Department Seminar: Compton J. Tucker
Title: Semi-arid carbon stocks of 9.9 billion trees from 326,000 commercial satellite images: A hyper-spatial study
Abstract: I will describe the combining of 50 cm spatial resolution commercial satellite data, machine learning, high performance computing, and field allometry that enabled the carbon content of ten billion semi-arid tree crowns to be mapped and converted into carbon at the tree level over 10 million km2 with an uncertainty of ±20%. Our results differ from all previous studies using satellite observations and from those based on numerical simulation models. Our voluminous output data required the development of a viewer for verification of our tree carbon mapping and for others to use our results, from tree(1) to tree(9.9 B). Because of our extensive training data, we found minimal improvement adding additional satellite data. We were also able to map hundreds of million trees under faint cirrus clouds and under moderate aerosol conditions. High-performance computing enabled the processing of our 40 trillion element input data array. Our results appeared in the March 2, 2023 issue of Nature.