pith. sign in

arxiv: 2507.18488 · v2 · submitted 2025-07-24 · 📊 stat.ME

INLA-RF: A Hybrid Modeling Strategy for Spatio-Temporal Environmental Data

Mario Figueira , Michela Cameletti , Luca Patelli This is my paper

Pith reviewed 2026-05-19 02:37 UTC · model grok-4.3

classification 📊 stat.ME
keywords hybrid modelingspatio-temporal dataINLA-SPDErandom forestuncertainty quantificationenvironmental statisticsBayesian inferenceprediction
0
0 comments X

The pith

A hybrid INLA-RF strategy combines Bayesian spatio-temporal models with random forests to improve predictions while propagating uncertainty between stages.

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

The paper sets out to show that complex environmental data with non-linear patterns can be handled more effectively by linking the flexibility of random forests to the uncertainty-aware structure of INLA-SPDE models. It introduces two concrete integration methods that feed random-forest output into the Bayesian model either as an offset or as direct corrections to latent nodes. A Kullback-Leibler stopping rule keeps the two-stage process from drifting. Simulation results indicate that the resulting predictions are stronger than either component alone while the uncertainty estimates remain coherent and the model stays interpretable.

Core claim

The authors establish that INLA-RF1, which treats random-forest predictions as an offset inside the INLA-SPDE linear predictor, and INLA-RF2, which lets the forest directly adjust selected nodes of the latent field, both allow uncertainty to flow from the first stage to the second. A Kullback-Leibler divergence criterion decides when the iterative loop has converged. Two simulation studies demonstrate that these hybrids deliver higher predictive accuracy for spatio-temporal processes than standalone INLA-SPDE or random forest while preserving calibration of posterior intervals.

What carries the argument

The INLA-RF iterative two-stage framework, in which random-forest output is either inserted as an offset or used to correct latent-field nodes inside an INLA-SPDE model, with a Kullback-Leibler divergence stopping rule that controls iteration.

If this is right

  • Spatio-temporal environmental forecasts become more accurate without sacrificing the ability to report credible intervals.
  • Modelers can retain the interpretability of a Bayesian latent-field description while borrowing the flexibility of tree-based learners for non-linear effects.
  • The same two-stage structure can be applied to other Bayesian geostatistical models that admit offsets or node-wise corrections.
  • A Kullback-Leibler stopping rule provides an objective, data-driven way to terminate the hybrid iteration.

Where Pith is reading between the lines

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

  • The approach could be extended to real-world air-quality or climate datasets where ground-truth values are only partially observed.
  • Similar hybrids might be tested in domains such as epidemiology or ecology where both spatial structure and abrupt local changes matter.
  • If the uncertainty propagation holds, the method could serve as a drop-in replacement for pure machine-learning models that currently lack calibrated intervals.

Load-bearing premise

That feeding random-forest predictions into the INLA-SPDE model as an offset or as node corrections produces statistically valid uncertainty propagation between the two stages.

What would settle it

A simulation in which the hybrid model's 95 percent credible intervals cover the true held-out values at a rate materially below 95 percent would falsify the claim of coherent uncertainty quantification.

Figures

Figures reproduced from arXiv: 2507.18488 by Luca Patelli, Mario Figueira, Michela Cameletti.

Figure 1
Figure 1. Figure 1: Spatial mesh for the Paraná state and simulated data in the case of [PITH_FULL_IMAGE:figures/full_fig_p019_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Posterior distributions of the linear predictor for the [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Posterior distributions of the linear predictor for the [PITH_FULL_IMAGE:figures/full_fig_p022_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Simulated time series with 2000 time points and 10 jumps (highlighted with red [PITH_FULL_IMAGE:figures/full_fig_p023_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Results from the INLA base model: Panel A shows the temporal random effect, [PITH_FULL_IMAGE:figures/full_fig_p038_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Nodes of the temporal random effect identified as stress points to be corrected, [PITH_FULL_IMAGE:figures/full_fig_p039_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison between the corrected latent field from the INLA-RF2 model (top) [PITH_FULL_IMAGE:figures/full_fig_p039_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Latent field values at the 100 stress points, comparing the corrected estimates [PITH_FULL_IMAGE:figures/full_fig_p040_8.png] view at source ↗
read the original abstract

