Exploring the Role of Wild Bird Species in the Transmission of Avian Influenza to Poultry

Abstract

Wild aquatic birds are crucial in maintaining the high pathogenicity avian influenza (HPAI) viruses. However, the HPAI dynamic at the wild-domestic interface is still poorly known, and a comprehensive understanding of species that could potentially act as a bridge between wetlands and poultry farms is still lacking. In this study, an eco-epidemiological framework was used to build species distribution models for 40 wild bird species camera-trapped at 10 poultry farms in northeastern Italy. The predicted wild bird favorability distributions were used to estimate HPAI outbreak occurrences in the area of interest, using an ensemble approach that included five methodologies: generalized linear model (GLM), generalized additive model (GAM), boosted regression trees (BRTs), random forest (RF), and maximum entropy (MaxEnt). The group of species that included most of the Ardeidae (i.e., great egret, purple heron, little egret, and cattle egret), one Galliformes (i.e., common pheasant), and one Gruiformes (i.e., common moorhen) showed the highest importance (IMP = 28%) in explaining the HPAI outbreak probability of occurrence in poultry, highlighting their potential bridging role between the reservoir species and the domestic populations. The second most important group of species (IMP = 17%) included one Anseriformes (i.e., mallard), two Charadriiformes (i.e., black-headed gull and yellow-legged gull), and one Ardeidae (i.e., gray heron), remarking their role in the disease ecology. These results underline the complex role of the wild-domestic interface in the epidemiology of HPAI, suggesting that a broader range of species than what is typically considered might be involved in HPAI virus ecology. Including these groups of species in targeted surveillance programs would help in fine-tuning sampling efforts and identifying early warning signals of possible transmission to poultry holdings.

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

1. Introduction

Avian influenza (AI) is a transboundary disease posing a severe threat to wildlife, poultry production, and public health [1, 2], impacting the development of the Global South and economies of the Global North [3]. Since 2013, the unprecedented spread of highly pathogenic AI (HPAI) H5Nx viruses within the 2.3.4.4 clade has gathered attention due to their worldwide dissemination and ability to infect a wide range of wild birds, poultry, and mammals, including humans [4, 5]. The virus’s ecology is rapidly evolving, challenging previously understood patterns and revealing significant gaps in our scientific understanding of its dynamics [6].

In 2021, AI persisted year-round in wild birds in Europe [7], causing mass mortality events in wild bird populations, threatening entire ecosystems [4, 8, 9] and affecting numerous mammal species [10, 11]. This shift in the virus’s behavior—particularly its ability to directly infect wild birds in its highly pathogenic form and ranging from the absence of symptoms to severe disease and mortality in species once considered reservoir hosts—demands a deeper investigation into the wild-domestic interface, which remains poorly understood. The interplay between wild bird populations and domestic poultry during the 2021–2022 epidemic exemplified this complexity, as Europe witnessed over 6615 HPAI virus detections across 37 countries, driven by the emergence of at least 19 different H5N1 clade 2.3.4.4b genotypes from various reassortment events [12–14].

The absence of new virus incursions during the 2022–2023 epidemic wave in Europe, along with the persistence of the virus in wild birds, allows to hypothesize that a new change in the disease dynamic occurred, with a prolonged persistence of the infection in the residential bird populations rather than new introduction events caused by the seasonal movements of migratory infected birds [4, 15, 16]. The emergence of new reassortants and the unpredictable shifts in the AI virus (AIV) ecology highlight the need to advance our understanding of its spread and transmission routes from wild populations to poultry [1].

