Pith. sign in

REVIEW 5 major objections 7 minor 32 references

Hybrid SINDy-EnKF in Learning Chikungunya Dynamics from Incomplete, Noisy or Partially Observed Data

T0 review · 5 major / 7 minor · reviewed 2026-07-30 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Pairing sparse equation discovery with ensemble filtering recovers Chikungunya host–vector trajectories from noisy, partial surveillance-style data.

desk verdict Useful synthetic demo of SINDy-as-forecast plus EnKF on a 10-compartment CHIKV model, but the clean-data “recovery” claim does not match Appendix C and Table 6 partly credits the filter for fixing a bad open-loop model. read the letter →

arxiv 2607.27137 v1 pith:WIDZUCBE submitted 2026-07-29 q-bio.PE

Bernard Asamoah Afful , Changhong Mou , Luis Gordillo This is my paper · ORCID
classification q-bio.PE
keywords ChikungunyavirusSINDyEnsembleKalmanFilterdataassimilationinfectiousdiseasedynamicspartialobservabilitycomputationalepidemiology
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Mechanistic models of Chikungunya spread need parameters and full compartment data that real surveillance rarely supplies. This paper shows that sparse identification of nonlinear dynamics (SINDy) can rediscover the governing mass-action equations from clean epidemic trajectories, but breaks down once observational noise is added, inventing spurious terms and distorting forecasts. Embedding the SINDy-derived model as the forecast operator inside an Ensemble Kalman Filter lets noisy partial observations correct the state and fill in unobserved compartments, including mosquito states never reported. In synthetic tests with only a few host compartments observed at 25% noise, the hybrid cut root-mean-square error by more than 98% for most states relative to noisy SINDy alone. The practical stake is a data-driven path to interpretable epidemic models that still work when surveillance is incomplete and noisy.

What carries the argument

Hybrid SINDy–EnKF: a sparsely identified bilinear ODE model serves as the EnKF forecast operator, while the filter’s ensemble covariance assimilates noisy partial observations and transfers information to unobserved compartments.

What would settle it

Apply the full pipeline to a held-out real CHIKV surveillance series (only the reported host compartments, with realistic noise and gaps) and check whether unobserved states and short-horizon forecasts stay within the error levels claimed on synthetic 25% noise data; systematic failure or non-physical reconstructed mosquito trajectories would falsify the central claim.

Watch

Extended reading notes

Core claim

Standalone SINDy recovers the CHIKV compartmental equations from noise-free data but fails under moderate-to-high noise; coupling that same SINDy model to an Ensemble Kalman Filter restores accurate trajectories for both observed and unobserved host and vector compartments under partial, noisy observations, with reported RMSE reductions above 98% for most of the ten states in the 25%-noise experiments.

Load-bearing premise

The method assumes that a hand-tuned sparsity threshold and a candidate library built from the same mass-action structure that generated the synthetic data produce a forecast model good enough for the filter—without real outbreak data or an automatic threshold rule.

Editorial extensions

If this is right

  • Interpretable CHIKV models can be learned and corrected online from partial noisy surveillance rather than fully specified a priori parameters.
  • Unobserved mosquito compartments can be inferred from host observations when host–vector couplings are retained in the forecast model.
  • Observation frequency bounds filter skill: high-frequency assimilation keeps NRMSE low; long gaps let SINDy model error dominate.
  • The same hybrid pattern extends, on the authors’ account, to weak-form or ensemble SINDy, Neural-ODE/PINN residuals, and joint state–parameter updates.
  • Outbreak monitoring and short-term forecasting become feasible targets once the framework is validated on real outbreak records.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If compartment-wise or scale-weighted sparsity thresholds replace a single global λ, the standalone SINDy failure mode on large vector populations may shrink before filtering is applied.
  • The same cross-covariance mechanism should transfer to other arboviruses with analogous host–vector mass-action structure (e.g., dengue, Zika) without redesigning the library class.
  • Real underreporting is often multiplicative and delayed, not additive Gaussian; testing that noise model is a natural next falsification step the paper leaves open.
  • Online updating of SINDy coefficients inside the filter would blur offline discovery and assimilation, potentially reducing dependence on the initial noisy identification.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 7 minor

