Air pollution exposure is associated with increased risk for multiple adverse health outcomes including neurological disorders, asthma, diabetes, ischemic heart disease, and death. Race/ethnicity and income are predictors for air pollution exposure across most cities in the United States due to past discriminatory housing practices such as redlining. Future public policies must address this Environmental Justice issue so that all socio-economic groups have access to clean air.Predicting air pollution exposure disparities under different future scenarios is a difficult task because non-linear chemistry governs the formation of many pollutants. Historical relationships between pollutant concentrations and land-use or even pollutant concentrations and emissions may therefore not be useful when predicting future conditions. Chemical Transport Models (CTMs) predict air pollution concentrations based on fundamental chemical and physical equations that can accurately transition to future conditions. The research in this thesis explores how to refine CTM inputs and configuring CTM spatial resolution to accurately quantify air pollution exposure disparities in the present day and in the future. In Chapter 2, ten major spatial surrogates describing the detailed locations of air pollution emissions in California are created/updated for the base year 2010 and future years from 2015 to 2040. The updated spatial surrogates generally improve CTM predictions for PM mass and EC concentrations in the Sacramento area (~10% for PM, ~3% for EC), the Bay Area (~3% for PM, ~1.5% for EC), and the region surrounding Los Angeles (~5% for PM, ~4% for EC). The updated spatial surrogates also improve predicted NOx concentrations in the core region of Los Angeles (~6%). Chapter 3 explores the relationship between domain size and spatial resolution that affects predicted air pollution disparities in present day and future simulations when data support from measurements is not available. Overall WRF/Chem CTM accuracy improves approximately 9% as spatial resolution increases from 4 km to 250 m in present-day simulations. Exposure disparity results are consistent with previous findings: minorities experience higher exposure than White residents. Predicted exposure disparities are found to be a function of the model configuration. CTM configurations that use spatial resolution/domain size of 1 km / 103 km2 and 4 km / 104 km2 over Los Angeles can detect a 0.5 µg m-3 exposure difference with statistical power greater than 90%. Chapter 4 conducts a comprehensive analysis of health co-benefits, racial disparities, and source / composition in air pollution exposure under six future energy scenarios and four future meteorology scenarios in California for future year 2050. Deeper reductions in the carbon intensity of energy sources progressively are found to reduce exposure to PM2.5 mass and PM0.1 mass for all California residents. The three energy scenarios that achieve an ~80% reduction in GHG emissions relative to 1990 levels simultaneously produce the greatest reduction in PM exposure for all California residents and the greatest reduction in the racial disparity of that exposure. The EJ assessment shows that adoption of low-carbon energy sources in the year 2050 reduces the race/ethnicity disparity in air pollution exposure in California by as much as 20% for PM2.5 mass and by as much as 40% for PM0.1 mass. Future studies should apply the methods developed in this thesis to other locations across the United States in order to better understand how future policies such as a transition to low carbon energy can help to reduce air pollution exposure disparities by race/ethnicity.