Anseriformes and Charadriiformes are the bird orders most frequently associated with AIVs and are generally considered the primary maintenance hosts in the virus’s natural ecology. While both groups harbor a very high viral diversity, knowledge gaps still persist regarding the differences between them. AIV surveillance and research have predominantly focused on Anseriformes, while much less is known about Charadriiformes, despite being the most abundant between AIV hosts and more closely linked to human activities [17–19]. Before 2022, Charadriiformes were predominantly infected with viruses belonging to the H13 and H16 subtypes, which were rarely isolated in poultry and to which reared birds are generally resistant [1, 20]. However, during the 2022–2023 HPAI H5N1 epidemic, Charadriiformes populations were heavily involved, experiencing extraordinary mass mortality events that resulted in the deaths of thousands of individuals. Since then, this order has gained significant attention, highlighting its role in the ecology of AI. Other types of birds, such as shorebirds, raptors, and waterfowl, were also involved in AI mass mortality events [14, 21], and the virus was detected in a wide range of species of several orders [22, 23]. If Anseriformes and Charadriiformes are considered to play a crucial role in the long-distance dissemination of the AIVs [24–26], the dynamic of the short-distance transmission events in the wild–domestic interface requires a broader perspective that considers the role of other species in the viral incursion into farms and in the maintenance of the viral circulation at a local scale [27].

Surveillance of wild birds for AIVs is essential for assessing prevalence, identifying strains circulating in a specific area, and providing effective early warning systems [28]; however, its effectiveness can be hampered due to a biased selection of the sampling sites, focusing almost only on species known to have a high AIV prevalence (i.e., waterfowl and seabirds), and underreporting of dead birds. Together, these factors can lead to a limited knowledge of the actual lineage circulation among avian populations [28, 29] which also might reflect a partial or even biased estimation of the HPAI risk of introduction into the poultry sector. Additional obstacles in achieving a comprehensive understanding of the wild-domestic interface include the limited knowledge of the complex interactions of wild birds with their natural environment, the broad spectrum of interactions among different wild avian populations (e.g., residential and migratory species), and the diversity of farming systems and reared species that wild birds may interact with.

In a previous study conducted in 2019 in northeastern Italy, a camera-trap survey was carried out in 10 poultry farms to investigate the composition of the wild ornithic populations living in the proximity of poultry premises [19] to generate some hypotheses on the wild bird species that might be potentially involved in the dynamics of HPAI spillover events. The area investigated through that survey is characterized by a high density of poultry farms (defined as densely populated poultry areas—DPPAs) and has been subject to recurrent outbreaks of HPAI [30–32]. In addition, the region is also characterized by the presence of extensive and widespread wetlands in close proximity to the DPPAs, which serve as ecological niches for various orders of wild birds. The findings of the previous study highlighted that the Anseriformes and Charadriiformes rarely approached the poultry facilities, mainly during the spring season, which is considered to be a lower-risk period for AIVs introduction into the poultry sector [20]. On the contrary, species belonging to the orders Passeriformes (i.e., magpies and blackbirds), Pelecaniformes (i.e., cattle egrets), Galliformes (i.e., pheasants), and Columbiformes (i.e., Eurasian collared doves and wood pigeons) were identified as the most inclined to come into close proximity to farming environments. We thus hypothesized that among these species, there might be the ones more likely capable of acting as “bridge hosts” for AIVs [33].

Herein, we present an eco-epidemiological approach to identify the wild bird species whose spatial distribution may help explain the occurrence of AI outbreaks in poultry. To achieve this, we used data obtained from the aforementioned ornithocoenosis study [33] and the location of primary outbreaks recorded during the 2017–2018 HPAI epidemic in Italy [31, 34]. The ultimate goal was the identification of the wild bird species and the quantification of their association with the probability of observing an HPAI outbreak in poultry at a given location, thus suggesting that these species may have played a significant role in the local transmission dynamic of the disease.

2. Materials and Methods

2.1. Study Area and Species Distribution Data

The study area is represented by the northern Italian regions, which include the provinces at higher risk of AIV introduction (Emilia Romagna, Friuli-Venezia Giulia, Lombardy, Piedmont, and Veneto) [35]. These regions encompass the DPPAs, characterized by the highest poultry farm density in Italy and recurrently affected by massive epidemic waves of AI [36]. Furthermore, extensive wetlands and marshes are in proximity to the DPPAs, serving as important reproductive grounds for various orders of wild aquatic birds.