Summary. The manuscript formulates a 10-compartment host–vector ODE model for chikungunya (2.4), then studies a pipeline in which SINDy with a constant/linear/bilinear library learns a surrogate from simulated trajectories at noise levels 0–50%, and the discovered system serves as the forecast operator of an EnKF that assimilates noisy observations of four host compartments (Vh, Ih, Jh, Th) to reconstruct the full state. Claims: (i) SINDy recovers (2.4) exactly from clean data; (ii) it degrades gracefully then badly with noise; (iii) the SINDy–EnKF hybrid reduces RMSE by >98% in most compartments (Table 6) and reconstructs unobserved states, with skill controlled by the assimilation interval dtobs (Fig. 7). All experiments are synthetic, generated from (2.4) with known observation operators. The pipeline is not definitionally circular, but the clean-data recovery claim is contradicted by the manuscript's own Appendix C, and the headline Table 6 numbers are computed against an open-loop baseline that any filter with frequent direct observations would beat.

Significance. A working recipe coupling sparse model discovery with sequential assimilation would be useful for compartmental epidemiology, where partial and noisy observation is the norm. The paper's strengths are a systematic noise-sensitivity sweep for SINDy on a realistic 10-compartment host–vector model (§3.2), an explicit dtobs-sensitivity analysis showing where the forecast model actually matters (Fig. 7), and publicly available code. However, all evidence is synthetic (self-generated RK4 data), the clean-recovery claim is internally contradicted, and the headline assimilation numbers lack a null-model baseline; as written, the contribution is a proof of concept rather than the surveillance-ready method the abstract suggests.

