Decide how to handle multiple types of spectral data

[tnigon]

When multiple types of spectral data are available (e.g, cropscan and sentinel), we will always have to make rather subjective decisions as to how these data sources are combined for training/testing/predicting, if at all. For example, is it appropriate to merge the datasets based on the closes center wavelength? I suspect this will depend on the problem/response variable trying to be predicted.

How should this be handled and who should be in control of this?

To do:

• 1. Create a function/utility that will merge multiple spectral datasets together given appropriate "instructions" (e.g., a config_spectral file?). Things to consider (list is not necessarily an inclusive list): wavelength threshold to consider bands from two sensors "similar", FWHM/bandwidth threshold to consider bands form two sensors "similar", spatial scale/pixel size of sensor platforms (should we be tracking this in all spectral datasets?), and post-processing that was done to the data (e.g., smoothing, segmentation, etc.).
• 2. Decide where this function should reside, more or less indicating who should control the "spectral data merging". Giving the ability to customer/partner might be preferred, but with the tradeoff to misuse/abuse the tool. Keeping it "in house" gives more quality control oversight, but seems to be more work/effort on our part.