NT4OP | Deep Learning in Earth and Climate Science from Satellite Imagery

Episode 4: Deep Learning in Earth and Climate Science from Satellite Imagery

Expert: Joachim Moortgat

Recent advances in satellite observational capabilities and computational power are greatly expanding our ability to monitor the Earth's changing surface. Machine Learning algorithms,  first developed for medical imaging, can be trained to automatically detect any feature of interest at the pixel-level, down to sub-meter resolutions and repeated over time. Obvious features of interest are roads, human infrastructure, rivers and other water bodies; but applications abound across disciplines: agricultural land use, energy resources, forestry/deforestation/wildfires, rock outcrops, glaciers and landslides, faulting, contaminant spills, tracking large (herds of) animals, etc. Professor Joachim Moortgat from the School of Earth Sciences develops and applies new deep learning algorithms that are optimized for vast collections of satellite imagery to tackle such problems in collaboration with a broad range of domain experts. He provides an overview of Earth and Climate science applications and capabilities, but also discuss the significant challenges associated with supervised deep learning.