pith. machine review for the scientific record. sign in

arxiv: 2604.18823 · v1 · submitted 2026-04-20 · 📊 stat.AP

Recognition: unknown

A Non-stationary, Amortized, Transfer Learning Approach for Modeling Italian Air Quality

Alessandro Fusta Moro , Antony Sikorski , Daniel McKenzie , Alessandro Fass\`o , Douglas Nychka
Authors on Pith no claims yet

Pith reviewed 2026-05-10 02:51 UTC · model grok-4.3

classification 📊 stat.AP
keywords air quality modelingspatial statisticstransfer learningnon-stationary processesNO2 concentrationsLatticeKrigchemical transport modelsItaly
0
0 comments X

The pith

A neural network learns non-stationary correlation structures from gridded models to improve air quality predictions from sparse stations.

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

The paper develops a spatial transfer learning method that combines full-coverage but smoothed Chemical Transport Model outputs with sparse but accurate ground station measurements of NO2 in Italy. It uses a neural architecture to estimate a non-stationary and anisotropic correlation structure from the gridded data in seconds, then transfers this structure to a basis function representation suitable for point data. This allows daily fine-scale predictions across the country with uncertainty estimates that better reflect Italy's varied geography than traditional stationary models. The approach matters because accurate air quality maps support public health decisions and environmental policy.

Core claim

The authors present a framework that first trains an image-to-image neural network on gridded CTM outputs to learn millions of spatially varying parameters defining a nonstationary anisotropic covariance. This covariance is then embedded into the LatticeKrig basis-function model and projected onto a finer grid while transferring to the change-of-support point data from stations. A final likelihood refinement adjusts the correlation range to capture fine-scale variability, resulting in predictions that outperform stationary alternatives on both data types.

What carries the argument

An image-to-image neural network that amortizes the estimation of a spatially varying precision matrix for the LatticeKrig basis functions, enabling transfer of the non-stationary correlation structure across data sources with different supports.

If this is right

  • Enables production of daily NO2 concentration maps on a fine grid with associated uncertainty.
  • Accommodates the complex, non-stationary spatial patterns induced by Italy's geography and terrain.
  • Demonstrates improved accuracy over stationary geostatistical models when validated on held-out station data.
  • Provides a scalable way to integrate gridded simulation outputs with point observations without requiring full re-estimation of the covariance.

Where Pith is reading between the lines

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

  • The amortized neural estimation could be extended to incorporate temporal dynamics for forecasting future air quality.
  • Similar transfer learning might apply to other environmental variables where high-resolution simulations exist alongside sparse observations.
  • The method's speed in learning the correlation structure suggests potential for near-real-time updating of maps as new data becomes available.

Load-bearing premise

That the correlation structure estimated from the smoothed CTM grids can be directly transferred to the finer-scale station data via the basis projection without substantial mismatch or bias.

What would settle it

Observing that a stationary model achieves equivalent or better predictive performance on cross-validated station measurements than the transferred non-stationary version would indicate that the non-stationarity transfer does not provide the claimed benefit.

Figures

Figures reproduced from arXiv: 2604.18823 by Alessandro Fass\`o, Alessandro Fusta Moro, Antony Sikorski, Daniel McKenzie, Douglas Nychka.

Figure 1
Figure 1. Figure 1: From the left: NO2 concentrations from ground monitoring stations, NO2 con￾centrations from chemical transport models, temperature from reanalysis (ERA5 datasets), NOx emission fluxes from CAMS-REG-ANT. Data for Italy, on 13th April 2023. 8 [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: 2023 annual averages of CTM-based parameter estimates. Top row from left to [PITH_FULL_IMAGE:figures/full_fig_p020_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: CTM NO2 reconstruction experiment for April 13, 2023. From left to right: the actual CTM NO2 field, masked data used for model fitting (pixels at locations with stations), the full prediction from the non-stationary model, and the difference between the true and predicted field. anisotropic LatticeKrig model, again using the same mean structure and CTM NO2 data, with covariance structure governed by the ST… view at source ↗
Figure 4
Figure 4. Figure 4: Left: time series of daily RMSE percent differences ((nonstationary / stationary [PITH_FULL_IMAGE:figures/full_fig_p022_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Differences between the range parameters estimated by the stationary model [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Estimated spatial component on November 1, 2023. Left: stationary model. [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Fine-grid predictions for November 1, 2023. Left: posterior mean NO [PITH_FULL_IMAGE:figures/full_fig_p027_7.png] view at source ↗
read the original abstract