major comments (5)
  1. [§3.1 and Appendix C (Table 7)] The claim that SINDy 'reproduces both the structural form and the parameter values of the original model (2.4) to within numerical precision' is contradicted by the manuscript's own appendix. The library (constant, linear, pairwise bilinear) cannot represent the rational forces of infection in (2.4), lambda_h = beta_v I_v/N_v and lambda_v = beta_h(eps_1 I_h+eps_2 J_h+eps_3 T_h)/N_h, with state-dependent N_v, N_h. Appendix C's 'Clean data' column confirms non-recovery: dV_h = 0.0000 and dT_h = 0.0000 identically, no I_v-dependent term in dS_h or dE_h, and spurious quadratics (e.g., +0.0133 E_h^2). Moreover, dV_h = 0 is inconsistent with Table 4's clean-data RMSE of 0.0001 for V_h, since the true dV_h ~ theta*S_h ~ O(10^2)/day with theta=0.4. Either the discovered-equation table is mislabeled or the RMSE table is. This must be resolved and the recovery claim corrected; the authors cite S S
  2. [§3.3 and Table 6] The headline >98% RMSE reduction is measured against open-loop integration of a badly misspecified 25%-noise SINDy model (Appendix C: dS_v ~ 12.03 S_h - 48.75 E_h - 289.6 S_h I_h + ...). With four channels directly observed at dtobs <= 1, essentially any propagator is pinned near the truth, so this number cannot distinguish 'SINDy-EnKF hybrid works' from 'EnKF with frequent partial observations pins the state down.' The I_h row already illustrates the metric's fragility (-279% 'reduction'). A control is needed: e.g., EnKF with a persistence/random-walk forecast or a deliberately wrong linear model at the same dtobs, plus reporting Table 6 across dtobs. Figure 7 suggests the forecast model genuinely matters only for dtobs >= 5 — quantify this rather than relying on the high-frequency regime.
  3. [Tables 4–5 captions; §3.2.1] The STLSQ threshold is hand-tuned per noise level, and the tuning is non-monotone (20% noise: lambda=1e-6, smaller than the 5% value 2e-5 and the 10% value 2.8e-5), which suggests selection against ground-truth error. In real surveillance no ground truth is available, so the pipeline that produces the forecast operator behind Table 6 currently relies on oracle information. A data-only selection rule (held-out trajectory cross-validation, Pareto-front selection, or ensemble-SINDy stability as in ref. [10]) should be specified and the main results shown to survive it.
  4. [§3.2 setup; §3.3.1; Table 3] Two load-bearing methodological gaps. (i) The derivative-estimation method for SINDy is never described: with dt = 0.0001 (Table 3) and 25–50% multiplicative noise, raw finite differences would produce enormous derivative noise, so any smoothing (Savitzky–Golay, weak form, etc.) must be stated. (ii) The EnKF configuration is unspecified: ensemble size N_e, model-error covariance Q_n in (2.16), calibration of R_o, and any inflation/localization are all absent, yet these control the results in §3.3. Without these details the noise-robustness and assimilation conclusions cannot be reproduced or independently assessed; the GitHub repository should also be cited with a URL.
  5. [Abstract; §3; §4] All validation is on synthetic RK4 trajectories generated from (2.4) itself, with a known linear observation operator and Gaussian noise. This is appropriate for a method stress test, but the abstract's framing around 'a common constraint in real-world epidemiological surveillance' overstates what is demonstrated. Either temper the abstract/conclusion to a proof-of-concept claim, or add at least one experiment with realistic surveillance features — underreporting, reporting delays, non-Gaussian or aggregated (weekly) counts, or a historical CHIKV outbreak dataset.
minor comments (7)
  1. [Supporting information / Appendices B–C] The Supporting Information lists 'Appendix B. Discovered SINDy equations; Appendix C. Numerical results ... partial and noisy data,' but the actual layout has them reversed (Appendix B contains the partial-data figure, Appendix C the equation table). Please fix the cross-referencing.
  2. [§2.3, Eq. (2.24)] After Eq. (2.24): 'where the observation operator G is assumed to be linear' — G is undefined; presumably H. Also, u^k_{m+1|m}, k=1,...,K for an ensemble of J members clashes with K already denoting the state dimension.
  3. [§3.3.1] The observed subset is described as 'vaccinated (V_h), infectious (I_h), treated (J_h), and recovered-under-treatment (T_h),' but per Table 1 J_h is asymptomatic infectious and T_h is treated. Please correct the labels.
  4. [Figure 7 caption] It is unclear what the plotted curves represent (per-compartment NRMSE/correlation? averages over host and vector groups?) — no legend or per-compartment identification is given.
  5. [Table 3] 'eps 0.2' appears without subscript — state whether eps_1 = eps_2 = eps_3 = 0.2, which sits oddly with the text's description of eps_i as distinct reduction factors.
  6. [Figures 8–10] Several panels have unlabeled y-axes and awkwardly placed 10^4 multipliers; row/column organization is hard to parse.
  7. [§3.2; Eq. (2.7)] 'we systematically add clean trajectories with state-dependent Gaussian noise' — grammatical error (presumably 'add ... to'). In Eq. (2.7), the state matrix is written with rows x_1^T,...,x_K^T although there are m time points and K states — the indexing is inconsistent.

Circularity Check

1 steps flagged · score 1.0 of 10

No load-bearing circularity: twin-experiment validation (ODE → SINDy surrogate → EnKF vs held-out truth) is self-contained; mild threshold peeking does not force the central claim.

  1. fitted input called prediction [§3.2.1 Noise model and threshold selection; captions of Tables 4–5]
    "The sparsity threshold λ̃ in (2.15) is tuned for each noise level to balance two competing risks: thresholds that are too small admit noise-driven spurious terms, whereas thresholds that are too large eliminate genuine physical terms. The values used are reported in the captions of Tables 4 and 5. Where λ = 0.0000000001 for clean data, λ = 0.00002 for 5% noise, ... λ = 0.0006 for 25% noise, ... and λ = 0.005 for 50% noise."