The results of the ornithocoenosis study by Martelli et al. [33] were used to identify the focal species of interest. Forty wild bird species were captured by camera-trapping in proximity to 10 commercial poultry farms located in the DPPAs [33]. Occurrence data for the detected species were obtained from the eBird repository [37]; only occurrence data recorded in the period during which the camera traps were operational (January 2019–December 2019) were included in the following analyses since necessary for the SDM construction to predict the occurrence probability of the selected species in the same time frame across the study area.

2.2. Environmental Data

Environmental data were handled through the Environmental Data for Veterinary Epidemiology (EVE) platform developed at the Istituto Zooprofilattico Sperimentale delle Venezie [38]. The considered variables included the Normalized Difference Vegetation Index (NDVI) [39]; the second level of the Corine Land Cover (CLC) 2018 [40]; the elevation (Digital Terrain Model—DTM) [41]; minimum, maximum, and average monthly measures of the land surface temperature (LST—°C) [42] and the monthly minimum, maximum and average precipitation (millimeters, mm), derived by ground sensors. Temperature and precipitation values were further processed to obtain a set of 19 bioclimatic variables (bioclim) [43]. Moreover, distances in meters (m) from the nearest wetland were calculated using the river network and waterbody information from the WISE-WFD database [44]. Finally, the CLC classification was transformed from a single qualitative variable into a set of quantitative variables, obtaining 15 raster layers with a spatial 3 km x3 km, in which each cell indicated the proportion of area occupied by each second-level CLC class.

The environmental variables were preliminarily assessed for multicollinearity to prevent redundancy, applying the Spearman method with a correlation threshold of 0.70 [45]. Bioclim layers bio8, bio9, bio18, and bio19 were also preventively excluded from further analysis due to known spatial artifacts [46]. Supporting Information 1 reports the complete list of environmental variables.

2.3. Outbreak Data

Epidemiological data on the 2017–2018 HPAI H5N8 epidemic in Italy were used to assess the association between outbreak occurrences and wild bird distributions. As the aim was to assess the potential spillover of HPAI viruses from wild bird populations to poultry farms, only primary domestic outbreaks (n = 49), as previously defined [32, 34], were included in the study. The 49 primary domestic outbreaks were used as occurrence points (coded as 1). A total of 2203 poultry farms with the following characteristics were considered absence points (coded as 0):

i. were not diagnosed as AIV positive during the HPAI H5N8 epidemic;

ii. were of the same productive types as the infected farms;

iii. were located within the same Regions where the outbreaks occurred;

iv. reared more than 250 poultry (Supporting Information 2).

2.4. Species Distribution Modeling

The distributions of the 40 wild bird species identified near the premises of 10 poultry farms in the north of Italy in 2019 through camera traps [33] were modeled using a BART approach. A common set of potential observation points was defined considering all locations in the eBird repository where at least one of the species was observed, as well as the 10 farms where the camera-trap survey was conducted. The eBird repository reports all the species detected at a single location, meaning that any species not listed were actually not detected, providing accurate information about their absence [37]. Therefore, in each model, the occurrence points varied according to the species considered, while the remaining points of the common set were assumed being true absence points. This approach significantly reduced geographical bias by excluding under-surveyed or nonsurveyed areas.

For the analysis, species occurrence or absence was considered at a consistent spatial unit level of 3 km x 3 km cells, regardless of the number of actual observations per unit. This approach mitigated the impact of sampling bias, addressing situations where certain easily accessible locations may have disproportionately higher survey efforts [47, 48].

A variable selection procedure was performed to select the best predictors set for each modeled species using the Boruta algorithm [49]. This approach allowed capturing all the important features, iteratively comparing the importance of each variable with the importance of “shadow” variables, which were synthetically created by randomly shuffling the original ones.