Air quality monitoring in Italy relies on sparse, irregular, ground-based stations that provide high-quality but incomplete measurements of pollution. Chemical transport models (CTMs) offer full spatial and temporal coverage but smooth over local variability. We develop a spatial transfer-learning framework that integrates these two data sources to produce daily, fine-grid predictions of nitrogen dioxide (NO$_2$) concentrations across Italy for 2023, with uncertainty quantification. The resulting maps provide a resource for decision making in downstream applications such as epidemiology and environmental policy. Our approach builds on the geostatistical LatticeKrig framework, which uses compactly supported basis functions and coefficients governed by a sparse precision matrix. We learn a nonstationary, anisotropic correlation structure from the gridded CTM outputs using an image-to-image neural architecture that estimates millions of spatially varying parameters in a matter of seconds. The basis-function representation enables this covariance structure to be transferred to the point-level station data and projected onto a finer prediction grid, a key extension for handling the change of support between data sources. A likelihood-based refinement step then adjusts the correlation range to recover fine-scale variability smoothed out by the gridded data. The proposed methodology results in a flexible, non-stationary, and anisotropic representation of the spatial process, better accommodating the complex geography of Italy. Performance is assessed through experiments on both gridded CTM outputs and point-level station measurements, demonstrating improvements over the stationary formulation.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The manuscript presents a non-stationary, amortized transfer-learning framework that extends the LatticeKrig model to integrate gridded chemical transport model (CTM) outputs with sparse point-level station measurements for daily NO2 concentration mapping across Italy in 2023. An image-to-image neural network learns millions of spatially varying parameters to capture anisotropic, non-stationary correlation structure from the CTM grids; this structure is projected onto irregular station locations and a finer prediction grid via compactly supported basis functions, followed by a likelihood-based refinement of the global correlation range to recover fine-scale variability, with uncertainty quantification. The central claim is that this yields a flexible representation better suited to Italy's complex geography and demonstrates improvements over stationary formulations in experiments on both gridded and point data.

Significance. If the transfer and refinement steps hold under scrutiny, the work offers a scalable, computationally efficient route to non-stationary spatial prediction that amortizes covariance estimation from large grids while explicitly handling change-of-support; this could be useful for environmental statistics applications requiring fine-scale maps with uncertainty, such as epidemiology or policy support. The combination of neural amortization with LatticeKrig basis projection is a technically interesting extension.

major comments (3)
  1. [Abstract] Abstract: the assertion that experiments 'demonstrate improvements over the stationary formulation' is unsupported by any quantitative metrics, error bars, cross-validation scores, or specific comparison results, which is load-bearing for the central performance claim.
  2. [§3] §3 (transfer and projection step): the non-stationary anisotropic structure learned from CTM grids is projected onto station locations via the compactly supported basis functions, but no quantitative diagnostic (e.g., comparison of projected local ranges or principal axes against empirical variograms computed directly from station data) is described to verify that CTM smoothing has not distorted the fine-scale structure; this assumption is central to the transfer-learning validity.
  3. [§4] §4 (experiments on station data): the likelihood refinement step tunes only a global range parameter after transfer; without reported sensitivity checks or hold-out validation showing that the spatially varying components remain appropriate at point support, it is unclear whether fine-scale variability is recovered without bias or overfitting across Italy's terrain.
minor comments (2)
  1. [Abstract] Abstract: the phrase 'amortized' is used in the title and text but is not explicitly defined in terms of what computation is being amortized (neural network weights versus per-prediction inference).
  2. [§2] Notation: the description of the sparse precision matrix and its relation to the neural-network outputs would benefit from an early equation reference to avoid ambiguity when the basis-function projection is introduced.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their insightful and constructive comments, which have helped us identify areas where the manuscript can be strengthened. We provide point-by-point responses to the major comments below, along with our plans for revision.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion that experiments 'demonstrate improvements over the stationary formulation' is unsupported by any quantitative metrics, error bars, cross-validation scores, or specific comparison results, which is load-bearing for the central performance claim.

