When Is It Safe to Put Birds Back? A New Model Maps the Avian Flu Recovery Timeline

The Science

When an avian flu outbreak tears through a poultry region, officials face two compounding crises. The first is stopping the epidemic: identifying infected farms, culling flocks, quarantining premises. The second is recovery—figuring out when it's safe to restock the empty barns without triggering a second wave. These questions usually get answered separately, by different tools, at different times. A team of researchers from Auckland University of Technology, University College Dublin, and Federal University of Technology Akure argue that's backwards.

They built a single modeling framework that handles both phases simultaneously—epidemic control and post-outbreak recovery—grounded in the specific mechanics of how avian flu actually spreads. Their model, described in a paper posted to arXiv in August 2026, tracks farms as epidemiological units organized by county and production type (broiler chickens, organic ducks, and other systems), then simulates how four distinct transmission pathways interact: direct local spread between neighboring farms, environmental contamination from virus persisting in water or soil, movement-mediated transmission through recorded animal or equipment transfers, and distance-dependent spatial spread across county boundaries (Fatoyinbo et al., 2026).

The synthetic outbreak they modeled occurred on fictional Jolly Island—a made-up country with 14 counties, 556 districts, and a high-risk zone along the eastern coast where migratory birds posed elevated introduction risk. Data arrived in three phases over 109 days, mimicking the real-world cadence of surveillance updates during an unfolding crisis. Across the entire observation period, 560 confirmed outbreaks were recorded, with 75 percent occurring in chicken farms and a peak of 24 new cases on a single day in late January.

Rather than treating all farms as identical, the model stratifies them by production system. Broiler-2 farms, organic duck operations, and everything else each have distinct transmission parameters, confinement multipliers, and susceptibility profiles. This matters because different farm types represent different biosecurity postures, bird densities, and contact patterns—variables that fundamentally shape epidemic dynamics.

What They Found

The epidemic was geographically concentrated and production-specific in ways that reveal the underlying transmission architecture.

Broiler farms bore the heaviest burden throughout the simulation. The cumulative infectious-farm-days—essentially the total "infectious exposure" experienced across all farms over time—varied dramatically by production class and region. Broiler operations in the northern cluster accumulated the highest regional burden, followed by organic duck farms in the same area, with other production systems showing comparatively lower transmission intensity (

,

,

,

).

Preventive culling—removing farms in high-risk neighborhoods before they show symptoms—proved its worth. Comparing simulations with and without the recorded preventive culls, the researchers found that the intervention trimmed the mean cumulative burden from 16,362.7 infectious-farm-days to 13,631.9, a reduction of 16.7 percent. Earlier confinement of broiler and organic duck farms also substantially compressed epidemic magnitude, while stronger environmental transmission amplified the peak. These aren't abstract percentages: each reduction in infectious-farm-days represents fewer farms infected, less virus circulating, and fewer animals lost.

Impact of Preventive Culling on Epidemic Burden

Label	Value
Without preventive culling	16,362.7
With preventive culling	13,631.9

Data from Table 1: Preventive culling impact analysis

The restocking analysis revealed something counterintuitive. As the epidemic wound down, the probability of triggering a rebound—defined as a post-restocking surge exceeding five new infections above the level present at restocking—naturally declined. But the transition wasn't smooth, and timing mattered more than intuition suggests.

Estimated Rebound Probability by Restocking Date

Label	Value
15 Mar 2026	0.9
22 Mar 2026	0.85
29 Mar 2026	0.78
5 Apr 2026	0.7
12 Apr 2026	0.62
19 Apr 2026	0.55
26 Apr 2026	0.48
3 May 2026	0.4

Rebound probability = P(post-restocking infections > restocking level + 5 farms)

Under their model assumptions, 24 May 2026 was the first candidate date satisfying a rebound-probability threshold of 0.20—that is, an 80 percent probability of avoiding a significant resurgence. For a restocking attempt on 15 March 2026, none of the tested restocking fractions met this criterion. This means that even when case counts appeared low, the underlying infectious pressure remained high enough that reintroducing susceptible birds carried meaningful risk.