The performances of the 40 models were assessed using three different metrics [50]: (i) the area under the receiver operating characteristic curve (AUC) to evaluate the overall model discrimination performance [51, 52]; (ii) the true skill statistic (TSS), to measure the classification performance [53]; and (iii) Miller’s calibration slope, to assess the model reliability [54, 55]. Thresholds were set to 0.70 for the AUC [56], 0.40 for the TSS [47], and between 0.50 and 1.50 (with an additional tolerance of 0.05) for Miller’s calibration slope [47]. The models of species that did not comply with these fixed thresholds were excluded from further analyses; however, the inclusion of species that did not meet the requirements was considered if prior knowledge emphasized their importance and their distributions aligned with the ecological knowledge about the species.

The predicted probability maps obtained from the remaining models were then transformed into favorability maps. The favorability is determined by both the presence probability and the ratio of presences to absences for the species within the modeled sample. This transformation mitigated the impact of species prevalence on the predicted values, thus allowing for direct comparison of predictions across the various species [57–59].

2.5. Outbreak Data Analysis and Modeling

A series of univariable models were built to explore the association between the individual wild bird species and the emergence of primary HPAI domestic outbreaks. A five-method ensemble approach was adopted to model the outbreak occurrences, including a generalized linear model (GLM), generalized additive model (GAM), random forest (RF), boosted regression tree (BRT), and maximum entropy (MaxEnt). Two hundred bootstrap reiterations were run for each model, resulting in a total of 1000 models, providing more accurate predictions than a single split of the data into training and test sets [60–62]. Three metrics were taken into account to measure the goodness of fit of the models: the correlation index (COR) [63], the AUC, and the TSS. For each univariable run per species, the median values of each performance metric were calculated from the 1000 bootstrap replicate models. Finally, the strength of the association of each single species with the estimated outbreak probability was ranked according to the three metrics (in order COR, AUC, and TSS).

Subsequently, a final multivariable model was built, encompassing the 22 selected wild bird species. Since many of the observed wild bird species are common and widespread in the study area, we assessed the collinearity among species’ favorability using the variance inflation factor (VIF) score and considering a threshold value of 10 [64]. A cluster analysis [64] and a nonmetric multidimensional scaling (nMDS) analysis [65] were used to further address collinearity problems, grouping species with highly related favorability. For each observed cluster, a single representative species was chosen based on the performance metrics values from the preliminary univariable models. The set of selected species was further checked for multicollinearity and used as explanatory variables in the final multivariable model [64]. If multiple species in a cluster had similar metrics indices, the choice of predictors was guided by ecological and epidemiological considerations. The same ensemble strategy used for the univariable models was employed also in the multivariable models. A final prediction map showing the HPAI H5N8 outbreak occurrence predicted probability across the entire study area was created. Marginal predictions (i.e., response curves) and variable importance scores were obtained for each predictor species included in the final outbreak probability model.

2.6. Notes on Data Processing and Software

All data cleaning and manipulation, modeling, and graphics were performed using the R software 4.2.2 [66] and RStudio [67], along with the following packages: “raster” [68] to build the environmental variables raster layers raster bricks, species distribution layers, and all the prediction maps; “dismo” [69] to obtain the 19 biovars; “terra” [70] for rasters manipulation; “fuzzySim” [71] to manage the gridded occurrence data and obtain the species favorability maps; “rgdal” [72] and “sp” [73, 74] to manipulate spatial objects; “Boruta” [51] for variables selection; “embarcadero” [75] and “dbarts” [76] for BART model analysis; “sdm” [77] for ensemble model approach. QGIS (version 3.36.1) software was used for rendering the map of the study area [78].

3. Results

3.1. Species Distribution Modeling

A total of 1501 3 km x3 km cells with at least one of the 40 wild bird species observed were included in the analyses. The species occurrence and absence points included in each of the models are shown in Supporting Information 3, while the numbers of presences and absences cells considered for each species are reported in Supporting Information 4.

The selected set of variables used as predictors in the regression BART models (i.e., one for each wild bird species) are reported in Supporting Information 5. Species distribution models that satisfied our criteria for inclusion in further analyses and relative performance metrics are presented in Supporting Information 6.

