About

Steven J. Greybush is a Professor of Meteorology and serves as the Associate Director of the Center for Advanced Data Assimilation and Predictability Techniques at Penn State. He is also a Co-Hire at the Institute for Computational and Data Sciences. His research focuses on atmospheric science, particularly in the areas of data assimilation, predictability, and atmospheric modeling. As a faculty member in the Department of Meteorology and Atmospheric Science, he contributes to advancing understanding and techniques related to atmospheric data analysis and weather prediction.

Research topics

• Geography
• Computer Science
• Meteorology
• Geology
• Environmental science
• Climatology
• Political Science
• Economics
• Development economics
• Medicine
• Remote sensing
• Physics
• Biology
• Economic geography
• Virology
• Regional science

Selected publications

• The Multi-University Consortium for Advanced Data Assimilation Research and Education (CADRE)

  Bulletin of the American Meteorological Society · 2026-04-22

  article

  Abstract The Multi-University Consortium for Advanced Data Assimilation Research and Education (CADRE) is a new initiative recently funded by the National Oceanic and Atmospheric Administration (NOAA) to accelerate data assimilation (DA) research, education, and workforce development. Unlike previous initiatives, CADRE fosters end-to-end, direct, and comprehensive collaboration between university faculty and government agencies. It supports innovative DA research, prepares the next-generation DA workforce, and facilitates the transition of DA research to operational applications. CADRE performs a broad scope of cutting-edge research to address multiscale, nonlinear, and coupled Earth system DA challenges to improve short-range (sub-hourly) to seasonal predictions. It achieves this task through innovative data assimilation algorithm development, novel applications of machine learning (ML) in DA, and optimizing the utilization of existing and new in-situ and remotely sensed observations. Beyond research, CADRE establishes a comprehensive education, workforce development, and community building program, which includes a novel graduate student advising model, new university class curriculum development, public training courses, community scientific workshops, an international exchange program, trans-disciplinary partnerships, an outreach program, and promotion of the open sharing of data, code, and educational materials. Through this unique holistic approach, CADRE is set to strengthen both the intellectual and software infrastructure in the broad community for DA research, increase the number of DA scientists with expanded skill sets, and revolutionize forecasting capabilities.

  PublisherDOI

• Evaluating the Performance of AI-based and Traditional NWP Guidance for High Impact Winter Storms over the U.S. East Coast

  Weather and Forecasting · 2026-04-17

  articleSenior author

  Abstract Over the last several years, AI-based numerical weather prediction (AI-NWP) has emerged to become a promising tool in weather forecasting. Many models, like the ECMWF AIFS or Google DeepMind’s Graphcast, have been shown to outperform traditional NWP for particular metrics and variables. However, questions remain over their reliability in day-to-day use and when applied to extreme weather events. Further analysis is critical to properly compare these systems with traditional methods. This study will analyze the performance of three leading AI-NWP systems for high impact winter weather along the U.S. East Coast with a goal of identifying systematic errors or improvements in predictions. Several key variables, including the quantitative precipitation forecast, are examined from both a deterministic and ensemble-based perspective. Evaluation of a collection of storms against reanalysis and other observations affirms the ability for AI-NWP to outperform traditional methods by up to 20-30% for standard metrics like RMSE over a regional domain. Additionally, AI-NWP is shown to be capable of recognizing key high amplitude winter storm features, like conveyor belts, even at long forecast lead times (7-10 days). Analysis of an ensemble sensitivity test applied to Graphcast track errors displays physically realistic upper-level drivers when compared against a physics-based ensemble. By examining the performance of traditional NWP and AI-NWP at predicting winter storm features as a function of lead time, the practical predictability of snowstorms is re-evaluated. As a result, this work can allow for meteorologists to have higher confidence when utilizing AI-NWP in an operational setting.