Environmental processes often exhibit complex, non-linear patterns and discontinuities across space and time, posing significant challenges for traditional geostatistical modeling approaches. In this paper, we propose a hybrid spatio-temporal modeling framework that combines the interpretability and uncertainty quantification of Bayesian models -- estimated using the INLA-SPDE approach -- with the predictive power and flexibility of Random Forest (RF). Specifically, we introduce two novel algorithms, collectively named INLA-RF, which integrate a statistical spatio-temporal model with RF in an iterative two-stage framework. The first algorithm (INLA-RF1) incorporates RF predictions as an offset in the INLA-SPDE model, while the second (INLA-RF2) uses RF to directly correct selected latent field nodes. Both hybrid strategies enable uncertainty propagation between modeling stages, an aspect often overlooked in existing hybrid approaches. In addition, we propose a Kullback-Leibler divergence-based stopping criterion. We evaluate the predictive performance and uncertainty quantification capabilities of the proposed algorithms through two simulation studies. Results suggest that our hybrid approach enhances spatio-temporal prediction while maintaining interpretability and coherence in uncertainty estimates.

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

2 major / 3 minor

Summary. The manuscript proposes INLA-RF, a hybrid spatio-temporal modeling framework that integrates the INLA-SPDE approach for Bayesian inference with Random Forest to capture complex non-linear patterns in environmental data. It defines two iterative algorithms: INLA-RF1, which inserts RF predictions as an offset into the INLA-SPDE linear predictor, and INLA-RF2, which overwrites selected nodes of the latent Gaussian field with RF corrections. A Kullback-Leibler divergence criterion is introduced to terminate the iteration. Performance is assessed via two simulation studies, with the central claim that the hybrids improve predictive accuracy while preserving interpretability and coherent uncertainty quantification.

Significance. The work targets a practical gap in spatio-temporal modeling where pure INLA-SPDE struggles with strong non-linearities and discontinuities. If the uncertainty-propagation mechanism can be shown to produce calibrated credible intervals, the approach would supply a usable compromise between the flexibility of tree-based methods and the structured uncertainty of Gaussian Markov random fields. The simulation-based evaluation and the explicit KL stopping rule are constructive elements that could be strengthened by coverage diagnostics.

major comments (2)
  1. [§3.1] §3.1 (INLA-RF1 algorithm): Treating RF point predictions as a fixed offset conditions the INLA posterior on those values without propagating RF variability into the linear predictor or hyperparameter posteriors. Standard INLA offset semantics treat the term as known; therefore the reported credible intervals are narrower than they would be under a joint model. The simulation studies must report empirical coverage of the nominal 95 % intervals under this construction to substantiate the coherence claim.
  2. [§3.2] §3.2 (INLA-RF2 algorithm): Directly replacing selected nodes of the latent field with RF values risks breaking the Markov property and the Gaussianity required by the SPDE approximation. The subsequent INLA step then computes marginals for an altered GMRF whose precision matrix is no longer consistent with the original mesh. A diagnostic (e.g., comparison of marginal variances before and after correction, or a small-scale joint-model benchmark) is needed to confirm that the final posteriors remain valid.
minor comments (3)
  1. [Simulation studies] The two simulation studies are referenced but their data-generating processes, mesh resolutions, and exact baseline comparators (pure INLA, pure RF, other hybrids) are not tabulated; adding a summary table of design parameters would improve reproducibility.
  2. [Figures] Figure captions should explicitly state whether plotted intervals are pointwise credible intervals or predictive intervals and should include the nominal coverage level used for calibration checks.
  3. [Stopping criterion] The KL-divergence stopping rule is introduced without a reference or derivation; a short appendix deriving its form from the iterative scheme would clarify its statistical justification.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which help clarify important aspects of uncertainty quantification in our hybrid framework. We respond to each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [§3.1] §3.1 (INLA-RF1 algorithm): Treating RF point predictions as a fixed offset conditions the INLA posterior on those values without propagating RF variability into the linear predictor or hyperparameter posteriors. Standard INLA offset semantics treat the term as known; therefore the reported credible intervals are narrower than they would be under a joint model. The simulation studies must report empirical coverage of the nominal 95 % intervals under this construction to substantiate the coherence claim.

