IberCloud-L

IberCloud-L is a dataset of 1,094 Landsat patches annotated for cloud and cloud shadow segmentation in four classes. It covers Landsat 5, 7, 8 and 9 imagery over the Iberian Peninsula, including Cantabrian, Mediterranean and southern coastal regions, an inland region and a mountainous region in the Pyrenees, with contrasting cloud regimes. The images are Landsat Collection 2 top-of-atmosphere (TOA) reflectance downloaded via Google Earth Engine. Annotations were produced in two stages. A model trained on CloudSEN12 generated initial masks, which an expert annotator then reviewed and corrected manually. Landsat 7 scenes affected by the Scan Line Corrector failure (after May 2003) were excluded. This dataset accompanies the article “NIMBUS: A multisensor cloud segmentation model for Sentinel-2 and Landsat coastal monitoring using RGB imagery”. Files train.zip val.zip test.zip Format Bands 1–3: RGB Band 4: GT mask (0 = clear sky, 1 = thick cloud, 2 = thin cloud, 3 = cloud shadow)