  PublisherDOI

• Bayesian Deep Learning for Convective Initiation Nowcasting Uncertainty Estimation

  Artificial Intelligence for the Earth Systems · 2026-03-03

  article

  Abstract This study evaluated the probability and uncertainty forecasts of five recently proposed Bayesian deep learning methods relative to a deterministic residual neural network (ResNet) baseline for 0–1-h convective initiation (CI) nowcasting using Geostationary Operational Environmental Satellite ( GOES-16 ) satellite infrared observations. Uncertainty was assessed by how well probabilistic forecasts were calibrated and how well uncertainty separated forecasts with large and small errors. Three Bayesian deep learning methods produced probabilistic forecasts that were comparable to those of the deterministic ResNet. Among them, the initial-weights ensemble + Monte Carlo (MC) dropout, a collection of deterministic ResNets with different initial weights to start training and dropout activated during inference, produced the most well-calibrated probability forecasts and the most reliable uncertainty estimates. The Bayesian ResNet ensemble performed worse than the deterministic ResNet at certain forecast times, likely due to the challenge of optimizing a larger number of parameters. To address this issue, the Bayesian–Model Priors with Empirical Bayes using Deep neural network (MOPED) ResNet ensemble was adopted, which constrained the hypothesis search near the deterministic ResNet solution and achieved forecast skill comparable to that of the deterministic ResNet. All Bayesian methods demonstrated well-calibrated uncertainty and effectively separated cases with large and small errors. In generalization tests, the initial-weights ensemble + MC dropout demonstrated better forecast skills than the Bayesian-MOPED ensemble and the deterministic ResNet on selected CI events in clear-sky regions but showed weaker generalization over broad clear-sky regions. In anvil cloud regions, all Bayesian methods produced skillful forecasts for the selected CI events but demonstrated poor generalization over the non-CI anvil regions.

  PublisherDOI

• Polarimetric Radar–Based Investigation of Microphysics in Dendritic and Needle Temperature Aggregation Zones during NASA IMPACTS

  Monthly Weather Review · 2025-08-11

  articleSenior author

  Abstract Microphysical processes that determine cloud particle habits, sizes, and concentrations can be inferred by specific polarimetric radar signatures, providing new perspectives to inform NWP models about hydrometeor types and their distributions in winter storms. In particular, areas of enhanced K DP sometimes occur in regions of aggregates within the dendritic growth zone (DGZ) and needle temperature zone (NTZ). Further, copolar correlation coefficient (CC) reductions are sometimes found in the DGZ. These polarimetric signatures are of interest because they suggest the presence of highly nonspherical particles among the otherwise isotropically scattering aggregates. In this study, in situ observations of cloud particle populations collected during the NASA Investigation of Microphysics and Precipitation for Atlantic Coast-Threatening Snowstorms (IMPACTS) field campaign are used to address debates in the previous literature about these signatures’ origins. In the regions of aggregates, cloud particle imager (CPI) cloud particle images show a variety of particle habits that contribute to enhancing K DP or reducing CC, suggesting these polarimetric signatures may result from a combination of different particle types. Particle size, concentration, and aspect ratio data from the two-dimensional stereo (2DS) and high-volume precipitation spectrometer (HVPS) probes determined that the K DP enhancement in the NTZ is produced by ice particles of 0.9–7.0-mm size, along with nonspherical aggregates with a mean aspect ratio < 0.6 at sizes between 4.6 and 20.0 mm. Increased K DP in the DGZ is attributed to ice particles of sizes 0.4–6.0 mm. Additionally, asymmetric particles at sizes 3.0–10.0 mm are evident in the DGZ. The DGZ’s CC reduction is attributed to ice particles with sizes 1.4–5.0 mm, while CPI images show a wide range of particle types including irregularly shaped crystals that would scatter anisotropically. Significance Statement This study determines the diversity of particle sizes, concentrations, and the potential habits that can contribute to observable changes in the polarimetric radar variables in winter storms. Knowledge about these radar signatures may benefit winter weather forecasting through future works on data assimilation.