Perhaps the most actionable finding involved how farms are restocked. Capacity-based restocking—replenishing farms only up to the capacity that was emptied by culling, rather than relative to the original population—reduced cumulative burden by 8.45 percent and cut rebound probability nearly in half, from 0.780 to 0.533. Phased, constrained restocking isn't just cautious; it's epidemiologically superior to aggressive repopulation.

Why This Changes Things

The standard playbook for agricultural epidemic response treats control and recovery as sequential problems solved by different players. Animal health officials manage culling, movement restrictions, and surveillance. Once incidence drops, economists and farm operators take over, negotiating restocking timelines based on market conditions and financial pressure. This division produces blind spots on both sides.

The modeling framework developed here shows what happens when you integrate those phases. By simulating the entire arc—from initial seeding through epidemic peak through waning through restocking—in a single stochastic framework, the researchers capture feedback effects that sequential analysis misses. Aggressive restocking doesn't just create economic exposure; it creates epidemiological exposure by concentrating susceptible animals in space and time, exactly when environmental contamination may still be elevated and surveillance intensity may have relaxed.

The production-system differentiation is equally important. A restocking strategy that works for organic duck farms—typically lower-density, outdoor-access operations with distinct environmental exposure profiles—may be inappropriate for intensive broiler facilities where bird density amplifies transmission. The data show these systems behave differently under identical control regimes: preventive culling reduced burden among Broiler-2 farms by a different magnitude than among organic duck operations, and environmental transmission affected them asymmetrically (

,

).

The rebound-probability framework offers something rare in epidemic modeling: a decision-relevant threshold. Rather than projecting whether cases will decline (a deterministic question already answered by surveillance), it answers what decision-makers actually need to know—what's the probability this restocking decision causes a second wave? A 0.20 threshold isn't arbitrary; it's a risk tolerance that could be adjusted based on viral fitness, economic context, or regulatory conservatism. The finding that 24 May 2026 first clears this bar, while 15 March 2026 does not, suggests that patience during the tail end of an outbreak carries compounding returns.

What's Next

Several operational gaps remain. The model aggregates farms within county-production strata, meaning individual farm-level variation in biosecurity, flock management, or building design is compressed into shared parameters. This is a practical necessity for computational tractability, but it means the framework may underestimate heterogeneity in real populations. Similarly, the environmental compartment is shared across production classes within each county; a farm that somehow isolates its birds from contaminated shared water or soil wouldn't be well-represented.

The analysis also relies on recorded movement data, which captures official farm-to-farm transfers but not informal contacts—personnel moving between operations, equipment shared across farms, vehicles visiting multiple sites. These pathways may be partially absorbed by the local and spatial transmission components, but their aggregate effect is uncertain.

Most importantly, the rebound probability was estimated under specific assumptions about post-restocking surveillance, biosecurity compliance, and regulatory enforcement. In practice, these factors introduce variability that the model treats as fixed. A restocking decision in a region with strong biosecurity culture and active surveillance will carry lower risk than the same decision in a region where monitoring has lapsed.

These caveats don't diminish the framework's value—they define the boundary of its applicability. The research demonstrates that integrating epidemic control and post-outbreak recovery within a single modeling architecture is both feasible and informative. Timely confinement, targeted preventive culling, and phased capacity-based restocking may reduce both immediate burden and resurgence risk. The specific thresholds—16.7 percent burden reduction from preventive culling, 24 May 2026 as the first safe restocking date under a 0.20 rebound threshold, 0.533 rebound probability under capacity-based versus 0.780 under baseline restocking—will need calibration against real outbreaks and local conditions.

What the framework offers is a way of thinking about agricultural epidemic response that makes the tradeoffs explicit. Control measures have post-recovery consequences; recovery decisions have control implications. Modeling both simultaneously forces those tradeoffs into view, giving decision-makers a clearer picture of what they're actually choosing between.