I want to create a land cover using the model in the link (https://www.arcgis.com/home/item.html?id=afd124844ba84da69c2c533d4af10a58). For this you need to mosaic Sentinel-2 images using the tool in the link (https://www.arcgis.com/home/item.html?id=e6e1f20cb0374d28a6eed24f5c2ff51b). When I run the land cover model using the mosaic, almost all land is classified as inland waters. I get more accurate results when I do the images without mosaic, just using XML file in Sentinel-2 image folder. However, the land cover model uses 60 m resolution images. The deep learning model is designed to use 10m resolution multispectral images by mosaic.
Has anyone used the model and experienced this problem before?