Machine‐Learned HASDM Thermospheric Mass Density Model With Uncertainty Quantification

Licata, Richard J.; Mehta, Piyush M.; Tobiska, W. Kent; Huzurbazar, Snehalata

doi:10.1029/2021sw002915

Cited by 29 publications

(51 citation statements)

References 54 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A higher noise level in the second iteration allows the estimated density to track the high‐frequency variations in the true density. During the 2003 Halloween storm, the errors in HASDM are around 20%–25% around the storm period along both CHAMP and GRACE‐A orbits (Licata, Mehta, Tobiska, & Huzurbazar, 2021). Using their machine‐learned global temperature model called EXTEMPLAR‐ML, Licata, Mehta, Wiemer, and Tobiska (2021) were able to reduce the mean density errors to 17.49% along CHAMP's orbit and 19.91% along GRACE‐A's orbit.…”

Section: Discussionmentioning

confidence: 99%

A Framework to Estimate Local Atmospheric Densities With Reduced Drag‐Coefficient Biases

et al. 2022

Self Cite

View full text Add to dashboard Cite

An accurate estimation of upper atmospheric densities is crucial for precise orbit determination (POD), prediction of low Earth orbit satellites, and scientific studies of the Earth's atmosphere. But densities estimated using satellite tracking data are always uncertain up to the drag‐coefficient assumed in the inversion method. This work develops a new framework to simultaneously estimate the density and drag‐coefficient for satellites with a time‐varying attitude. We do so by leveraging Fourier drag‐coefficient models, previously developed by the authors, and physical models of the drag‐coefficient. The method is tested with synthetic data for different geomagnetic activities, altitude levels, and errors in the gas‐surface interaction parameters. We report an improvement of up to 70% in density estimates for the simulations. Finally, POD data from Spire satellites are used for validation. An improvement of around 29% is obtained in the filter density estimates over NRLMSISE‐00 and 49% over JB2008 compared to the High Accuracy Satellite Drag Model densities.

show abstract

Section: Discussionmentioning

confidence: 99%

A Framework to Estimate Local Atmospheric Densities With Reduced Drag‐Coefficient Biases

et al. 2022

Self Cite

View full text Add to dashboard Cite

show abstract

“… are temporal PCA coefficients and are orthogonal modes—also called basis functions. The choice order of truncation ( r = 10) was chosen from analyses in previous work as it allows for 90% of the variance to be captured and only results in % truncation error 23 , 54 . The orthogonal modes are derived through U consists of orthogonal vectors representing the modes of variability.…”

Section: Methodsmentioning

confidence: 99%

“…6 ). Licata et al found that the mean square error loss function resulted in underestimated uncertainty estimates in surrogate modeling for the HASDM dataset 23 .…”

Section: Methodsmentioning

confidence: 99%

“…Bruinsma et al developed an uncertainty-based version of DTM2020 using polynomials to describe the 1 − sigma uncertainties as a function of the inputs 22 . Licata et al used MC dropout to obtain uncertainty estimates for a global density modeling application with good calibration, providing baseline performance 23 . In this work, we leverage machine learning (ML) to generate predictive density models for the thermosphere that also provide robust and reliable uncertainty estimates.…”

Section: Introductionmentioning

confidence: 99%

“…The combination of numbers refers to the average of that index over that many hours prior to epoch. The HASDM input set originates from Licata et al 23 .…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Uncertainty quantification techniques for data-driven space weather modeling: thermospheric density application

Licata

Mehta

2022

Sci Rep

Self Cite

View full text Add to dashboard Cite

Machine learning (ML) has been applied to space weather problems with increasing frequency in recent years, driven by an influx of in-situ measurements and a desire to improve modeling and forecasting capabilities throughout the field. Space weather originates from solar perturbations and is comprised of the resulting complex variations they cause within the numerous systems between the Sun and Earth. These systems are often tightly coupled and not well understood. This creates a need for skillful models with knowledge about the confidence of their predictions. One example of such a dynamical system highly impacted by space weather is the thermosphere, the neutral region of Earth’s upper atmosphere. Our inability to forecast it has severe repercussions in the context of satellite drag and computation of probability of collision between two space objects in low Earth orbit (LEO) for decision making in space operations. Even with (assumed) perfect forecast of model drivers, our incomplete knowledge of the system results in often inaccurate thermospheric neutral mass density predictions. Continuing efforts are being made to improve model accuracy, but density models rarely provide estimates of confidence in predictions. In this work, we propose two techniques to develop nonlinear ML regression models to predict thermospheric density while providing robust and reliable uncertainty estimates: Monte Carlo (MC) dropout and direct prediction of the probability distribution, both using the negative logarithm of predictive density (NLPD) loss function. We show the performance capabilities for models trained on both local and global datasets. We show that the NLPD loss provides similar results for both techniques but the direct probability distribution prediction method has a much lower computational cost. For the global model regressed on the Space Environment Technologies High Accuracy Satellite Drag Model (HASDM) density database, we achieve errors of approximately 11% on independent test data with well-calibrated uncertainty estimates. Using an in-situ CHAllenging Minisatellite Payload (CHAMP) density dataset, models developed using both techniques provide test error on the order of 13%. The CHAMP models—on validation and test data—are within 2% of perfect calibration for the twenty prediction intervals tested. We show that this model can also be used to obtain global density predictions with uncertainties at a given epoch.

show abstract

Qualitative and Quantitative Assessment of the SET HASDM Database

Licata

Mehta

Tobiska

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

Machine‐Learned HASDM Thermospheric Mass Density Model With Uncertainty Quantification

Cited by 29 publications

References 54 publications

A Framework to Estimate Local Atmospheric Densities With Reduced Drag‐Coefficient Biases

A Framework to Estimate Local Atmospheric Densities With Reduced Drag‐Coefficient Biases

Uncertainty quantification techniques for data-driven space weather modeling: thermospheric density application

Qualitative and Quantitative Assessment of the SET HASDM Database

Contact Info

Product

Resources

About