    λ is chosen per noise level with access to how well the discovered model matches the known true trajectories (component-wise and relative RMSE in Tables 4–5). That is ground-truth-informed hyperparameter fitting for the forecast operator later used in the EnKF experiments. It is only mild: the Table 6 / Figs. 8–10 claims are still scored on sequential assimilation of partial observations against held-out full state, not on re-reporting the fitted λ or in-sample regression residual as a ‘prediction.’

full rationale

The paper’s derivation chain is a standard synthetic twin experiment, not a closed definitional loop. Trajectories are generated from the known CHIKV ODE (2.4); SINDy builds a sparse surrogate from a fixed bilinear library; EnKF assimilates independent noisy partial observations and is scored against held-out true states via RMSE/NRMSE/correlation (Eqs. 3.1–3.2, Table 6, Figs. 7–10). Success is not forced by construction: a bad propagator can still fail (and the Ih row of Table 6 even shows EnKF increasing error). There is no self-citation uniqueness theorem, no renaming of a known empirical law, and no claim that a fitted scalar is itself the reported prediction. The only mild circularity-adjacent practice is hand-tuning of the STLSQ threshold λ per noise level using knowledge of reconstruction error against truth (captions of Tables 4–5). That peeks at ground truth for model selection, but it does not make the EnKF state-reconstruction metrics tautological—the filter still has to track unobserved compartments from partial observations. Library choice matched to mass-action structure is a modeling assumption (and may be inadequate for the rational incidence terms in (2.4)), which is a correctness risk, not circularity under the stated rules. Score 1 reflects that single minor threshold-tuning issue; central hybrid claim remains independently testable within the paper’s own protocol.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The central hybrid claim rests on standard ODE epidemic modeling and standard SINDy/EnKF algorithms, plus several paper-specific modeling and tuning choices: a fixed bilinear library, per-noise sparsity thresholds, synthetic truth from one parameterized CHIKV system, and a chosen partial observation set. No new physical entities are postulated; free parameters are the usual epidemiological rates plus method knobs (λ, ensemble settings, Ro).

free parameters (5)
  • STLSQ sparsity threshold λ (per noise level) = e.g. 1e-10 clean; 2e-5 (5%); 2.8e-5 (10%); 1e-6 (20%); 6e-4 (25–30%); 1e-3 (40%); 5e-3 (50%)
    Hand-chosen differently for clean and each ε in {5…50}% (captions of Tables 4–5); controls which terms survive and is load-bearing for both standalone SINDy quality and the forecast model fed to EnKF.
  • CHIKV mechanistic parameters (Table 3) = Fixed table values (e.g. βh=0.5, βv=0.4, ρ=0.7, θ=0.4, η=0.7)
    Λh, Λv, βh, βv, σh, σv, γs, γa, γT, τ, θ, η, ω, μh, μv, δh, ρ, ϵ, etc. define the synthetic truth and forces of infection; not inferred from real data in this study.
  • Relative noise amplitude ε and Ro = ε up to 50%; Ro “calibrated to a representative noise level”
    Observation noise scale in Y = X + ε σ(X) N(0,I) and EnKF observation covariance; set by experiment design, not estimated from field data.
  • Observation interval dtobs and observed subset = dtobs swept in [0.1, 10]; four host compartments observed
    Which compartments are seen (Vh, Ih, Jh, Th) and how often assimilation occurs strongly affect NRMSE/correlation (Fig. 7); design choices, not learned.
  • Ensemble size Ne and initial ensemble perturbation
    EnKF Monte Carlo approximation quality depends on Ne and prior spread; mentioned as Ne members but not numerically fixed in the extracted experimental text.
assumptions (6)
  • domain assumption Mass-action compartmental ODE structure (2.4) with forces λh=βv Iv/Nv and λv=βh(ϵ1 Ih+ϵ2 Jh+ϵ3 Th)/Nh adequately represents CHIKV host–vector transmission for the study.
    Section 2.1 builds the entire synthetic truth and the SINDy library around this structure.
  • domain assumption f admits a sparse representation in a library of constants, linear terms, and pairwise bilinear products (SINDy hypothesis).
    Section 2.2.1; standard SINDy modeling assumption, here aligned with mass-action.
  • domain assumption EnKF with sample covariances and (possibly nonlinear) observation ensemble updates yields a usable sequential estimator for this nonlinear ODE (Gaussian-ish error heuristics).
    Section 2.3 state-space form (2.16)–(2.26); classical EnKF working assumptions.
  • domain assumption Noisy measurements are zero-mean white (state-dependent Gaussian in experiments) with known/usable Ro.
    Eqs. (2.11), (2.17) and §3.2.1 noise model.
  • standard math STLSQ with a global threshold sufficiently approximates the ℓ0-regularized sparse regression (2.10)/(2.15).
    Section 2.2; standard algorithmic stand-in for sparse identification.
  • ad hoc to paper Synthetic full-state trajectories plus known H are an adequate proxy for “real-world epidemiological surveillance” constraints when arguing reconstruction skill.
    Abstract and §3 claim relevance to surveillance while Availability states no real-world data were used.
invented entities (1)
  • Hybrid SINDy–EnKF forecast–assimilate loop for CHIKV (SINDy model as An inside EnKF)
    purpose: Use sparse discovered ODEs as the EnKF propagator to filter noise and infer unobserved compartments.
    Methodological coupling emphasized in abstract and Fig. 2; components preexist, the packaged epi workflow is the paper’s construct.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hybrid SINDy-EnKF in Learning Chikungunya Dynamics from Incomplete, Noisy or Partially Observed Data." pith.science (2026). https://pith.science/paper/WIDZUCBE