    Authors: We agree that the abstract would be improved by including specific quantitative support for the performance claim. While Section 4 of the manuscript reports cross-validation results, RMSE reductions, and log-likelihood improvements from the station data experiments, these details were not summarized numerically in the abstract. We will revise the abstract to incorporate key quantitative metrics, such as the observed RMSE improvement and cross-validation scores, to make the claim fully supported. revision: yes

  2. Referee: [§3] §3 (transfer and projection step): the non-stationary anisotropic structure learned from CTM grids is projected onto station locations via the compactly supported basis functions, but no quantitative diagnostic (e.g., comparison of projected local ranges or principal axes against empirical variograms computed directly from station data) is described to verify that CTM smoothing has not distorted the fine-scale structure; this assumption is central to the transfer-learning validity.

    Authors: This is a fair observation on the need for explicit validation of the transfer step. Section 3 describes the projection of the learned non-stationary structure onto station locations using the LatticeKrig basis functions, but does not include direct quantitative comparisons to station-derived empirical variograms. In the revision we will add such diagnostics, for example by comparing the projected local ranges and anisotropy axes at station sites against variograms estimated from station residuals, to confirm that the transferred structure preserves relevant fine-scale features. revision: yes

  3. Referee: [§4] §4 (experiments on station data): the likelihood refinement step tunes only a global range parameter after transfer; without reported sensitivity checks or hold-out validation showing that the spatially varying components remain appropriate at point support, it is unclear whether fine-scale variability is recovered without bias or overfitting across Italy's terrain.

    Authors: We appreciate the referee drawing attention to the validation of the refinement step. The current procedure in Section 4 fixes the spatially varying parameters transferred from the CTM and optimizes only the global range parameter. To address concerns about suitability at point support, we will expand the experiments in the revised manuscript to include sensitivity analyses (e.g., perturbing the fixed parameters within plausible ranges) and additional hold-out validation on withheld stations, stratified by terrain type, to demonstrate that fine-scale variability is recovered without systematic bias or overfitting. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained

full rationale

The paper trains an image-to-image network on external CTM gridded outputs to estimate spatially varying non-stationary parameters, projects the resulting covariance via LatticeKrig compactly supported basis functions onto irregular station locations, and performs a separate likelihood refinement on the point data to adjust the global range. None of these steps reduce the final predictions or uncertainty quantification to the inputs by construction; the neural network outputs are not redefined in terms of the station likelihood, and the basis projection is a fixed linear operation independent of the target variable. Prior LatticeKrig work is invoked for the basis representation but is not a self-citation load-bearing the central claim, as the non-stationary structure itself is learned from CTM data and validated on held-out station measurements. Experiments explicitly compare against the stationary baseline on both data types, confirming independent content.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on the transferability of CTM-derived non-stationary structure to station data and the ability of likelihood refinement to recover fine-scale signal; these are domain assumptions rather than derived results.

free parameters (2)
  • neural network weights
    Millions of spatially varying parameters estimated from CTM image-to-image training; fitted to gridded data.
  • correlation range adjustment
    Likelihood-based scalar or local adjustment applied after transfer to match station variability.
axioms (2)
  • domain assumption CTM-derived correlation structure is transferable to station data via basis-function projection without introducing systematic bias from smoothing or change of support.
    Invoked in the transfer step and refinement; central to the claim of improved fine-scale predictions.
  • standard math The LatticeKrig sparse precision matrix representation remains valid when the correlation parameters are replaced by the neural-network output.
    Background assumption from the cited LatticeKrig framework.

