Reduced-Order Data Assimilation for Thermospheric Density Using Physics-informed SINDyc Models
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.
Rosa: We're looking at the paper: "Reduced-Order Data Assimilation for Thermospheric Density Using Physics-informed SINDyc Models".
Dev: Here's the gist of the paper. Orbit prediction and space situational awareness require accurate thermospheric mass density, which responds nonlinearly to solar and geomagnetic forcing across several orders of magnitude <ref:2604.24646#pg2> The method described here pairs a simplified, physics-informed atmospheric model with satellite measurements to produce observationcalibrated density estimates <ref:2604.24646#pg6> Incorporating satellite data reduces density errors relative to running the model open-loop, most clearly during disturbed conditions <ref:2604.24646#pg2> The approach remains stable when observations are sparse and runs at a small fraction of the cost of full general-circulation propagation, supporting the update cadence used in operational drag prediction <ref:2604.24646#pg6> A record of upper-atmosphere density from August 2000 to December 2025 is released alongside this work <ref:2604.24646#pg2>.
Rosa: Model Architecture and Dynamics The core of the method involves a SINDyc-AR reduced-order model that propagates thermospheric density in a PCA latent space driven by solar and geomagnetic indices <ref:2604.24646#pg2> An Extended Kalman Filter assimilates in situ LEO density measurements, correcting the reduced-order latent state in real time <ref:2604.24646#pg2>.
Dev: The model is built using PCA to compress the high-dimensional thermospheric state into a low-dimensional latent representation that retains the dominant modes of global density variability <ref:2604.24646#pg2> The dynamics are modeled as z˙(t) = f(z(t), u(t)) <ref:2604.24646#pg2>.
Rosa: This framework uses both SINDyc-AR as the nonlinear reduced-order model and DMDc as a linear reference it is compared against <ref:2604.24646#pg2>.
Paper summary: Rosa: This paper, "Reduced-Order Data Assimilation for Thermospheric Density Using Physics-informed SINDyc Models," is tackling the problem that orbit prediction and space situational awareness need accurate thermospheric mass density, which responds nonlinearly to solar and geomagnetic forcing across several orders of magnitude >
Dev: They introduce a SINDyc-AR reduced-order model that propagates this thermospheric density in a PCA latent space driven by solar and geomagnetic indices >
Taro: Then they use an Extended Kalman Filter to assimilate in situ LEO density measurements, correcting that reduced-order latent state in real time >
Rosa: The core idea is using Principal Component Analysis, or PCA, to compress the high-dimensional thermospheric state into a low-dimensional latent representation that keeps the dominant modes of global density variability >
Dev: The dynamics are modeled as z dot t equals f of z t and u t, where SINDyc-AR is used for the nonlinear reduced-order model and DMDc is compared against it as a linear reference >
Taro: The assimilation framework uses an augmented companion state zeta k defined in Eq. (thirteen), which correctly propagates the cross-covariances between the current and lagged latent states that show up in any AR process >
Rosa: When they get a log10 density measurement yk, the innovation is calculated as nu k equals yk minus Haug hat zeta kk-one minus mu, where Haug is augmented as H plus zero >
Dev: This formulation correctly handles those cross-covariances that arise in an AR process, and the observation operator is augmented to include that log10 density measurement >
Conclusion: Rosa: So we've been looking at how this paper uses SINDyc models and data assimilation to get better estimates of thermospheric density, and now we’re heading into the conclusion to see what it actually means for us out here on the ground or in space.
