The raw images for the paper "Estimation of deracinated trees area in temperate forest with satellite images employing machine learning methods". The images for the learning algorithms were collected from satellite images using Landsat 8 images from the U.S. Geological Survey (USGS) Landsat Look Viewer (https://landsatlook.usgs.gov). Landsat 8 (formally, the Landsat Data Continuity Mission) was launched on February 11, 2013. The spatial resolution of Landsat 8 is 30 m per pixel. We set the following four training data categories: deracinated tree area, forest, soil (such as agricultural field), and others (e.g., waterbodies and cities). We physically searched the Landsat images for Japan during 2018, and collected 50 images (300 × 300 pixels) for each category.