Multi-satellite Burned Area Change Detection Dataset (Pre-fire Sentinel-2 and Landsat-8; Post-fire PlanetScope)

This dataset is designed for burned area change detection using multi-satellite imagery, including PlanetScope (PS), Landsat-8 (L8), and Sentinel-2 (S2). The dataset is organized into two main subsets: a patch-wise dataset and a region-wise dataset. 1) Patch-wise dataset The patch-wise dataset consists of multiple configurations:- PS_RGB_L8_RGB (Pre-fire Landsat-8, Post-fire PlanetScope)- PS_RGB_S2_RGB (Pre-fire Sentinel-2, Post-fire PlanetScope)- PS_RNG_L8_RNG (Pre-fire Landsat-8, Post-fire PlanetScope)- PS_RNG_S2_8_RNG (Pre-fire Sentinel-2, Post-fire PlanetScope)- PS_RNG_S2_8a_RREG (Pre-fire Sentinel-2, Post-fire PlanetScope) Each configuration contains three splits: train, val, and test. 2) Region-wise dataset The region-wise dataset also consists of multiple configurations:- PS_RNG_L8_RNG (Pre-fire Landsat-8, Post-fire PlanetScope)- PS_RNG_S2_8_RNG (Pre-fire Sentinel-2, Post-fire PlanetScope)- PS_RNG_S2_8a_RREG (Pre-fire Sentinel-2, Post-fire PlanetScope) Each configuration contains five regions in Korea and is divided into train, val, and test splits. Across all subsets, the data are structured as paired pre- and post-fire images with corresponding ground truth masks:- A: pre-fire images - B: post-fire images - label: corresponding ground truth masks for burned areas All images are provided as cropped samples for model training and evaluation. The dataset supports both change detection and semantic segmentation tasks. For change detection, models can utilize paired inputs (A and B) with the corresponding labels. For semantic segmentation, models can be trained using post-fire images (B) and the corresponding ground truth labels.