Currently, when passed a multiband raster, predictor_transform() will apply the transformation to each layer individually. This is a sensible default behaviour because many multiband rasters will contain separate variables in different layers.
However, there are occasions when it is desirable to keep the same scaling across layers (e.g., a time series of a single variable). In this case, the transformation should be done using a single set of global statistics (e.g., mean/sd, or min/max combining ALL layers).
Proposed changes:
- Add a flag to the arguments that will control whether transformation is done per layer (the default) or globally.
- Ensure that this maintains compatibility with the
state option (should check that there is one statistic instead of one per layer).
Currently, when passed a multiband raster,
predictor_transform()will apply the transformation to each layer individually. This is a sensible default behaviour because many multiband rasters will contain separate variables in different layers.However, there are occasions when it is desirable to keep the same scaling across layers (e.g., a time series of a single variable). In this case, the transformation should be done using a single set of global statistics (e.g., mean/sd, or min/max combining ALL layers).
Proposed changes:
stateoption (should check that there is one statistic instead of one per layer).