@misc{pith2026260727137,
  author       = {Pith},
  title        = {Pith review of: Hybrid SINDy-EnKF in Learning Chikungunya Dynamics from Incomplete, Noisy or Partially Observed Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WIDZUCBE}},
  note         = {Machine review of arXiv:2607.27137}
}
read the original abstract

Current mechanistic models for the transmission dynamics of the Chikungunya virus (CHIKV) rely on uncertain parameters or partially observed data. This limitation challenges the use of theoretical models for understanding and forecasting disease spread. Here we present a hybrid, data-driven model framework that combines Sparse Identification of Nonlinear Dynamics (SINDy) with the Ensemble Kalman Filter (EnKF) for sequential data assimilation. Our numerical experiments show that this approach improves prediction accuracy and provides a good reconstruction of unobserved trajectories under partial observability, a common constraint in real-world epidemiological surveillance. SINDy can be applied to epidemic trajectories, recovering the underlying equations in noise-free conditions. However, standalone SINDy is highly sensitive to noise, leading to spurious terms and poor performance. Hence, we embed the identification procedure within an EnKF framework, which assimilates noisy observations to correct forecast states from the SINDy-derived model and to infer unobserved state variables.

Figures

Figures reproduced from arXiv: 2607.27137 by the authors.

Figure 1
Figure 1. Illustration of the Chikungunya virus (CHIKV) model. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Illustration of learning chikungunya transmission dynamics proposed in this paper: a sparse [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Time evolution of the host and vector compartments obtained by numerically solving the original [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Numerical simulation of the SINDy-identified CHIKV model using noisy state measurements. The [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: Numerical simulation of the SINDy-identified CHIKV model using noisy state measurements. The [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Numerical simulation of the SINDy-identified CHIKV model using noisy state measurements. The [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Sensitivity of Ensemble Kalman Filter to the observational time interval [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: EnKF reconstruction of observed host compartments. Comparison of the 25% noisy SINDy trajectory, forecast prior mean, true clean trajectory, filtered posterior mean, and observations for the observed compartments Vh, Ih, Jh, and Th. 18 [PITH_FULL_IMAGE:figures/full_fi…
Figure 9
Figure 9. Figure 9: EnKF reconstruction of host compartments. Comparison of the 25% noisy SINDy trajectory, forecast prior mean, true clean trajectory, and filtered posterior mean for all host compartments. The EnKF corrects the noisy SINDy forecast and closely tracks the clean reference …
Figure 10
Figure 10. Figure 10: EnKF reconstruction of vector compartments. Comparison of the 25% noisy SINDy trajectory, forecast prior mean, true clean trajectory, and filtered posterior mean for the vector compartments Sv, Ev, and Iv. The filtered estimates remain close to the true clean trajecto…
Figure 11
Figure 11. Figure 11: Numerical simulations show the trajectories obtained when SINDy is trained on only the first [PITH_FULL_IMAGE:figures/full_fig_p025_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 1 linked inside Pith