Extra-solar planets, brown dwarfs, and low-mass stars are distinct astronomical objects that encompass a wide range of masses, occupying the lower end of the mass and temperature spectrum within the vastness of the Universe. These cool objects’ peak emission occurs at longer wavelengths in the red and infrared regions. They present abundant avenues for observational and theoretical exploration, facilitating the study of their evolutionary pathways. It is believed that stars are born from gravitational collapse of molecular clouds of gas and dust that are scattered throughout most galaxies over thousands to millions of years. The hot core of the collapsing cloud will form a protostar with fast rotation. Some of these spinning clouds of collapsing gas and dust break up into two or three blobs that each form stars which later become multiple systems. The crucial factor setting these objects apart lies in their initial masses, which provide valuable insights into their formation mechanisms and their post-formation evolution. Stars (≥ 80MJup), including low-mass stars like M dwarfs, are self-luminating entities that sustain nuclear fusion in their cores, generating energy through the fusion of hydrogen atoms. Brown dwarfs occupy a mass range between that of the heaviest gas giant planets and the lightest stars (∼ 15− ∼ 80MJup), they fall short of the mass required for sustained hydrogen burning and do not possess a stable energy-generating mechanism. They briefly undergo deuterium burning during their early formation stages, but this process is not sustainable in the long term. Planets (∼ < 15MJup), distinct from stars and brown dwarfs, lack an inherent energy generation process and primarily emit radiation received from their host star and residual internal heat from their formation. Their energy output is primarily determined by the absorbed and reflected light from their parent star. The distinction between these objects is crucial as it helps us classify and study the vast diversity of objects beyond our solar system, providing insights into their formation, evolution, and the conditions that allow for the existence of habitable planets. If one can determine the mass of a detected object empirically and independently, it not only helps confirm the nature of the object but also tests and informs evolutionary models of substellar or sometimes stellar evolution. The work in this thesis expands the sample of planets and brown dwarfs whose dynamical masses and orbits are empirically measured, in order to chart the vast substellar evolutionary landscape spanning the entire range of masses.

Before the Gaia satellite, the detection of directly imaged brown dwarfs and giant planets orbiting main-sequence stars was a rare occurrence. Merely a small collection of approximately thirty such objects had been identified, and only a fraction of them possessed independently measured masses. These masses, determined through the observation of orbital dynamics, are known as dynamical masses and are regarded as the most reliable and independent measurements, as they do not rely on cooling or evolutionary models. The work undertaken in this thesis focuses specifically on attaining precise mass measurements using the synergistic capabilities of the Hipparcos and Gaia astrometric missions. By combining multiple observation data types, namely Radial Velocity (RV), relative astrometry, and absolute astrometry, one can determine the 3D Keplerian orbit and the precise dynamical mass of a potential, presumably faint companion orbiting a star by measuring the reflex motion of the star in the sky as it responds to the gravitational tug exerted by the unseen companion in orbit.

In Chapter 2, we demonstrate the powerful combination of radial velocities, direct imaging, and Hipparcos and Gaia absolute astrometry via the Hipparcos-Gaia Catalog of Accelerations (HGCA) to achieve highly accurate dynamical mass measurements. To accomplish this, we utilize the Markov-Chain-Monte-Carlo code orvara, along with the epoch astrometry fitting code htof, which enable us to effectively constrain orbits using these three distinct sources of data. By employing orvara, we successfully address the inherent degeneracy in determining an exoplanet’s mass and inclination encountered in RV-only studies by incorporating absolute astrometry in conjunction with radial velocity. Moving on to Chapter 3, we focus on astrometry-only data, using Gaia EDR3 to calibrate ground-based relative and absolute astrometry from VLT/ESO, to precisely determine the barycentric orbit of ε Indi B, one of the closest and the first discovered binary T dwarf systems. These measurements establish them as the most accurately characterized binary brown dwarfs to date, and enable both a relative and absolute test of currently available substellar cooling models. In Chapter 4 and Chapter 5, we present the detection results from our pilot observation program conducted at the Keck/NIRC2 instrument, aiming to identify and image substellar companions around nearby accelerating stars. By utilizing the valuable precursor information from orvara and HGCA, we are able to predict the precise location of the companion a priori, detect the companion with direct imaging, and jointly fit orbits to their data to obtain precise dynamical masses. Our work significantly enhances the success rate of high-contrast imaging surveys, and can serve as a powerful tool for selecting promising follow-up candidates for interferometry and spectroscopic characterizations with instruments like JWST and GRAVITY/KPIC.

This thesis focuses on the domain of multi-instrumental music generation using deep learning. In particular, this thesis focuses on the application of the attention-based Transformer neural network architecture in the aforementioned domain, along with exploring different music representations, and the application of transfer learning in the context of music generation.

Background

Methods

Results

Conclusions