A5013616568- Meteorological Phenomena and Simulations
- Climate variability and models
- Precipitation Measurement and Analysis
- Hydrological Forecasting Using AI
- Plant Water Relations and Carbon Dynamics
- Solar Radiation and Photovoltaics
- Wind and Air Flow Studies
- Oceanographic and Atmospheric Processes
- Flood Risk Assessment and Management
- Computational Physics and Python Applications
- Reservoir Engineering and Simulation Methods
- Explainable Artificial Intelligence (XAI)
- Musicology and Musical Analysis
- Atmospheric aerosols and clouds
- Cryospheric studies and observations
- Energy Load and Power Forecasting
- Neural Networks and Applications
- Tropical and Extratropical Cyclones Research
- Advanced Data Processing Techniques
- Fire effects on ecosystems
- Ethics and Social Impacts of AI
- Landslides and related hazards
- Scientific Computing and Data Management
- Music Technology and Sound Studies
- Atmospheric and Environmental Gas Dynamics
Colorado State University
2024
University of Washington
2021-2024
NSF National Center for Atmospheric Research
2016-2024
Cooperative Institute for Research in Environmental Sciences
2021-2024
University of Oklahoma
2011-2024
Pennsylvania State University
2024
Ames Research Center
2023
Computational & Information Systems Laboratory
2022-2023
NOAA Earth System Research Laboratory
2021
University of Colorado Boulder
2021
Abstract This paper synthesizes multiple methods for machine learning (ML) model interpretation and visualization (MIV) focusing on meteorological applications. ML has recently exploded in popularity many fields, including meteorology. Although been successful meteorology, it not as widely accepted, primarily due to the perception that models are “black boxes,” meaning thought take inputs provide outputs but yield physically interpretable information user. introduces demonstrates MIV...
Abstract High-impact weather events, such as severe thunderstorms, tornadoes, and hurricanes, cause significant disruptions to infrastructure, property loss, even fatalities. events can also positively impact society, the on savings through renewable energy. Prediction of these has improved substantially with greater observational capabilities, increased computing power, better model physics, but there is still room for improvement. Artificial intelligence (AI) data science technologies,...
Abstract Stochastic parameterizations account for uncertainty in the representation of unresolved subgrid processes by sampling from distribution possible forcings. Some existing stochastic utilize data‐driven approaches to characterize uncertainty, but these require significant structural assumptions that can limit their scalability. Machine learning models, including neural networks, are able represent a wide range distributions and build optimized mappings between large number inputs...
Abstract Deep learning models, such as convolutional neural networks, utilize multiple specialized layers to encode spatial patterns at different scales. In this study, deep models are compared with standard machine approaches on the task of predicting probability severe hail based upper-air dynamic and thermodynamic fields from a convection-allowing numerical weather prediction model. The data for study come patches surrounding storms identified in NCAR ensemble runs 3 May June 2016....
Abstract Clouds are critical for weather and climate prediction. The multiple scales of cloud processes make simulation difficult. Often models measurements used to develop empirical relationships large‐scale be computationally efficient. Machine learning provides another potential tool improve our parameterizations clouds. To explore these opportunities, we replace the warm rain formation process in a General Circulation Model (GCM) with detailed treatment from bin microphysical model that...
Abstract Forecasting severe hail accurately requires predicting how well atmospheric conditions support the development of thunderstorms, growth large hail, and minimal loss mass to melting before reaching surface. Existing forecasting techniques incorporate information about these processes from proximity soundings numerical weather prediction models, but they make many simplifying assumptions, are sensitive differences in model configuration, often not calibrated observations. In this...
Abstract Many statistical downscaling methods require observational inputs and expert knowledge thus cannot be generalized well across different regions. Convolutional neural networks (CNNs) are deep-learning models that have generalization abilities for various applications. In this research, we modify UNet, a semantic-segmentation CNN, apply it to the of daily maximum/minimum 2-m temperature (TMAX/TMIN) over western continental United States from 0.25° 4-km grid spacings. We select...
Abstract This paper describes the use of convolutional neural nets (CNN), a type deep learning, to identify fronts in gridded data, followed by novel postprocessing method that converts probability grids objects. Synoptic-scale are often associated with extreme weather midlatitudes. Predictors 1000-mb (1 mb = 1 hPa) wind velocity, temperature, specific humidity, wet-bulb potential and/or geopotential height from North American Regional Reanalysis. Labels human-drawn Weather Prediction Center...
Abstract Statistical downscaling (SD) derives localized information from larger-scale numerical models. Convolutional neural networks (CNNs) have learning and generalization abilities that can enhance the of gridded data (Part I this study experimented with 2-m temperature). In research, we adapt a semantic-segmentation CNN, called UNet, to daily precipitation in western North America, low resolution (LR) 0.25° high (HR) 4-km grid spacings. We select LR precipitation, HR climatology,...