Dev: Yeah, so the title itself, "Reduced-Order Data Assimilation for Thermospheric Density Using Physics-informed SINDyc Models," it tells you a lot about how they tackled this problem—it’s about taking a complex physics model and making it fast enough to actually use with real satellite data >
Rosa: Right. It means they took these high-dimensional atmospheric states, right? And instead of running a super slow general circulation model every time, they used this reduced-order model to get the main stuff quickly >
Dev: Exactly. And then the data assimilation part—the Extended Kalman Filter—that’s where you take those actual measurements from satellites and feed them back into the system to correct the model in real time >
Rosa: So, what does that correction actually do? It means we can get a much more accurate picture of how dense the atmosphere is right where a satellite is, especially when things are getting rough with solar activity or geomagnetic storms >
Dev: That’s the core idea. The results show that this approach consistently lowers the errors in density estimation compared to just running the model on its own, particularly when you look at those tracks where they actually have good data >
Rosa: And they released a big dataset, a record from two thousand to two thousand twenty-five spanning August through December, which is pretty substantial for anyone trying to study long-term atmospheric trends >
Dev: It’s a good dataset, but we gotta remember what the paper flags as limitations. They admit that using just ten PCA basis functions for the model might miss some of the smaller, more localized density structures during really extreme storm conditions >
Rosa: So it’s not perfect everywhere—it's good where they have data, but they know it might struggle in those super chaotic moments or at very high latitudes >
Dev: That points toward future work, right? They suggest extending this to incorporate other density fields directly into the filter and even using TLE-based estimates for energy dissipation rates when we look ahead >
Rosa: Right. So, while this framework gives us a powerful way to get good density estimates now, it’s also showing us where the next big steps in refining these models need to happen for really tough conditions >
Department of Mechanical, Materials and Aerospace Engineering, West Virginia University
eess.SY, cs.SY
Submitted: 2026-04-27
Updated: 2026-10-29
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 80/100
The gist: Orbit prediction and space situational awareness require accurate thermospheric mass density, which responds nonlinearly to solar and geomagnetic forcing across several orders of magnitude…
Key concepts
- SINDyc-AR
- This is a reduced-order model that uses Sparse Identification of Nonlinear Dynamics (SINDy) combined with an Autoregressive (AR) structure. It simplifies the complex thermospheric density dynamics by finding a sparse set of coefficients that describe how the density changes based on solar and geomagnetic forcing.
- Extended Kalman Filter (EKF)
- The EKF is used to assimilate real-time satellite measurements into the reduced-order model. It corrects the model's predicted state in real time by comparing predictions with actual measurements, effectively reducing errors in the density estimate.
- PCA Latent Space
- Principal Component Analysis (PCA) is used to compress a high-dimensional thermospheric state into a much lower-dimensional space. This latent representation captures the most important modes of global density variability, making the model easier to manage and propagate.
Terminology
Summary
Orbit prediction and space situational awareness require accurate thermospheric mass density, which responds nonlinearly to solar and geomagnetic forcing across several orders of magnitude <ref:2604. p. 1 manuscript submitted to Space Weather
Abstract
Orbit prediction and space situational awareness require accurate thermospheric mass
density, which responds nonlinearly to solar and geomagnetic forcing across several orders of magnitude. Physics-based general circulation models resolve that response but
are computationally expensive, while empirical models run cheaply and carry no timeevolving atmospheric state. Physics-based models describe these processes in detail but
are too slow for routine use, while simpler empirical models run quickly but track changing space-weather conditions less faithfully. The method described here pairs a simplified, physics-informed atmospheric model with satellite measurements to produce observationcalibrated density estimates. The simplified model represents the main patterns of thermospheric behavior and how they change in response to space weather; a data assimilation technique updates the model using density observations from satellites such as