  PublisherDOI

• Mesoscale and Microphysical Characteristics of Elevated Convection and Banded Precipitation over an Arctic Cold Front: A Case Study from IMPACTS

  Journal of the Atmospheric Sciences · 2025-04-23 · 1 citations

  articleOpen access

  Abstract The mesoscale and microphysical structure of a cloud system associated with an Arctic front is analyzed using data from two research aircraft, two WSR-88D radars, the HYSPLIT model, and initialization fields from the RAP model. The flights, conducted during the NASA Investigation of Microphysics and Precipitation in Atlantic Coast-Threatening Snowstorms (IMPACTS) campaign, collected in situ and remote sensing data as the cloud system moved across Illinois. The system developed within an air mass that, based on back trajectory analysis, originated over the subtropical eastern Pacific before being lifted over the Arctic front. This led to a region of potential instability extending upward over the frontal zone. The ascending flow triggered the release of the instability that manifested as elevated convection in the storm’s southern sector. In the convective region, supercooled water was found in cloud towers, leading to saturated conditions that supported growth of a range of particle habits and growth by riming. Within this region, and in shallower clouds between convective towers, needle particle habits, supercooled water, and high ice particle concentrations implied active secondary ice processes. Two snowbands formed north of the convective region, with radar evidence suggesting that precipitation within these bands originated in cloud towers at altitudes of 4–6 km in a near-neutral to weakly unstable region. Water saturated conditions, evidenced by supercooled water at the sampling level, permitted the growth of a range of particle habits. Despite ice particle concentrations < 15 L −1 within the bands, some aggregated particles exceeding a centimeter in maximum dimension were observed at −5°C, likely contributing to the 21–27 dB Z e reflectivity characteristic of the bands.

  PublisherOA PDFDOI

• Bayesian Deep Learning for Convective Initiation Nowcasting Uncertainty Estimation

  ArXiv.org · 2025-07-22

  preprintOpen access

  This study evaluated the probability and uncertainty forecasts of five recently proposed Bayesian deep learning methods relative to a deterministic residual neural network (ResNet) baseline for 0-1 h convective initiation (CI) nowcasting using GOES-16 satellite infrared observations. Uncertainty was assessed by how well probabilistic forecasts were calibrated and how well uncertainty separated forecasts with large and small errors. Most of the Bayesian deep learning methods produced probabilistic forecasts that outperformed the deterministic ResNet, with one, the initial-weights ensemble + Monte Carlo (MC) dropout, an ensemble of deterministic ResNets with different initial weights to start training and dropout activated during inference, producing the most skillful and well-calibrated forecasts. The initial-weights ensemble + MC dropout benefited from generating multiple solutions that more thoroughly sampled the hypothesis space. The Bayesian ResNet ensemble was the only one that performed worse than the deterministic ResNet at longer lead times, likely due to the challenge of optimizing a larger number of parameters. To address this issue, the Bayesian-MOPED (MOdel Priors with Empirical Bayes using Deep neural network) ResNet ensemble was adopted, and it enhanced forecast skill by constraining the hypothesis search near the deterministic ResNet hypothesis. All Bayesian methods demonstrated well-calibrated uncertainty and effectively separated cases with large and small errors. In case studies, the initial-weights ensemble + MC dropout demonstrated better forecast skill than the Bayesian-MOPED ensemble and the deterministic ResNet on selected CI events in clear-sky regions. However, the initial-weights ensemble + MC dropout exhibited poorer generalization in clear-sky and anvil cloud regions without CI occurrence compared to the deterministic ResNet and Bayesian-MOPED ensemble.

  PublisherOA PDFDOI

• Existence and stability of equilibria in infectious disease dynamics with behavioral feedback

  Physical review. E · 2025-01-27 · 1 citations

  articleOpen access

  Mathematical models have provided a general framework for understanding the dynamics and control of infectious disease. Many compartmental models are limited in that they do not account for the range of behavioral feedbacks that have been observed in the response to emerging infections. Here we expand on the SIR compartmental model framework by introducing a general class of behavioral feedbacks that encompasses both individual responses and nonpharmaceutical interventions. By linking transmission dynamics and behavior, this class of models can capture the interplay of disease incidence, behavioral response, and controls such as vaccination. We prove mathematically the existence of two endemic equilibria depending on the vaccination rate: one in the presence of low vaccination but with reduced societal activity (the "new normal"), and one with return to normal activity but with vaccination rate below that required for disease elimination. Establishing the existence and stability of these equilibria is a precursor to designing control strategies that may exploit them.

