Bulletin of the American Physical Society
APS March Meeting 2023
Volume 68, Number 3
Las Vegas, Nevada (March 5-10)
Virtual (March 20-22); Time Zone: Pacific Time
Session W53: Data Science for ClimateFocus
|
Hide Abstracts |
Sponsoring Units: GDS GPC DFD Chair: William Ratcliff, National Institute of Standards and Technology Room: Room 307 |
Thursday, March 9, 2023 3:00PM - 3:36PM Author not Attending |
W53.00001: Leveraging Interpretable Machine Learning for Climate Physics Invited Speaker: Laure Zanna In this presentation, I will describe the complex and multiscale nature of the climate system and how machine learning can be leveraged to deepen our understanding of key physical climate processes. I will focus on advances in interpretable and physics-aware machine learning methods that have the potential to accelerate scientific discovery in climate physics and modeling. In particular, I will discuss examples of interpretable and generalizable machine learning models that capture ocean turbulence processes (horizontal scale of 10 km-100 km) and how these turbulent features can impact large-scale ocean currents (1000’s of kms). The machine-learned models of turbulent processes are shown to improve coarse-resolution climate simulations by faithfully capturing the complex multiscale dynamical properties in the climate system. |
Thursday, March 9, 2023 3:36PM - 3:48PM |
W53.00002: Probabilistic learning for predictive modeling of climate variability Balu Nadiga While comprehensive climate models are skilful at predicting the response of the climate system to external forcing, they are less skilful when it comes to predicting the natural variability of climate. A variety of probabilistic machine learning techniques ranging from Reservoir Computing to Generative Adversarial Networks to Bayesian Neural Networks are considered in the latter context of predicting natural variability of climate. These models are seen to improve upon the Linear Inverse Modeling (LIM) approach which has itself been sometimes thought of as capturing the bulk of the predictable component of natural variability. |
Thursday, March 9, 2023 3:48PM - 4:00PM |
W53.00003: Learning fire spread dynamics with physics-constrained machine learning Jatan Buch, Aniket Jivani, Xun Huan, A. Park Williams, Pierre Gentine Recent years have seen a dramatic increase in the extent and intensity of area burned by large wildfires in different parts of the world. However, a complete theoretical understanding of the dynamics driving the spread of fires across a landscape is still elusive, leading to significant uncertainty in predicting the spread of active wildfires as well as mitigating their impact. In this talk, I will present preliminary results from a physics-constrained machine learning (ML) model of fire spread dynamics. Our ML model consists of an Ensemble Kalman Filter (EnKF) data assimilation (DA) algorithm applied to the latent space of a conditional Variational Autoencoder (cVAE). The cVAE is trained on simulation data from an Hamilton-Jacobi PDE advected by a fire spread rate parameterized by the Rothermel and Balbi models, whereas for the DA step we incorporate fire geometries observed at half-day time steps by the Visible Infrared Imaging Radiometer Suite (VIIRS) instrument. We show that including the surface fire spread dynamics simulated based on an idealized representation of the interaction between ignition patterns, fuel characteristics, and meteorological conditions dramatically improves the performance of the DA-cVAE model when compared to a purely data driven approach. Using the ML model as a simulation-based inference technique allows us to robustly quantify the uncertainty in the parameters of a fire spread rate model. Altogether, our model provides quick forecasts of wildland fire spread to facilitate risk minimization, while serving as a diagnostic framework for the limitations of the current theoretical paradigm. |
Thursday, March 9, 2023 4:00PM - 4:12PM |
W53.00004: Long-term instability of deep learning-based digital twins of the climate system: Cause and solution Ashesh K Chattopadhyay, Pedram Hassanzadeh As the need for skillful, long lead-time predictions of extreme events increases, deep learning-based digital twins of the Earth system have shown promises to deliver fast, accurate, computationally efficient forecasts. While these digital twins' short-term weather forecasts are increasingly becoming better, even competitive with those of numerical models, they often become unstable/unphysical when integrated for long time scales, e.g., beyond 20 days. Long-term stability of digital twins is a desirable property since, if the simulation has the right mean and variability, it would allow us to generate a large number of ensembles of physically-consistent climate simulations at a fraction of the computational cost of traditional climate models. This would enable us to gather better insights into the physics of extreme events, their causal triggers, and how their distribution changes over time, amongst other things. However, currently, the cause of the instability in deep learning-based digital twins is largely unknown, and hence, most remedies are ad-hoc and often empirical. Using two-layer quasi-geostrophic turbulence and ERA5 data as test cases, for the first time, we reveal a causal mechanism for this instability through the lenses of both deep learning theory and physics. We then provide an architecture-agnostic, physics-inspired solution to stabilize deep learning-based models. We show improvement in short-term forecasts, as well as long-term stable emulations for hundreds of years with accurate mean and variability. |
Thursday, March 9, 2023 4:12PM - 4:24PM |
W53.00005: Global and direct solar irradiance estimation using deep learning and selected spectral satellite images Shanlin Chen To fully exploit the spectral information of modern geostationary satellites, a deep learning framework based on convolutional neural network and attention mechanism is proposed for 5-minute ground-level global horizontal irradiance (GHI) and direct normal irradiance (DNI) estimations. Correlation analysis is performed to select the representative satellite bands, which can improve the modeling efficiency without accuracy loss compared with the usage of all spectral bands. The results show that the proposed model produces GHI estimation with a normalized root mean squared error (nRMSE) of 20.57% and a normalized mean bias error (nMBE) of -2.04%. The DNI estimation has a nRMSE of 23.63% and the nMBE is 0.36%. Compared with the National Solar Radiation Database (NSRDB) based on the physical solar model, the proposed method produces a GHI estimation with the nRMSE reduction of 5.15%. As for DNI estimation, the proposed method shows a nRMSE reduction of 13.77%. Meanwhile, the proposed methods generally yield better GHI and DNI estimations under different intervals of clear-sky index than NSRDB. The combination of deep learning and remote sensing shows potentials in better extracting the cloud information via multispectral satellite images, which can better support solar resourcing and forecasting applications especially under cloudy conditions. |