From the 40 initial wild bird species, 22 were selected, belonging to 8 different orders: Pelecaniformes, Charadriiformes, Anseriformes, Galliformes, Passeriformes, Accipitriformes, Gruiformes, and Columbiformes. Eighteen species were selected based on the model performances, while four additional species (the Eurasian Blackbird—Turdus merula, the Eurasian Blue Tit—Cyanistes caeruleus, the Barn Swallow—Hirundo rustica, and the Common Buzzard—Buteo buteo) were still retained even if not entirely compliant with the model metrics thresholds (TSS < 0.40). This decision was supported by prior knowledge of their importance and the alignment of their distributions with the species’ ecological understanding. The final prediction maps with the species distributions and the favorability maps are shown in Supporting Information 7 and Supporting Information 8, respectively.

The probability values varied greatly between species. The species with the lowest values was the Mediterranean Gull (Larus melanocephalus) (median, M = 0.012, interquartile range, IQR = 0.03), while the species with the highest values was the Eurasian Blackbird (M = 0.57, IQR = 0.20). The favorability values, which represent the degree to which local conditions lead to a higher or lower probability than the overall prevalence of the species presence in the area, were more uniform across species. Again, the Mediterranean Gull had the lowest average values (M = 0.19, IQR = 0.31), while the Hooded Crow (Corvus cornix) had the highest (M = 0.65, IQR = 0.46).

3.2. Outbreaks Models

3.2.1. Univariable Models

The median COR values ranged between 0.05 and 0.13, the AUC between 0.58 and 0.73, and the TSS between 0.24 and 0.41. RF generally obtained significantly higher values for all three metrics. BRT provided good metrics values. GLM, GAM, and MaxEnt returned good classification and discrimination metrics (AUC and TSS) but relatively poor COR values (Supporting Information 9).

In general, the species showed a modest to low association with the occurrence of AI outbreaks. The response curves of the individual species are positively, even though weakly, associated with an increase in the probability of AI outbreak occurrences, except for the Eurasian Blackbird and the Eurasian Blue tit, where a negative association is observed. The species that exhibited higher correlations belonged mainly to the Pelecaniformes and Charadriiformes orders (Figure 1).

[figure(s) omitted; refer to PDF]

3.2.2. Species Clusters

The cluster analysis identified seven different groups of species that exhibited similar predicted distributions, which means high collinearity between the species’ favorability predictions (Figure 2a). The distribution of the species among the various clusters aligns well with the ecology and taxonomic order of these species (Supporting Information 10). Water birds are distributed across three clusters (i.e., Clusters 1, 3, and 4), while terrestrial species are present in the remaining four clusters (Clusters 2, 5, 6, and 7), except for the common pheasant (Phasianus colchicus), which falls within Cluster 3. Most Pelecaniformes belong to Cluster 3, while Anseriformes and Charadriiformes are grouped together in Cluster 4. A few exceptions are represented by the black-winged stilt (Himantopus himantopus, Cluster 3), the gray heron (Ardea cinerea, Cluster 4), and the Mediterranean gull, which does not fall in any cluster. As for terrestrial birds, they are predominantly grouped in Cluster 2. The Eurasian blackbird and the Eurasian blue tit cluster together, while the barn swallow and common buzzard do not cluster with any other species. The nMDS analysis (Figure 2b) confirms the groupings previously obtained from the cluster analysis and shows that Cluster 2, Cluster 3, and Cluster 4 are quite close to each other, indicating that these species have more similar favorability distributions. The species chosen as representative of the whole cluster and their favorability maps are shown in Figure 3.

[figure(s) omitted; refer to PDF]

3.2.3. Multivariable Model

The multivariate ensemble model considered the representative species for each of the seven clusters as the “best regressors” for a comprehensive description of the association between the outbreak occurrence probability and the wild bird species’ favorability. The model also estimated the importance of each species group, helping identify which group contributed the most in HPAI spillover events into poultry (Figure 4). The preliminary multicollinearity assessment showed acceptable VIF scores to keep all the selected predictors (all VIF < 10; maximum Pearson correlation coefficient of 0.76 between Cluster 3 and Cluster 4).