  PublisherOA PDFDOI

• Preconception and Prenatal Environment and Growth Faltering Among Children in Uganda

  JAMA Network Open · 2025-03-19 · 6 citations

  articleOpen access

  Importance: Children with growth faltering are more susceptible to infections and may experience cognitive, physical, and metabolic developmental impairments. Objective: To assess whether prenatal and preconception meteorological and environmental factors are associated with village-level rates of childhood growth outcomes in Uganda. Design, Setting, and Participants: This cross-sectional study used data collected between June 20, 2015, and December 16, 2016, from the 2016 Ugandan Demographic and Health Survey for individuals aged 0 to 59 months with available anthropometric measures (weight and length or height). Data analysis was conducted from October 2020 to April 2024. Exposures: Factors assessed included meteorological information, such as drought index (Standardized Precipitation-Evapotranspiration Index [SPEI]), Aridity Index, rainfall, temperature, and vegetation indices; demographic and economic development factors (nighttime light emissions, driving time to the nearest city); and land topography (slope angle, elevation above sea level). Main Outcomes and Measures: The main outcomes were height-for-age z score (HAZ), weight-for-age z score (WAZ), and weight-for-height z score (WHZ). Spatial resolution estimates, at 1 km × 1 km of childhood growth faltering indicators, were created. Results: Of the 5219 individuals aged 0 to 59 months included in the analysis, 2633 (50%) were female; mean (SD) age was 29 (17) months. Of these individuals, 30.22% (95% CI, 29.36%-30.98%) had stunting, 12.23% (95% CI, 11.55%-12.91%) had underweight, and 3.63% (95% CI, 3.46%-3.80%) had wasting. Large disparities in the burden of childhood growth faltering existed within Uganda at smaller and larger spatial scales; villages in the northeastern and southwestern areas of the country had the highest prevalence of all forms of growth faltering (stunting, >40%; underweight, >16%; and wasting, >6%). Higher SPEI at 3 months before birth was positively associated with all childhood growth outcomes: HAZ (β, 0.06; 95% CI, 0.02-0.10), WAZ (β, 0.04; 95% CI, 0.01-0.07), and WHZ (β, 0.03; 95% CI, 0.001-0.06). Higher location mean rainfall 11 months before birth was also positively associated with HAZ (β, 0.06; 95% CI, 0.01-0.10). Aridity Index associations with WAZ (β, 0.09; 95% CI, 0.04-0.13) and WHZ (β, 0.09; 95% CI, 0.02-0.16) were consistent with findings for SPEI. Conclusions and Relevance: In this study of 5219 individuals 0 to 59 months of age in Uganda, rainfall and long-term availability of water at preconception and during gestation were positively associated with nutritional child growth outcomes. Understanding the relative contributions of meteorological environment factors on the spatial distribution of undernutrition at various spatial scales within Uganda (from the village to the district level) may help in the design of more cost-effective delivery of precision public health programs.

  PublisherOA PDFDOI

• Analyzing Self-Organizing Maps of Modeled U.S. Coastal Wind Regimes with a Comparison to Observations