CHAMP, GRACE, GRACE-FO, GOCE, and Swarm. Incorporating satellite data reduces
density errors relative to running the model open-loop, most clearly during disturbed
conditions. Incorporating satellite data reduces
density errors relative to running the model open-loop, most clearly during disturbed
conditions. The approach remains stable when observations are sparse and runs at a small
fraction of the cost of full general-circulation propagation, supporting the update cadence
used in operational drag prediction. An observation-calibrated
global density record spanning August 2000 to December 2025 is released alongside this
work. 1 Plain Language Summary
Predicting the density of the thermosphere, the upper layer of Earth’s atmosphere,
is important for tracking satellites and managing space operations.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Model Architecture and Dynamics
The core of the method involves a SINDyc-AR reduced-order model that propagates thermospheric density in a PCA latent space driven by solar and geomagnetic indices <ref:2604.24646#pg2> An Extended Kalman Filter assimilates in situ LEO density measurements, correcting the reduced-order latent state in real time <ref:2604. p. 0 Key Points:
• A SINDyc-AR reduced-order model propagates thermospheric density in a PCA
latent space driven by solar and geomagnetic indices. • An Extended Kalman Filter assimilates in situ LEO density measurements, correcting the reduced-order latent state in real time. The framework developed below combines three components. PCA compresses the
high-dimensional thermospheric state into a low-dimensional latent representation that
retains the dominant modes of global density variability. mehta@mail. edu
–1–
arXiv:2604. mehta@mail. edu
–1–
arXiv:2604. The fullstate estimate is recovered via the inverse projection
xˆfull,k = Wzˆk + µ0. 3 Library Construction and SINDyc-AR
With the thermospheric state represented in latent coordinates, the dynamics are
modeled as
z˙(t) = f(z(t), u(t)), (8)
where z(t) ∈ R
r
is the latent state and u(t) ∈ R
ndrivers is the vector of external drivers. , 2026); the same computational economy that
makes ROM attractive also makes an ensemble of dynamics models tractable at an emulator’s compute budget rather than a general circulation model’s, with the spread across
those models providing uncertainty quantification and evolving ROM into a reduced order probabilistic emulator (ROPE) (Licata & Mehta, 2023, 2024; Tapedia & Mehta, 2026). This work uses both: SINDyc-AR as the nonlinear reduced-order model, and DMDc as
the linear reference it is compared against.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Data Assimilation Framework
The Extended Kalman Filter (EKF) operates on the augmented companion state ζk defined in Eq. (13) <ref:2604. p. 0 mehta@mail. edu
–1–
arXiv:2604. (13), with
the associated augmented covariance P
aug
k|k
∈ R
naug×naug . This formulation correctly propagates the cross-covariances between the current and lagged latent states that arise in
any AR process. Log-density is easier to treat statistically and expresses relative variations directly, which is how density behavior is best characterized (Emmert & Picone,
2010). Observation operator
For a satellite at position (l, τ, a) (latitude, local time, altitude), let ϕ ∈ R
d denote the vector of tri-linear interpolation weights at that location. mehta@mail. edu
–1–
arXiv:2604. 5 Log-Density Measurement Model
Onboard satellite accelerometers and neutral mass spectrometers provide measurements of thermospheric density that span several orders of magnitude. Rather than assimilating these observations in linear density units, which would produce poorly scaled
filter gains, measurements are expressed in log10 density.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Model Identification and Extension
The SINDyc-AR model identifies a sparse set of coefficients Ξ so the dynamics are a linear combination of nonlinear basis functions drawn from a candidate library <ref:2604.24646#pg2>. The autoregressive extension adds lagged state and driver values, so that temporal lags act as lifting coordinates in the Koopman sense <ref:2604. p. 3 The SINDy with control extension (SINDyc) (Brunton et al. , 2016b) adds driverdependent and cross terms to capture state–driver interactions, and the autoregressive
extension adopted here, SINDyc-AR, adds lagged state and driver values, so that temporal lags act as lifting coordinates in the Koopman sense (Arbabi & Mezi´c, 2017; Pan
& Duraisamy, 2020). With this convention the
single-snapshot and matrix forms of the dynamics carry the coefficient matrix on the same
side, z˙ = Ξ θ(z, u) and Z˙ = Ξ Θ(Z, U), so the ordering of the matrix product is fixed
by the shapes and no transposition is implied between them. The library matrix is constructed by evaluating a set of candidate functions over
the available data:
Θ(Z, U) =
1
zi
zizj
ua
uaub
ziua
.Where the paper says it
Where the paper says it
Performance and Output
The framework was evaluated across three assimilation scenarios during both geomagnetically active and quiet periods <ref:2604.24646#pg2>. Assimilation consistently reduces density estimation errors along assimilated orbit tracks relative to the open-loop SINDyc-AR forecast <ref:2604. p. 1 The simplified model represents the main patterns of thermospheric behavior and how they change in response to space weather; a data assimilation technique updates the model using density observations from satellites such as
CHAMP, GRACE, GRACE-FO, GOCE, and Swarm. Incorporating satellite data reduces
density errors relative to running the model open-loop, most clearly during disturbed
conditions. manuscript submitted to Space Weather
This paper contributes: (i) an EKF that operates on the companion-form lifting
of the SINDyc-AR emulator, so the filter propagates the cross-lag covariances the autoregressive structure implies; (ii) an analytical Jacobian propagation for the nonlinear
library, keeping the error covariance consistent with the first-order expansion of the AR
process so the process noise need not absorb linearization error; (iii) a log-density observation model whose measurement Jacobian is state-independent, giving a well-conditioned
update across the full dynamic range of density; (iv) a multi-satellite extension that assimilates geometrically distinct tracks simultaneously in the shared latent space; and (v)
an assimilated thermospheric density dataset spanning 2000–2025, released with this work. The reduced state is what makes this affordable: a 60-state EKF update costs on
the order of a 60 × 60 covariance propagation rather than a computing allocation, so
the covariance that carries the uncertainty on the density is propagated exactly rather
than sampled from an ensemble (Section 4.Where the paper says it
Where the paper says it
Limitations and Future Directions
The framework has several limitations, including the choice of a fixed r = 10 PCA basis being heuristic and may underrepresent mesoscale and high-latitude density structures during extreme storm conditions <ref:2604.24646#pg2>. Cross-orbit generalization to withheld orbits remains partial, as the correction is biased toward the altitude and local-time regions that the assimilated tracks sample <ref:2604.24646#pg2>. Future work includes extending this framework to assimilate HASDMderived density fields directly within the filter <ref:2604. p. 4 The SINDycAR model operates within this latent space to identify a compact representation of the
latent-state evolution under external solar and geomagnetic forcing. An EKF operates
in the same latent space to assimilate onboard satellite density measurements, correcting forecast errors and propagating uncertainty estimates alongside the state. Physics-based models describe these processes in detail but
are too slow for routine use, while simpler empirical models run quickly but track changing space-weather conditions less faithfully. The method described here pairs a simplified, physics-informed atmospheric model with satellite measurements to produce observationcalibrated density estimates. , 2018) gives efficient state estimation without adjusting the external
drivers directly, though it does not resolve how well model error is characterized: process noise still needs tuning, and the reduced model inherits whatever biases its parent
carries. What it changes is the size of the problem: a covariance of order 101–102
states
can be propagated exactly, with no need to approximate it by an ensemble.Where the paper says it
Where the paper says it
Where the paper says it
Supporting Information
The Supporting Information provides detailed diagnostics, including the normalized permode change in the latent state due to assimilation and the 400 km global density panels with the assimilated-minus-open-loop difference field <ref:2604.24646#pg2>. The authors also archive their model code and analysis scripts on Zenodo under a CC BY 4. p. 2 ROM trades a bounded amount of physical fidelity for several orders of magnitude
of compute reduction (Qian et al. , 2014; Mehta & Linares, 2017), compressing that output into a low-dimensional latent space, usually by proper orthogonal decomposition (POD)
or principal component analysis (PCA) (Golub & Reinsch, 1970; Rowley et al. mehta@mail. edu
–1–
arXiv:2604. This work
couples the two: a quasi-physical ROM built with both DMDc and an autoregressive SINDyc
formulation, trained on TIE-GCM simulations, with an Extended Kalman Filter (EKF)
