A5071655569- Meteorological Phenomena and Simulations
- Climate variability and models
- Hydrological Forecasting Using AI
- Computational Physics and Python Applications
- Combustion and flame dynamics
- Scientific Computing and Data Management
- Tropical and Extratropical Cyclones Research
- Model Reduction and Neural Networks
- Precipitation Measurement and Analysis
- Advanced Combustion Engine Technologies
- Neural Networks and Applications
- Flood Risk Assessment and Management
- Nonlinear Dynamics and Pattern Formation
- Seismology and Earthquake Studies
- Topological and Geometric Data Analysis
- Image and Signal Denoising Methods
- Cryospheric studies and observations
- Remote Sensing and Land Use
- Advanced Thermodynamics and Statistical Mechanics
- Advanced Image Processing Techniques
- Advanced Vision and Imaging
- Data Visualization and Analytics
- Fluid Dynamics and Turbulent Flows
- Remote Sensing and LiDAR Applications
- Time Series Analysis and Forecasting
Nvidia (United States)
2023-2024
Allen Institute for Artificial Intelligence
2024
Nvidia (United Kingdom)
2024
Seattle University
2023
Lawrence Berkeley National Laboratory
2015-2022
National Energy Research Scientific Computing Center
2018-2021
University of Cambridge
2010-2014
Machine learning (ML) provides novel and powerful ways of accurately efficiently recognizing complex patterns, emulating nonlinear dynamics, predicting the spatio-temporal evolution weather climate processes. Off-the-shelf ML models, however, do not necessarily obey fundamental governing laws physical systems, nor they generalize well to scenarios on which have been trained. We survey systematic approaches incorporating physics domain knowledge into models distill these broad categories....
Abstract. The Atmospheric River Tracking Method Intercomparison Project (ARTMIP) is an international collaborative effort to understand and quantify the uncertainties in atmospheric river (AR) science based on detection algorithm alone. Currently, there are many AR identification tracking algorithms literature with a wide range of techniques conclusions. ARTMIP strives provide community information different methodologies guidance most appropriate for given question or region interest. All...
FourCastNet, short for Fourier Forecasting Neural Network, is a global data-driven weather forecasting model that provides accurate to medium-range predictions at $0.25^{\circ}$ resolution. FourCastNet accurately forecasts high-resolution, fast-timescale variables such as the surface wind speed, precipitation, and atmospheric water vapor. It has important implications planning energy resources, predicting extreme events tropical cyclones, extra-tropical rivers. matches accuracy of ECMWF...
Abstract Atmospheric rivers (ARs) are now widely known for their association with high‐impact weather events and long‐term water supply in many regions. Researchers within the scientific community have developed numerous methods to identify track of ARs—a necessary step analyses on gridded data sets, objective attribution impacts ARs. These different been answer specific research questions hence use criteria (e.g., geometry, threshold values key variables, time dependence). Furthermore,...
While deep learning has shown tremendous success in a wide range of domains, it remains grand challenge to incorporate physical principles systematic manner the design, training, and inference such models. In this paper, we aim predict turbulent flow by its highly nonlinear dynamics from spatiotemporal velocity fields large-scale fluid simulations relevance turbulence modeling climate modeling. We adopt hybrid approach marrying two well-established simulation techniques with learning....
Extreme weather amplified by climate change is causing increasingly devastating impacts across the globe. The current use of physics-based numerical prediction (NWP) limits accuracy and resolution due to high computational cost strict time-to-solution limits.
Numerous facets of scientific research implicitly or explicitly call for the estimation probability densities. Histograms and kernel density estimates (KDEs) are two commonly used techniques estimating such information, with KDE generally providing a higher fidelity representation function (PDF). Both methods require specification either bin width bandwidth. While exist choosing bandwidth optimally objectively, they computationally intensive, since repeated calculation KDE. A solution...
Abstract Thermoacoustic systems can oscillate self-excitedly, and often non-periodically, owing to coupling between unsteady heat release acoustic waves. We study a slot-stabilized two-dimensional premixed flame in duct via numerical simulations of $G$ -equation coupled with acoustics. examine the bifurcations routes chaos for three control parameters: (i) position duct, (ii) length (iii) mean flow velocity. observe period-1, period-2, quasi-periodic chaotic oscillations. For certain...