pith-pipeline@v0.9.0 · 5569 in / 1511 out tokens · 45312 ms · 2026-05-10T02:51:53.784967+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

69 extracted references · 7 canonical work pages · 3 internal anchors

  1. [1]

    PyTorch: An Imperative Style, High-Performance Deep Learning Library

    Pytorch: An imperative style, high-performance deep learning library , author=. arXiv preprint arXiv:1912.01703 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 1912

  2. [2]

    Decoupled Weight Decay Regularization

    Decoupled weight decay regularization , author=. arXiv preprint arXiv:1711.05101 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  3. [3]

    Chen, Jieneng and Lu, Yongyi and Yu, Qihang and Luo, Xiangde and Adeli, Ehsan and Wang, Yan and Lu, Le and Yuille, Alan L and Zhou, Yuyin , journal=

  4. [4]

    Geostatistics for Large Datasets , isbn =

    Sun, Ying and Li, Bo and Genton, Marc , year =. Geostatistics for Large Datasets , isbn =

  5. [5]

    Journal of the Korean Statistical Society , volume=

    A modeling approach for large spatial datasets , author=. Journal of the Korean Statistical Society , volume=. 2008 , publisher=

    2008

  6. [6]

    Environmetrics: The official journal of the International Environmetrics Society , volume=

    Spatial modelling using a new class of nonstationary covariance functions , author=. Environmetrics: The official journal of the International Environmetrics Society , volume=. 2006 , publisher=

    2006

  7. [7]

    Models-3 community multiscale air quality (

    Binkowski, Francis S and Roselle, Shawn J , journal=. Models-3 community multiscale air quality (. 2003 , publisher=

    2003

  8. [8]

    Computational Statistics & Data Analysis , volume=

    Neural networks for parameter estimation in intractable models , author=. Computational Statistics & Data Analysis , volume=. 2023 , publisher=

    2023

  9. [9]

    Banesh, Divya and Panda, Nishant and Biswas, Ayan and Van Roekel, Luke and Oyen, Diane and Urban, Nathan and Grosskopf, Michael and Wolfe, Jonathan and Lawrence, Earl , booktitle=. Fast. 2021 , organization=

    2021

  10. [10]

    Task-agnostic amortized inference of

    Liu, Sulin and Sun, Xingyuan and Ramadge, Peter J and Adams, Ryan P , journal=. Task-agnostic amortized inference of

  11. [11]

    Annual Review of Statistics and Its Application , volume=

    Neural methods for amortized inference , author=. Annual Review of Statistics and Its Application , volume=. 2024 , publisher=

    2024

  12. [12]

    Environmental and Ecological Statistics , volume=

    A non-stationary spatial model of PM2.5 with localized transfer learning from numerical model output , author=. Environmental and Ecological Statistics , volume=. 2026 , publisher=

    2026

  13. [13]

    LatticeVision: Image to Image Networks for Modeling Non-Stationary Spatial Data

    LatticeVision: Image to Image Networks for Modeling Non-Stationary Spatial Data , author=. arXiv preprint arXiv:2505.09803 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    A spatiotemporal analysis of

    Fioravanti, Guido and Cameletti, Michela and Martino, Sara and Cattani, Giorgio and Pisoni, Enrico , journal=. A spatiotemporal analysis of. 2022 , publisher=

    2022

  15. [15]

    A multiresolution

    Nychka, Douglas and Bandyopadhyay, Soutir and Hammerling, Dorit and Lindgren, Finn and Sain, Stephan , journal=. A multiresolution. 2015 , publisher=

    2015

  16. [16]

    Package ‘

    Nychka, Douglas and Hammerling, Dorit and Sain, Stephan and Lenssen, Nathan and Nychka, Maintainer Douglas , year=. Package ‘

  17. [17]

    An explicit link between

    Lindgren, Finn and Rue, H. An explicit link between. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2011 , publisher=

    2011

  18. [18]

    Multivariate low-rank state-space model with

    Rodeschini, Jacopo and Tedesco, Lorenzo and Finazzi, Francesco and Otto, Philipp and Fass. Multivariate low-rank state-space model with. Spatial Statistics , pages=. 2026 , publisher=

    2026

  19. [19]

    Lindgren, Finn and Bolin, David and Rue, H. The. Spatial Statistics , volume=. 2022 , publisher=

    2022

  20. [20]

    2013 , publisher=

    Spatial variation , author=. 2013 , publisher=

    2013

  21. [21]

    Biometrika , pages=

    On stationary processes in the plane , author=. Biometrika , pages=. 1954 , publisher=

    1954

  22. [22]

    Advances and challenges in space-time modelling of natural events , pages=

    Geostatistics for large datasets , author=. Advances and challenges in space-time modelling of natural events , pages=. 2011 , publisher=

    2011

  23. [23]

    Modeling spatial data using local likelihood estimation and a

    Wiens, Ashton and Nychka, Douglas and Kleiber, William , journal=. Modeling spatial data using local likelihood estimation and a. 2020 , publisher=

    2020

  24. [24]

    arXiv preprint arXiv:2508.20067 , year=

    Neural Conditional Simulation for Complex Spatial Processes , author=. arXiv preprint arXiv:2508.20067 , year=

    work page arXiv

  25. [25]

    Environmetrics , volume=

    Fast parameter estimation of generalized extreme value distribution using neural networks , author=. Environmetrics , volume=. 2024 , publisher=

    2024

  26. [26]

    Fast covariance parameter estimation of spatial

    Gerber, Florian and Nychka, Douglas , journal=. Fast covariance parameter estimation of spatial. 2021 , publisher=

    2021

  27. [27]

    Likelihood-free parameter estimation with neural

    Sainsbury-Dale, Matthew and Zammit-Mangion, Andrew and Huser, Rapha. Likelihood-free parameter estimation with neural. The American Statistician , volume=. 2024 , publisher=

    2024

  28. [28]

    Modeling Spatial Extremes using Non-

    Rai, Sweta and Nychka, Douglas W and Bandyopadhyay, Soutir , journal=. Modeling Spatial Extremes using Non-

  29. [29]

    Advances in neural information processing systems , volume=

    Deep sets , author=. Advances in neural information processing systems , volume=

  30. [30]

    Forlani, Chiara and Bhatt, Samir and Cameletti, Michela and Krainski, Elias and Blangiardo, Marta , journal=. A joint. 2020 , publisher=

    2020

  31. [31]

    , title =

    Chang, Howard H. , title =. Handbook of Spatial Epidemiology , editor =

  32. [32]

    Biometrics , volume=

    Space-time data fusion under error in computer model output: an application to modeling air quality , author=. Biometrics , volume=. 2012 , publisher=

    2012

  33. [33]

    Journal of agricultural, biological, and environmental statistics , volume=

    A spatio-temporal downscaler for output from numerical models , author=. Journal of agricultural, biological, and environmental statistics , volume=. 2010 , publisher=

    2010

  34. [34]

    , title =

    Berrocal, Veronica J. , title =. Handbook of Environmental and Ecological Statistics , editor =

  35. [35]

    Model evaluation and spatial interpolation by

    Fuentes, Montserrat and Raftery, Adrian E , journal=. Model evaluation and spatial interpolation by. 2005 , publisher=

    2005

  36. [36]

    Journal of the American Statistical Association , volume=

    Combining incompatible spatial data , author=. Journal of the American Statistical Association , volume=. 2002 , publisher=

    2002

  37. [37]

    Estimating surface

    Shetty, Shobitha and Schneider, Philipp and Stebel, Kerstin and Hamer, Paul David and Kylling, Arve and Berntsen, Terje Koren , journal=. Estimating surface. 2024 , publisher=

    2024

  38. [38]

    Inferring surface

    Sun, Wenfu and Tack, Frederik and Clarisse, Lieven and Schneider, Rochelle and Stavrakou, Trissevgeni and Van Roozendael, Michel , journal=. Inferring surface. 2024 , publisher=

    2024

  39. [39]

