Semantic content-based image retrieval in semantic EO data cubes

Abstract

Earth observation data cubes are widely used for time series analysis of continuous variables such as reflectance or vegetation indices, but many applications require selecting suitable images before and after an event. This contribution presents a semantic content-based image retrieval (SCBIR) approach that operates on semantic EO data cubes, where each observation has at least one categorical interpretation. Users define an area of interest, an event date and search criteria (e.g. cloud cover thresholds), and an automated inference selects suitable pre- and post-event scenes at AOI level. Using a forest fire event on Tenerife in August 2023 with semantically enriched Sentinel-2 images (SIAM-based color33), the workflow automatically identifies cloud-free images and supports the derivation of an impact mask, and can be reused for similar events elsewhere.