    Authors: We agree that the use of RF predictions as a fixed offset in INLA-RF1 does not propagate RF variability into the INLA posterior or hyperparameters, and that the resulting credible intervals are therefore narrower than those from a fully joint model. The iterative structure provides limited feedback, but does not fully resolve this issue. To substantiate the coherence claim, we will add empirical coverage diagnostics for the nominal 95% intervals to the simulation studies in the revised manuscript. revision: yes

  2. Referee: [§3.2] §3.2 (INLA-RF2 algorithm): Directly replacing selected nodes of the latent field with RF values risks breaking the Markov property and the Gaussianity required by the SPDE approximation. The subsequent INLA step then computes marginals for an altered GMRF whose precision matrix is no longer consistent with the original mesh. A diagnostic (e.g., comparison of marginal variances before and after correction, or a small-scale joint-model benchmark) is needed to confirm that the final posteriors remain valid.

    Authors: We recognize that node replacement in INLA-RF2 can affect the Markov property and Gaussianity assumptions of the SPDE approximation, potentially altering the precision matrix. While the corrections are applied selectively, this remains an approximation. We will add the recommended diagnostics, including comparisons of marginal variances before and after correction, to the revised manuscript to confirm posterior validity. revision: yes

Circularity Check

0 steps flagged

No significant circularity: new hybrid integration method is self-contained

full rationale

The paper proposes two explicit algorithmic constructions (INLA-RF1 using RF predictions as offset; INLA-RF2 correcting selected latent nodes) inside an iterative loop with a KL-divergence stopping rule. These are presented as novel methodological choices built on the established INLA-SPDE and Random Forest frameworks rather than derived from equations that reduce to their own fitted parameters or prior self-citations. No step equates a claimed prediction or uncertainty-propagation property to a quantity defined in terms of itself; the uncertainty-coherence claim is an asserted property of the proposed integration, not a tautological renaming or fitted-input prediction. The derivation chain therefore remains independent of the target results.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Only the abstract is available, so the ledger is limited to high-level assumptions stated or implied there. No specific free parameters, invented entities, or detailed axioms are extractable beyond the core modeling premises.

axioms (2)
  • domain assumption INLA-SPDE provides interpretable Bayesian inference and uncertainty quantification for spatio-temporal processes
    Invoked as the base statistical component that the hybrid framework builds upon and preserves.
  • domain assumption Random Forest can effectively capture non-linear patterns and discontinuities in environmental data
    Used to justify why RF is integrated to complement the statistical model.

pith-pipeline@v0.9.0 · 5729 in / 1386 out tokens · 107287 ms · 2026-05-19T02:37:37.191027+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

45 extracted references · 45 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

    work page

  2. [2]

    Cressie , author C

    author N. Cressie , author C. K. Wikle , title Statistics for Spatio-Temporal data , publisher Wiley , year 2011

    work page 2011

  3. [3]

    author E. T. Krainski , author V. Gómez-Rubio , author H. Bakka , author A. Lenzi , author D. Castro-Camilo , author D. Simpson , author F. Lindgren , author H. Rue , title Advanced Spatial Modeling with Stochastic Partial Differential Equations Using R and INLA , publisher CRC-Press , year 2018

    work page 2018

  4. [4]

    author C. K. Wikle , author A. Zammit-Mangion , author N. Cressie , title Spatio-Temporal Statistics with R , publisher CRC-Press , year 2019

    work page 2019

  5. [5]

    and Lv, J

    author H. Rue , author S. Martino , author N. Chopin , title Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations , journal Journal of the Royal Statistical Society. Series B: Statistical Methodology volume 71 ( year 2009 ) pages 319--392 . :10.1111/j.1467-9868.2008.00700.x

    work page doi:10.1111/j.1467-9868.2008.00700.x 2009

  6. [6]

    Lindgren , author H

    author F. Lindgren , author H. Rue , author J. Lindström , title An explicit link between G aussian F ields and G aussian M arkov R andom F ields: the S tochastic P artial D ifferential E quation approach , journal J. Royal Stat. Soc.: Series B (Statistical Methodology) volume 73 ( year 2011 ) pages 423--498

    work page 2011

  7. [7]

