Pith. sign in

REVIEW 4 major objections 5 minor 68 references

Dynamic Bayesian Predictive Stacking via Markovian Spatiotemporal Propagation

T0 review · 4 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read The paper claims that dynamic Bayesian predictive stacking—running J conjugate matrix-variate dynamic linear models in parallel and combining them with leave-future-out weights—preserves closed-form forward-filter-backward-sampler conjugacy

desk verdict Promising architecture, but Algorithm 3 does not sample the stacked posterior it claims, and that flaw is load-bearing. read the letter →

arxiv 2602.08544 v2 pith:6ZJIYWAK submitted 2026-02-09 stat.ME stat.CO

Luca Presicce , Sudipto Banerjee This is my paper · ORCID
classification stat.MEstat.CO MSC 62M3062M1062F15
keywords Bayesianpredictivestackingdynamiclinearmodelsmatrix-variatedistributionsspatiotemporalleave-future-outcross-validationonlineinferenceforwardfilteringbackwardsamplinglarge-scalegeostatistics
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

Dynamic Bayesian predictive stacking is proposed as a way to run Bayesian multivariate spatiotemporal analysis online, without Markov chain Monte Carlo. The paper's central claim: by running J conjugate matrix-variate dynamic linear models in parallel—one per fixed spatial hyperparameter combination—and averaging their one-step-ahead forecasts with leave-future-out weights, the forward-filter-backward-sampler keeps its closed-form updating while the stacked mixture remains available for inference. A sympathetic reader would care because this makes posterior inference, forecasting, and spatial interpolation feasible for data with hundreds of thousands of observations on ordinary hardware, and because the same machinery can generate training data for amortized neural forecasting.

What carries the argument

The load-bearing object is the stacked posterior mixture p̂(·|Dτ)=Σj ŵj p(·|Dτ,Mj), with each component a matrix-variate dynamic linear model in the Gaussian-Wishart conjugate family; the weights are computed per spatial location by leave-future-out log-score optimization over one-step-ahead predictive densities. The mechanism that preserves conjugacy is the parallel-forward-propagation design: each model's conditional posterior is propagated forward separately, so the posterior-to-prior update in FFBS remains a closed-form conjugate step, and only the backward pass samples models from the stacked weights.

What would settle it

Generate data from a mixture of two spatial regimes, e.g., α=0.7/ϕ=2 in half the locations and α=0.9/ϕ=6 in the other half; run dynbps using aggregated weights and compare per-location smoothed credible-interval coverage with nominal 95%. If coverage falls well below nominal in either regime—or if coverage using per-location weights is materially better than coverage using global/consensus weights—the aggregation assumption is falsified.

Watch

Extended reading notes

Core claim

The central claim is that predictive stacking can be inserted into the forward-filter-backward-sampler without breaking its conjugacy, provided the information passed from one time shard to the next is the set of J conditional posteriors p(·|Dτ,Mj), not the stacked mixture. With stacking weights ŵj(τ) obtained by maximizing the log leave-future-out score of each location's one-step-ahead predictive density, the stacked posterior p̂(·|Dτ)=Σj ŵj(τ)p(·|Dτ,Mj) delivers posterior samples, forecasts, and interpolated maps by first drawing a model from the current weights and then drawing from that model's closed-form posterior. The paper demonstrates this on 240,000 satellite-derived climate obser

Load-bearing premise

The central claim rests on a single per-time-shard weight vector—global average or consensus—being able to represent the spatially varying model preferences that the per-location weights detect; if it cannot, the backward-smoothed stacked posterior and its credible intervals are miscalibrated even though each conditional posterior is exact.

Editorial extensions

If this is right

  • Full posterior inference, k-step forecasts, and spatial interpolation are all obtained from closed-form matrix-variate Student-t draws, so no MCMC burn-in, tuning, or convergence checks are needed.
  • The procedure is naturally online: when a new temporal shard arrives, one parallel forward pass plus a weighted backward pass updates the entire posterior, making it suitable for streaming geospatial data.
  • The leave-future-out scoring extends ordinary stacking to time-dependent data, which is the right target when the goal is forecasting rather than interpolation across independent observations.
  • On the case study, 240,000 observations across 500 locations, 4 variables, and 120 months are analyzed in about 12 minutes on a laptop, suggesting the method can serve as a routine engine for large climate datasets.
  • Because weights are location-specific before aggregation, the method also gives a diagnostic map of which spatial covariance model each location's data favor.

