I am a veterinary epidemiology researcher interested to benefit livestock in a desert. As water is an important resource for herders therefore I have become interested to map water bodies.I learned GEE and used sentinel 2 data . After pansharpening , I calculated water indices and found ranges of e.g., mndwi for some water bodies to decide a threshold range. To some extent I was successful but I get false positive results toward periphery of desert probably at inference between desert and agriculture or green land is not well remarked (shapefile of desert is a bit rough).
Water bodies (tobas in local language) in desert are are often small about 20 m*20m , rain-fed , muddy to light green in colour , and shallow . Waterbodies are located in compact flat soil areas (dahar in local language )total barren . There is some green vegetation around around water bodies which is source of palatable forage for livestock (like ring of green around water)
What could be possible reasons for false positive results ? Any how to overcome this problem.
I burned out but attempted object based image analysis and even supervised classification too