    Deep learning estimation of daily ground-level

    Ghahremanloo, Masoud and Lops, Yannic and Choi, Yunsoo and Yeganeh, Bijan , journal=. Deep learning estimation of daily ground-level. 2021 , publisher=

    2021

  40. [40]

    Estimation of daily

    Stafoggia, Massimo and Bellander, Tom and Bucci, Simone and Davoli, Marina and De Hoogh, Kees and De'Donato, Francesca and Gariazzo, Claudio and Lyapustin, Alexei and Michelozzi, Paola and Renzi, Matteo and others , journal=. Estimation of daily. 2019 , publisher=

    2019

  41. [41]

    Journal of Geophysical Research: Machine Learning and Computation , volume=

    Air quality estimation and forecasting via data fusion with uncertainty quantification: theoretical framework and preliminary results , author=. Journal of Geophysical Research: Machine Learning and Computation , volume=. 2024 , publisher=

    2024

  42. [42]

    2015 , publisher=

    Statistics for spatial data , author=. 2015 , publisher=

    2015

  43. [43]

    Journal of the American Statistical Association , volume=

    Spatial modeling with spatially varying coefficient processes , author=. Journal of the American Statistical Association , volume=. 2003 , publisher=

    2003

  44. [44]

    2003 , publisher=

    Hierarchical modeling and analysis for spatial data , author=. 2003 , publisher=

    2003

  45. [45]

    Journal of Computational and Graphical Statistics , volume=

    Covariance tapering for interpolation of large spatial datasets , author=. Journal of Computational and Graphical Statistics , volume=. 2006 , publisher=

    2006

  46. [46]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Estimation and model identification for continuous spatial processes , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 1988 , publisher=

    1988

  47. [47]

    Journal of the American Statistical Association , volume=

    Approximate likelihood for large irregularly spaced spatial data , author=. Journal of the American Statistical Association , volume=. 2007 , publisher=

    2007

  48. [48]

    Covariance tapering for multivariate

    Bevilacqua, Moreno and Fass. Covariance tapering for multivariate. Statistical Methods & Applications , volume=. 2016 , publisher=

    2016

  49. [49]

    The Dataset of Daily Air Quality for the Years 2013-2023 in

    Alessandro. The Dataset of Daily Air Quality for the Years 2013-2023 in. 2026 , eprint=

    2013

  50. [50]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Fixed rank kriging for very large spatial data sets , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2008 , publisher=

    2008

  51. [51]

    Journal of agricultural, biological and environmental Statistics , volume=

    A case study competition among methods for analyzing large spatial data , author=. Journal of agricultural, biological and environmental Statistics , volume=. 2019 , publisher=

    2019

  52. [52]

    Hierarchical nearest-neighbor

    Datta, Abhirup and Banerjee, Sudipto and Finley, Andrew O and Gelfand, Alan E , journal=. Hierarchical nearest-neighbor. 2016 , publisher=

    2016

  53. [53]

    doi:10.24381/7cc0465a , url =

    2021 , publisher =. doi:10.24381/7cc0465a , url =

    work page doi:10.24381/7cc0465a 2021

  54. [54]

    Kuenen, J. J. P. and Dellaert, S. and Visschedijk, A. and Jalkanen, J.-P. and Super, I. and Denier van der Gon, H. A. C. , title =. Earth System Science Data , year =

  55. [55]

    Earth System Science Data , year =

    Guevara, Marc and Jorba, Oriol and Tena, Carlos and. Earth System Science Data , year =. doi:10.5194/essd-13-367-2021 , url =

    work page doi:10.5194/essd-13-367-2021 2021

  56. [56]

    Earth system science data , volume=

    Mu. Earth system science data , volume=. 2021 , publisher=

    2021

  57. [57]

    Copernicus climate change service (c3s) climate data store (cds) , volume=

    Hersbach, Hans and Bell, Bill and Berrisford, Paul and Biavati, Gionata and Hor. Copernicus climate change service (c3s) climate data store (cds) , volume=

  58. [58]