    Pichler , author F

    author M. Pichler , author F. Hartig , title Machine learning and deep learning—a review for ecologists , journal Methods in Ecology and Evolution volume 14 ( year 2023 ) pages 994--1016 . :https://doi.org/10.1111/2041-210X.14061

    work page doi:10.1111/2041-210x.14061 2023

  8. [8]

    author C. K. Wikle , author A. Zammit-Mangion , title Statistical deep learning for spatial and spatiotemporal data , journal Annual Review of Statistics and Its Application volume 10 ( year 2023 ) pages 247--270

    work page 2023

  9. [9]

    author I. E. Agbehadji , author I. C. Obagbuwa , title Systematic review of machine learning and deep learning techniques for spatiotemporal air quality prediction , journal Atmosphere volume 15 ( year 2024 )

    work page 2024

  10. [10]

    author E. W. Fox , author J. M. Ver Hoef , author A. R. Olsen , title Comparing spatial regression to random forests for large environmental data sets , journal PLoS One volume 15 ( year 2020 ) pages e0229509

    work page 2020

  11. [11]

    author P. S. G. de Mattos Neto , author G. D. C. Cavalcanti , author D. S. de O. Santos Júnior , author J. A. de Souza , author F. Ren , title Hybrid systems using residual modeling for sea surface temperature forecasting , journal Scientific Reports volume 12 ( year 2022 ) pages 487 . https://doi.org/10.1038/s41598-021-04238-z. :10.1038/s41598-021-04238-z

    work page doi:10.1038/s41598-021-04238-z 2022

  12. [12]

    author D. P. Johnson , author N. Ravi , author G. Filippelli , author A. Heintzelman , title A Novel Hybrid Approach: Integrating Bayesian SPDE and Deep Learning for Enhanced Spatiotemporal Modeling of PM2.5 Concentrations in Urban Airsheds for Sustainable Climate Action and Public Health , journal Sustainability volume 16 ( year 2024 ). https://www.mdpi....

    work page 2024

  13. [13]

    Kakouri , author T

    author A. Kakouri , author T. Kontos , author G. Grivas , author G. Filippis , author M.-B. Korras-Carraca , author C. Matsoukas , author A. Gkikas , author E. Athanasopoulou , author O. Speyer , author C. Chatzidiakos , author E. Gerasopoulos , title Spatiotemporal modeling of long-term PM2.5 concentrations and population exposure in Greece, using machin...

    work page doi:10.1016/j.scitotenv.2024.178113 2025

  14. [14]

    Di , author H

    author Q. Di , author H. Amini , author L. Shi , author I. Kloog , author R. Silvern , author J. T. Kelly , author M. B. Sabath , author C. Choirat , author P. Koutrakis , author A. Lyapustin , et al., title An ensemble-based model of PM2.5 concentration across the contiguous United States with high spatiotemporal resolution , journal Environment Internat...

    work page 2019

  15. [15]

    author I. A. Gheyas , author L. S. Smith , title A novel neural network ensemble architecture for time series forecasting , journal Neurocomputing volume 74 ( year 2011 ) pages 3855--3864 . :https://doi.org/10.1016/j.neucom.2011.08.005

    work page doi:10.1016/j.neucom.2011.08.005 2011

  16. [16]

    Saad , author J

    author F. Saad , author J. Burnim , author C. Carroll , et al., title Scalable spatiotemporal prediction with bayesian neural fields , journal Nature Communications volume 15 ( year 2024 ) pages 7942 . https://doi.org/10.1038/s41467-024-51477-5. :10.1038/s41467-024-51477-5

    work page doi:10.1038/s41467-024-51477-5 2024

  17. [17]

    MacBride , author V

    author C. MacBride , author V. Davies , author D. Lee , title A spatial autoregressive random forest algorithm for small-area spatial prediction , journal The Annals of Applied Statistics volume 19 ( year 2025 ) pages 485 -- 504 . :10.1214/24-AOAS1969

    work page doi:10.1214/24-aoas1969 2025

  18. [18]

    big n problem

    author G. Jona Lasinio , author G. Mastrantonio , author A. Pollice , title Discussing the “big n problem” , journal Statistical Methods and Applications volume 22 ( year 2013 ) pages 97--112 . https://doi.org/10.1007/s10260-012-0207-2. :10.1007/s10260-012-0207-2

    work page doi:10.1007/s10260-012-0207-2 2013

  19. [19]

    Lindgren , author D

    author F. Lindgren , author D. Bolin , author H. Rue , title The SPDE approach for Gaussian and non-Gaussian fields: 10 years and still running , journal Spatial Statistics volume 50 ( year 2022 ). :https://doi.org/10.1016/j.spasta.2022.100599, note special Issue: The Impact of Spatial Statistics

    work page doi:10.1016/j.spasta.2022.100599 2022

  20. [20]

    Lindgren , author H

    author F. Lindgren , author H. Rue , title Bayesian Spatial Modelling with R - INLA , journal Journal of Statistical Software volume 63 ( year 2015 ) pages 1--25 . :10.18637/jss.v063.i19

    work page doi:10.18637/jss.v063.i19 2015

  21. [21]

