The CloudCast Dataset

Dataset Characteristics

• Image Resolution: The raw satellite images have a resolution of 3712 x 3712 pixels, where each pixel represents a 3×3 km area.
• Time Intervals: Images are captured at 15-minute intervals.
• Spatial Resolution: The annotated dataset has a spatial resolution of 928 x 1530 pixels.
• Cloud Types: The dataset includes 10 different cloud types, offering a diverse range of cloud formations for analysis.

Data Processing

• Pre-Processing: EUMETSAT performs initial processing, including airplane removal, before making the images public.
• Annotation: Images are annotated at the pixel level using a segmentation algorithm to accurately label cloud types.

Usage The CloudCast dataset is ideal for training and benchmarking computer vision methods in cloud classification and weather forecasting. It provides a standardized dataset for researchers and developers to develop and test machine learning models for meteorological applications.

Additional Information

• Full-Disk Images: The dataset includes full-disk images of Earth, centered at zero degrees longitude.
• Grayscale Dataset: A subset of the dataset offers full-resolution grayscale images centered and projected over Europe (728×728 pixels).

Applications The dataset can be used in various applications, including:

• Cloud classification
• Weather forecasting
• Climate modeling
• Atmospheric research