Habitat loss exacerbates pathogen spread: An Agent-based model of avian influenza infection in migratory waterfowl

Habitat availability determines the distribution of migratory waterfowl along their flyway, which further influences the transmission and spatial spread of avian influenza viruses (AIVs). The extensive habitat loss in the East Asian-Australasian Flyway (EAAF) may have potentially altered the virus transmission and spread, but those consequences are rarely studied. We constructed 6 fall migration networks that differed in their level of habitat loss, wherein an increase in habitat loss resulted in smaller networks with fewer sites. The networks were integrated with an agent-based model and a susceptible-infected-recovered model to simulate waterfowl migration and AIV transmission. We found that extensive habitat loss in the EAAF can 1) relocate the outbreaks northwards responding to the distribution changes of wintering waterfowl geese, 2) increase the outbreak risk in remaining sites due to larger bird congregations, and 3) facilitate AIV transmission among wintering geese. Our modelling output suggested that there was a certain system resilience of migration network to confront the site removal. In addition, the outputs were in line with the predictions from the concept of “migratory escape”, affecting the pattern of infection prevalence in the waterfowl population. Our modelling shed light on the potential consequences of habitat loss in transmitting and spreading AIV at the flyway scale, and suggested the driving mechanisms behind these effects, advocating the importance of nature conservation in changing spatial and temporal patterns of AIV outbreak. Author summary What are the possible consequences of extensive habitat loss on the transmission and spread of avian influenza viruses (AIVs)? We used a logistic regression model to select the suitable habitats of Greater white-fronted goose in the East Asian-Australasian Flyway and treated these habitats as sites to construct 6 fall migration networks that differed in their level of habitat loss (i.e., site removal). We then simulate geese migration in these networks, and explore the impacts of habitat loss on habitat connectivity and AIV transmission. We found the extensive habitat loss can cause relocation of the outbreaks and increase the outbreak risk and AIV transmission. Our modelling outputs suggested a certain network resilience to confront the site loss, and a “migratory escape” to change the spatial and temporal pattern of infection prevalence in the population. Overall, our study showed that land use changes and habitat loss can affect disease distribution and prevalence, suggested the importance of habitat conservation in changing the spatial and temporal pattern of AIVs transmission and spread.

Virus spread over the sites 141 The GLM showed that the basic reproduction number R 0 increased with weighted 142 in-degree (Fig 3A and Table 1; p ≤ 0.001), indicating that outbreak risk at site increased 143 under higher connectivity and with more birds migrating from other sites. Moreover, the R 0 is 144 significantly greater in the last two scenarios, compared to the complete network ( Fig 3B and 145 Table 1; p ≤ 0.01), indicating that site removal increased the outbreak risk in the remaining 146 sites.  Virus transmission in the population 161 The infection prevalence showed similar patterns that one striking infection peak followed 162 by another gentle peak (Fig 4). The first striking peak occurred when the majority geese still 166 times greater than the direct transmission (Fig 4C and D).

167
The infection prevalence did not differ strongly among the scenarios except for the 168 extreme scenario (Fig 4). Specifically, the second infection peak (i.e., the peak in wintering

178
To gain a better understanding of the effects of anthropogenic disturbance on the spatial 179 and temporal patterns of zoonotic disease outbreaks, we used models to explore the potential

302
We constructed directional links from sites with a higher latitude to sites with a lower 303 latitude between centres of each pair, if the geographical distance D ij (i.e., geographic 304 distance between site i and j) was equal or shorter than the migration step length L step (i.e., the 305 maximum migration distance without rest). In total, we generated 6 theoretical migration 306 networks, by removing the sites with different levels of habitat loss using increments of 10%.
307 These theoretical networks are the complete network, and five network scenarios with 308 increments in habitat loss (i.e., the scenario of removing sites with >40% habitat loss, 309 and >30%, >20%, >10% and >0%). The 6 theoretical networks are shown in S1 Fig, with 310 their corresponding basic network metrics listed in S1 Table. Simulation of bird migration 312 We applied a migratory flow network to simulate the moving geese over the sites [41].
313 Each site was assigned a variable, site attractiveness A i t to represent the suitability of the site i 314 at given time t. Each link was assigned two variables, migration resistance R ij to represent the 315 difficulty for travelling from site i to j, and the migration probability MP ij to represent the 316 likelihood for travelling from site i to j. Moreover, we assumed the attractiveness A i t was 317 negatively influenced by bird density  i t , whereas the migration resistance R ij was positively 318 influenced by geographical distance D ij . These variables at time step t were calculated as: where k 1 and k 2 are scaling parameters. In general, the decision was determined by bird 323 density and distance between the sites (S1 Appendix), and the bird prefers to select the link 324 with greatest migration probability MP ij .

325
A total of 10,000 geese were simulated as agents in our model. Each goose was randomly 326 assigned body mass m, according to a gaussian distribution at the beginning of simulation. At 327 each time step t, the body mass dynamic was calculated as: (4) where a is the accumulation rate during resting at a site, c is the body mass consumption 330 rate during flying, s is the flying speed. When a resting bird cumulated enough body mass 331 (i.e., above a threshold ), the bird selected a site to migrate to in next time step.

332
Satellite tracking revealed that Greater white-fronted goose migrates within a narrow 333 corridor (i.e., longitude range) and makes stops for rest and refuelling during fall migration 334 [22,23]. Therefore, we setup two variables, the corridor width w, and the expected number of 335 rests n, to constrain the sites selection. The corridor width w constrains the birds to migrate 336 within a range of longitudes, and the number of rests n regulates the number of stopover sites 337 before arriving at the wintering site. The detailed decision-making rules are explained in S1 338 Appendix. For simplification, we did not include any goal-oriented behaviour or mortality or 339 reproduction in the model.

340
Simulation of pathogen transmission 341 We applied an SIR model to simulate the AIV transmission in the migratory population.

357
The amount of virus V t in the environment at site i is calculated as: where  is the virus decaying rate in the environment, and  is the virus shedding rate. We 360 divided the equation by shedding rate  to obtain: