pith. sign in

arxiv: 2512.01279 · v2 · submitted 2025-12-01 · 📊 stat.ME

A Dynamical Model for Spatio-Temporal Processes Motivated by Second-Order Partial Differential Equations

Yutong Zhang , Xiao Liu This is my paper

Pith reviewed 2026-05-17 03:33 UTC · model grok-4.3

classification 📊 stat.ME
keywords spatio-temporal processessecond-order SPDEstate-space modelsGalerkin approximationstochastic differential equationscovariance estimationfinite-dimensional approximation
0
0 comments X

The pith

A finite-dimensional dynamical model is derived from second-order SPDEs through an infinite state-space SDE and Galerkin projection for spatio-temporal processes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a new statistical model for processes that vary across both space and time by starting from second-order stochastic partial differential equations. It first constructs an infinite-dimensional linear state-space form in which the evolution follows a proposed stochastic differential equation. Galerkin's method then projects this onto a finite set of basis functions, producing a practical model with finitely many states. The resulting model supplies an explicit space-time covariance function and bounds the difference from the exact covariance of the original SPDE. Numerical examples on wave equations, advection-diffusion, and aerosol transport illustrate how the finite model supports computation and parameter estimation.

Core claim

An infinite-dimensional linear state-space representation is obtained where the state transition is governed by a proposed SDE motivated by the second-order SPDE. Galerkin's method then yields a finite-dimensional approximation to this SDE, giving a dynamical model with finite states that facilitates computation and parameter estimation. The space-time covariance of the approximated model is derived explicitly, and the error between the approximate and exact covariance matrices is quantified.

What carries the argument

Galerkin projection of the infinite-dimensional SDE representation onto a finite basis to produce a tractable state-space model whose covariance approximates that of the motivating second-order SPDE.

If this is right

  • The finite-state model permits direct maximum-likelihood or Bayesian parameter estimation for spatio-temporal data.
  • Exact expressions for the space-time covariance become available for the approximated process.
  • Quantified covariance error provides a practical guide for choosing the number of basis functions.
  • The framework applies to physical systems governed by wave, advection-diffusion, and similar second-order equations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same projection technique could be tested on boundary conditions or forcing terms that differ from those used in the numerical examples.
  • Embedding the finite model inside a larger hierarchical Bayesian framework might allow joint inference on both the SPDE parameters and the basis functions.
  • If the error bounds remain tight under mesh refinement, the model could serve as a computationally cheap surrogate for uncertainty quantification in large-scale environmental simulations.

Load-bearing premise

The Galerkin finite-dimensional approximation preserves the key dynamics and covariance structure of the original second-order SPDE without introducing large uncontrolled errors in practical regimes.

What would settle it

A simulation or real-data experiment in which the covariance error between the finite model and a high-resolution reference solution grows without bound or in which parameter estimates obtained from the finite model deviate substantially from those obtained by direct discretization of the SPDE.

Figures

Figures reproduced from arXiv: 2512.01279 by Xiao Liu, Yutong Zhang.

Figure 1
Figure 1. Figure 1: GOES-18 ABI Level-2 AOD distribution on January 9, 2025, with the red dot [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The proposed approach for obtaining the dynamical model representation of [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A simulated advection-diffusion process with a change in the source term at [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Filtered processes from the proposed (above) and the conventional (below) models, [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Filtered αk(t) for selected spatial frequencies from the proposed approach. Some important observations are obtained from Figures 5 and 6: ⋄ It is ineffective, if not impossible, to detect the change of the source term by simply looking at the trend of the estimated overall mean (i.e., at the wavenumber k = (0, 0)). For k = (0, 0), the coefficient αk=(0,0)(t) is exactly the filtered overall spatial mean of… view at source ↗
Figure 6
Figure 6. Figure 6: Filtered βk(t) for selected spatial frequencies from the proposed approach. αk(t) at time step 20 indicates the change; see [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Filtered αk(t) for selected spatial frequencies from the existing approach [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Filtered βk(t) for selected spatial frequencies from the existing approach. 3.2 Example-II: Application to Remote-Sensing AOD Data In Example-II, we re-visit the motivating example presented in Section 1.1, and apply the proposed approach to model the remote-sensing AOD process and detect spatial areas with rapid AOD changes during the 2025 Los Angeles Kenneth Fire episode. The fire occurred in January 202… view at source ↗
Figure 9
Figure 9. Figure 9: Ten evenly spaced (every 20 minutes) snapshots of AOD data streams from the [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Filtered processes from the proposed model [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Filtered processes from the conventional models [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Estimated first-order time derivative of the process from the proposed model [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Estimated growth-decay from the conventional model [PITH_FULL_IMAGE:figures/full_fig_p021_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Comparison of αk(t) across some selected wavenumbers, (1, 0),(1, 1),(1, 2),(1, 3). The vertical dashed line marks 2:07 PM which indicates the time when the wildfire started [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Comparison of βk(t) at the selected wavenumber, (0, 0). The vertical dashed line (2:07 PM ) indicates the onset of the Kenneth fire which grew rapidly by 5:30 PM PST fanned by strong Santa Ana winds the reported onset of the Kenneth fire near the border of Los Angeles and Ventura counties. More interestingly, the filtered βk(t) from the proposed model shows a clear late afternoon surge (after 4:00 PM ) in… view at source ↗
read the original abstract