Reading between the lines

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

  • An obvious extension the paper does not pursue: use the per-location weights, rather than the global average, in the backward pass for location-level quantities, reserving global weights for shared parameters; this would test whether aggregation is the only source of miscalibration.
  • The same stacking-and-parallel-propagation logic should transfer to non-Gaussian observations if a conjugate or conditionally conjugate family exists for the observation equation, though this is speculative and the closed-form speed would depend on that family.
  • The weight aggregation step (global average vs consensus) is the natural stress point for spatially nonstationary data; a simulation where model preference varies sharply by region would separate the two aggregation strategies.
  • The 12-minute runtime on a laptop only covers the closed-form DLM family; scaling the same stacking approach to more flexible state-space models would need approximate backward sampling, so the speed and the model class are coupled.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper develops a scalable online inference framework ("dynamic Bayesian predictive stacking") for matrix-variate dynamic linear spatiotemporal models. It fixes a grid of J spatial hyperparameter pairs {α, φ}, propagates J parallel conjugate matrix-variate Gaussian-Wishart forward filters through time, computes leave-future-out one-step predictive stacking weights per location at each time, aggregates these to global or consensus vectors, and then uses a weighted backward sampling pass to obtain smoothed posterior samples. The same machinery is used for multi-step forecasting and spatial interpolation. The authors report simulation studies and an analysis of 500 locations × 4 variables × 120 months of Copernicus data, run in about 12 minutes on a laptop, plus an amortized neural-network forecast experiment.

Significance. The proposed idea is attractive and the forward-filtering component is sound: parallel conjugate filtering over fixed hyperparameters is a sensible way to avoid MCMC, and leave-future-out stacking is a principled scoring choice for temporal data. The paper ships reproducible code and gives a serious large-scale case study. However, the central "exact inference" claim is currently not supported by the backward-sampling algorithm as written, and two supporting approximations (weight aggregation and the case-study hyperparameter grid) are not validated at the level the claims require. With a corrected backward pass and a more cautious framing as an approximate stacked smoother, the method could be a useful contribution.

major comments (4)
  1. [Section 3, Algorithm 3, Eq. (13)] The weighted backward pass does not sample from the stacked posterior (12). In standard FFBS, Θ_t | Θ_{t+1}, D_T, M_j is a conditional sampler for a fixed model. Algorithm 3 re-draws j ∼ Multinom(1, w^{(t)}) independently at every t, so Eq. (13) is a product of time-indexed conditional mixtures. Expanding it gives J^T terms with product weights ∏_t w^{(t)}_{j_t}, not Σ_j w_j p(Θ_{0:T}|D_T,M_j). Consequently even the marginal of Θ_t is not the target mixture, because Θ_{t+1} may have been generated under a different model. The empirical coverage in Figures 4–5 does not rescue this: it only shows approximate calibration. Please either (a) sample one model index j at t=T and keep it fixed through the backward recursion, which exactly samples from (12), or (b) if the current time-varying resampling is intended, call it an approximation, state the target distribution explicitly, and remove "e
  2. [Section 2.2, Eqs. (11)–(12)] The aggregation of per-location LFO weights to a single global/consensus vector is an unvalidated approximation. The only mathematical support (Supplement S1.2) is that the averaged/count vectors remain probability vectors; it says nothing about whether the resulting mixture approximates any meaningful posterior or preserves spatially varying model preferences. The statement that joint space-time scoring is "infeasible" because "the notion of a point rapidly loses sense" is an informal argument, not a rigorous justification. Since Algorithm 3 uses the same aggregated weights for all locations, the smoothed posterior and credible intervals may be miscalibrated in the presence of heterogeneous model preferences; Section 4.2 only compares closed vs. open settings and does not check this directly.
  3. [Section 2.2, Eq. (10)] The key analytical posterior predictive for spatial interpolation is not derived; the text refers to "similar arguments used in Presicce and Banerjee (2024, Supplement 1.1)". This distribution underpins Algorithm S2 and the spatial interpolation results, so the derivation (or at least a verification of matrix dimensions and parameters) should appear in the supplement rather than being delegated to an unpublished preprint.
  4. [Section 5, Appendix S4] The candidate grid {α=0.999, φ∈{0.180,0.212,0.229,0.308}} was obtained from variograms fitted to the same training residuals used for stacking and in-sample interpolation. This uses the data twice to select and then evaluate models, so the reported excellent fit and in-sample interpolated maps are expected to be optimistic. The 24-month holdout only validates one-step-ahead forecasting, not smoothing or interpolation. Please provide an independent validation (e.g., a spatial holdout or a fully pre-specified grid) and report out-of-sample interpolation metrics.
minor comments (5)
  1. [Abstract / Section 2.2] "Interoperability" appears where "interpretability" seems intended; please check both occurrences.
  2. [Section 2.2] Typo: "Let us aware the reader aware" should read "Let us make the reader aware" or similar.
  3. [Eq. (7)] The off-diagonal block of R*(k) appears asymmetric (F_{T+k}R_T(k) vs. R_T(k)F^T_{T+k}); please verify and correct the block definition.
  4. [Section 4.1 / Section 6] The GitHub URL appears as /lucapresicce/lucapresicce/... in one place and /lucapresicce/... in another; please unify.
  5. [Section 4.1 / Section 6] The text refers to an Rcpp-based spFFBS package, then later says the software is "currently being migrated to R"; please clarify the current status.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the posterior-update chain is standard conjugate FFBS; noted issues are approximation/correctness or empirical choices, not reductions by construction.