  Artificial Intelligence for the Earth Systems · 2025-03-28

  articleOpen access

  Abstract Wind offshore of the northeastern United States is a vast and plentiful resource. However, wind variability needs accounting for when planning, installing, and operating offshore wind farms. Therefore, increased knowledge of four general areas becomes vital: 1) common coastal wind regimes and their impact on wind energy production, 2) common regime transitions, 3) how near-surface wind shear varies between wind-shear regimes, and 4) whether numerical forecast model skill is regime dependent. A self-organizing map (SOM) clusters hub-height (80 m) wind data from the High-Resolution Rapid Refresh (HRRR) model covering the northeast coast to address areas 1–3. The SOM identifies three general wind pattern types: unidirectional flow, confluent/diffluent flow, and cyclonic/anticyclonic flow. The strongest mean HRRR wind speeds offshore of New York are associated with a low pressure system near Maine (12 m s −1 ) and wintertime–springtime westerlies (11 m s −1 ), while the weakest winds are associated with a nearby high (≤3 m s −1 ) and a diffluence zone (4 m s −1 ). Using a separate SOM trained on 10–80-m wind differences, warm air advection over cooler northern waters typically leads to lower-level stabilization and thus increased shear. Regarding area 4, modeled winds are compared to buoy lidar observations for each SOM regime. As in observations, the HRRR monthly averaged wind speed decreased in the summer. HRRR generally underforecasts wind speed near the buoys, although the monthly averaged bias decreased over 2 years from 1.4 to 0.1–0.2 m s −1 . Greater bias occurred for regimes representing nearby pressure systems, indicating that HRRR skill can be regime dependent. Significance Statement Optimal wind energy utilization requires accurate wind forecasts, which in turn require an understanding of regional wind regimes. Our study used a machine learning method called a self-organizing map to identify the main types of weather regimes that affect offshore wind power for the northeastern U.S. coast: unidirectional flow, confluent/diffluent flow, and nearby pressure systems. In general, a unidirectional wind or a low pressure system north of the domain is a high-wind-speed regime beneficial for offshore wind energy, whereas confluence/diffluence zones or pressure systems within the domain are generally low-wind-speed regimes that are less beneficial for offshore wind energy. Future studies can apply this analysis for regime-based forecasting methods.

  PublisherOA PDFDOI

• Implications of AI for Atmospheric Predictability of Convection and Winter Storms

  2025-03-15

  preprintOpen access1st authorCorresponding

  Recent advances in artificial intelligence (AI), specifically with applications of deep learning, have brought paradigm-shifting changes to Numerical Weather Prediction.  Recent AI-based NWP systems have rivaled traditional physics-based global NWP systems according to some verification metrics.  However, the performance of these systems for extreme events, and their implications for atmospheric predictability, has not yet been fully explored.    In this study, the practical predictability for winter storms in eastern North America will be compared using forecasts generated by several traditional NWP and AI-NWP systems.   In addition to domain-wide verification statistics, the realism of cyclone structure and evolution will be evaluated at different forecast lead times.  We plan to discuss the ensemble predictability of events, evaluating the sensitivity of the AI-NWP systems to initial condition perturbations, with implications for data assimilation.  Finally, at the mesoscale, we will demonstrate a convection initiation nowcasting system that utilizes deep learning to generate probabilities of new convection forming at lead times under one hour, which we interpret using explainable AI and uncertainty quantification.

  PublisherDOI

Recent grants

• Collaborative Research: SI2-SSI: Big Weather Web: A Common and Sustainable Big Data Infrastructure in Support of Weather Prediction Research and Education in Universities

  NSF · $100k · 2015–2018

• Lake-effect Snow: Understanding Predictability and Dynamics through Ensemble-Based Convective-Permitting Data Assimilation, Modeling, and Sensitivity Analysis

  NSF · $493k · 2018–2023

Frequent coauthors

• R. J. Wilson

  Ames Research Center

  38 shared

• Eugenia Kalnay

  University of Maryland, College Park

  30 shared

• Takemasa Miyoshi

  27 shared

• Ross N. Hoffman

  NOAA Center for Satellite Applications and Research

  19 shared

• George S. Young

  The Francis Crick Institute

  19 shared

• Matthew J. Hoffman

  Rochester Institute of Technology

  17 shared

• H. E. Gillespie

  Jet Propulsion Laboratory

  14 shared

• Kayo Ide

  University of Maryland, College Park

  13 shared

Labs

• Center for Advanced Data Assimilation and Predictability Techniques (ADAPT)PI

Awards & honors

• Alumni Society Award (2011)