Rachel Stutz, Orion Space Solutions; Jeffrey Steward, Orion Space Solutions; Matthew Cooper, Orion Space Solutions; Connor Johnstone, Orion Space Solutions; Tristan Clark, Orion Space Solutions; John Noto, Orion Space Solutions; Anastasia Newheart, Orion Space Solutions; Junk Wilson, Orion Space Solutions; Geoff Crowley, Arcfield
Keywords: thermosphere, ionosphere, space weather, drag, data assimilation, HPC, high performance computing, ensemble, uncertainty
Abstract:
Satellites in Low Earth Orbit (LEO) experience a variety of orbital dynamics that are well understood. When modeled properly, the relatively small residual force of drag can be quantified and provide an opportunistic observation of thermospheric density and neutral winds. These observations have been shown to improve specification of modeled initial conditions of the thermosphere and the solar forcing conditions through the process of data assimilation. Data assimilation relies upon models of uncertainty in estimates of satellite drag. As has been repeatedly shown, the more accurate these uncertainty models, the better the data assimilation can utilize the information. A successful method employed in weather operational centers around the world (in particular, terrestrial weather centers) utilizes an ensemble of model simulations to provide “errors-of-the-day.” This ensemble approach involves running tens or even hundreds of high-resolution models with slightly different initial conditions, where each ensemble member is thought to be an equally likely “true” system, giving a rich, non-parameterized covariance model of background errors. However, such simulations require extensive high performance computing (HPC) resources, which can be a challenge outside of major research centers.
This presentation will illustrate recent preliminary results from applying a hybrid ensemble-variational data assimilation algorithm to update a physics-based model of the ionosphere-thermosphere (I-T) system with hundreds of ensemble members using first-principles physics. Since the ionosphere and thermosphere are tightly coupled, observations of one system can be propagated to the other through the underlying I-T model. Satellite ephemerides can be used as opportunistic observations of the neutral atmosphere, and these, as well as observations of the ionosphere, are assimilated onto a physics-based background model of the I-T system. Ionospheric observations include Total Electron Content (TEC) from GNSS ground and Radio Occultation (RO) receivers as well as ionosonde measurements. The use of ensembles in this data assimilation method allows for capturing uncertainty in thermospheric density and winds, which are then used to update those fields and calculate uncertainty in propagated LEO satellite positions. These techniques can apply to various Space Domain Awareness (SDA) use cases, including precision in orbital prediction for object tracking and maneuver detection, as well as multiple-object conjunction assessment.
Along with preliminary data assimilation results of this system, the focus of this presentation is the computational architecture used to perform this analysis in the cloud. Running high-rank ensembles of global space weather models, along with sophisticated assimilation techniques involving large-scale covariance models, can put a strain on compute resources for users without significant on-premise resources like supercomputing clusters. This study uses a Slurm-based HPC architecture in the cloud, which allows for on-demand access to cloud resources and reduces compute costs compared to operating them continuously. In this setup, data assimilation workflows can be executed via Parsl in a Jupyter notebook, allowing for interactive and iterative workflow development, and the status of tasks in the workflow can be visualized for ease of workflow monitoring and debugging. The use of the aforementioned setup and tools for developing and executing the data assimilation workflow in this study is shown. Finally, an analysis of the cost and runtime required for these workflow executions is performed, including a comparison to alternative computing architectures. As the need for HPC grows in the space domain with increasingly complex problems being solved, the ability to make efficient use of computational resources becomes yet more important. The lessons learned in this study will help inform not only the state of the art in assimilative space weather modeling but also the way that powerful compute resources can be harnessed within the broader SDA community.
Date of Conference: September 16-19, 2025
Track: Atmospherics/Space Weather