full rationale

The central derivation is not circular. The forward filtering recursions (Eqs. 2–6) are the standard matrix-variate DLM/FFBS conjugacy updates, and the stacking weights in Eq. (11) are defined by an explicit leave-future-out log-score optimization over the one-step-ahead predictive densities. The weighted posterior Eq. (12) is then a mixture of these exact conjugate posteriors with data-driven weights; no parameter in the derivation is fitted to the quantity later called a prediction. The spatial-interpolation formula Eq. (10) is stated in closed form and, although its derivation is deferred to the authors' own preprint, it is a parameter-free distributional result with stated assumptions, not a fitted or self-referential constraint. The case-study hyperparameter grid is informed by variograms on the training data, but the reported forecasting and interpolation evaluations are on held-out months/locations; this is empirical model-selection practice rather than a circular prediction. The amortized RNN experiment explicitly supervises the network on dynbps outputs, so agreement with dynbps on held-out simulations is an emulator check, not a claim that dynbps is derived from the RNN. The more serious concern—Algorithm 3's time-varying independent resampling of model indices—means the distribution actually sampled is a product-of-mixtures (Eq. 13) rather than the joint stacked posterior (Eq. 12); this is a mathematical approximation/correctness issue in the claimed 'exact' smoothing, not a circular reduction of the derivation to its inputs. No fitted input is renamed as a prediction, no load-bearing result reduces to a self-citation, and no definition is equivalent to the target result by construction. Hence the circularity score is 0.

Assumptions & free parameters 5 free parameters · 8 assumptions · 0 invented entities

The method is assembled largely from standard conjugate DLM theory (West & Harrison 1997) and established stacking/LFO methodology (Yao et al. 2018; Bürkner & Vehtari 2020). The paper's own additions are the per-location dynamic weights and the parallel/weighted FFBS scheme. The heaviest unstated inputs are: (i) the analytical predictive (10) imported from the authors' own 2024 preprint; (ii) the assumption that aggregating per-location weights to one vector preserves calibration, justified only by an infeasibility assertion; (iii) a data-informed case-study hyperparameter grid. No genuinely new entities are postulated.