[figure(s) omitted; refer to PDF]

Cluster 3 (including most of the Pelecaniformes, the common pheasant, the black-winged stilt, and the common moorhen—Gallinula chloropus) showed the highest average importance (Imp = 28%) and a strong positive association with the emergence of new outbreaks. A moderate association was also observed for Cluster 4, which accounts for the species typically considered reservoirs of the disease (i.e., Anseriformes and Charadriiformes), although its average importance is lower than Cluster 3 (Imp = 17%). Cluster 1, which contains only the Mediterranean gull species, also shows a moderate positive association and moderate importance (Imp = 11%). Cluster 5 (Eurasian blackbird and Eurasian blue tit) and Cluster 2 showed general negative response curves and were ranked second (Imp = 23%) and third (Imp = 20%) by importance, respectively. Both Cluster 6 (Common Buzzard) and Cluster 7 (Barn Swallow) were associated with an increased outbreak probability occurrence just at higher values of species favorability.

Figure 5a shows the predicted probability of outbreaks occurrences in poultry based on the distributions of wild bird species. The range of values is moderate and does not exceed 0.50. The areas at higher risk of spillover are the ones located near wetlands, both coastal and inland. The DPPA is the area where the likelihood of observing an HPAI outbreak generally is higher, compared to the rest of the northern Italy regions. The metrics for the goodness of fit (Figure 5b) reflect good discrimination and classification abilities of the final model, while the COR index shows a modest performance (0.19), although better than the values obtained from the univariable analyses.

[figure(s) omitted; refer to PDF]

4. Discussion

This study examined the association between the distributions of wild bird species observed in proximity to a number of poultry farms located in the DPPA [33] and the occurrence of HPAI H5N8 outbreaks in Northern Italy during the 2017–2018 epidemic.

The univariable models allowed to preliminarily explore which wild bird species had a stronger association with potential spillover events. Species with the strongest association with primary outbreaks in poultry resulted in belonging to the Pelecaniformes, and Charadriiformes orders. Among the Pelecaniformes species, the favorability distributions of birds belonging to the Ardeidae family resulted in having the strongest correlation with AI outbreak occurrences. However, only the cattle egret (Bubulcus ibis) was also frequently observed in proximity to farms during the 2019 camera trap survey, while species such as the little egret (Egretta garzetta) and the purple heron (Ardea purpurea) were detected with a much lower frequency [33]. Although interest in the role of these species in AI epidemiology and their susceptibility to AIVs, as testified by several recent studies, a comprehensive understanding on the impact of Ardeids on AI circulation dynamics is still lacking [79–81]. This growing interest is mainly due to the land use of Ardeidae species, as they are frequently observed both in wetlands, thus sharing habitat with known reservoir species such as birds belonging to the Anseriformes order, and in urbanized and agricultural areas [17], suggesting a potential role as bridge species. Furthermore, since the emergence of the clade 2.3.4.4 HPAIV in 2013, there has been an increase in the number of reported positive cases in herons in Asian and European countries [79], including Italy [47], also accounting for seroconversion cases [80], In particular, Soda et al. [79] specifically investigated the susceptibility of the Ardeidae by testing gray heron, little egret, night heron (Nycticorax nycticorax), and intermediate egret (Ardea intermedia) after experimental infections. The study demonstrated that most of these species are susceptible to H5 HPAI viruses, potentially with lethal outcomes, and can efficiently shed the virus via oral secretions. The authors also hypothesized that the regurgitating and fishing behavior of these species, coupled with the sharing of wetlands normally frequented by AIVs reservoir species, may also facilitate the interspecies viral transmission. The role of cattle egret in the disease ecology and epidemiology is still uncertain; while AI viral RNA has been detected in nature in cattle egret [82], and a single AI outbreak was reported in proximity to an infected broiler farm [81], other studies indicated no AI positive birds belonging to this species [17, 27]. However, Wu et al. [83] found a weak positive association between cattle egret distribution and the presence of HPAI outbreaks in Taiwan.

