This dataset contains landform classification maps for the Bislak (16/01/2017), Laoag (10/07/2018) and Abra (21/11/2019) Rivers in northwest Luzon, the Philippines. River landforms (water, vegetated bars, unvegetated bars) were classified using the SVM machine learning model established on Sentinel-2 Level 2A data. The classification training/testing dataset have been enveloped in LibSVM format (.txt) for reading and processing in Python 3.7 using Scikit-Learn Tools. In addition to the landform classification maps, the dataset includes rasters (.tif) from three downscaling approaches (area-to-point regression kriging, nearest neighbour resampling and super-resolution method) and their associated image segmentation shapefiles (.shp) for the Bislak River on 1 January 2018. The area-to-point regression kriging (ATPRK) method was processed using codes developed by Wang et al (2016). The nearest neighbour resampling downscaling images were generated in SNAP software. The super-resolution downscaling images were obtained using the Sen2Res plugin in SNAP software.