Lukas Kondmann, Aysim Toker, Marc Rußwurm, Andrés Camero, Devis Peressuti, Grega Milcinski, Pierre-Philippe Mathieu, Nicolas Longepe, Timothy Davis, Giovanni Marchisio, Laura Leal-Taixé, Xiaoxiang Zhu
Recent advances in remote sensing products allow near-real time monitoring of the Earth’s surface. Despite increasing availability of near-daily time-series of satellite imagery, there has been little exploration of deep learning methods to utilize the unprecedented temporal density of observations. This is particularly interesting in crop monitoring where time-series remote sensing data has been used frequently to exploit phenological differences of crops in the growing cycle over time. In this work, we present DENETHOR: The DynamicEarthNET dataset for Harmonized, inter-Operabel, analysis-Ready, daily crop monitoring from space. Our dataset contains daily, analysis-ready Planet Fusion data together with Sentinel-1 radar and Sentinel-2 optical time-series for crop type classification in Northern Germany. Our baseline experiments underline that incorporating the available spatial and temporal information fully may not be straightforward and could require the design of tailored architectures. The dataset presents two main challenges to the community: Exploit the temporal dimension for improved crop classification and ensure that models can handle a domain shift to a different year.