A5090320439- Meteorological Phenomena and Simulations
- Climate variability and models
- Computational Physics and Python Applications
- Hydrological Forecasting Using AI
- Oceanographic and Atmospheric Processes
- Geophysics and Gravity Measurements
- Cryospheric studies and observations
- Tropical and Extratropical Cyclones Research
- Precipitation Measurement and Analysis
- Methane Hydrates and Related Phenomena
- Atmospheric aerosols and clouds
- Neural Networks and Applications
- Reservoir Engineering and Simulation Methods
- Distributed and Parallel Computing Systems
- Energy Load and Power Forecasting
- Scientific Research and Discoveries
- Atmospheric and Environmental Gas Dynamics
- Air Quality Monitoring and Forecasting
- Advanced MRI Techniques and Applications
- Complex Network Analysis Techniques
- Scientific Computing and Data Management
- Functional Brain Connectivity Studies
- Advanced Clustering Algorithms Research
- Manufacturing Process and Optimization
- Experimental Learning in Engineering
Nvidia (United States)
2023-2024
Nvidia (United Kingdom)
2024
Seattle University
2023
Allen Institute for Artificial Intelligence
2021-2023
Vulcan (United States)
2019-2022
Allen Institute
2022
University of Washington
2018-2021
NOAA Geophysical Fluid Dynamics Laboratory
2021
New York University
2015-2017
Courant Institute of Mathematical Sciences
2015-2016
Functional connectivity analysis of resting state blood oxygen level–dependent (BOLD) functional MRI is widely used for noninvasively studying brain networks. Recent findings have indicated, however, that even small (≤1 mm) amounts head movement during scanning can disproportionately bias estimates, despite various preprocessing efforts. Further complications interregional estimation from time domain signals include the unaccounted reduction in BOLD degrees freedom related to sensitivity...
The brain is the body's largest energy consumer, even in absence of demanding tasks. Electrophysiologists report on-going neuronal firing during stimulation or task regions beyond those primary relationship to perturbation. Although biological origin consciousness remains elusive, it argued that emerges from complex, continuous whole-brain collaboration. Despite converging evidence suggesting whole continuously working and adapting anticipate actuate response environment, over last 20 y,...
Abstract Weather and climate models approximate diabatic sub‐grid‐scale processes in terms of grid‐scale variables using parameterizations. Current parameterizations are designed by humans based on physical understanding, observations, process modeling. As a result, they numerically efficient interpretable, but potentially oversimplified. However, the advent global high‐resolution simulations observations enables more robust approach machine learning. In this letter, neural network‐based...
Abstract General circulation models (GCMs) typically have a grid size of 25–200 km. Parametrizations are used to represent diabatic processes such as radiative transfer and cloud microphysics account for subgrid‐scale motions variability. Unlike traditional approaches, neural networks (NNs) can readily exploit recent observational data sets global cloud‐system resolving model (CRM) simulations learn subgrid This article describes an NN parametrization trained by coarse‐graining near‐global...
Abstract Neural networks are a promising technique for parameterizing subgrid-scale physics (e.g., moist atmospheric convection) in coarse-resolution climate models, but their lack of interpretability and reliability prevents widespread adoption. For instance, it is not fully understood why neural network parameterizations often cause dramatic instability when coupled to fluid dynamics. This paper introduces tools interpreting behavior that customized the parameterization task. First, we...
Abstract Due to limited resolution and inaccurate physical parameterizations, weather climate models consistently develop biases compared the observed atmosphere. Using FV3GFS model at coarse resolution, we propose a method of machine learning corrective tendencies from hindcast simulation nudged toward observational analysis. We show that random forest can predict nudging this with moderate skill using only state as input. This is then coupled FV3GFS, adding temperature, specific humidity...
Abstract Global atmospheric “ storm‐resolving ” models with horizontal grid spacing of less than 5 km resolve deep cumulus convection and flow in complex terrain. They promise to be reference that could used improve computationally affordable coarse‐grid global climate across a range climates, reducing uncertainties regional precipitation temperature trends. Here, machine learning nudging tendencies as functions column state is correct the physical parameterization temperature, humidity,...
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...
Weather and climate models approximate diabatic sub-grid-scale processes in terms of grid-scale variables using parameterizations. Current parameterizations are de- signed by humans based on physical understanding, observations process modeling. As a result, they numerically efficient interpretable, but potentially over-simplified. However, the advent global high-resolution simulations enables more robust approach machine learning. In this letter, neural network (NN) parameterization is...
Abstract Parameterization of subgrid‐scale processes is a major source uncertainty in global atmospheric model simulations. Global storm‐resolving simulations use finer grid (less than 5 km) to reduce this by explicitly resolving deep convection and details orography. This study uses machine learning replace the physical parameterizations heating moistening rates, but not wind tendencies, coarse‐grid (200 atmosphere model, using training data obtained spatially coarse‐graining 40‐day...