    Spatiotemporal modelling of

    Otto, Philipp and Fusta Moro, Alessandro and Rodeschini, Jacopo and Shaboviq, Qendrim and Ignaccolo, Rosaria and Golini, Natalia and Cameletti, Michela and Maranzano, Paolo and Finazzi, Francesco and Fass. Spatiotemporal modelling of. Environmental and Ecological Statistics , volume=. 2024 , publisher=

    2024

  59. [59]

    Agrimonia: a dataset on livestock, meteorology and air quality in the

    Fass. Agrimonia: a dataset on livestock, meteorology and air quality in the. Scientific Data , volume=. 2023 , publisher=

    2023

  60. [60]

    Air pollution removal by green infrastructures and urban forests in the city of

    Bottalico, Francesca and Chirici, Gherardo and Giannetti, Francesca and De Marco, Alessandra and Nocentini, Susanna and Paoletti, Elena and Salbitano, Fabio and Sanesi, Giovanni and Serenelli, Chiara and Travaglini, Davide , journal=. Air pollution removal by green infrastructures and urban forests in the city of. 2016 , publisher=

    2016

  61. [61]

    Machine learning reveals that prolonged exposure to air pollution is associated with

    Gatti, Roberto Cazzolla and Velichevskaya, Alena and Tateo, Andrea and Amoroso, Nicola and Monaco, Alfonso , journal=. Machine learning reveals that prolonged exposure to air pollution is associated with. 2020 , publisher=

    2020

  62. [62]

    Air pollution impact on pregnancy outcomes in

    Capobussi, Matteo and Tettamanti, Roberto and Marcolin, Luca and Piovesan, Luca and Bronzin, Silvia and Gattoni, Maria Elena and Polloni, Ilaria and Sabatino, Giuliana and Tersalvi, Carlo A and Auxilia, Francesco and others , journal=. Air pollution impact on pregnancy outcomes in. 2016 , publisher=

    2016

  63. [63]

    The association between air pollution and the incidence of idiopathic pulmonary fibrosis in

    Conti, Sara and Harari, Sergio and Caminati, Antonella and Zanobetti, Antonella and Schwartz, Joel D and Bertazzi, Pietro A and Cesana, Giancarlo and Madotto, Fabiana , journal=. The association between air pollution and the incidence of idiopathic pulmonary fibrosis in. 2018 , publisher=

    2018

  64. [64]

    Short-term exposure to particulate matter (

    Orellano, Pablo and Reynoso, Julieta and Quaranta, Nancy and Bardach, Ariel and Ciapponi, Agustin , journal=. Short-term exposure to particulate matter (. 2020 , publisher=

    2020

  65. [65]

    European Respiratory Journal , volume=

    Nitrogen dioxide and mortality: review and meta-analysis of long-term studies , author=. European Respiratory Journal , volume=. 2014 , publisher=

    2014

  66. [66]

    The effects of air pollution on

    Coker, Eric S and Cavalli, Laura and Fabrizi, Enrico and Guastella, Gianni and Lippo, Enrico and Parisi, Maria Laura and Pontarollo, Nicola and Rizzati, Massimiliano and Varacca, Alessandro and Vergalli, Sergio , journal=. The effects of air pollution on. 2020 , publisher=

    2020

  67. [67]

    Multivariate Low-Rank State-Space Model with

    Rodeschini, Jacopo and Tedesco, Lorenzo and Finazzi, Francesco and Otto, Philipp and Fass. Multivariate Low-Rank State-Space Model with. arXiv preprint arXiv:2509.12825 , year=

    work page arXiv

  68. [68]

    Maximum likelihood estimation of the multivariate hidden dynamic geostatistical model with application to air quality in

    Calculli, Crescenza and Fass. Maximum likelihood estimation of the multivariate hidden dynamic geostatistical model with application to air quality in. Environmetrics , volume=. 2015 , publisher=

    2015

  69. [69]

    Stochastic environmental research and risk assessment , volume=

    Kriging with external drift for functional data for air quality monitoring , author=. Stochastic environmental research and risk assessment , volume=. 2014 , publisher=

    2014