An important class of spatio-temporal models is constructed by leveraging the hierarchical structure of dynamical (or, state-space) models. This paper proposes a new statistical dynamical model for spatio-temporal processes motivated by second-order stochastic partial differential equations (SPDE). In particular, an infinite-dimensional linear state-space representation is obtained where the state transition is governed by a proposed SDE. Then, using the Galerkin's method, a finite-dimensional approximation to the infinite-dimensional SDE is obtained, yielding a dynamical model with finite states that facilitates computation and parameter estimation. The space-time covariance of the approximated dynamical model is obtained, and the error between the approximate and exact covariance matrices is quantified. Comprehensive numerical investigations, including 2D wave equation, seismic wave propagation, advection-diffusion equations and wildfire aerosol propagation processes, are performed to demonstrate the application of the proposed model. Code is available.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes a new statistical dynamical model for spatio-temporal processes motivated by second-order stochastic partial differential equations (SPDEs). It constructs an infinite-dimensional linear state-space representation in which the state transition is governed by a proposed SDE (via an auxiliary velocity-like component), applies Galerkin's method to obtain a finite-dimensional approximation, derives the space-time covariance of the resulting model, quantifies the error between the approximate and exact covariance matrices, and validates the approach through numerical experiments on the 2D wave equation, seismic wave propagation, advection-diffusion equations, and wildfire aerosol propagation, with accompanying code.

Significance. If the error bounds hold and the finite-dimensional approximation faithfully retains the covariance structure of the original second-order SPDE, the framework would supply a computationally tractable, physically motivated state-space model that facilitates parameter estimation and prediction for complex spatio-temporal data. The provision of code is a clear strength that supports reproducibility of the reported numerical results.

major comments (3)
  1. [Derivation of infinite-dimensional state-space representation] The derivation of the infinite-dimensional linear state-space representation (via the proposed SDE and auxiliary component) is load-bearing for the subsequent Galerkin step and error analysis; the manuscript must explicitly show how the second-order time derivative is converted while preserving the original SPDE's noise structure and boundary conditions, because any mismatch directly affects the claimed covariance error quantification.
  2. [Galerkin approximation and covariance error quantification] The error quantification between approximate and exact covariance matrices assumes that the Galerkin projection commutes with the second-order time operator and preserves the correct inner-product structure. For non-self-adjoint spatial operators (advection term in advection-diffusion and seismic examples) or processes with sharp gradients (wildfire aerosol), standard polynomial or Fourier bases can introduce spurious dispersion or damping that grows with time, potentially violating the stated error bounds outside the specific numerical setups shown.
  3. [Numerical investigations] In the numerical investigations section, the reported approximation errors for the wildfire aerosol and advection-diffusion cases should be accompanied by explicit comparisons to the derived theoretical bounds; without this, it is difficult to confirm that the finite-state model remains useful for parameter estimation when boundary conditions or sharp features are present.
minor comments (2)
  1. [Abstract] The abstract would benefit from a brief mention of the form of the error bound or the typical magnitude observed in the experiments.
  2. [Throughout the derivation sections] Notation for the infinite-dimensional operators and the finite basis should be introduced with consistent symbols and referenced in every subsequent equation that uses them.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their insightful comments, which have helped us improve the clarity and rigor of our manuscript. We address each of the major comments in detail below.