Earth and Space Science Open Archive This preprint has been submitted to is under consideration at Geophysical Research Letters. ESSOAr a venue for early communication or feedback before peer review. Data may be preliminary.Learn more about preprints preprintOpen AccessYou are viewing the latest version by default [v1]Correcting weather climate models machine learning nudged historical simulationsAuthorsOliverWatt-MeyeriDNoah DominoBrenowitzSpencer KonciusClarkBrianHenniDAnnaKwaJeremy...
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 Since the weather is chaotic, it necessary to forecast an ensemble of future states. Recently, multiple AI models have emerged claiming breakthroughs in deterministic skill. Unfortunately, hard fairly compare ensembles forecasts because variations ensembling methodology become confounding and baseline data volume immense. We address this by scoring lagged initial condition ensembles—whereby can be constructed from a library hindcasts. This allows first parameter‐free intercomparison...
Since the weather is chaotic, forecasts aim to predict distribution of future states rather than make a single prediction. Recently, multiple data driven models have emerged claiming breakthroughs in skill. However, these mostly been benchmarked using deterministic skill scores, and little known about their probabilistic Unfortunately, it hard fairly compare AI sense, since variations choice ensemble initialization, definition state, noise injection methodology become confounding. Moreover,...
Climate models are complicated software systems that approximate atmospheric and oceanic fluid mechanics at a coarse spatial resolution. Typical climate forecasts only explicitly resolve processes larger than 100 km any process occurring below this scale (e.g. thunderstorms) using so-called parametrizations. Machine learning could improve upon the accuracy of some traditional physical parametrizations by from global cloud-resolving models. We compare performance two machine models, random...
Abstract Bretherton et al. (2022, https://doi.org/10.1029/2021MS002794 ) demonstrated a successful approach for using machine learning (ML) to help coarse‐resolution global atmosphere model with real geography (a ∼200 km version of NOAA's FV3GFS) evolve more like fine‐resolution model, at the scales resolved by both. This study extends that work application in multiple climates and multi‐year ML‐corrected simulations. Here four (∼25 km) 2 year reference simulations are run FV3GFS...
Abstract One approach to improving the accuracy of a coarse‐grid global climate model is add machine‐learned (ML) state‐dependent corrections prognosed tendencies, such that evolves more like reference fine‐grid storm‐resolving (GSRM). Our past work demonstrating this was trained with short (40‐day) simulations GFDL's X‐SHiELD GSRM 3 km horizontal grid spacing. Here, we extend span full annual cycle by training and testing our ML using new year‐long simulation. corrective models are learning...
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 (\textit{CorrDiff}), where UNet prediction mean corrected diffusion step. Akin to Reynolds decomposition in fluid dynamics, this isolates...
Existing ML-based atmospheric models are not suitable for climate prediction, which requires long-term stability and physical consistency. We present ACE (AI2 Climate Emulator), a 200M-parameter, autoregressive machine learning emulator of an existing comprehensive 100-km resolution global model. The formulation allows evaluation laws such as the conservation mass moisture. is stable 100 years, nearly conserves column moisture without explicit constraints faithfully reproduces reference...
Abstract We present a parsimonious deep learning weather prediction model to forecast seven atmospheric variables with 3‐hr time resolution for up 1‐year lead times on 110‐km global mesh using the Hierarchical Equal Area isoLatitude Pixelization (HEALPix). In comparison state‐of‐the‐art (SOTA) machine (ML) models, such as Pangu‐Weather and GraphCast, our DLWP‐HPX uses coarser far fewer prognostic variables. Yet, at 1‐week times, its skill is only about 1 day behind both SOTA ML models...
Modern climate projections lack adequate spatial and temporal resolution due to computational constraints. A consequence is inaccurate imprecise predictions of critical processes such as storms. Hybrid methods that combine physics with machine learning (ML) have introduced a new generation higher fidelity simulators can sidestep Moore's Law by outsourcing compute-hungry, short, high-resolution simulations ML emulators. However, this hybrid ML-physics simulation approach requires...
Abstract We present a machine learning based emulator of microphysics scheme for condensation and precipitation processes (Zhao‐Carr) used operationally in global atmospheric forecast model (FV3GFS). Our tailored architecture achieves high skill (≥94%) predicting condensate amounts maintains low global‐average bias (≤4%) 1 year continuous simulation when replacing the Fortran scheme. The stability success this stems from key design decisions. By separating emulation processes, we can better...
Abstract. Simulation software in geophysics is traditionally written Fortran or C++ due to the stringent performance requirements these codes have satisfy. As a result, researchers who use high-productivity languages for exploratory work often find hard understand, modify, and integrate with their analysis tools. fv3gfs-wrapper an open-source Python-wrapped version of NOAA (National Oceanic Atmospheric Administration) FV3GFS (Finite-Volume Cubed-Sphere Global Forecast System) global...
Cloud microphysical parameterizations in atmospheric models describe the formation and evolution of clouds precipitation, a central weather climate process. Cloud-associated latent heating is primary driver large small-scale circulations throughout global atmosphere, have important interactions with radiation. Clouds are ubiquitous, diverse, can change rapidly. In this work, we build first emulator an entire cloud parameterization, including fast phase changes. The performs well offline...