assimilating in situ mass density from CHAMP, GRACE, GRACE-FO, GOCE and Swarm. It is evaluated on withheld orbits across geomagnetic conditions and single-, dual- and
multi-satellite configurations, against open-loop reduced-order predictions, NRLMSIS 2. mehta@mail. edu
–1–
arXiv:2604. mehta@mail. edu
–1–
arXiv:2604. mehta@mail. edu
–1–
arXiv:2604. manuscript submitted to Space Weather
Reduced-Order Data Assimilation for Thermospheric
Density Using Physics-informed SINDyc Models
Sriram Narayanan1∗
, Daniele Sicoli1
, and Piyush Mehta1
1Department of Mechanical, Materials and Aerospace Engineering, West Virginia University, Morgantown,
WV, USA. Key Points:
• A SINDyc-AR reduced-order model propagates thermospheric density in a PCA
latent space driven by solar and geomagnetic indices. manuscript submitted to Space Weather
Reduced-Order Data Assimilation for Thermospheric
Density Using Physics-informed SINDyc Models
Sriram Narayanan1∗
, Daniele Sicoli1
, and Piyush Mehta1
1Department of Mechanical, Materials and Aerospace Engineering, West Virginia University, Morgantown,
WV, USA. Key Points:
• A SINDyc-AR reduced-order model propagates thermospheric density in a PCA
latent space driven by solar and geomagnetic indices. (3)
The thermosphere is primarily forced by two external drivers: the solar 10. 7 cm flux F10. (3)
The thermosphere is primarily forced by two external drivers: the solar 10. 7 cm flux F10. 7 cm flux F10. 7
(in solar flux units, sfu), a proxy for solar extreme-ultraviolet radiation, and the planetary geomagnetic index Kp, which characterizes the level of geomagnetic activity. 7 cm flux F10. 7
(in solar flux units, sfu), a proxy for solar extreme-ultraviolet radiation, and the planetary geomagnetic index Kp, which characterizes the level of geomagnetic activity. mehta@mail. edu
–1–
arXiv:2604. mehta@mail. edu
–1–
arXiv:2604. This work
couples the two: a quasi-physical ROM built with both DMDc and an autoregressive SINDyc
formulation, trained on TIE-GCM simulations, with an Extended Kalman Filter (EKF)
assimilating in situ mass density from CHAMP, GRACE, GRACE-FO, GOCE and Swarm. It is evaluated on withheld orbits across geomagnetic conditions and single-, dual- and
multi-satellite configurations, against open-loop reduced-order predictions, NRLMSIS 2. This work
couples the two: a quasi-physical ROM built with both DMDc and an autoregressive SINDyc
formulation, trained on TIE-GCM simulations, with an Extended Kalman Filter (EKF)
assimilating in situ mass density from CHAMP, GRACE, GRACE-FO, GOCE and Swarm. It is evaluated on withheld orbits across geomagnetic conditions and single-, dual- and
multi-satellite configurations, against open-loop reduced-order predictions, NRLMSIS 2. For completeness, neural networks are also used to model the dynamics of the reduced state (Tapedia et al. , 2026); the same computational economy that
makes ROM attractive also makes an ensemble of dynamics models tractable at an emulator’s compute budget rather than a general circulation model’s, with the spread across
those models providing uncertainty quantification and evolving ROM into a reduced order probabilistic emulator (ROPE) (Licata & Mehta, 2023, 2024; Tapedia & Mehta, 2026). For completeness, neural networks are also used to model the dynamics of the reduced state (Tapedia et al. , 2026); the same computational economy that
makes ROM attractive also makes an ensemble of dynamics models tractable at an emulator’s compute budget rather than a general circulation model’s, with the spread across
those models providing uncertainty quantification and evolving ROM into a reduced order probabilistic emulator (ROPE) (Licata & Mehta, 2023, 2024; Tapedia & Mehta, 2026). 1 (Emmert
et al. , 2021, 2022) and the High Accuracy Satellite Drag Model (HASDM) (Storz et al. 1 (Emmert
et al. , 2021, 2022) and the High Accuracy Satellite Drag Model (HASDM) (Storz et al. mehta@mail. edu
–1–
arXiv:2604. The dataset released with this work, described fully in Section 6, is a continuous
observation-calibrated global density record covering August 2000 to December 2025, built
by running the filter forward across the period and ingesting whatever subset of CHAMP,
GRACE, GRACE-FO, GOCE and Swarm measurements is available at each epoch. It
ships in HDF5 format as the assimilated and open-loop reduced-order states on a 60-
second time base, with a per-step log of contributing satellites, filter-consistency diagnostics and a notebook that reconstructs the density field on the full grid. The dataset released with this work, described fully in Section 6, is a continuous
observation-calibrated global density record covering August 2000 to December 2025, built
by running the filter forward across the period and ingesting whatever subset of CHAMP,
