Charting the Future of Earth Observation 2024

Charting the Future of Earth Observation: Technology Innovation for Climate Intelligence14Enhancing data visualization and decision-making – ML-based vs geospatial AI models TABLE 1ML-based models GeoAI foundational models What are the primary applications and what type of outputs do they generate?Uses classification, regression and clustering techniques to make localized predictions, scenario analysis and risk assessmentsProduces global predictions, multiscale models and general- purpose representations How do they enhance data visualization and decision-making?Heatmaps, 3D models and dashboardsInteractive maps, simulation tools and AR/VR experiencesML-based models Geospatial AI foundational models What are the primary applications and what type of outputs do they generate?Uses classification, regression and clustering techniques to make localized predictions, scenario analysis and risk assessmentsProduces global predictions, multiscale models and general-purpose representations How do they enhance data visualization and decision-making?Heatmaps, 3D models and dashboards Interactive maps, simulation tools and AR/VR experiences The use of EO data in ML-based and foundational models can decrease the build time of a flood map by as much as 80%,18 helping to convert data into more precise predictions of floods. EO data was proven to increase the accuracy of predicting flood susceptibility by up to 20%,19 and using EO data with ML algorithms has led to a near-100 times larger dataset when modelling storms and hurricanes compared to traditional methods.20 In flood forecasting, these models could help city planners recognize flood-prone areas and prioritize the deployment of temporary designs like sandbags and other immediate flood defences. Longer-term measures, such as improved evacuation strategies, may simultaneously be implemented. In the case of flash flooding, a 12-hour lead time can potentially reduce damage by up to 60%. In comparison, just a one-hour advance lead time can reduce damage by 20%.21,22 Better flood vulnerability maps facilitate preventative action that can reduce the cost of a major flood. Annual investments in flood defences of $50 billion for coastal cities could cut expected losses in 2050 from up to $1 trillion a year to $60-63 billion,23 with a further $12-71 billion required by 2100 to address sea-level rise.24 AI-driven, detailed flood mapping and accurate forecasting can help direct resources to the most vulnerable regions. While ML has significantly improved weather forecasting, a challenge remains in applying the same techniques to long-term climate predictions of more than 10 years. Climate systems are more complex with longer timescales, requiring more advanced models that can capture the intricate feedback mechanisms and long-term trends. Additionally, the lack of transparency in foundation models remains an issue compared to traditional weather and climate models. Combining these models through physics- informed neural networks (PINNs) might allow users to achieve the best of both worlds, integrating the advanced pattern recognition of foundation models with the interpretability and physical consistency of traditional models.Technology pipeline: Digital twins Digital twins represent an advanced simulation model that integrates EO data to simulate various climate scenarios. These models can be used by city planners and policy-makers for simulations of possible strategies and their impacts on the virtual outcome. Digital twins use advanced analytics, ML and AI to analyse data from multiple sources, predict environmental changes and simulate complex “what if” scenarios. They achieve this by creating a dynamic, digital replica of the Earth’s system. This novel approach enables high-fidelity climate simulations, providing a crucial sandbox for testing and refining climate-related strategies. The Destination Earth (DestinE) initiative from the European Commission is an example of the pioneering efforts in developing a detailed digital twin of the Earth. Using advances in high-performance computing, EO data and ML techniques, DestinE will provide predictions of global environmental change scenarios and their impacts, drawing from capabilities of the European Centre for Medium-Range Weather Forecasts (ECMWF), the European Space Agency (ESA) and the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT). The first prototype has been delivered at EuroHPC LUMI Supercomputer Centre in Kajaani, Finland.25 Currently in its second phase, DestinE’s Open Core Service Platform offers a user-friendly, secure cloud- based digital modelling and open simulation platform to complete a full digital replica of Earth by 2030. In another demonstration of PPPs, Lockheed Martin and NVIDIA are advancing AI-driven EO digital twins for the National Oceanic and Atmospheric Administration (NOAA). Through Lockheed Martin’s open Rosetta 3D for AI-driven data fusion and NVIDIA’s Omniverse Nucleus for seamless data sharing, these digital twins are increasing the precision of environmental monitoring by integrating diverse data from NOAA’s extensive collections on the cryosphere, land, atmosphere, space weather and ocean domains. Both satellite EO and ground- based observations are used to create an exact representation of the Earth at high resolution. Digital twins use advanced analytics, ML and AI to analyse data from multiple sources, predict environmental changes and simulate complex “what if” scenarios.