I am working on a project that requires classifying Landsat imagery using relatively minor variances in pixel values (i.e. more subtle than simply identifying "forest"/"non-forest"). A seam between Landsat WRS paths runs right up the middle of my ROI. Compositing multiple years of observations for each path still results in significant differences in reflectance values on either side of the seam, which is proving fatal to my classifier. My training data is skewed toward a single WRS path, so training and running a classifier on each path individually would be a problem.
Currently, the best solution I've been able to reach is to create mosaics for each Landsat WRS path in my ROI, export them and download them locally to my computer, mosaic them together using the Orfeo Toolbox's "Mosaic" tool (which adjusts the input images' histograms so that they approximate one another), and then re-upload the merged image to Earth Engine to continue the analysis. While this can get the job done, it's inelegant and makes it harder to transparently document the analytical workflow.
Is there a way I can combine EE's reducers such that these adjacent images' reflectance values can be harmonized?
Something like a "histogram matching followed by mean reducer" tool is what I'm imagining, but I'm not literate enough to know whether the numerous reducer tools described in the API documentation can be combined to that effect.