read point-by-point responses

  1. Referee: [Derivation of infinite-dimensional state-space representation] The derivation of the infinite-dimensional linear state-space representation (via the proposed SDE and auxiliary component) is load-bearing for the subsequent Galerkin step and error analysis; the manuscript must explicitly show how the second-order time derivative is converted while preserving the original SPDE's noise structure and boundary conditions, because any mismatch directly affects the claimed covariance error quantification.

    Authors: We agree with the referee that the derivation is central to our framework. In the revised manuscript, we will provide a more detailed, step-by-step exposition in Section 2.1, explicitly demonstrating the introduction of the auxiliary velocity component to convert the second-order SPDE into a first-order system. We will show that the noise term remains unchanged in distribution and discuss how boundary conditions are incorporated through the choice of the function space and the projection. This expansion will ensure transparency regarding the preservation of the original SPDE properties. revision: yes

  2. Referee: [Galerkin approximation and covariance error quantification] The error quantification between approximate and exact covariance matrices assumes that the Galerkin projection commutes with the second-order time operator and preserves the correct inner-product structure. For non-self-adjoint spatial operators (advection term in advection-diffusion and seismic examples) or processes with sharp gradients (wildfire aerosol), standard polynomial or Fourier bases can introduce spurious dispersion or damping that grows with time, potentially violating the stated error bounds outside the specific numerical setups shown.

    Authors: The referee raises an important point about the applicability to non-self-adjoint operators. Our error analysis in Theorem 3.2 relies on the properties of the chosen basis functions and the projection operator. In the examples involving advection-diffusion and seismic waves, we selected bases that are appropriate for the problem, and the numerical results indicate that the errors remain controlled. To address this concern, we will include in the revised paper a discussion in Section 3 on the conditions under which the commutation holds and potential remedies such as using operator-adapted bases for cases with strong advection or sharp gradients. We believe this will clarify the scope of the error bounds. revision: partial

  3. Referee: [Numerical investigations] In the numerical investigations section, the reported approximation errors for the wildfire aerosol and advection-diffusion cases should be accompanied by explicit comparisons to the derived theoretical bounds; without this, it is difficult to confirm that the finite-state model remains useful for parameter estimation when boundary conditions or sharp features are present.

    Authors: We appreciate this recommendation for strengthening the numerical validation. In the updated Section 4, we will add explicit comparisons of the observed approximation errors in the wildfire aerosol propagation and advection-diffusion examples against the theoretical error bounds derived in Section 3. This will include quantitative assessments and visualizations to demonstrate that the errors are consistent with the bounds, supporting the model's utility for parameter estimation in these challenging scenarios. revision: yes

Circularity Check

0 steps flagged

Derivation from second-order SPDE via auxiliary velocity state and Galerkin projection is self-contained and independent of fitted parameters

full rationale

The paper constructs an infinite-dimensional linear state-space model by introducing an auxiliary component to convert the second-order SPDE into a first-order system, proposes an SDE for the state transition, and applies the standard Galerkin method to obtain a finite-dimensional approximation whose covariance error is quantified from projection properties. None of the load-bearing steps reduce to self-definition, fitted inputs renamed as predictions, or self-citation chains; the approach relies on established numerical techniques for SPDEs without importing uniqueness theorems or ansatzes from the authors' prior work. The central claim therefore remains independent of its own outputs and is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Based solely on the abstract, the model introduces a proposed SDE whose specific form is not detailed and relies on Galerkin projection whose convergence assumptions are standard but unstated here; no explicit free parameters or invented entities are enumerated.

axioms (1)
  • domain assumption Galerkin method yields a convergent finite-dimensional approximation to the infinite-dimensional SDE
    Invoked when moving from infinite to finite states; standard in numerical PDE literature but treated as given.
invented entities (1)
  • Proposed SDE for state transition no independent evidence
    purpose: Governs the evolution in the infinite-dimensional linear state-space representation
    Introduced as the core of the new dynamical model; no independent evidence provided in abstract.