    Bakka , author H

    author H. Bakka , author H. Rue , author G.-A. Fuglstad , author A. Riebler , author D. Bolin , author J. Illian , author E. Krainski , author D. Simpson , author F. Lindgren , title Spatial modeling with r-inla: A review , journal WIREs Computational Statistics volume 10 ( year 2018 ) pages e1443 . :https://doi.org/10.1002/wics.1443

    work page doi:10.1002/wics.1443 2018

  22. [22]

    Gómez-Rubio , title Bayesian Inference with INLA , publisher Chapman & Hall/CRC Press , year 2020

    author V. Gómez-Rubio , title Bayesian Inference with INLA , publisher Chapman & Hall/CRC Press , year 2020 . :10.1201/9781315175584

    work page doi:10.1201/9781315175584 2020

  23. [23]

    Van Niekerk , author H

    author J. Van Niekerk , author H. Rue , title Low-rank Variational Bayes correction to the Laplace method , journal Journal of Machine Learning Research volume 25 ( year 2024 ) pages 1--25

    work page 2024

  24. [24]

    Van Niekerk , author E

    author J. Van Niekerk , author E. Krainski , author D. Rustand , author H. Rue , title A new avenue for Bayesian inference with INLA , journal Computational Statistics & Data Analysis volume 181 ( year 2023 ) pages 107692 . :10.1016/j.csda.2023.107692

    work page doi:10.1016/j.csda.2023.107692 2023

  25. [25]

    Rue , author L

    author H. Rue , author L. Held , title Gaussian Markov Random Fields , publisher Chapman and Hall/CRC , year 2005 . :10.1201/9780203492024

    work page doi:10.1201/9780203492024 2005

  26. [26]

    Lindgren , author H

    author F. Lindgren , author H. Rue , author J. Lindström , title An explicit link between Gaussian fields and Gaussian Markov random fields: the stochastic partial differential equation approach , journal Journal of the Royal Statistical Society: Series B (Statistical Methodology) volume 73 ( year 2011 ) pages 423--498 . :https://doi.org/10.1111/j.1467-98...

    work page doi:10.1111/j.1467-9868.2011.00777.x 2011

  27. [27]

    Breiman , title Random forests , journal Machine learning volume 45 ( year 2001 ) pages 5--32

    author L. Breiman , title Random forests , journal Machine learning volume 45 ( year 2001 ) pages 5--32

    work page 2001

  28. [28]

    Sasi Kiran Gaddipati et al

    author B. Efron , title Bootstrap Methods: Another Look at the Jackknife , journal The Annals of Statistics volume 7 ( year 1979 ) pages 1 -- 26 . https://doi.org/10.1214/aos/1176344552. :10.1214/aos/1176344552

    work page doi:10.1214/aos/1176344552 1979

  29. [29]

    Breiman , author J

    author L. Breiman , author J. Friedman , author C. J. Stone , author R. Olshen , title Classification and regression tree analysis , publisher CRC Press , year 1984

    work page 1984

  30. [30]

    James , author D

    author G. James , author D. Witten , author T. Hastie , author R. Tibshirani , title An Introduction to Statistical Learning: with Applications in R , edition 2nd ed., publisher Springer US , year 2021

    work page 2021

  31. [31]

    Patelli , author M

    author L. Patelli , author M. Cameletti , author N. Golini , author R. Ignaccolo , title A Path in Regression Random Forest Looking for Spatial Dependence: A Taxonomy and a Systematic Review , publisher Springer Nature Switzerland , address Cham , year 2024 , pp. pages 467--489 . :10.1007/978-3-031-69111-9_23

    work page doi:10.1007/978-3-031-69111-9_23 2024

  32. [32]

    Saha , author S

    author A. Saha , author S. Basu , author A. Datta , title Random forests for spatially dependent data , journal Journal of the American Statistical Association volume 118 ( year 2023 ) pages 665--683

    work page 2023

  33. [33]

    author M. J. Heaton , author A. Millane , author J. S. Rhodes , title Adjusting for spatial correlation in machine and deep learning , year 2024 . https://arxiv.org/abs/2410.04312. http://arxiv.org/abs/2410.04312 arXiv:2410.04312

    work page arXiv 2024

  34. [34]

