Agenda

Speakers: Niccolò Maffezzoli, University of Venice and University of California Irvine

Link Zoom: https://unive.zoom.us/j/87081697778?pwd=qmFzTgYoLdBg7zbkvVLQWs1BIpb4Py.1

Abstract:
Accurate knowledge of ice volumes is essential for predicting future sea level rise, managing freshwater resources, and assessing impacts on societies, from regional to global. Efforts to better constrain ice volumes face challenges due to sparse thickness measurements, uncertainties in model input variables, and limitations in ice flow model parameterizations. Glaciers currently account for approximately 20% of global sea level rise. Available modeled estimates of glacier volumes vary widely, particularly in arid regions where billions rely on glacier-fed freshwater. On ice sheets, despite Synthetic Aperture Radar enabling unprecedented mapping of surface ice velocity from space, thickness inversion methods still yield significant errors - especially along coastal regions where complex bathymetry and fjord systems hinder mass conservation approaches, often requiring spatial interpolation techniques. In West Antarctica, ocean-forcing at the grounding lines has triggered the retraction and acceleration of major outlet glaciers, threatening the stability of the whole West Antarctic Ice Sheet. This region alone holds over 3 meters of potential sea level rise and is at risk of collapse even under a +1.5°C warming scenario. Improved ice thickness estimates and bedrock mapping near grounding lines are thus crucial to improve ice flow models and reduce the uncertainties in future sea level rise projections. Millions of sparse thickness measurements have been collected over decades of surveys over iced bodies on Earth. Approximately 4 million measurements exist for glaciers worldwide, 20 million in Greenland and 80 million in Antarctica, largely thanks to NASA’s IceBridge mission. Yet, leveraging this extensive dataset through machine learning has remained largely unexplored to this day.

To close this gap, we have developed a global machine learning framework to model the ice thickness of individual glaciers on Earth. The model combines two gradient-boosted decision tree schemes trained on 39 numerical features. It integrates both traditional physics-based variables, such as ice velocity and mass balance, and geometrical features commonly used in area-volume scaling approaches. We find that the system outperforms existing models, by up to 30-40% at high latitudes where most ice is stored, and generalizes well in the ice sheet peripheries. I will present the rationale, benefits, and limitations of machine learning approaches, along with our strategy towards for a complete machine learning-based map of the Greenland and Antarctic ice sheets.

Bio Sketch:
Niccolò Maffezzoli is a postdoc at University of Venice and University of California Irvine, as well as Associate Member of the National Research Council. He obtained a PhD in geophysics at the University of Copenhagen in 2017, working on ice core science. He then transitioned to machine learning applied to the cryosphere domain, winning two Marie Curie Postdoctoral Fellowships at Ca’ Foscari and a Climate Change AI Investigator Grant. He enjoys developing hybrid physics-machine learning models for pressing climate challenges. He collaborates with researchers at the Niels Bohr Institute, Dartmouth College and NASA’s Jet Propulsion Laboratory.
Website: https://nmaffe.github.io/