In our department we are currently developing approaches to derive vegetation parameters across large areas from mostly optical dataset (Sentinel-2 and Landsat).
However, a commonly faced challenge is to derive some sort of mosaics that allows the application of a regression or classification algorithm across large areas without a substantial part of the investigated areas being affected by artefacts caused for example by clouds, phenological differences, shadows, etc.
I am aware of algorithms like STAR-FM (combining Landsat and MODIS) and some other Landsat-interpolation approaches (e.g., applying the Google Earth Engine to calculate median values for a certain time period etc.).
However, the options for Sentinel-2 are still sparse and many of the existing algorithms seem to be not readily available on open-source platforms (STAR-FM might be an exception).
So my question: What kind of appraoches do you use for large-scale applications of optical satellite data like Landsat and Sentinel? Anyone aware of some interesting new approaches currently being developed?