free parameters (5)
  • Stacking weights ŵ(τ)_j (per-location, global, consensus) = Time-varying, estimated at each τ; simulations prefer {α=0.8, φ=4} in M-closed settings
    Estimated by maximizing the leave-future-out log score (Eq. 11); these are the actual fitted quantities of the method and carry all model-averaging weight.
  • Nugget/spatial-variance proportion α grid = Simulations: {0.7, 0.8, 0.9}; case study: 0.999 (variogram-informed)
    α parametrizes V_t(α)=((1−α)/α)I_n in Eq. (8). In the case study it is set from an empirical variogram fitted on the same data (Appendix S4) rather than swept, a data-informed choice.
  • Spatial range φ grid = Simulations: {2,4,6} or M-open uniform draws in [1,50]; case study: {0.180, 0.212, 0.229, 0.308}
    Candidate values of the exponential spatial correlation kernel; chosen by hand in simulations and by data-informed variograms in the case study.
  • W_B^t (p×p row-covariance of dynamic regression coefficients) = Not specified
    Appears in W_t(φ)=diag(W_B^t, R_t(S,S;φ)) in Eq. (8) but its specification or update is never given; the filter cannot be run without it.
  • Neural network parameters ψ (~49M) = Trained on 100 dynbps-generated datasets, Section 4.1
    Required for the amortized-forecast demonstration; not part of the core statistical method, but the main fitted object in the transfer-learning experiment.
assumptions (8)
  • standard math Matrix-variate normal-inverse-Wishart conjugacy updates (Eqs. 2–6) are correct.
    Textbook DLM results (West & Harrison 1997); acceptable background.
  • domain assumption For fixed {α,φ}, the model in Eq. (8) is a standard DLM with closed-form posteriors.
    Consequence of construction; requires V_t(α) positive definite, i.e. α∈(0,1).
  • domain assumption The per-location predictive log score aggregated over time (Eq. 11) approximates the true predictive score.
    Standard leave-future-out reasoning (Bürkner & Vehtari 2020); assumed transferable to the multivariate spatial setting without proof.
  • ad hoc to paper Aggregating per-location weights to a global/consensus vector preserves inferential validity.
    Section 2.2 asserts simultaneous space-time evaluation is 'infeasible' without proof; Supplement S1.2 proves only that the aggregated vectors are probability vectors, not that they yield calibrated posteriors.
  • ad hoc to paper The analytical posterior predictive in Eq. (10) holds.
    Deferred to 'similar arguments used in Presicce and Banerjee, 2024, Supplement 1.1' (Section 2.1) — the authors' own unpublished preprint; the derivation is not reproduced here.
  • domain assumption B_t ⊥⊥ Ω_t at all t and G_t is block-diagonal.
    Modeling assumption separating regression coefficients from the spatial process evolution (Eq. 8).
  • ad hoc to paper The case-study candidate grid (α=0.999, φ∈{0.180,0.212,0.229,0.308}) derived from variograms on the same data does not materially inflate performance.
    Appendix S4 selects the grid from empirical spatiotemporal variograms of the same training data — a data-informed model-class choice whose effect on the reported 'striking' forecasts is unquantified.
  • domain assumption Markovian state-space dependence across temporal shards.
    Core model structure (Eq. 1); standard for DLMs, but it is the property that makes parallel/sequential shard propagation valid.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dynamic Bayesian Predictive Stacking via Markovian Spatiotemporal Propagation." pith.science (2026). https://pith.science/paper/6ZJIYWAK

@misc{pith2026260208544,
  author       = {Pith},
  title        = {Pith review of: Dynamic Bayesian Predictive Stacking via Markovian Spatiotemporal Propagation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6ZJIYWAK}},
  note         = {Machine review of arXiv:2602.08544}
}
read the original abstract

This manuscript develops computationally efficient online learning for multivariate spatiotemporal models. The proposed framework relies on matrix-variate Gaussian distributions, dynamic linear models, and Bayesian predictive stacking to efficiently share information across temporal data shards. The model facilitates effective information propagation over time while seamlessly integrating spatial components within a dynamic framework, building a Markovian dependence structure between datasets at successive time instants. This structure supports flexible, high-dimensional modeling of complex dependence patterns, as commonly found in spatiotemporal phenomena, where computational challenges arise rapidly with increasing dimensions. The proposed approach further manages exact inference through predictive stacking, enhancing robustness and interoperability. Combining sequential and parallel processing of temporal shards, each unit passes assimilated information forward and then back-smooths it to improve posterior estimation, incorporating all available information. This framework advances the scalability and adaptability of spatiotemporal modeling, making it suitable for dynamic, multivariate, and data-rich environments. Simulation experiments and an extracted data analysis from the Copernicus Data Space Ecosystem (CDSE) help evaluate and illustrate the framework.

Figures

Figures reproduced from arXiv: 2602.08544 by the authors.

