Modeling and outbreak analysis map African swine fever in South Korean farms

African swine fever outbreaks in South Korean pig farms showed a combination of regional clustering and wider geographic dispersion in early 2026, with the second, third and fourth confirmed outbreaks located about 178, 126 and 190 kilometers from the nearest previously confirmed outbreak, respectively. A study combining a stochastic agent-based model of within-farm transmission with a retrospective analysis of confirmed farm outbreaks from 2019 through March 2026 found that environmental transmission was the strongest modeled influence on the basic reproductive number, while the infectious period, direct pig-to-pig transmission and environmental transmission were associated with epidemic duration, mortality and peak prevalence. The researchers emphasized that the distances described spatial proximity among confirmed outbreaks, not proven farm-to-farm transmission links, and that several possible mechanisms could underlie the dispersed pattern.

The study modeled individual pigs in closed farm populations ranging from 500 to 2,500 animals, with one infectious pig introduced at the start of each simulated outbreak. Implemented in NetLogo, the susceptible–exposed–infectious–removed model advanced in daily steps and represented infection through two routes: direct contact between pigs and indirect exposure to environmental contamination generated by viral shedding. A continuous contamination variable represented contaminated floors, feeders, drinkers, pen walls, equipment and other fomites. Parameters were drawn from published estimates or biologically plausible assumptions through 1,000 Latin hypercube simulations, with model outputs including the basic reproduction number, epidemic duration, cumulative mortality and peak prevalence. Partial rank correlation coefficients, accompanied by 95% bootstrap confidence intervals from 2,000 resamples, were used to assess the direction and strength of parameter associations.

Under baseline management conditions, the modeled reproduction number had a median of 2.0 and a wide 95% prediction interval of 0.00 to 13.33. In 70.3% of simulations it exceeded 1, indicating that within-farm transmission was likely to be sustained under most sampled combinations, although stochastic extinction remained possible. Environment-to-pig transmission was the strongest positive association with the reproduction number, with a partial rank correlation coefficient of 0.901, followed by pig-to-pig transmission at 0.770. The environmental viral decay rate was negatively associated at −0.427, supporting the modeled role of prolonged environmental survival in maintaining opportunities for infection. The authors cautioned against direct numerical comparisons with reproduction numbers reported for farm systems elsewhere, because the quantity depends on production conditions, outbreak stage, herd structure and control practices.

Simulated outbreaks also showed a broad range of severity. Median epidemic duration was eight days, but the 95% prediction interval extended from one to 245 days, illustrating the possibility of either rapid fade-out or prolonged transmission. The infectious period had the strongest positive association with duration, followed by pig-to-pig and environment-to-pig transmission. Cumulative mortality had a median of 0.95% and a prediction interval of 0.09% to 100%, producing a strongly bimodal pattern: many simulations caused little mortality because infection stopped early, whereas successful transmission chains sometimes affected nearly the entire herd. Median peak prevalence was only 0.10%, but its prediction interval ranged from 0.00% to 31.31%. Direct pig-to-pig transmission was most strongly associated with peak prevalence, followed by the infectious period and environmental transmission.

The modeled control scenarios produced different patterns. An enhanced biosecurity scenario, representing preventive farm segmentation, improved hygiene, lower environmental transmission, reduced shedding and faster viral decay, shifted the median reproduction number below one and sharply reduced mortality, epidemic duration and peak prevalence in most simulations. A reactive outbreak-response scenario incorporated detection delays of zero to seven days, movement restrictions, enhanced cleaning and culling triggered by the presence of an infectious pig. It also reduced outbreak magnitude and duration relative to baseline, but the median reproduction number remained close to one and outcomes varied more than under enhanced biosecurity. These are counterfactual model results based on literature-informed multipliers, not evaluations of a particular intervention trial, and the within-farm model was not calibrated against detailed Korean farm outbreaks.

Between farms, the analysis documented a clear change in the national outbreak pattern since South Korea confirmed its first domestic pig outbreak in September 2019. Early cases were concentrated in border-adjacent municipalities in Gyeonggi-do and Incheon and in northern inland areas of Gangwon-do. Later years showed expansion into central and southern provinces, including Chungcheongnam-do, Jeollanam-do, Jeollabuk-do and Gyeongsangnam-do. Annual counts remained between two and seven from 2020 through 2022, rose to 10 and 11 in 2023 and 2024, fell to six in 2025, and then reached 24 by March 2026. Because the 2026 dataset covers only the first quarter, it cannot represent the full year or establish a final annual total.

