate. It would be nice to have xarray throw a warning or error to prevent me chomping up all the RAM in my system if I do try to do this sort of thing though.

I'll let @shoyer give a definitive answer here, but I don't think that .where is meant to check whether the input makes "sense" or not. What happens is related to how xarray chooses to broadcast non matching dimensions:

```python import xarray as xr import numpy as np d1 = xr.DataArray(np.arange(3), coords={'t1': np.linspace(0, 1, 3)}, dims='t1') d2 = xr.DataArray(np.arange(4), coords={'t2': np.linspace(0, 1, 4)}, dims='t2')

d2 * d1 <xarray.DataArray (t2: 4, t1: 3)> array([[0, 0, 0], [0, 1, 2], [0, 2, 4], [0, 3, 6]]) Coordinates: * t2 (t2) float64 0.0 0.3333 0.6667 1.0 * t1 (t1) float64 0.0 0.5 1.0

d2.where(d1 == 1) <xarray.DataArray (t2: 4, t1: 3)> array([[ nan, 0., nan], [ nan, 1., nan], [ nan, 2., nan], [ nan, 3., nan]]) Coordinates: * t2 (t2) float64 0.0 0.3333 0.6667 1.0 * t1 (t1) float64 0.0 0.5 1.0 ``` which "makes sense", but is going to have a huge memory consumption if your arrays are large.

Thanks for the report! It would be great if you could be a bit more specific: - if data1 and data2 are unrelated, why do you want to apply where on both variables? What is your expectation on the output? - do you have the possibility to produce a minimal, self-contained working example?