Figure 1
Figure 1. Data shards dynamics dependencies representation which vary continuously in space and discretely in time. We automate inference propagation using a dynamic Bayesian predictive stacking framework. 2.1 Matrix-variate Dynamic linear models Let {Yt , t ∈ T } be a matrix-valued time series indexed by discrete time points in T = {0, ±1, ±2, ±3, · · · }, where each Yt = n Y (t) i,j o is n × q with the (i, j)-th element Y (… view at source ↗
Figure 2
Figure 2. Leave-future-out cross-validation across time and locations is infeasible. As the number of points increases, the density lies in a higher￾dimensional space, where the notion of a point rapidly loses sense, and its probability goes to zero even faster. Then, as the number of locations grows, the evaluation of the predictive density for the whole matrix Yt results in numerical instabilities. Nevertheless, individual … view at source ↗
Figure 3
Figure 3. Spatial process Ω one-step ahead forecast surface interpolations: true spatial process Ω (leftmost column), 50-th quantile dynbps forecast (center column), 50-th quantile amortized forecast (rightmost column); each row corresponds to a different outcome in Ω These yield 100 simulated pairs {Z (i) , W(i)} that we use to supervise the training of the neural network gψ, characterized by the set of parameters ψ. Each Z … view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Posterior metrics at each time point for regression coefficients B, i.e. Θ1:p,1:q, under the considered model configu￾rations and 50 replications: average empirical bias (top left), average posterior standard deviation (top right), average credible interval coverage (b…
Figure 5
Figure 5. Figure 5: Posterior metrics at each time point for each component of the multivariate spatial process Ω, under the considered model configurations and 50 replications: average empirical bias (top left), average posterior standard deviation (top right), average credible interval …
Figure 6
Figure 6. Figure 6: One-step ahead average monthly forecast for selected spatial points large-scale domains. The temporal structure via monthly indexing enables the model to learn seasonal and interannual dynamics; this setup facilitated both interpolation and prediction tasks within a co…
Figure 7
Figure 7. Figure 7: Spatial surface interpolation at unobserved time (out of sample) variability for predictive intervals, even though the empirical coverage reports very good results. Notwithstanding, state-space models usually provide predictive intervals that grow linearly with t; in t…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 5 linked inside Pith

  1. [1]

    Banerjee, G

    Abdalla, N., S. Banerjee, G. Ramachandran, and S. Arnold (2020). Bayesian state space modeling of physical processes in industrial hygiene. Technometrics\/ 62\/ (2), 147--160

    2020

  2. [2]

    Ali, M. M. (1979, December). Analysis of stationary spatial-temporal processes: Estimation and prediction. Biometrika\/ 66\/ (3), 513--518

    1979

  3. [3]

    Allaire, J. J., T. Kalinowski, D. Falbel, D. Eddelbuettel, Y. Tang [aut, cph, N. Golding, G. I. E. Tutorials), , Posit, and PBC (2024, April). tensorflow: R Interface to ' TensorFlow '

    2024

  4. [4]

    Banerjee, S. (2020). Modeling massive spatial datasets using a conjugate Bayesian linear modeling framework. Spatial Statistics\/ 37 , 100417

    2020

  5. [5]

    Banerjee, S., X. Chen, I. Frankenburg, and D. Zhou (2025). Dynamic Bayesian Learning for Spatiotemporal Mechanistic Models . Journal of Machine Learning Research\/ 26\/ (146), 1--43

    2025

  6. [6]

    Banerjee, S., A. E. Gelfand, and B. P. Carlin (2025, September). Hierarchical modeling and analysis for spatial data\/ (3 ed.). London, England: Chapman and Hall

    2025

  7. [7]

    Breiman, L. (1996). Stacked Regressions . Machine Learning\/ 24\/ (1), 49--64

    1996

  8. [8]

    Sugasawa, M

    Cabel, D., S. Sugasawa, M. Kato, K. Takanashi, and K. McAlinn (2025, January). Bayesian Spatial Predictive Synthesis . arXiv:2203.05197 [stat]

    arXiv 2025

Show all 68 references
  1. [9]

    Carter, C. K. and R. Kohn (1994, September). On Gibbs sampling for state space models. Biometrika\/ 81\/ (3), 541--553

    1994

  2. [10]

    Zhong, H

    Chen, L., X. Zhong, H. Li, J. Wu, B. Lu, D. Chen, S.-P. Xie, L. Wu, Q. Chao, C. Lin, Z. Hu, and Y. Qi (2024, July). A machine learning model that outperforms conventional global subseasonal forecast models. Nature Communications\/ 15\/ (1), 6425

    2024

  3. [11]

    Chen, Q., R. Han, F. Ye, and W. Li (2011). Spatio-temporal ecological models. Ecological Informatics\/ 6\/ (1), 37--43

    2011

  4. [12]

    van Merriënboer, C

    Cho, K., B. van Merriënboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y. Bengio (2014, October). Learning Phrase Representations using RNN Encoder – Decoder for Statistical Machine Translation . In A. Moschitti, B. Pang, and W. Daelemans (Eds.), Proceedings of th...

