Abstract: Realistic sonars radiate spherically spreading waves and have directivity. Therefore, they insonify a target over a finite number of Fresnel zones and span a continuum of oblique incident angles, even when the center of the beam is at normal incidence. These effects strongly influence both the overall scattered pressure levels and resonances. For example, because of the spreading of the beam and associated oblique insonification within the beam, normal modes associated with axially propagating guided waves are excited that would not have otherwise existed for an idealized incident plane wave. This thesis analyzes acoustic scattering by solid elastic cylinders insonified by realistic sonars both theoretically and experimentally. A theoretical model to predict scattering by arbitrary-length cylinders is derived based on the apparent volume flow accounting for the above-mentioned practical sonar properties, namely, spherical spreading and directionality. The formulation is first bench-marked against the formally exact T-matrix solution and tested against previously published laboratory data for finite cylinders. It is found that the formulation outperforms the T-matrix solution in predicting laboratory observations at near-normal incidence. Laboratory experiments are then conducted on arbitrary length smooth cylinders insonified by a directional sonar, with a small number of Fresnel zone excited, to evaluate the theory for monostatic as well as bistatic geometries. The formulation is found to outperform the classical scattering models in predicting the new measurements. For example, resonances associated with axially propagating guided waves excited at broadside incidence observed in the experiments are predicted by the proposed formulation but not by the classical models. The measurements are found to agree well with predictions in terms of overall scattering levels and resonance locations. However, the resonance shapes exhibit inconsistent agreement with data. In addition to testing the predictions, the bistatic laboratory observations presented herein substantiate the significant effects on scattering due to the properties of the incident field from practical sonars. The comparison between theoretical and experimental results is then extended for the more complex case involving statistically rough elastic cylinders with one-dimensional Gaussian roughness. The roughness is found to have a considerable impact on all aspects of scattering—overall levels as well as locations and shapes of resonances. General agreement is found between the theoretically predicted and measured ensemble averaged scattered pressure. Both the theory and data reveal two main observations in the ensemble-averaged scattered field: overall scattered pressure levels are seen to decrease, and resonance effects are diminished compared to the corresponding case of smooth cylinders. Effect of various statistical properties of the rough cylinder, namely, different root mean square (RMS) roughness for fixed correlation length and different correlation lengths for fixed RMS roughness on the scattered field are investigated. Finally, the fluctuations of the scattered field are analyzed using the derived formulation.

Abstract: AI-driven weather forecast models are now more accurate than the best physics-based models. Using similar technology, the open-source Ai2 Climate Emulator (ACE) has been trained to accurately emulate both the daily weather variability (including rainfall extremes) and climate of two leading global atmospheric models, at 100-1000x smaller computational cost. Unlike reanalysis-trained AI weather forecast models, ACE is designed to be trained and deployed across multiple climates; for this purpose, it can be coupled to a simple ‘slab ocean’ model. In the near future, ACE coupled to a companion emulator of a full-physics ocean will affordably emulate hyper-realistic but hyper-expensive ‘digital twin’ models of the atmosphere and ocean with km-scale grids. Digital twin models avoid key uncertainties in present-day coarser-grid climate models and may help us more reliably predict regional trends in precipitation and other climate extremes. The combination of ACE and digital twins promises to provide more reliable local information for climate-sensitive decision making without the complexity of dynamical downscaling.

Biography: Chris Bretherton directs a climate modeling group at the Allen Institute for AI (Ai2) in Seattle which uses machine learning trained on global storm-resolving model output and observational data to improve climate model simulations. He is an Emeritus Professor of the Atmospheric Science and Applied Mathematics Departments at the University of Washington, where for 35 years he studied cloud formation and turbulence and improved their simulation in atmospheric models. He is an American Meteorological Society Charney Award winner, IPCC author, AMS, and AGU Fellow, and a member of the National Academy of Sciences.

Abstract: In recent years, there has been substantial interest in using deep neural networks (NNs) to improve the modeling and prediction of Earth’s climate, a complex, nonlinear, multi-scale dynamical system. For 1 to 10-day weather forecasting, fully data-driven NN-based models have already transformed the state-of-the-art and are on their way to becoming operational. Promising results on building NN-based models for longer time-scale, and even emulators to study climate change are emerging. Significant progress has been also made in developing data-informed subgrid-scale parameterizations (closures) for hybrid weather and climate modeling. However, challenges with understanding the learning process of these models and concerns about their ability to provide early warning or statistics of the rarest, most impactful extreme weather events (the so-called gray swans) slow down progress and widespread adaptation of AI-based models.  I will discuss examples of successes and failures of AI weather and climate models, and present ideas around using tools from math and physics, e.g., conducting Fourier analysis of NNs and leveraging rare-event sampling methods, to address them.

Biography: Prof. Pedram Hassanzadeh leads the University of Chicago’s Climate Extreme Theory and Data Group and is an Associate Professor at the Department of Geophysical Sciences, Committee on Computational and Applied Math, and the Data Science Institute (DSI). He also leads the DSI’s new AI for Climate Initiative (AICE). He received his MA (in applied math) and PhD (working on geophysical turbulence) from UC Berkeley in 2013. He was a Ziff Environmental Fellow at Harvard University before joining Rice University in 2016 and moving to the University of Chicago in 2024. His research is at the intersection of scientific machine learning, computational and applied math, climate change, extreme weather, and geophysical fluid dynamics. He has received an NSF CAREER Award, ONR Young Investigator Award, and Early Career Fellowship from the National Academies Gulf Research Program.

Abstract: Earth’s climate is chaotic and noisy. Finding usable signals amidst all of the noise can be challenging: be it predicting if it will rain, knowing which direction a hurricane will go, understanding the implications of melting Arctic ice, or detecting the impacts of humans on the earth’s surface. Here, I will demonstrate how explainable artificial intelligence (XAI) techniques can sift through vast amounts of climate data and push the bounds of scientific discovery: allowing scientists to ask “why?” but now with the power of machine learning.

Biography: Dr. Elizabeth (Libby) Barnes is a Professor of Atmospheric Science at Colorado State University. Libby’s research has centered on climate variability, predictability, and change and the data analysis tools used to better understand the Earth system (including machine learning and causal discovery). She teaches graduate courses on statistical analysis methods, machine learning for Earth scientists, and data-driven forecasting from days-to-decades. See Barnes Group.