Also, Charadriiformes resulted positively associated with AI outbreaks; however, they were rarely detected in proximity to the 10 poultry farms in the previous study [33]. Despite these species being observed only on a few days and mainly in springtime, their potential role in the AI dynamics has been reported in the literature for both HPAI and LPAI viruses [84]. Furthermore, the HPAI H5N1 subtype that massively circulated in gull populations in 2022–2023 was identified in multiple domestic outbreaks as well, indicating potential direct and indirect introductions from gulls to poultry farms [14, 85].

As for Anseriformes, the distribution of Mallards (Anas platyrhynchos), notably one of the most important reservoirs of AIVs [18], showed relatively weaker associations. In fact, Martelli et al. [33] detected a scant presence of mallards in the proximity of poultry farms, especially during autumn–winter seasons. This was observed in particular for larger farms, which are generally avoided by mallards, most probably because they are rarely located in proximity to water or wetlands, suitable environments for mallards [17]. However, the presence of ponds, channels, and surface waters within or near poultry farms could still attract Anseriformes species [86], and the impact of these species on the AIVs transmission dynamics may largely depend on the environment surrounding the farms.

Negative associations were observed only for two Passeriformes species: the Eurasian blackbird and the Eurasian blue tit. These species, together with the common buzzard and the barn swallow, were included in the final model despite their model performance did not satisfy the selected threshold for the metrics. In fact, the Eurasian blackbird was retained due to its frequent detection near the farms’ premises, as previously described in Martelli et al. [33], while the common buzzard is included in the EFSA list of target species for passive surveillance for AI [22]. Besides, the Eurasian blue tit and the barn swallow were considered species of interest, as they proved capable of carrying and transmitting AIVs [17].

The performance metrics of the models were modest, indicating an overall weak influence of individual species on AI outbreak emergence in poultry. This was not entirely unexpected as highly complex events such as sporadic primary AIV introductions in poultry farms were extremely simplified as a function of the distribution of species highly abundant across the whole study area. However, those preliminary findings were further refined by complementary analyses.

Cluster analysis allowed reducing the 22 species into groups, homogeneous for the characteristics of their favourability distribution, thus permitting to reduce the number of variables to be included in the multivariable analysis. Even though Cluster 2, Cluster 3, and Cluster 4 resulted similar to each other, marked differences were observed in their response curves, hence showing different capacities of affecting the outbreak occurrence probability. Cluster 3, grouping most of the Pelecaniformes species, resulted in having the highest importance amongst those showing a positive association with AI outbreak occurrence. Similarly, Cluster 4 included species most frequently considered as HPAI reservoirs. Cluster 2, which mostly encompassed terrestrial bird species, had a variable response curve, showing a marked negative association where the favourability of the species was higher. This might be due to the marked variability in the response curves of the single species included in Cluster 2. Although the distributions of species belonging to these three clusters showed generally similar distribution ranges and patterns when observed at a regional scale level, the different responses to specific environmental variables permitted differentiating their favorability maps on a local scale, ultimately allowing to highlight different levels of association with the occurrence of AI outbreaks.

The common pheasant, by itself, showed a relatively weak association with the occurrence of AI outbreaks. However, its inclusion in Cluster 3 indicated a strong collinearity with the little egret, the purple heron, and the cattle egret, which had the strongest association with the HPAI occurrences. Furthermore, the common pheasant is a species bred to be released for hunting purposes; as such, it has been considered to represent a risk of AIVs transmission to wild bird populations [87]. This is also corroborated by recent findings of HPAI H5N8 and H5N1 virus in both farmed and released pheasants on a hunting estate in Finland in 2021, with multiple HPAI cases in various species of wild birds in the ensuing months [88]. Furthermore, several studies have demonstrated that pheasants can transmit both HPAI and LPAI viruses [89, 90], stressing its potential role in disease spread and prompting for the implementation of strict monitoring and control measures of this species.