    2014

  5. [13]

    Simpson, L

    Cooper, A., D. Simpson, L. Kennedy, C. Forbes, and A. Vehtari (2023). Cross-validatory model selection for Bayesian autoregressions with exogenous regressors. arXiv preprint\/ . arXiv:2301.08276 [stat.ME]

    2023 arXiv

  6. [14]

    Cressie, N. A. C. and C. K. Wikle (2011). Statistics for spatio-temporal data . Wiley series in probability and statistics. Hoboken, N.J: Wiley

    2011

  7. [15]

    Czaran, T. and S. Bartha (1992, February). Spatiotemporal dynamic models of plant populations and communities. Trends in Ecology & Evolution\/ 7\/ (2), 38--42

    1992

  8. [16]

    Elkhouly, M. and M. A. R. Ferreira (2021). Dynamic multiscale spatiotemporal models for multivariate Gaussian data. Spatial Statistics\/ 41 , 100475

    2021

  9. [17]

    Felsche, E. and R. Ludwig (2021, December). Applying machine learning for drought prediction in a perfect model framework using data from a large ensemble of climate simulations. Natural Hazards and Earth System Sciences\/ 21\/ (12), 3679--3691

    2021

  10. [18]

    Gamerman, D., H. F. Lopes, and E. Salazar (2008). Spatial dynamic factor analysis. Bayesian Analysis\/ 3\/ (4), 759 -- 792

    2008

  11. [19]

    Jain, and U

    Ganguly, A., S. Jain, and U. Watchareeruetai (2023, October). Amortized Variational Inference : A Systematic Review . Journal of Artificial Intelligence Research\/ 78 , 167--215

    2023

  12. [20]

    Gelfand, A. E., S. Banerjee, and D. Gamerman (2005). Spatial process modelling for univariate and multivariate dynamic spatial data. Environmetrics\/ 16\/ (5), 465--479

    2005

  13. [21]

    Gelfand, A. E., P. Diggle, P. Guttorp, and M. Fuentes (2010). Handbook of Spatial Statistics . Taylor & Francis

    2010

  14. [22]

    Gneiting, T. and A. E. Raftery (2007). Strictly Proper Scoring Rules , Prediction , and Estimation . Journal of the American Statistical Association\/ 102\/ (477), 359--378

    2007

  15. [23]

    Hefley, T. J., M. B. Hooten, E. M. Hanks, R. E. Russell, and D. P. Walsh (2017). Dynamic spatio-temporal models for spatial data. Spatial Statistics\/ 20 , 206--220

    2017

  16. [24]

    Idjigbèrou, S. E., E. S. P. Assédé, S. Biaou, G. N. Gouwakinnou, A. K. Natta, and S. S. H. Biaou (2025). Spatio-temporal dynamics of Isoberlinia -dominated woodlands in disturbance-prone landscapes over 15 years. Global Ecology and Conservation\/ 59 , e03512

    2025

  17. [25]

    Ivanovic, B. and M. Pavone (2019, October). The Trajectron : Probabilistic Multi - Agent Trajectory Modeling With Dynamic Spatiotemporal Graphs . In Proceedings of the IEEE / CVF International Conference on Computer Vision ( ICCV )

    2019

  18. [26]

    Jacobs, J. M., B. Lowry, M. Choi, and C. H. Bolster (2009, March). GOES Solar Radiation for Evapotranspiration Estimation and Streamflow Prediction . Journal of Hydrologic Engineering\/ 14\/ (3), 293--300

    2009

  19. [27]

    Jiménez, J. C. and C. A. d. B. Pereira (2021). Assessing dynamic effects on a Bayesian matrix-variate dynamic linear model: An application to task-based fMRI data analysis. Computational Statistics & Data Analysis\/ 163 , 107297

    2021

  20. [28]

    Falbel, J

    Kalinowski, T., D. Falbel, J. J. Allaire, F. Chollet, RStudio, Google, Y. Tang [ctb, cph, W. V. D. Bijl, M. Studer, and S. Keydana (2024, April). keras: R Interface to ' Keras '

    2024

  21. [29]

    Vehtari, and A

    Kennedy, L., A. Vehtari, and A. Gelman (2024). Model validation for aggregate inferences in out-of-sample prediction. arXiv preprint\/ . arXiv:2312.06334 [stat.ME]

    2024 arXiv

  22. [30]

    Landim, F. and D. Gamerman (2000). Dynamic hierarchical models: an extension to matrix-variate observations. Computational Statistics & Data Analysis\/ 35\/ (1), 11--42

    2000

  23. [31]

    , Knut Are, N

    McAlinn, K., A. , Knut Are, N. , Jouchi, , and M. West (2020, July). Multivariate Bayesian Predictive Synthesis in Macroeconomic Forecasting . Journal of the American Statistical Association\/ 115\/ (531), 1092--1110

    2020

  24. [32]