pith-pipeline@v0.9.0 · 5445 in / 1428 out tokens · 48980 ms · 2026-05-17T03:33:07.024198+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages · 1 internal anchor

  1. [1]

    and Farhat, C

    Amsallem, D. and Farhat, C. (2014). On the stability of reduced-order linearized compu- tational fluid dynamics models based on pod and galerkin projection: descriptor vs non- descriptor forms. InReduced order methods for modeling and computational reduction, pages 215–233. Springer

    work page 2014

  2. [2]

    Anderson, B. D. (1982). Reverse-time diffusion equation models.Stochastic Processes and their Applications, 12(3):313–326

    work page 1982

  3. [3]

    Benner, P., Gugercin, S., and Willcox, K. (2015). A survey of projection-based model reduction methods for parametric dynamical systems.SIAM review, 57(4):483–531

    work page 2015

  4. [4]

    E., K., K

    Brown, P. E., K., K. F., Gareth, O. R., and Stefano, T. (2000). Blur-generated non-separable space-time models.Journal of the Royal Statistical Society. Series B (Statistical Method- ology), 62(4):847–860

    work page 2000

  5. [5]

    Clarotto, L., Allard, D., Romary, T., and Desassis, N. (2024). The spde approach for spatio- temporal datasets with advection and diffusion.Spatial Statistics, 62:100847

    work page 2024

  6. [6]

    and Wikle, C

    Cressie, N. and Wikle, C. (2011).Statistics for Spatio-Temporal Data. John Wiley & Sons

    work page 2011

  7. [7]

    Deng, X., Kang, L., and Lin, C. D. (2025). Design of experiments for emulations: A selective review from a modeling perspective.arXiv preprint arXiv:2505.09596

    work page arXiv 2025

  8. [8]

    Gneiting, T. (2002). Nonseparable, stationary covariance functions for space–time data. Journal of the american statistical association, 97(458):590–600. 23

    work page 2002

  9. [9]

    Gramacy, R. B. (2020).Surrogates: Gaussian process modeling, design, and optimization for the applied sciences. Chapman and Hall/CRC

    work page 2020

  10. [10]

    and S¨ arkk¨ a, S

    Hartikainen, J. and S¨ arkk¨ a, S. (2010). Kalman filtering and smoothing solutions to temporal Gaussian process regression models. In2010 IEEE International Workshop on Machine Learning for Signal Processing (MLSP), pages 379–384. IEEE

    work page 2010

  11. [11]

    Hector, E. C. and Reich, B. J. (2024). Distributed inference for spatial extremes modeling in high dimensions.Journal of the American Statistical Association, 119(546):1297–1308

    work page 2024

  12. [12]

    N., Eck, T

    Holben, B. N., Eck, T. F., Slutsker, I., and et al. (1998). AERONET—a federated instrument network and data archive for aerosol characterization.Remote Sensing of Environment, 66(1):1–16

    work page 1998

  13. [13]

    Hu, Z., Cao, Q., Kawaguchi, K., and Karniadakis, G. E. (2025). Deepomamba: State-space model for spatio-temporal pde neural operator learning.Available at SSRN 5149007. IPCC (2021). Annex vii: Glossary. In Zhai, P. e. a., editor,Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the IPCC, pag...

    work page 2025

  14. [14]

    and Sch¨ afer, F

    Katzfuss, M. and Sch¨ afer, F. (2024). Scalable bayesian transport maps for high- dimensional non-gaussian spatial fields.Journal of the American Statistical Association, 119(546):1409–1423

    work page 2024

  15. [15]

    Lan, S., Pasha, M., Li, S., and Shen, W. (2025). Spatiotemporal besov priors for bayesian inverse problems.Journal of the American Statistical Association, (just-accepted):1–30

    work page 2025

  16. [16]

    Li, Z., Yang, S., and Wu, C. J. (2024). Parameter inference based on gaussian processes informed by nonlinear partial differential equations.SIAM/ASA Journal on Uncertainty Quantification, 12(3):964–1004

    work page 2024

  17. [17]

    Lin, Z., Li, Y., Yin, F., Maro˜ nas, J., and Thi´ ery, A. H. (2025). Efficient transformed gaussian process state-space models for non-stationary high-dimensional dynamical systems.arXiv preprint arXiv:2503.18309

    work page arXiv 2025

  18. [18]