  1. [10]

    Ensemble-sindy.Proceedings: Mathematical, Physical and Engineering Sciences, 478(2260):1–20, 2022

    U Fasel, JN Kutz, BW Brunton, and SL Brunton. Ensemble-sindy.Proceedings: Mathematical, Physical and Engineering Sciences, 478(2260):1–20, 2022

    2022

  2. [1]

    Joint estimation of hand-foot-mouth disease model and prediction in korea using the ensemble kalman filter

    Wasim Abbas, Sieun Lee, and Sangil Kim. Joint estimation of hand-foot-mouth disease model and prediction in korea using the ensemble kalman filter. PLOS Computational Biology , 21(4):e1012996, 2025

    2025

  3. [2]

    Machine learning- based new sparse algorithms and reinforcement learning for computational fluid dynamics

    Hayder Alrazen, Izzat Najmi Yaacob, Norkhairunnisa Mazlan, and Noorfaizal Yidris. Machine learning- based new sparse algorithms and reinforcement learning for computational fluid dynamics. InArtificial Intelligence for Computational Fluid Dynamics , pages 313–343. Elsevier, 2026

    2026

  4. [3]

    Automated model discovery based on covid-19 epidemiologic data.medRxiv, pages 2026–02, 2026

    Morteza Babazadeh Shareh, Florian Kleiner, Michael Böhme, Corinna Hägele, Petra Dickmann, and Rainer Heintzmann. Automated model discovery based on covid-19 epidemiologic data.medRxiv, pages 2026–02, 2026

    2026

  5. [4]

    Parameter estimation in a stochastic seihr model of covid-19 using the enkf: A case study based on real-world data.Chaos, Solitons & Fractals , 208:118225, 2026

    Youssef Belgraoui, Aziza Belmaati, and Mustapha Kabil. Parameter estimation in a stochastic seihr model of covid-19 using the enkf: A case study based on real-world data.Chaos, Solitons & Fractals , 208:118225, 2026

    2026

  6. [5]

    Sparse identification of nonlinear dynamics for stochastic delay differential equations.Journal of Computational and Applied Mathematics, page 117247, 2025

    Dimitri Breda, Dajana Conte, Raffaele D’Ambrosio, Ida Santaniello, and Muhammad Tanveer. Sparse identification of nonlinear dynamics for stochastic delay differential equations.Journal of Computational and Applied Mathematics, page 117247, 2025

    2025

  7. [6]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems.Proceedings of the National Academy of Sciences , 113(15):3932–3937, 2016

    Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems.Proceedings of the National Academy of Sciences , 113(15):3932–3937, 2016

    2016

  8. [7]

    Chikungunya virus–vector interactions

    Lark L Coffey, Anna-Bella Failloux, and Scott C Weaver. Chikungunya virus–vector interactions. Viruses, 6(11):4628–4663, 2014

    2014

Show all 32 references
  1. [8]

    On the definition and the computation of the basic reproduction ratio r 0 in models for infectious diseases in heterogeneous populations

    Odo Diekmann, Johan Andre Peter Heesterbeek, and Johan Anton Jacob Metz. On the definition and the computation of the basic reproduction ratio r 0 in models for infectious diseases in heterogeneous populations. Journal of Mathematical Biology , 28:365–382, 1990

    1990

  2. [9]

    The ensemble kalman filter: Theoretical formulation and practical implementation.Ocean dynamics, 53(4):343–367, 2003

    Geir Evensen. The ensemble kalman filter: Theoretical formulation and practical implementation.Ocean dynamics, 53(4):343–367, 2003