Tracking and predicting extreme events in large-scale spatio-temporal climate data are long standing challenges science. In this paper, we propose Convolutional LSTM (ConvLSTM)-based models to track predict hurricane trajectories from data; namely, pixel-level history of tropical cyclones. To address the tracking problem, model time-sequential density maps trajectories, enabling capture not only temporal dynamics but also spatial distribution trajectories. Furthermore, introduce a new...
Abstract. Identifying, detecting, and localizing extreme weather events is a crucial first step in understanding how they may vary under different climate change scenarios. Pattern recognition tasks such as classification, object detection, segmentation (i.e., pixel-level classification) have remained challenging problems the sciences. While there exist many empirical heuristics for detecting events, disparities between output of these methods even single event are large often difficult to...
Abstract. To manage Earth in the Anthropocene, new tools, institutions, and forms of international cooperation will be required. Virtualization Engines is proposed as an federation centers excellence to empower all people respond immense urgent challenges posed by climate change.
Abstract Predictions of weather hazard require expensive km-scale simulations driven by coarser global inputs. Here, a cost-effective stochastic downscaling model is trained from high-resolution 2-km over Taiwan conditioned on 25-km ERA5 reanalysis. To address the multi-scale machine learning challenges data, we employ two-step approach Corrector Diffusion ( CorrDiff ), where UNet prediction mean corrected diffusion step. Akin to Reynolds decomposition in fluid dynamics, this isolates...
Data assimilation of observational data into full atmospheric states is essential for weather forecast model initialization. Recently, methods deep generative have been proposed which allow using new input without retraining the model. They could also dramatically accelerate costly process used in operational regional models. Here, a central US testbed, we demonstrate viability score-based context realistically complex km-scale weather. We train an unconditional diffusion to generate...
Abstract Numerous studies have shown that atmospheric models with high horizontal resolution better represent the physics and statistics of precipitation in climate models. While it is abundantly clear from these high‐resolution increases rate extreme precipitation, not whether added events are “realistic”; they occur simulations response to same forcings drive similar reality. In order understand increasing results improved model fidelity, a hindcast‐based, multiresolution experimental...
Synchronization is a universal concept in nonlinear science but has received little attention thermoacoustics. In this numerical study, we take dynamical systems approach to investigating the influence of harmonic acoustic forcing on three different types self-excited thermoacoustic oscillations: periodic, quasi-periodic and chaotic. When periodic system forced, find that: (i) at low amplitudes, it responds both frequency natural (self-excited) frequency, as well their linear combinations,...
We propose MESHFREEFLOWNET, a novel deep learning-based super-resolution framework to generate continuous (grid-free) spatio-temporal solutions from the low-resolution inputs. While being computationally efficient, MESHFREEFLOWNET accurately recovers fine-scale quantities of interest. allows for: (i) output be sampled at all resolutions, (ii) set Partial Differential Equation (PDE) constraints imposed, and (iii) training on fixed-size inputs arbitrarily sized domains owing its fully...
We present an efficient convolution kernel for Convolutional Neural Networks (CNNs) on unstructured grids using parameterized differential operators while focusing spherical signals such as panorama images or planetary signals. To this end, we replace conventional kernels with linear combinations of that are weighted by learnable parameters. Differential can be efficiently estimated one-ring neighbors, and parameters optimized through standard back-propagation. As a result, obtain extremely...
Fourier Neural Operators (FNOs) have proven to be an efficient and effective method for resolution-independent operator learning in a broad variety of application areas across scientific machine learning. A key reason their success is ability accurately model long-range dependencies spatio-temporal data by global convolutions computationally manner. To this end, FNOs rely on the discrete transform (DFT), however, DFTs cause visual spectral artifacts as well pronounced dissipation when...
Abstract. Identifying weather patterns that frequently lead to extreme events is a crucial first step in understanding how they may vary under different climate change scenarios. Here, we propose an automated method for recognizing atmospheric rivers (ARs) data using topological analysis and machine learning. The provides useful information about features (shape characteristics) statistics of ARs. We illustrate this by applying it outputs version 5.1 the Community Atmosphere Model (CAM5.1)...
Abstract. There is growing interest in data-driven weather prediction (DDWP), e.g., using convolutional neural networks such as U-NET that are trained on data from models or reanalysis. Here, we propose three components, inspired by physics, to integrate with commonly used DDWP order improve their forecast accuracy. These components (1) a deep spatial transformer added the latent space of capture rotation and scaling transformation for spatiotemporal data, (2) data-assimilation (DA)...