wpm() accepts zero-base/negative inputs producing silent zero/NaN

[brendancol]

Summary

wpm() (Weighted Product Model) in xrspatial/mcda/combine.py does not validate that input criterion values are positive. The product prod(x_i ** w_i) is undefined for x_i <= 0 when w_i is non-integer:

• 0 ** 0.5 == 0, so a single zero collapses the entire product to zero, masking bugs in upstream standardization.
• (-1) ** 0.5 == nan, so negative inputs silently propagate NaN through the output without an error.

The function docstring states inputs should be standardized 0-1, but no runtime check enforces this.

Reproduction

from xrspatial.mcda.combine import wpm
import xarray as xr, numpy as np

data = xr.Dataset({
    'a': xr.DataArray(np.array([[0.0, 1.0]])),
    'b': xr.DataArray(np.array([[-1.0, 2.0]])),
})
weights = {'a': 0.5, 'b': 0.5}
r = wpm(data, weights)
# returns [[nan, 1.41]] with no warning

Context

This is a sibling Cat 3 finding from the mcda security audit. The NaN-weight Cat 3 was fixed in #1311 / #1312. PR #1312 noted in its description that wpm accepting zero-base/negative-weight combinations was a deferred follow-up. This issue tracks that work.

Proposed fix

Validate criterion values before the product. Raise ValueError naming the offending variable when any value is <= 0 (NaN should still propagate, since that is the documented behavior tested in test_wpm_nan).