    Kullback , title Information Theory and Statistics , A Wiley publication in mathematical statistics, publisher Dover Publications , year 1997

    author S. Kullback , title Information Theory and Statistics , A Wiley publication in mathematical statistics, publisher Dover Publications , year 1997

    work page 1997

  35. [35]

    Cameletti , author F

    author M. Cameletti , author F. Lindgren , author D. Simpson , author H. Rue , title Spatio-temporal modeling of particulate matter concentration through the SPDE approach , journal AStA Advances in Statistical Analysis volume 97 ( year 2013 ) pages 109--131 . https://doi.org/10.1007/s10182-012-0196-3. :10.1007/s10182-012-0196-3

    work page doi:10.1007/s10182-012-0196-3 2013

  36. [36]

    Moraga , title Geospatial Health Data: Modeling and Visualization with R‑INLA and Shiny , publisher Chapman & Hall/CRC , year 2019

    author P. Moraga , title Geospatial Health Data: Modeling and Visualization with R‑INLA and Shiny , publisher Chapman & Hall/CRC , year 2019

    work page 2019

  37. [37]

    Fioravanti , author S

    author G. Fioravanti , author S. Martino , author M. Cameletti , author G. Cattani , title Spatio-temporal modelling of PM10 daily concentrations in Italy using the SPDE approach , journal Atmospheric Environment volume 248 ( year 2021 ) pages 118192

    work page 2021

  38. [38]

    Fioravanti , author M

    author G. Fioravanti , author M. Cameletti , author S. Martino , author G. Cattani , author E. Pisoni , title A spatiotemporal analysis of NO concentrations during the Italian 2020 COVID-19 lockdown , journal Environmetrics volume 33 ( year 2022 ) pages e2723

    work page 2020

  39. [39]

    Fioravanti , author S

    author G. Fioravanti , author S. Martino , author M. Cameletti , author A. Toreti , title Interpolating climate variables by using INLA and the SPDE approach , journal International Journal of Climatology volume 43 ( year 2023 ) pages 6866--6886

    work page 2023

  40. [40]

    Otto , author A

    author P. Otto , author A. F. Moro , author J. Rodeschini , author Q. Shaboviq , author R. Ignaccolo , author N. Golini , author M. Cameletti , author P. Maranzano , author F. Finazzi , author A. Fass\`o , title Spatiotemporal modelling of PM _ 2.5 concentrations in Lombardy (Italy): a comparative study , journal Environmental and Ecological Statistics vo...

    work page doi:10.1007/s10651-023-00589-0 2024

  41. [41]

    Simpson , author H

    author D. Simpson , author H. Rue , author A. Riebler , author T. G. Martins , author S. H. Sørbye , title Penalising model component complexity: A principled, practical approach to constructing priors , journal Statistical Science volume 32 ( year 2017 ) pages 1--28 . :10.1214/16-STS576

    work page doi:10.1214/16-sts576 2017

  42. [42]

    Journal of the American Statistical Association , author =

    author G.-A. Fuglstad , author D. Simpson , author F. Lindgren , author H. Rue , title Constructing Priors that Penalize the Complexity of Gaussian Random Fields , journal Journal of the American Statistical Association volume 114 ( year 2019 ) pages 445--452 . :10.1080/01621459.2017.1415907

    work page doi:10.1080/01621459.2017.1415907 2019

  43. [43]

    author D. R. Roberts , author V. Bahn , author S. Ciuti , author M. S. Boyce , author J. Elith , author G. Guillera‐Arroita , author S. Hauenstein , author J. J. Lahoz‐Monfort , author B. Schröder , author W. Thuiller , author D. I. Warton , author B. A. Wintle , author F. Hartig , author C. F. Dormann , title Cross‐validation strategies for data with tem...

    work page doi:10.1111/ecog.02881 2017

  44. [44]

    Meyer , author C

    author H. Meyer , author C. Reudenbach , author T. Hengl , author M. Katurji , author T. Nauss , title Improving performance of spatio-temporal machine learning models using forward feature selection and target-oriented validation , journal Environmental Modelling & Software volume 101 ( year 2018 ) pages 1--9 . :10.1016/j.envsoft.2017.12.001, note publis...

    work page doi:10.1016/j.envsoft.2017.12.001 2018

  45. [45]

    note Large language model

    author OpenAI , title Chatgpt (gpt-4-turbo, accessed june 2025) , howpublished https://chat.openai.com , year 2025 . note Large language model

    work page 2025