The low Earth orbit (LEO) environment (200-2000 km altitude) is set to become progressively crowded with an ever-increasing number of satellite and launches mega-constellation launches, and orbital debris. Accurately forecasting orbit positions is necessary for avoiding collisions and is especially difficult during geomagnetic storms due to highly variable neutral density and the associated uncertainty on atmospheric drag. It is critical to improve the predictive capabilities of neutral densities by advancing scientific understanding of the ionosphere-thermosphere (I-T)'s storm-time response. However, current physics-based I-T models are inadequate due to biases and misspecified external forcing, and with gaps in neutral observations, robust observation-model integration is challenging. Data assimilation techniques seek to address these limitations by systematically synthesizing physics-based models and observations to enable a more complete depiction of the I-T system. Additionally, more work is needed to quantify storm-time neutral density uncertainties and their impact on orbit positions. This thesis develops new physics-based data assimilation capabilities for LEO neutral density to advance 1) scientific understanding of the thermosphere's storm-time response and 2) uncertainty quantification of LEO position errors due to geophysical variability.

This thesis first makes important contributions towards neutral density specification by using strongly coupled I-T data assimilation of electron density profiles (EDPs) retrieved from radio occultation (RO) observations to constrain neutral states. Dietrich et al. (2022) demonstrates the benefits of assimilated RO EDPs to reduce orbit position errors by 70%, accomplished through improved neutral density biases with estimated neutral temperature and neutral winds. This data assimilation framework is furthermore applied to a geomagnetic storm event to reduce overall neutral density biases by 20%, including storm recovery cooling effects. A comprehensive observing system simulation (OSSE) approach also investigates RO constellation designs as a global I-T monitoring system (Dietrich et al., 2024).

The thesis second contributes to quantifying non-Gaussian neutral density uncertainties, using a new application of a particle filter framework in a reduced forcing parameter space that incorporates a physics-based I-T model. End-to-end uncertainty quantification is accomplished using a Monte Carlo approach, wherein forcing parameter uncertainties are mapped to uncertainties of neutral density. This second component of the thesis thus opens the possibility for incorporating non-linear I-T storm dynamics into atmospheric drag and orbit position uncertainties to support future orbit positioning and conjunction analysis operations.