GRACE, GRACE-FO, GOCE and Swarm measurements is available at each epoch. It
ships in HDF5 format as the assimilated and open-loop reduced-order states on a 60-
second time base, with a per-step log of contributing satellites, filter-consistency diagnostics and a notebook that reconstructs the density field on the full grid. mehta@mail. edu
–1–
arXiv:2604. mehta@mail. edu
–1–
arXiv:2604. , 2022; Mezi´c, 2005). Dynamic Mode Decomposition (DMD) computes a best-fit linear operator in the original
state space, a finite-rank Koopman approximation on a linear observable subspace (Schmid,
2010; Rowley et al. It
ships in HDF5 format as the assimilated and open-loop reduced-order states on a 60-
second time base, with a per-step log of contributing satellites, filter-consistency diagnostics and a notebook that reconstructs the density field on the full grid. The remainder of this paper is organized as follows. The method described here pairs a simplified, physics-informed atmospheric model with satellite measurements to produce observationcalibrated density estimates. The simplified model represents the main patterns of thermospheric behavior and how they change in response to space weather; a data assimilation technique updates the model using density observations from satellites such as
CHAMP, GRACE, GRACE-FO, GOCE, and Swarm.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Robust density estimation during geomagnetic storms through observation-calibrated state correction. The system can correct density errors relative to open-loop prediction,
most visibly during geomagnetic storms and under singlesatellite coverage.
-
Efficient real-time thermospheric state tracking using a reduced-order model. By propagating the state in a low-dimensional latent space, the system achieves
the same computational economy that makes ROM attractive also makes an ensemble of dynamics models tractable at an emulator’s compute budget rather than a general circulation model’s.
-
Improved accuracy across diverse observational geometries via multi-satellite assimilation. The system can assimilate
geometrically distinct tracks simultaneously in the shared latent space,
leading to improved estimates along every assimilated track while providing across-orbit generalization
benefit when validating withheld orbits. -
Scale-invariant measurement processing using log-density observations. The observation model allows for assimilation of measurements expressed in log10 density, which is
easier to treat statistically and expresses relative variations directly.
-
Long-term, continuous thermospheric density product generation. The framework can provide
a continuous observation-calibrated global density record covering August 2000 to December 2025,
which includes per-step diagnostics for assessing filter behavior over the entire observational period.
Abstract
Orbit prediction and space situational awareness require accurate thermospheric mass density, which responds nonlinearly to solar and geomagnetic forcing across several orders of magnitude. Physics-based general circulation models resolve that response but are computationally expensive, while empirical models run cheaply and carry no time-evolving atmospheric state. An autoregressive Sparse Identification of Nonlinear Dynamics with control (SINDyc-AR) model, derived from the Thermosphere-Ionosphere-Electrodynamics General Circulation Model (TIE-GCM), captures the dominant modes of variability and their dependence on the drivers at a fraction of the parent model's cost. An extended Kalman filter assimilates in situ observations from CHAMP, GRACE, GRACE-FO, GOCE, and Swarm into that model across several orbital configurations and geomagnetic conditions, with a linear DMDc model as reference. Assimilation reduces density error relative to open-loop prediction, most visibly during geomagnetic storms and under single-satellite coverage. The two reduced-order models perform comparably throughout. NRLMSIS 2.1 and HASDM can outperform the assimilated model far from the assimilated track, so results are framed as improvements over the open-loop forecast. An observation-calibrated global density record spanning August 2000 to December 2025 is released alongside this work.
Sources
Related papers
- Industrial Load Modeling and Optimization for Market-Based Interaction with Power Systems
- Exact and Approximate Convex Reformulation of Linear Stochastic Optimal Control with Chance Constraints
- Certified Set Convergence for Piecewise Affine Systems
- Symbolic Control for Stochastic Systems via Finite Parity Games
- Multiobjective optimization-based design and dispatch of islanded, hybrid microgrids for remote, off-grid communities in sub-Saharan Africa
- Physiology-informed layered sensing for intelligent human-exoskeleton interaction