Our results showed that satellite telemetry and remote sensing data may be combined to model indicators of key epidemiological parameters and their temporal variability with high spatial and temporal resolution. We were able to quantify seasonal variation in locations and timing of potential indirect contacts between wild ducks and chickens. We identified a critical period at the beginning of the rainy season that may have the highest potential for transmission and spread of pathogens in the IND due to regional movement. Our method can therefore be implemented at a local scale to assess the potential contacts at the wildlife-livestock interface in remote tropical areas.
The potential for contacts between comb ducks and chickens during the dry season may be explained by environmental dynamics. We observed that correspondence between predicted suitable cells for comb ducks and natural ponds increased during the dry season indicating that wild birds were likely increasingly using natural ponds. Dessication of these natural ponds in the vicinity of villages explains lower predicted contacts with chickens during the dry season. At the end of the dry season and the beginning of the rainy season, the agreement between predicted habitat used by comb ducks and natural ponds is very low (kappa < 0.2) indicating that the birds stopped using these habitats when they dried out. Wild birds moved to smaller ponds or irrigated areas during this period, increasing potential contacts with chickens.
The potential spread of pathogens by comb ducks is related to drivers leading to regional movements of wild birds at the beginning of the rainy season. Two main factors influenced the movement behaviour of comb ducks. First, refilling of many water bodies with seasonal rainfall led to the emergence of areas rich in food resources surrounding the IND . Second, the breeding behaviour of the birds influenced their movements. Comb ducks, as many other tropical waterbirds species, breed mainly during the rainy season . Their breeding sites, remote water bodies or flooded forests, are the only favourable habitats during the rainy season. The use of favourable habitats for both feeding and breeding during the rainy season led to regional movements that increased the potential spread of avian-borne pathogens by wild birds.
Validation and extrapolation of our results are limited by the type of data and the relatively small sample size used in our study. No validation (e.g cross-validation) method was convenient due to the spatial auto-correlation of the GPS data. We only showed the training AUC automatically calculated by Maxent. But a careful interpretation of these good scores of AUC (mean = 0.95, sd = 0.05) is required. First, AUC can be criticized for its reliability to accurately assess the performance of niche models [26, 27]. And second, the spatial-autocorrelation of our data likely leads to an overestimation of the performance of the model. The absoltute value of potential contact rate estimated in this study would be hard to generalize to other species or other ecosystems due to the limited number of individuals from a single species we could monitored. Although we obtained numerous detailed GPS locations on four individuals, these individuals may have only represented movements of only one sub-population as the four tagged comb ducks went to the same pond in the study area. However, a field observation a year after the release of the tagged birds showed this pond was a major roosting for comb ducks and white-faced whistling ducks. One would expect birds from different family groups to congregate on this pond with other groups. Furthermore, the behaviour of the tagged comb ducks was in compliance with what is known about their ecology and what local hunters reported during our field work: they congregated on the remnant water bodies during the dry season and performed a regional movement to reach different breeding grounds during the rainy season [20, 28]. That is why we believe these results can be extrapolated to other wild ducks in the study area. However, we would not extrapolate our results to other African wetlands without deploying more satellite transmitters in these areas. Studies with more transmitters would be more informative but their cost could be a major limitation. We believe that our study with few transmitters is interesting to locally assess the potential contacts between wild birds and chickens, and can be easily implemented, even in remote areas .
Contact rate between hosts, being either direct or indirect, is one of the main parameters of the transmission dynamics of infectious diseases. However several other parameters may modulate the probability for a contact to produce an effective transmission of pathogen. First, transmission is likely to be influenced by the density of hosts . Evidence of density-dependent transmission of avian influenza has been shown for wild birds [30, 31]. Second, transmission depends on the probability that the contact occurs between a susceptible and an infected hosts. Consequently, the proportion of infected hosts in the population and the level of population immunity are crucial parameters . Finally, the duration of persistence of the pathogen in the environment will control the probability of transmission. This is particularly true for water-borne transmitted viruses like AIV for which temperature, pH, and salinity may reduce the duration of survival and infectivity of the virus in the environment [8, 21, 22, 24].
Therefore, one has to consider the possibility that the period with the greatest contact rate between wild and domestic hosts may not be a period of maximum transmission.
In classical SIR models, transmission parameters are usually assembled in the parameter β [6
]. Different components of the β could be considered separately, like contact rate and transmission rate. Estimating and modelling these two distinct components of the β separately should provide a better understanding of the transmission dynamics of infectious diseases. It would also allow the use of a direct measure of these parameters rather than estimating a global β, which is usually the case [32
]. Our main results, the proportion of villages in potential contact with comb ducks, may be used to refine a seasonal forcing of the contact rate. In a SIR model with a density-dependent transmission, the force of infection λ would be expressed as:
where p(t) is the proportion of villages in potential contact with comb ducks as a function of time, β is the probability that a domestic bird is infected with AIV following contact with a wild bird, I is the number of infected wild birds, and N is the size of the wild bird population [7, 33]. Thus, our method could potentially be integrated into epidemiological models aiming to take into account the dynamic of contact rates between hosts. It would improve the efficiency of these models when contact rates are explicit parameters of the models . Finally, our approach could be used to identify the villages with an increased risk of indirect contact with wild ducks. This would allow implementing risk-based surveillance in areas by targeting the villages with the highest risk of contact.