Charting the Future of Earth Observation 2024

Charting the Future of Earth Observation: Technology Innovation for Climate Intelligence9 In post-disaster events, where there is an urgent need to accelerate damage assessments, high- resolution EO data and ML models use “semantic segmentation” to categorize each pixel of a satellite image, not unlike object recognition.9 This allows detailed and accurate damage assessments that once took weeks to be completed in hours or minutes. Traditionally, geospatial damage assessments take 30 seconds per structure, which means evaluating 4.1 million structures after a disaster like Hurricane Ian would take one person 16 years. In the case of Hurricane Ian, ML algorithms were used, resulting in rapid evaluation and $80 million in expedited assistance for disaster survivors.10 Recently, for example, Microsoft partnered with Planet to use AI models and satellite EO imagery to assess the extent of damage from Hurricane Beryl in Carriacou, Grenada. Microsoft trained a change detection model that analysed both pre- and post-event EO imagery and intersected this data with building footprint layers from an incomplete OpenStreetMap dataset. While the analysis only covered a portion of the affected area, this approach allowed the model to quickly identify 888 damaged buildings out of 1,937 in the area analysed. Post-disaster building assessment in response to Hurricane Beryl in Carriacou, Grenada FIGURE 3 Note: Pre-disaster satellite EO imagery (top) and damage assessment visualizer showing AI-predicted damage layer in red overlay (bottom). Source: Esri World Imagery. (2024). Microsoft AI for Good - Damage Assessment Visualizer for Hurricane Beryl in Grenada (July, 2024). https://satelliteimagerydemostg.z5.web.core.windows.net/damage-assessment/hurricane_beryl_grenada_07_02_2024.html. 0-20% damage 20-40% damage 40-60% damage 60-80% damage 80-100% damagePre-disaster Post-disaster including damage assessments