    McAlinn, K. and M. West (2019, May). Dynamic Bayesian predictive synthesis in time series forecasting. Journal of Econometrics\/ 210\/ (1), 155--169

    2019

  25. [33]

    Nobre, A. A., A. M. Schmidt, and H. F. Lopes (2005, May). Spatio-temporal models for mapping the incidence of malaria in Pará . Environmetrics\/ 16\/ (3), 291--304

    2005

  26. [34]

    Zhang, J

    Pan, S., L. Zhang, J. R. Bradley, and S. Banerjee (2025, May). Bayesian Inference for Spatial - Temporal Non - Gaussian Data Using Predictive Stacking . arXiv:2406.04655 [stat]

    2025

  27. [35]

    Paul-Christian Bürkner, J. G. and A. Vehtari (2020). Approximate leave-future-out cross-validation for Bayesian time series models. Journal of Statistical Computation and Simulation\/ 90\/ (14), 2499--2523

    2020

  28. [36]

    Hass, and A

    Pherwani, P., N. Hass, and A. Yanchenko (2024). Spatiotemporal Modeling and Forecasting at Scale with Dynamic Generalized Linear Models . In Proceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection , GeoAnomalies '24, New York, NY, USA, pp.\...

    2024

  29. [37]

    Ferreira, and M

    Prado, R., M. Ferreira, and M. West (2021). Time Series : Modeling , Computation , and Inference , Second Edition . Chapman & Hall / CRC Texts in Statistical Science . CRC Press

    2021

  30. [38]

    Presicce, L. and S. Banerjee (2024). Bayesian Transfer Learning for Artificially Intelligent Geospatial Systems : A Predictive Stacking Approach . arXiv preprint\/ . arXiv:2410.09504 [stat.ME]

    2024 arXiv

  31. [39]

    Quintana, J. M. and M. West (1987). An Analysis of International Exchange Rates Using Multivariate DLM 's. Journal of the Royal Statistical Society. Series D (The Statistician)\/ 36\/ (2/3), 275--281

    1987

  32. [40]

    Radev, S. T., U. K. Mertens, A. Voss, L. Ardizzone, and U. K \"o the (2020). BayesFlow : Learning complex stochastic models with invertible neural networks. IEEE transactions on neural networks and learning systems\/ 33\/ (4), 1452--1466

    2020

  33. [41]

    Ruiz Maraggi, L. M., L. W. Lake, and M. P. Walsh (2021, July). Using Bayesian Leave - One - Out and Leave - Future - Out Cross - Validation to Evaluate the Performance of Rate - Time Models to Forecast Production of Tight - Oil Wells . In Proceedings of the 9th Unconventional ...

    2021

  34. [42]

    Zammit-Mangion, J

    Sainsbury-Dale, M., A. Zammit-Mangion, J. Richards, and R. Huser (2024). Neural Bayes Estimators for Irregular Spatial Data using Graph Neural Networks . arXiv:2310.02600

    2024 arXiv

  35. [43]

    Sandefur, J. (1990). Discrete Dynamical Systems : Theory and Applications . Clarendon Press

    1990

  36. [44]

    Schmidt, A. M. and H. F. Lopes (2019). Dynamic models. In Handbook of Environmental and Ecological Statistics , pp.\ 57--80. Chapman & Hall

    2019

  37. [45]

    Stein, M. L. (1988). Asymptotically Efficient Prediction of a Random Field with a Misspecified Covariance Function . The Annals of Statistics\/ 16\/ (1), 55--63

    1988

  38. [46]

    Stein, M. L. and M. S. Handcock (1989). Some asymptotic properties of kriging when the covariance function is misspecified. Mathematical Geology\/ 21 , 171--190

    1989

  39. [47]

    Stroud, J. R., P. Müller, and B. Sansó (2001). Dynamic models for spatiotemporal data. Journal of the Royal Statistical Society: Series B (Statistical Methodology)\/ 63\/ (4), 673--689

    2001

  40. [48]

    Tallman, E. and M. West (2023, October). Bayesian predictive decision synthesis. Journal of the Royal Statistical Society Series B: Statistical Methodology\/ 86\/ (2), 340--363

    2023

  41. [49]

    Zhang, and S

    Tang, W., L. Zhang, and S. Banerjee (2021). On identifiability and consistency of the nugget in Gaussian spatial process models. Journal of the Royal Statistical Society: Series B (Statistical Methodology)\/ 83\/ (5), 1044--1070

    2021

  42. [50]