    2003

  3. [11]

    Urban Fasel, J Nathan Kutz, Bingni W Brunton, and Steven L Brunton. Ensemble-sindy: Robust sparse model discovery in the low-data, high-noise limit, with active learning and control.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , 478(2260), 2022

    2022

  4. [12]

    Sparseidentificationofnonlineardynamics with low-dimensionalized flow representations.Journal of Fluid Mechanics , 926:A10, 2021

    KaiFukami, TakaakiMurata, KaiZhang, andKojiFukagata. Sparseidentificationofnonlineardynamics with low-dimensionalized flow representations.Journal of Fluid Mechanics , 926:A10, 2021

    2021

  5. [13]

    Lin Guo, Xiaokai Yang, Zhonghua Zheng, Nicole Riemer, and Christopher W Tessum. Uncertainty quantification in reduced-order gas-phase atmospheric chemistry modeling using ensemble sindy.Journal of Geophysical Research: Machine Learning and Computation , 1(4):e2024JH000358, 2024

    2024

  6. [14]

    Spatial modes for transmission of chikungunya virus during a large chikungunya outbreak in italy: a modeling analysis

    Giorgio Guzzetta, Francesco Vairo, Alessia Mammone, Simone Lanini, Piero Poletti, Mattia Manica, Roberto Rosa, Beniamino Caputo, Angelo Solimini, Alessandra Della Torre, et al. Spatial modes for transmission of chikungunya virus during a large chikungunya outbreak in italy: a ...

    2020

  7. [15]

    Modeling and prediction of the transmission dynamics of covid-19 based on the sindy-lm method.Nonlinear Dynamics, 105(3):2775–2794, 2021

    Yu-Xin Jiang, Xiong Xiong, Shuo Zhang, Jia-Xiang Wang, Jia-Chun Li, and Lin Du. Modeling and prediction of the transmission dynamics of covid-19 based on the sindy-lm method.Nonlinear Dynamics, 105(3):2775–2794, 2021. 31

    2021

  8. [16]

    Automatic differentiation to simulta- neously identify nonlinear dynamics and extract noise probability distributions from data

    Kadierdan Kaheman, Steven L Brunton, and J Nathan Kutz. Automatic differentiation to simulta- neously identify nonlinear dynamics and extract noise probability distributions from data. Machine Learning: Science and Technology , 3(1):015031, 2022

    2022

  9. [17]

    Kadierdan Kaheman, J Nathan Kutz, and Steven L Brunton. Sindy-pi: a robust algorithm for parallel implicit sparse identification of nonlinear dynamics.Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , 476(2242), 2020

    2020

  10. [18]

    An application of the ensemble kalman filter in epidemiological modeling

    Rajnesh Lal, Weidong Huang, and Zhenquan Li. An application of the ensemble kalman filter in epidemiological modeling. Plos one, 16(8):e0256227, 2021

    2021

  11. [19]

    Uncovering dynamical equations of stochastic decision models using data-driven sindy algorithm.Neural Computation, 37(3):569–587, 2025

    Brendan Lenfesty, Saugat Bhattacharyya, and KongFatt Wong-Lin. Uncovering dynamical equations of stochastic decision models using data-driven sindy algorithm.Neural Computation, 37(3):569–587, 2025

    2025

  12. [20]

    Reemergence of chikungunya virus.Journal of virology, 88(20):11644–11647, 2014

    Thomas E Morrison. Reemergence of chikungunya virus.Journal of virology, 88(20):11644–11647, 2014

    2014

  13. [21]

    Data- driven identification of biological systems using multi-scale analysis

    Ismaila Muhammed, Dimitris M Manias, Dimitris A Goussis, and Haralampos Hatzikirou. Data- driven identification of biological systems using multi-scale analysis. PLOS Computational Biology , 21(11):e1013193, 2025

    2025

  14. [22]

    Derivative-free domain- informed data-driven discovery of sparse kinetic models.Industrial & Engineering Chemistry Research , 64(5):2601–2615, 2025

