Drylands cover approximately 40% of the Earth’s land surface and are home to over a quarter of the global population. Despite the deficit of surface water, rare but strong precipitation events are the fundamental driver for geomorphic activity in arid regions. A quantification of the frequency and magnitude of episodic river discharge is essential for a robust characterization of flood hazards and, thus, better understanding of the poorly studied hydromorphodynamics in deserts. However, observation data from gauges are sparsely distributed and, if existent, often do not cover a sufficiently long seamless time series or feature extensive gaps. This applies, for instance, to the remote Northwest Namibia, where more than a dozen ephemeral rivers drain the Kunene Highlands towards the Skeleton Coast, yet daily river flow data for a period of several decades is only available from the Hoanib.

Hence, we propose a workflow based on the Landsat multispectral satellite imagery archive to detect flood events and their spatial impact since 1984 in a high resolution (30 m) for the entire Kunene Region (~144 km²). To cater for the limitations related to a revisit time of 16 days and potential impracticality of scenes due to cloud cover, we calculated spectral indices allowing for the detection of both inundated areas during flooding (e.g., Normalized Difference Water Index) and effects sustained after flood recession (e.g., Tasseled Cap Wetness to detect increased soil moisture). The large remote sensing dataset is processed via cloud computing using the Google Earth Engine. As a novel approach, we try to implement a frequency analysis directly in the Google Earth Engine environment after attributing the spectral imprints of floods to their magnitudes. For this purpose, a statistical relationship is developed between the daily record of the gauging station at the Hoanib and the spatiotemporal multispectral surface characteristics along the river course and floodplains. By transferring this relationship to the other ephemeral streams, spatially highly resolved recurrence intervals for areas affected by floods of different magnitudes can be derived for the entire Kunene Region.