Multispectral tools: convert data to float32 dtype before doing calculations

[thuydotm]

Fixes #726.

TODO:

• add tests

[codecov-commenter]

Codecov Report

Base: 79.88% // Head: 79.88% // No change to project coverage 👍

Coverage data is based on head (bb04f35) compared to base (af42544).
Patch coverage: 100.00% of modified lines in pull request are covered.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #755   +/-   ##
=======================================
  Coverage   79.88%   79.88%           
=======================================
  Files          19       19           
  Lines        4170     4170           
=======================================
  Hits         3331     3331           
  Misses        839      839           

Impacted Files	Coverage Δ
xrspatial/multispectral.py	69.76% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[brendancol]

I know f32 is ideal for performance. But this sounds a little dangerous if the user is assuming there data stays as the same type during calculation. Could add a little more detail on why this is important to cast?

[thuydotm]

Almost our tools return data of dtype f32. Although the calculations on f32 can cause some minor numerical error if the original data is of f64, we save the result in f32 I think those errors are acceptable. In case of issue #726, converting from uint16 to floating point dtype (f32 specifically) is necessary to make sure result of a calculation does not exceed the bounds of unsigned int data type.

[brendancol]

@thuydotm Can you follow up with a PR with documentation of how these types are handled?

#766