I am trying to create a LULC time series from 1990-2023. To generate my time series I picked four dates that all use different satellites (Landsat 5, 7, 8 and Sentinel-1 and -2). I only have training data for 2023. Currently I am simply retraining the model with the same 2023 points used on historic data. I feel like this is leading to some inconsitencies in my classifications. I want to be able to run my trained classifier on all the dates, however there are some severe band inconsitencies. Is there any way to make the model accomodate these changes in bands?

The inconsitencies come from differences in bands between satellites but also in my pre-porccessing steps for different years i.e. pan-sharpening on landsat data using the panchromatic bands, and sentinel-1 as a whole (doesnt collect RBG).

I am working in Google Earth Engine API (JavaScript)

Any help would be greatly apreciated

I tried using the same model (trained on 2023 sentinel data) for both sentinel and landsat8 after removing several bands, however my LULC map for the landsat8 data was completely wrong (to a quite a large degree, based on my visual assessment and ground truth data).