    Climatic covariates for the frequency analysis of heavy rainfall in the Mediterranean region

    Tramblay, Y., L. Neppel, and J. Carreau (2011, September). Brief communication " Climatic covariates for the frequency analysis of heavy rainfall in the Mediterranean region". Natural Hazards and Earth System Sciences\/ 11\/ (9), 2463--2468

    2011

  43. [51]

    Uijlenhoet, and A

    van Osnabrugge, B., R. Uijlenhoet, and A. Weerts (2019, March). Contribution of potential evaporation forecasts to 10-day streamflow forecast skill for the Rhine River . Hydrology and Earth System Sciences\/ 23\/ (3), 1453--1467

    2019

  44. [52]

    Gelman, and J

    Vehtari, A., A. Gelman, and J. Gabry (2017). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC . Statistics and Computing\/ 27\/ (5), 1413--1432

    2017

  45. [53]

    Mononen, V

    Vehtari, A., T. Mononen, V. Tolvanen, T. Sivula, and O. Winther (2016). Bayesian Leave - One - Out Cross - Validation Approximations for Gaussian Latent Variable Models . Journal of Machine Learning Research\/ 17\/ (103), 1--38

    2016

  46. [54]

    (1990, October)

    Werbos, P. (1990, October). Backpropagation through time: what it does and how to do it. Proceedings of the IEEE\/ 78\/ (10), 1550--1560

    1990

  47. [55]

    West, M. and J. Harrison (1997, February). Bayesian Forecasting and Dynamic Models ( Springer Series in Statistics ) . Springer-Verlag

    1997

  48. [56]

    Wikle, C. K. and M. B. Hooten (2010, November). A general science-based framework for dynamical spatio-temporal models. TEST\/ 19\/ (3), 417--451

    2010

  49. [57]

    Wolpert, D. H. (1992). Stacked generalization. Neural Networks\/ 5\/ (2), 241--259

    1992

  50. [58]

    Vehtari, D

    Yao, Y., A. Vehtari, D. Simpson, and A. Gelman (2018). Using Stacking to Average Bayesian Predictive Distributions (with Discussion ). Bayesian Analysis\/ 13\/ (3), 917--1007

    2018

  51. [59]

    Sainsbury-Dale, and R

    Zammit-Mangion, A., M. Sainsbury-Dale, and R. Huser (2024). Neural Methods for Amortised Inference

    2024

  52. [60]

    Sainsbury-Dale, and R

    Zammit-Mangion, A., M. Sainsbury-Dale, and R. Huser (2025, March). Neural Methods for Amortized Inference . Annual Review of Statistics and Its Application\/ 12\/ (Volume 12, 2025), 311--335

    2025

  53. [61]

    Zammit-Mangion, A. and C. K. Wikle (2020, June). Deep integro-difference equation models for spatio-temporal forecasting. Spatial Statistics\/ 37 , 100408

    2020

  54. [62]

    Zhang, H. (2004). Inconsistent estimation and asymptotically equal interpolations in model-based geostatistics. Journal of the American Statistical Association\/ 99\/ (465), 250--261

    2004

  55. [63]

    Zhang, L. and S. Banerjee (2022). Spatial factor modeling: A Bayesian matrix-normal approach for misaligned data. Biometrics\/ 78\/ (2), 560--573

    2022

  56. [64]

    Tang, and S

    Zhang, L., W. Tang, and S. Banerjee (2025). Bayesian geostatistics using predictive stacking. Journal of the American Statistical Association\/ (In press)

    2025

  57. [65]

    Grant, M. and S. Boyd (2008). Graph implementations for nonsmooth convex programs. In V. Blondel, S. Boyd, and H. Kimura (Eds.), Recent Advances in Learning and Control , Lecture Notes in Control and Information Sciences , pp.\ 95--110. Springer-Verlag Limited

    2008

  58. [66]

    Grant, M. C. (2005). Disciplined convex programming . PhD Thesis , Stanford University

    2005

  59. [67]

    Guhaniyogi, R. and S. Banerjee (2018). Meta- Kriging : Scalable Bayesian Modeling and Inference for Massive Spatial Datasets . Technometrics\/ 60\/ (4), 430--444

    2018

  60. [68]

    Snijders, T. A. B. (1988). On Cross - Validation for Predictor Evaluation in Time Series . In T. K. Dijkstra (Ed.), On Model Uncertainty and its Statistical Implications , Berlin, Heidelberg, pp.\ 56--69. Springer Berlin Heidelberg

    1988

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.