The 2026 epidemic began with confirmed outbreaks in Gangneung-si on January 17, Anseong-si on January 23 and Yeonggwang-gun on January 26. Additional separated outbreaks followed in late January and early February, while February brought more apparent clustering in Gyeonggi-do and adjacent Chungcheongnam-do, including Pocheon, Hwaseong, Pyeongtaek and Dangjin. Outbreaks also continued in distant parts of Jeollanam-do, Gyeongsangnam-do and Gyeongsangbuk-do. Monthly analysis found repeated activity in January from 2023 through 2026 and additional clustering in March during 2023, 2025 and 2026. Winter and early-spring concentration was observed, while summer and autumn cases were more sporadic, although the descriptive data do not establish a seasonal mechanism or causal effect.

Nearest-neighbor analysis was possible for 23 of the 24 outbreaks confirmed in early 2026 because the first had no earlier outbreak within that year for comparison. Distances from each newly confirmed farm to the nearest previously confirmed farm had a first quartile of 15.0 kilometers, a median of 31.3 kilometers and a third quartile of 58.3 kilometers; the mean was 52.1 kilometers and the range extended from 0.8 to about 190 kilometers. Three outbreaks were less than 10 kilometers from the nearest earlier outbreak, 13 were 10 to 50 kilometers away, three were 50 to 100 kilometers away and four exceeded 100 kilometers. As shorter distances became more frequent later in the epidemic, the combined evidence points to initial multifocal occurrence followed by regional clustering and continuing dispersion, rather than a pattern confined to a narrow geographic front.

The researchers considered several unconfirmed explanations for the long distances, including multiple introductions, delayed detection, undetected intermediate infections and non-local movement through production networks. Shared animal, feed, vehicle, equipment and personnel routes could connect distant farms, and slaughterhouses or feed-distribution systems may function as high-connectivity interfaces. However, the study lacked movement records, trace-back investigations, genomic data, complete locations for uninfected farms and direct evidence of slaughterhouse or feed links. Detection of African swine fever virus DNA in porcine plasma-based feed ingredients has also been reported, but DNA detection does not establish viable or infectious virus, and feed-associated transmission was not demonstrated in this study. Geographically separated confirmations may therefore mark missed intermediate infections or differences between infection and confirmation dates rather than direct long-distance spread.

Together, the simulations and outbreak observations support a cautious, combined control approach for South Korea’s pig sector. The within-farm results are consistent with the importance of limiting environmental contamination, shortening the time that pigs remain infectious, rapid recognition and removal of infected animals, and rigorous cleaning and disinfection. The between-farm results indicate that measures centered only on close neighbors of confirmed farms may miss risks extending beyond local outbreak zones. The authors therefore advocate local containment and strong on-farm biosecurity alongside broader surveillance and risk management for transport systems, shared service networks, feed distribution and slaughterhouse-associated movements. Because the 2026 data covered only three months and the model simplified farm-specific housing, production stages and biosecurity practices, future work should integrate susceptible-farm locations, movement and trace-back records, service networks and genomic evidence to test specific transmission pathways directly.

Subject of Research: Agent-based and spatiotemporal analysis of African swine fever in South Korean pig farms

Article Title: Understanding African swine fever dynamics in South Korean pig farms through agent-based modeling and spatiotemporal analysis of farm outbreaks

Article References: Tanveer, M., Hirwa, F., Ntakiyisumba, E., & Won, G. (2026). Understanding African swine fever dynamics in South Korean pig farms through agent-based modeling and spatiotemporal analysis of farm outbreaks. Frontiers in Veterinary Science, 13, Article 1945299. https://doi.org/10.3389/fvets.2026.1945299

Image Credits: AI Generated

DOI: 10.3389/fvets.2026.1945299

Keywords: African swine fever, South Korea, pig farms, agent-based modeling, spatiotemporal analysis, farm outbreaks, biosecurity, environmental transmission, disease surveillance, viral epidemiology, Understanding, African