The issue was that the NumPy tolist function originally used converts to to the nearest compatible builtin Python type, however it does not convert nan to Python None as nan is a valid Python float:

>>> type(float('nan'))
<class 'float'>

So any nan values are left as-is, and persisted to the JSON output causing invalid JSON.

import numpy as np
arr = np.array([1.0, np.nan, 3.0])
arr.tolist()
[1.0, nan, 3.0]

This PR ensures nan are converted to None and then when converted to JSON are returned as null. NaN is not a valid JSON value. tolist takes care of converting other NumPy types such as np.uint32(2) (so no changes are required in json_serial in util.py.

cc @tomkralidis

Maybe estonia_light.tif can be added as a test (it is 11KB).