WASCAT: An all-sky cloud segmentation dataset to support operational meteorology in West Africa

WASCAT (West African All-Sky Cloud Segmentation Dataset) is a rigorously validated, ground-based cloud imaging dataset captured in Kumasi, Ghana. It addresses the critical lack of publicly available, locally representative tropical sky datasets for Sub-Saharan Africa, enabling robust computer vision, semantic segmentation, and meteorological research in data-scarce regions. Release Scope (v1.0): 845 expert-validated image-mask pairs, shipped as RAW ORIGINALS at native capture geometry (predominantly 1920×1280). Zero processing applied: No glare-inpainting, disk-clipping, or resizing has been applied to the released files. Complete Benchmark Package: Includes a 25-column metadata ledger, session-disjoint splits, a certified Random Forest draft generator, and multi-architecture benchmark results. Reproducibility: MIT-licensed Python code is included to reproduce the harmonisation pipeline used to train the benchmark models.