Charting the Future of Earth Observation 2024

Training data Verification and optimizationAlgorithm training Data source: – Satellite EO data – Historical climate data – In-situ measurementsData preprocessing: – Data cleaning – Data labelling – Data normalization – Data integrationML model selectionDecision supportAI Capability of machines to perform tasks that typically require human intelligence ML Focuses on enabling systems to learn from data and improve over time DL Made up of neural networks to study intricate data patterns Charting the Future of Earth Observation: Technology Innovation for Climate Intelligence 8Muon Space, a start-up developing small satellites to monitor Earth’s climate, announced a partnership with the non-profit Earth-Fire Alliance to build a constellation of 50 satellites focused on wildfire prevention and monitoring. The first phase of the three satellites will be launched in 2026 and is equipped with six-band multispectral infrared instruments that provide high-fidelity data to detect fires faster than on-the-ground observations. The multispectral infrared instruments will differentiate genuine wildfire events from false positives and enhance wildfire detection and assessment of fire intensity. FireSat will operate in low Earth orbit with an observation swath of 1,500 kilometres and an average ground sample size of 80 metres. This will allow it to detect fire ignition sites as small as 25 metres squared (m²) with a revisit time of 20 minutes. This can significantly improve initial response and monitoring, especially in remote areas. For comparison, NASA’s Fire Information for Resource Management System (FIRMS) uses satellite observations from the Moderate Resolution Imaging Spectroradiometer (MODIS) and the Visible Infrared Imaging Radiometer Suite (VIIRS) instruments to detect active fires and thermal anomalies. MODIS typically detects both flaming and smouldering fires of 1,000m² in size. In very good conditions, high-quality observations can detect fires one-tenth of this size (100m²), and under optimal and extremely rare conditions, even smaller fires of 50m² can be detected.8 FireSat provides a 100% improvement over MODIS under optimal conditions and up to 300% in good conditions. With increasing global demand for timely climate intelligence, the time between data acquisition and actionable insights needs to be significantly reduced. Integrating AI and edge computing into EO platforms is revolutionizing data processing and analysis. These advances greatly increase computational efficiency and enable real-time analysis to improve the speed and accuracy of climate-related decision-making. Technology pipeline: AI, machine learning and deep learning models The increased availability and complexity of high-resolution EO data has called for the use of advanced AI, ML and DL. In the context of satellite EO data, AI executes the broad analysis of data, while ML, a subset of AI, focuses on the recognition of patterns and anomalies within large EO datasets. A further subset is DL, which facilitates the extraction of detailed features from high-resolution satellite imagery. These algorithms can process large amounts of complex EO data in almost real time, resulting in less time between data collection and the generation of insights. By improving computational efficiency, ML technologies overcome several limitations of traditional data processing methods.1.2 Reduced EO data processing time to enable near-real-time climate response Data processing framework – using AI to transform EO data into actionable climate insights FIGURE 2 Integrating AI and edge computing into EO platforms is revolutionizing data processing and analysis. AI = artificial intelligence, ML = machine learning, DL = deep learning