By Ioannis Papoutsis, DeepCube Coordinator – Associate Researcher, Institute for Astronomy Astrophysics Space Applications & Remote Sensing, National Observatory of Athens

One thing is certain, DeepCube does not emerge in a vacuum. The big picture is the European Commission’s Destination Earth. This initiative has a 10-year horizon and aspires to unlock the potential of digital modelling of the Earth’s physical resources and related environmental phenomena, providing us with knowledge about the effects of climate change, the state of the oceans, the cryosphere, biodiversity, land use, and natural resources. This knowledge, coupled with socio-economic information, can help us plan and prepare better for major environmental degradation and climate-related disasters.

To support this ambitious initiative, the European Space Agency has put forward the concept of Digital Twin Earth as a key ingredient of Destination Earth. The concept of Digital Twin Earth is to develop a dynamic, digital replica of our planet. In other words, a model that is powered by observations and provides an accurate representation of the past, present and future changes on our planet. Digital Twin Earth will help visualize, monitor and forecast natural and human activity. Ultimately, we will be able to monitor the health of the planet and perform simulations of Earth’s systems and their interconnections with human behavior. This way, we will reinforce Europe’s efforts for sustainable development and a better environment, in order to respond to the urgent challenges addressed by European environmental policies and the Green Deal. [Source: European Space Agency]

Image source: European Centre for Medium-Range Weather Forecasts (ECMWF)

We see DeepCube as a showcase of the Digital Twin Earth potential, addressing all of its ingredients, from data to infrastructure, to technology, to R&D, to new business models, with Artificial Intelligence (AI) and big Copernicus data at its core. DeepCube’s contribution to the community is twofold. Firstly, we will develop an innovative, open and interoperable platform that will enable the deployment of end-to-end Deep Learning pipelines on big Earth Observation data, regardless of the underlying cloud infrastructure. Secondly, we will design and develop new AI architectures, to address emerging Earth Observation problems that imply high environmental and societal impact. We aspire to enhance our understanding of Earth’s processes that are correlated with the current and future Climate emergency and feasibly generate high business value.

The DeepCube platform consists of technologically mature components and constitutes a unique project legacy. It will be developed as an open-source project that can be deployed in different cloud environments. The platform will provide novel solutions for all phases on an Earth Observation based AI pipeline, from data ingestion, to data organization in data cubes, feature engineering, semantic annotation, distributed Deep Learning, semantic reasoning and visualization.

DeepCube is driven by the scientific and business questions behind its Use Cases. What makes the DeepCube applications different is that they serve non-traditional use cases and penetrate untapped markets, exploiting unique datasets and employing new AI architectures. Specifically, DeepCube will address problems that require quantitative estimation of geophysical variables. Such problems cannot provide models defined exclusively by data and observations, but we have to somehow re-implant physics to the modeling, and also be able to explain what our model has learnt. To this end, we test a hybrid modeling approach for Earth System Science, where we combine data-driven modeling bound by physical parameters, further enhanced through eXplainable AI for “physics-aware” AI applications and Causality. In addition, we address problems that exploit non-space data, linking other information sources, such as social media, industrial and socio-economic data, to create new value chains. Lastly, we address AI problems that make use of interferometric Synthetic Aperture Radar (SAR) data. The Copernicus Sentinel-1 archive of SAR data, which amounts to petabytes of information, is the richest asset that remains hugely under-exploited by the scientific community when it comes to AI-based applications.

So, why DeepCube? Now more than ever, we need to be prepared to respond to the biggest societal and environmental challenges our planet is facing, employing every scientific solution possible. The combination of Artificial Intelligence and big Earth Observation data can be a game changer in the fight against the rapid Climate Change and its devastating consequences.