arXiv subjects

Maryam Golbazi

Publications and source records attributed to Maryam Golbazi.

4 recordsLinked to original sources

arXiv2026

Apeliotes: A Diffusion-Based Modeling Framework for km-scale Multi-Level Atmospheric Fields

High-resolution atmospheric data are required to resolve mesoscale and localized meteorological structures, however such datasets remain limited in many regions of the world. Existing high-resolution weather products are typically produced through dynamical downscaling, which is computationally expensive and difficult to scale across locations, variables, and forecast scenarios. These limitations motivate machine-learning-based downscaling systems that can generate multiple weather variables stochastically while producing new high-resolution fields directly. In this paper we present Apeliotes, a framework for high-resolution weather forecasting. Built on the global re-analysis atmospheric data, a pre-trained global weather foundation model, and a regionally trained generative diffusion model, Apeliotes not only provides accurate kilometer-scale weather variables, but also multi-level atmospheric fields which are not directly available in the existing global atmospheric data. Our comprehensive evaluation demonstrates that Apeliotes achieves highly competitive performance. The model predicts vertical wind profile with less than 3\% error between truth and predicted fields, achieving correlations of 0.91 for 10-m wind speed and 0.99 for 2-m temperature, with NRMSE values of 0.42 and 0.17, respectively.

cs.LG↗

arXiv2026

Learning to model pediatric asthma exacerbation from multiple risk factors: a case study in coastal Virginia

Childhood asthma is a common illness exacerbated by air pollution as well as meteorological and neighborhood-level socioeconomic factors. Modeling asthma exacerbation (AE) in large spatiotemporal datasets requires disentangling impacts from multiple contributors. In this case study, we compared three techniques that balance predictive power with interpretability to predict AE in Hampton Roads, a coastal Virginia region comprising 7 cities and over 1.5 million people. After collating ambient air pollution measurements, weather data, and measures of neighborhood opportunity, we modeled zip code-level acute AE visits to a regional children's hospital and affiliated providers from 2018-2023. Generalized linear models (GLM) provided a baseline while neural networks (NN) served as a maximally predictive target. To bridge between statistical models and deep learning, we developed a framework based on sparse dictionary learning to identify and interpret parsimonious nonlinear interacting equations. After comparing each model's predictive performance, we estimated relative risks for AE due to input exposure variables and found consensus across frameworks. Our work links statistical and interpretable machine learning models to highlight possible synergistic interactions influencing AE, and may enable future studies to guide public health interventions in coastal Virginia.

cs.LG↗

arXiv2020

Changes in air quality and human mobility in the U.S. during the COVID-19 pandemic

The first goal of this study is to quantify the magnitude and spatial variability of air quality changes in the US during the COVID-19 pandemic. We focus on two federally regulated pollutants, nitrogen dioxide (NO2), and fine particulate matter (PM2.5). Observed concentrations at all available ground monitoring sites (240 and 480 for NO2 and PM2.5, respectively) were compared between April 2020 and April of the prior five years, 2015-2019, as the baseline. Large statistically significant decreases in NO2 concentrations were found at more than 65% of the monitoring sites, with an average drop of 2 ppb when compared to the mean of the previous five years. The same patterns are confirmed by satellite-derived NO2 column totals from NASA OMI. PM2.5 concentrations from the ground monitoring sites, however, were more likely to be higher. The second goal of this study is to explain the different responses of the two pollutants during the COVID-19 pandemic. The hypothesis put forward is that the shelter-in-place measures affected peoples' driving patterns most dramatically, thus passenger vehicle NO2 emissions were reduced. Commercial vehicles and electricity demand for all purposes remained relatively unchanged, thus PM2.5 concentrations did not drop significantly. To establish a correlation between the observed NO2 changes and the extent to which people were sheltering in place, we use a mobility index, which was produced and made public by Descartes Labs. This mobility index aggregates cell phone usage at the county level to capture changes in human movement over time. We found a strong correlation between the observed decreases in NO2 concentrations and decreases in human mobility. By contrast, no discernible pattern was detected between mobility and PM2.5 concentrations changes, suggesting that decreases in personal-vehicle traffic alone may not be effective at reducing PM2.5 pollution.

physics.ao-ph↗

arXiv2019

Methods to Estimate Surface Roughness Length for Offshore Wind Energy

The northeastern coast of the U.S. is projected to expand its offshore wind capacity from the existing 30 MW to over 22 GW in the next decade. Yet, only a few wind measurements are available in the region and none at hub height, thus extrapolations are needed to estimate wind speed as a function of height. A common method is the log-law, which is based on surface roughness length (z0). No reliable estimates of z0 for the region have been presented in the literature. Here, we fill this knowledge gap using two field campaigns that were conducted in the Nantucket Sound at the Cape Wind (CW) platform. We tested three different methods to calculate z0: 1) analytical, dependent on friction velocity u* and a stability function psi; 2) the Charnock relationship between z0 and u*; and 3) a statistical method based on wind speed observed at the three levels. The first two methods are physical, whereas the statistical method is purely mathematical. Comparing mean and median of z0, we find that the median is a more robust statistics. In general, the median z0 exhibits little seasonal variability and a weak dependency on atmospheric stability, which was predominantly unstable (54-67%). The statistical method, despite delivering unrealistic z0 values at times, gives the best estimates of 60-m winds. The unrealistic z0 values are caused by non-monotonic wind speed profiles, occurring about 41% of the time, and should not be rejected because they produce realistic fits. In summary, if wind speed data from multiple levels are available, the statistical method is recommended. If multi-level wind speeds are not available but advanced sonic anemometry is available at one level, the analytical method is recommended over the Charnock's. If a constant value of z0 is sought after to characterize the region, we recommend the median from the statistical method, i.e., 6.09*10^(-3) m.

physics.ao-ph↗