    Siddharth Prabhu, Nick Kosir, Mayuresh V Kothare, and Srinivas Rangarajan. Derivative-free domain- informed data-driven discovery of sparse kinetic models.Industrial & Engineering Chemistry Research , 64(5):2601–2615, 2025

    2025

  15. [23]

    Ekf–sindy: Empow- ering the extended kalman filter with sparse identification of nonlinear dynamics.Computer Methods in Applied Mechanics and Engineering , 431:117264, 2024

    Luca Rosafalco, Paolo Conti, Andrea Manzoni, Stefano Mariani, and Attilio Frangi. Ekf–sindy: Empow- ering the extended kalman filter with sparse identification of nonlinear dynamics.Computer Methods in Applied Mechanics and Engineering , 431:117264, 2024

    2024

  16. [24]

    Online learning in bifurcating dynamic systems via sindy and kalman filtering.arXiv preprint arXiv:2411.04842 , 2024

    Luca Rosafalco, Paolo Conti, Andrea Manzoni, Stefano Mariani, and Attilio Frangi. Online learning in bifurcating dynamic systems via sindy and kalman filtering.arXiv preprint arXiv:2411.04842 , 2024

    2024 arXiv

  17. [25]

    Alessandro Maria Selvitella and Elliot Allen. On the effectiveness of sparse identification methods to detect nonlinear models of oscillatory dynamics in psychology and the life sciences.Nonlinear Dynamics, Psychology & Life Sciences , 30(1), 2026

    2026

  18. [26]

    Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission.Mathematical biosciences, 180(1-2):29–48, 2002

    Pauline Van den Driessche and James Watmough. Reproduction numbers and sub-threshold endemic equilibria for compartmental models of disease transmission.Mathematical biosciences, 180(1-2):29–48, 2002

    2002

  19. [27]

    The law of mass action in epidemiology

    Edwin B Wilson and Jane Worcester. The law of mass action in epidemiology. Proceedings of the National Academy of Sciences , 31(1):24–34, 1945

    1945

  20. [28]

    Basic reproduction numbers for a class of reaction-diffusion epidemic models

    Chayu Yang and Jin Wang. Basic reproduction numbers for a class of reaction-diffusion epidemic models. Bulletin of mathematical biology , 82(8):111, 2020

    2020

  21. [29]

    On the convergence of the sindy algorithm.Multiscale Modeling & Simulation, 17(3):948–972, 2019

    Linan Zhang and Hayden Schaeffer. On the convergence of the sindy algorithm.Multiscale Modeling & Simulation, 17(3):948–972, 2019

    2019

  22. [30]

    Ensemble kalman method for learning turbulence models from indirect observation data.Journal of Fluid Mechanics , 949:A26, 2022

    Xin-Lei Zhang, Heng Xiao, Xiaodong Luo, and Guowei He. Ensemble kalman method for learning turbulence models from indirect observation data.Journal of Fluid Mechanics , 949:A26, 2022

    2022

  23. [31]

    Global resurgence of chikungunya virus: outbreak drivers and emerging solutions

    Yi Zhang, Jing Wu, Xiaoyang Cheng, Yuxuan Yang, Xinyu Wang, Xiaoyu Zhao, Xiaoyan Wang, Huiling Ouyang, Jingwen Ai, and Wenhong Zhang. Global resurgence of chikungunya virus: outbreak drivers and emerging solutions. Emerging Microbes & Infections , 15(1):2603714, 2026

    2026

  24. [32]

    Chikungunya virus in 2025: Epidemiology, immunopathogenesis, and vaccine development—a narrative review.Infection and Drug Resistance, pages 1–14, 2026

    Chenxi Zhao, Ziruo Ge, Tingyu Zhang, Zhouling Jiang, Di Tian, and Zhihai Chen. Chikungunya virus in 2025: Epidemiology, immunopathogenesis, and vaccine development—a narrative review.Infection and Drug Resistance, pages 1–14, 2026. 32

    2025

Pith tools

Reviewed July 30, 2026 · model on record in the stance chip above.