Given the growing use of Artificial Intelligence (AI) and machine learning (ML) methods across all aspects environmental sciences, it is imperative that we initiate a discussion about ethical responsible AI. In fact, much can be learned from other domains where AI was introduced, often with best intentions, yet led to unintended societal consequences, such as hard coding racial bias in criminal justice system or increasing economic inequality through financial system. A common misconception...
Abstract Benchmark datasets and benchmark problems have been a key aspect for the success of modern machine learning applications in many scientific domains. Consequently, an active discussion about benchmarks has also started atmospheric sciences. Such allow comparison tools approaches quantitative way enable separation concerns domain scientists. However, clear definition weather climate is missing with result that scientists are confused. In this paper, we equip sciences recipe how to...
Abstract Probabilistic quantitative precipitation forecasts challenge meteorologists due to the wide variability of amounts over small areas and their dependence on conditions at multiple spatial temporal scales. Ensembles convection-allowing numerical weather prediction models offer a way produce improved estimates forecast uncertainty. These allow for individual convective storms model grid, but they often displace in space, time, intensity, which results added Machine learning methods can...
Abstract This paper describes the development of convolutional neural networks (CNN), a type deep-learning method, to predict next-hour tornado occurrence. Predictors are storm-centered radar image and proximity sounding from Rapid Refresh model. Radar images come Multiyear Reanalysis Remotely Sensed Storms (MYRORSS) Gridded NEXRAD WSR-88D dataset (GridRad), both which multiradar composites. We train separate CNNs on MYRORSS GridRad data, present an experiment optimize CNN settings, evaluate...
Abstract Flows in the atmospheric boundary layer are turbulent, characterized by a large Reynolds number, existence of roughness sublayer and absence well-defined viscous layer. Exchanges with surface therefore dominated turbulent fluxes. In numerical models for flows, fluxes must be specified at surface; however, not known priori parametrized. Atmospheric flow models, including global circulation, limited area large-eddy simulation, employ Monin–Obukhov similarity theory (MOST) to...
Abstract Demands to manage the risks of artificial intelligence (AI) are growing. These demands and government standards arising from them both call for trustworthy AI. In response, we adopt a convergent approach review, evaluate, synthesize research on trust trustworthiness AI in environmental sciences propose agenda. Evidential conceptual histories reveal persisting ambiguities measurement shortcomings related inconsistent attention contextual social dependencies dynamics trust....
Abstract Artificial intelligence (AI) can be used to improve performance across a wide range of Earth system prediction tasks. As with any application AI, it is important for AI developed in an ethical and responsible manner minimize bias other effects. In this work, we extend our previous work demonstrating how go wrong weather climate applications by presenting categorization the sciences. This assist developers identify potential biases that affect their model throughout development life...
Artificial Intelligence (AI) weather prediction (AIWP) models are powerful tools for medium-range forecasts but often lack physical consistency, leading to outputs that violate conservation laws. This study introduces a set of novel physics-based schemes designed enforce the global dry air mass, moisture budget, and total atmospheric energy in AIWP models. The highly modular, allowing seamless integration into wide range AI model architectures. Forecast experiments conducted demonstrate...
Machine learning has shown promise in reducing bias numerical weather model predictions of wind gusts. Yet, they underperform to predict high gusts even with additional observations due the right-skewed distribution Uncertainty quantification (UQ) addresses this by identifying when are reliable or needs cautious interpretation. Using data from 61 extratropical storms Northeastern USA, we introduce evidential neural network (ENN) as a novel approach for UQ gust predictions, leveraging...
Severe weather, including tornadoes, thunderstorms, wind, and hail annually cause significant loss of life property. We are developing spatiotemporal machine learning techniques that will enable meteorologists to improve the prediction these events by improving their understanding fundamental causes phenomena building skillful empirical predictive models. In this paper, we present enhancements our Spatiotemporal Relational Probability Trees autonomous discovery relationships as well with...
Abstract Three diagnostic fields were examined to assess their ability act as surrogates for tornadoes in a convection-allowing ensemble system run during the spring of 2015. The diagnostics included midlevel (2–5 km AGL) updraft helicity (UH25), low-level (0–3 (UH03), and (1 vertical relative vorticity (RVORT1). RVORT1 was used direct measure rotation strength. Each storm’s magnitude near-storm environment properties extracted from each hour’s forecasts using an object-based approach....
Abstract This is a test case study assessing the ability of deep learning methods to generalize future climate (end 21st century) when trained classify thunderstorms in model output representative present‐day climate. A convolutional neural network (CNN) was strongly rotating from current created using Weather Research and Forecasting at high‐resolution, then evaluated against found perform with skill comparatively both climates. Despite training labels derived threshold value severe...
Abstract Secondary organic aerosols (SOA) are formed from oxidation of hundreds volatile compounds (VOCs) emitted anthropogenic and natural sources. Accurate predictions this chemistry key for air quality climate studies due to the large contribution submicron aerosol mass. Currently, only explicit models, such as Generator Explicit Chemistry Kinetics Organics in Atmosphere (GECKO‐A), can fully represent chemical processing thousands species. However, their extreme computational cost...