    Lindgren, F., Rue, H., and Lindstr¨ om, J. (2011). An explicit link between gaussian fields and gaussian markov random fields: the stochastic partial differential equation approach. Journal of the Royal Statistical Society Series B: Statistical Methodology, 73(4):423–498

    work page 2011

  19. [19]

    Liu, X., Yeo, K., Hwang, Y., Singh, J., and Kalagnanam, J. (2016). A statistical modeling approach for air quality data based on physical dispersion processes and its application to ozone modeling.The Annals of Applied Statistics, pages 756–785

    work page 2016

  20. [20]

    M., and Lu, S

    Liu, X., Yeo, K. M., and Lu, S. Y. (2022). Statistical modeling for spatio-temporal data from stochastic convection-diffusion processes.Journal of the American Statistical Association (Theory and Methods), 117:1482–1499. 24

    work page 2022

  21. [21]

    Morris, M. D. (2012). Gaussian surrogates for computer models with time-varying inputs and outputs.Technometrics, 54(1):42–50. Pf¨ ortner, M., Steinwart, I., Hennig, P., and Wenger, J. (2022). Physics-informed gaussian process regression generalizes linear pde solvers.arXiv preprint arXiv:2212.12474

    work page arXiv 2012

  22. [22]

    Polonelli, T., Brunelli, D., Bartolini, A., and Benini, L. (2019). A lorawan wireless sen- sor network for data center temperature monitoring. arXiv preprint arXiv:1902.09400. High-resolution, battery-operated wireless sensors deployed continuously for six months to detect hotspots and optimize cooling

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  23. [23]

    Quick, H., Banerjee, S., and Carlin, B. P. (2015). Bayesian modeling and analysis for gradients in spatiotemporal processes.Biometrics, 71(3):575–584

    work page 2015

  24. [24]

    A., Kaufman, Y

    Remer, L. A., Kaufman, Y. J., Tanr´ e, D., and et al. (2005). The modis aerosol algorithm, products, and validation.Journal of the Atmospheric Sciences, 62(4):947–973

    work page 2005

  25. [25]

    Richardson, R., Kottas, A., and Sans´ o, B. (2020). Spatiotemporal modelling using integro- difference equations with bivariate stable kernels.Journal of the Royal Statistical Society Series B: Statistical Methodology, 82(5):1371–1392

    work page 2020

  26. [26]

    and Hartikainen, J

    Sarkka, S. and Hartikainen, J. (2012). Infite-dimensional kalman filtering approach to spatio- temporal gaussian process regression.International Conference on Artificial Intelligence and Statistics (AISTATS)

    work page 2012

  27. [27]

    Sigrist, F., Kunsch, H., and Stahel, W. (2015). Stochastic partial differential equation basedmodelling of large space–time data sets.Journal of the Royal Statistical Society, Series B, 77:3–33

    work page 2015

  28. [28]

    Wang, X., Liu, X., Tsung, F., and Yeo, K. (2022). Statistical monitoring of spatio-temporal processes with application to warehouse temperature management.Journal of Quality Technology, 54(2):215–235

    work page 2022

  29. [29]

    Wei, G., Liu, X., and Barton, R. (2024). An extended pde-based statistical spatio-temporal model that suppresses the gibbs phenomenon.Environmetrics, 35(2):e2831

    work page 2024

  30. [30]

    Whittle, P. (1954). On stationary processes in the plane.Biometrika, pages 434–449

    work page 1954

  31. [31]

    Wikle, C. K. and Hooten, M. B. (2010). A general science-based framework for dynamical spatio-temporal models.Test, 19(3):417–451

    work page 2010

  32. [32]

    Zhang, Q., Chien, P., Liu, Q., Xu, L., and Hong, Y. (2021). Mixed-input gaussian process emulators for computer experiments with a large number of categorical levels.Journal of Quality Technology, 53(4):410–420

    work page 2021

  33. [33]

    Zhu, S. X. and Xie, Y. (2022). Spatiotemporal-textual point processes for crime linkage detection.The Annals of Applied Statistics, pages 1151–1170. 25

    work page 2022