Mapping Urban Green Space Dynamics: a Semantic Earth Observation Data Cube Approach

Abstract

Urban green space mapping based on satellite imagery is now possible more frequently and over shorter timespans thanks to dense time-series of open and free Earth observation (EO) images (e.g. the Copernicus Sentinel-2 mission). Despite this data availability, many approaches still focus on identifying the annual maximum extent of urban green spaces instead of utilising the entire dense image stack to characterise seasonal dynamics. We aim to temporally inform urban green space delineations, which could be relevant for applications like urban heat mitigation or citizens’ urban green perception. We present a semantic EO data cube approach that allows ad-hoc, browser-based vegetation mapping for custom areas and timespans using transferable semantic models. We demonstrate the approach using a Sentinel-2 semantic EO data cube covering Austria, which makes use of every available Sentinel-2 observation since 2015 and where non-valid observations (e.g. cloud) can be masked out on an individual pixel basis to increase the number of valid observations for shorter timespans rather than relying on image-wide metadata. While we show results for the city of Vienna, the approach is transferrable to anywhere in Austria using the same infrastructure, or any other similar semantic EO data cube worldwide.