EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Learning spatial transmission maps from sparse outbreak snapshots with diffusion models

Andrew Baggaley iD

Journal of the Royal Society Interface · 04.11.2026 · 23 (244)

DOI: 10.1098/rsif.2026.0413 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of the Royal Society Interface, online erschienen am 11.11.2026. Verfasst von Andrew Baggaley. Beteiligte Einrichtung: School of Mathematical Sciences, Lancaster University , Lancaster, UK. Hauptthemen laut MEDLINE: Disease Outbreaks, Models, Biological, Plant Diseases. Die Arbeit nennt 29 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Research England.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Spatial heterogeneity in disease transmission is a key driver of epidemic dynamics, yet inferring spatially varying transmission maps from partial outbreak observations remains challenging. Standard mechanistic models often assume homogeneous transmission, while data-driven approaches struggle with the sparse, stochastic nature of surveillance data. We demonstrate that conditional diffusion models provide an effective framework for solving this inverse problem in spatial epidemiology. Using a spatially explicit stochastic SIR model on a lattice with heterogeneous transmission fields, we train a conditional elucidated diffusion model to generate conditional samples of the transmission landscape from sparse infection snapshots at four time points. We also develop an approach to accommodate observation uncertainty through a Bernoulli observation model that accounts for imperfect detection during field surveillance. Evaluation on independent test realizations shows strong reconstruction accuracy, with uncertainty estimates that are well calibrated for landscapes drawn from the training distribution. Crucially, the method remains robust even when over 80% of infections are unobserved. Short-horizon forecasting experiments demonstrate that diffusion-inferred heterogeneous transmission fields yield improved probabilistic predictions compared to homogeneous baselines, with more accurate spatial risk maps. This work establishes generative diffusion models as a promising tool for spatial epidemiological inference, with potential applications to plant disease surveillance and, more speculatively, to vector-borne disease mapping and adaptive management in heterogeneous landscapes.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Disease OutbreaksModels, BiologicalPlant DiseasesHumans

Stichwörter der Autor:innen

generative modellingplant diseasespatial epidemiology

Autor:innen und Institutionen

  • School of Mathematical Sciences, Lancaster University , Lancaster, UK.

ORCID-Profile: Andrew Baggaley

Förderung

  • Research England

Zitieren

APA

Baggaley, A. (2026). Learning spatial transmission maps from sparse outbreak snapshots with diffusion models. Journal of the Royal Society Interface, 23(244). https://doi.org/10.1098/rsif.2026.0413

Vancouver

Baggaley A. Learning spatial transmission maps from sparse outbreak snapshots with diffusion models. Journal of the Royal Society Interface. 2026;23(244). doi:10.1098/rsif.2026.0413

BibTeX
@article{Baggaley2026,
  author = {Baggaley, Andrew},
  title = {Learning spatial transmission maps from sparse outbreak snapshots with diffusion models},
  journal = {Journal of the Royal Society Interface},
  year = {2026},
  volume = {23},
  number = {244},
  doi = {10.1098/rsif.2026.0413}
}
Kostenlos starten →