Thursday, March 9, 2023 4:24PM - 4:36PM Author not Attending |
W53.00006: Integrating the spectral analyses of neural networks and climate physics for stable, explainable, and generalizable models Pedram Hassanzadeh, Yifei Guan, Adam Subel, Ashesh K Chattopadhyay The atmospheric and oceanic turbulent circulations involve a variety of nonlinearly interacting physical processes spanning a broad range of spatial and temporal scales. To make simulations of these turbulent flows computationally tractable, processes with scales smaller than the typical grid size of weather/climate models have to be parameterized. Recently, there has been substantial interest (and progress) in using deep learning techniques to develop data-driven subgrid-scale (SGS) parameterizations for the climate system. Another approach that is rapidly gaining popularity is to learn the entire spatio-temporal variability of the climate system from data, i.e., developing fully data-driven forecast models or emulators. For either of these approaches to be useful and reliable in practice, a number of major challenges have to be addressed. These include 1) instabilities or unphysical drifts, 2) learning in the small-data regime, 3) interpretability, and 4) extrapolation to different parameters. Using several setups of 2D turbulence, two-layer quasi-geostrophic turbulence, Rayleigh-Benard convection, and ERA5 reanalysis, we introduce methods to address (1)-(4). The key aspect of some of these methods is combining the spectral analyses of deep neural networks and turbulence/nonlinear physics, as well as leveraging recent advances in theory and applications of deep learning. We will show how these spectral analyses shed light on the inner workings of the deep neural networks and connect them to the underlying physics, providing a general framework for interpreting and understanding deep neural networks when applied to nonlinear dynamical systems such as the climate system. |
Thursday, March 9, 2023 4:36PM - 4:48PM |
W53.00007: Reduced-order modeling of Arctic Amplification feedbacks Adam Rupe, Craig Bakker, Derek DeSantis, Jian Lu High latitude regions are warming at an accelerated rate compared to other regions of the Earth. This Arctic Amplification (AA) has significant impacts within and outside the arctic. While it is known from observations and Earth system models that multi-component feedbacks contribute to AA, disentangling the full effects of these complex nonlinear interactions is a major challenge. Here we present initial work on data-driven reduced-order models to analyze the key feedback between ice albedo and surface temperature. Our preliminary analysis utilizes best-fit linear models based on the Dynamic Mode Decomposition with control (DMDc). We show that DMDc captures known linear feedbacks between ice albedo and surface temperature, and includes dominant spatial patterns of variability. We also show how to generalize to nonlinear reduced-order models and how reduced-order models may be used to capture causal relations from data. |
Thursday, March 9, 2023 4:48PM - 5:24PM |
W53.00008: Physics-informed and Equality-constrained Artificial Neural Networks with Applications to Partial Differential Equations and Multi-fidelity Data Assimilation Invited Speaker: shamsulhaq basir Understanding the complexity of the physical universe in its details is a challenging endeavor. However, many physical processes integral to engineering applications can be concisely described by a set of unified governing laws. In this study, we investigate the application of artificial neural networks to mining physics. Particularly, we discuss the challenges associated with integrating observational data with known laws of physics. We then present a neural network-based approach that recasts solving a phenomena governed by known physics as a constrained optimization problem. Our approach is noise-aware, physics-informed, equality-constrained and adept at multi-fidelity data fusion. We demonstrate the efficacy and versatility of our approach by applying it to the solution of several challenging problems governed by linear and non-linear partial differential equations. |
Thursday, March 9, 2023 5:24PM - 5:36PM |
W53.00009: Energy harvesting by an intelligent body from turbulence Yagmur Kati, sinan gundogdu, Bruno Andreis, Sabine Klapp Turbulence is a ubiquitous phenomenon that is encountered in nature ranging from astrophysical to biophysical scales. Still, it is widely appreciated as one of the key unsolved problems in modern physics and engineering. Despite nearly all fluid flows being turbulent, the energy in turbulence has not been utilized to its full potential today. Energy harvesting from turbulence has been challenging due to the inability to study the process accurately since turbulence has an unpredictable and disordered spatiotemporal structure. On the other hand, we know that birds frequently use turbulent energy in the atmosphere to subsidize soaring flight. Fish also can take advantage of the flow irregularity to power their rotation or directional propulsion. An interesting question is whether we can exploit this fluid erratic motion by teaching a body how to react in a smart way to generate energy. Here we report numerical and analytical evidence on how the rotational dynamics of a neutrally buoyant body can conspire to allow the harvesting of energy from the turbulent fluid motion efficiently. We suggest addressing this problem with machine learning to find an optimal way of controlling the body so that it can tow itself for long distances and harvest energy. |
Thursday, March 9, 2023 5:36PM - 5:48PM |
W53.00010: Numerical proof of shell model turbulence closure Giulio Ortali, Alessandro Corbetta, Gianluigi Rozza, Federico Toschi We focus on the development of turbulence subgrid closure models that, employed in an LES approach, exhibit intermittent effects and energy cascade dynamics that are statistically indistinguishable from those of the fully resolved turbulent system. Due to the massive amount of data needed to reach converged statistics of high order statistical moments, we consider the setting of Shell Models of Turbulence, reduced dynamical systems of Ordinary Differential Equations that have been shown to rather faithfully mimic the phenomenology of the energy cascade of Homogeneous Isotropic Turbulence in Fourier space |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700