The modest association between Cluster 2, which includes the main species of synanthropic terrestrial birds, and the spillover probability should not lead to the exclusion of these birds from the list of potential spreaders of AIVs [17]. In fact, many of these species have been frequently observed near poultry farms [33]. Their ubiquitous presence across the study area could have flatted the variability in the distribution of the species, potentially masking their real impact on the transmission dynamics of the disease. In fact, these species may have a role as the final link in the wild–domestic interface. For instance, the Columbiformes are acknowledged as being a mechanical vector for AIVs, as they have scarce susceptibility to the infection [91]; however, effective viral replication and transmission were proven for HPAI H5N6 virus in experimental conditions [8, 92], and a naturally infected wood pigeon (Columba palumbus) were found dead in Germany in 2022 [93]. Conversely, corvid species such as the Eurasian magpie (Pica pica) and the hooded crow were found to be highly susceptible to the infection [17], and virus antigens were detected in both the respiratory and digestive tract of a naturally infected magpie [94, 95].

The final outbreak prediction map shows the highest probabilities of HPAI outbreak occurrence along the Adriatic coastal area, which is characterized by coastal wetlands and marshes. However, the favorability of species with the strongest association with HPAI in poultry (Cluster 3) mostly depends on the presence of inland wetlands, which are in close proximity to the DPPAs recurrently affected by AI epidemic waves. Therefore, while the coastal area can be assumed to be the basin of reservoir species (i.e., Anseriformes and Charadriiformes), the inland areas can be hypothesized to host the bird species acting as a bridge between wetlands and poultry premises. AI dynamics and the transmission chain between the wild and the domestic compartments are complex phenomena with multiple actors involved. If the presence of a reservoir species (e.g., Cluster 4) is necessary to trigger the spill-over to the domestic sector, this might be subordinate to the presence of one or more bridge species (e.g., Cluster 3) to transport the virus from the wetlands areas to the farms. Moreover, terrestrial birds such as Corvids and Columbiformes are abundantly distributed in areas typically affected by AI circulation but also in environments where waterbirds are highly present, suggesting that their role in the final AIVs transmission to poultry farms should not be completely ruled out.

Despite the valuable insights provided by this study, some inherent limitations should be acknowledged. The species occurrence data for the 40 wild bird species observed in proximity to farms allowed for reliably defining the distribution only for 22 species, based on the selected performance metrics, hence including just a partial set of species in further analyses. The species excluded had a limited presence in the study area, and their inclusion would have potentially biased the results and conclusions of the study. The species seasonality and the temporal association between the outbreak occurrence and the species’ favorability in the same period were not considered. Associating the favorability of a species defined on a seasonal basis with the time points of outbreak occurrence might have led to a more precise assessment of each species’ contribution. However, the occurrence data available would likely be insufficient to construct robust favorability maps for each species and each season, thus limiting both the number of included species in the analysis and the reliability of the results. The DPPA, where most outbreaks occurred, is characterized by a marked environmental uniformity, which affects the distribution of wild bird species and the degree of association between the wild birds’ distributions and the occurrence of AI outbreaks.

Although several mammal species were observed during the camera trap survey in close proximity to poultry premises (i.e., Rattus spp., Myocastor coypus, and Erinaceus europaeus), they were not included in the present study. The reason is that the role of small mammal species is assumed to be limited to that of mechanical vectors [96, 97], potentially acting as the final link between external areas and farming facilities where the animals are housed. Rather than focusing on the potential role of mammals, the specific aim of this study was to identify which wild bird species distribution (the most likely involved in the dynamics of AIV transmission from wetland areas to poultry farms) was more strongly associated with outbreak occurrence in the domestic sector. Ecological drivers of AI dynamics have recently attracted interest due to the rapidly shifting epidemiology characteristics of AIVs. While several studies have investigated potential correlations between environmental variables and outbreaks of AI [98, 99], studies on possible relationships between the distributions of wild birds and the occurrence of AI outbreaks are scarce [83, 100]. The findings of our study emphasize the need to thoroughly investigate the virus-shedding patterns and transmission dynamics of AIVs, focusing on the wild bird species that appear to be most strongly associated with the circulation of AI outbreaks. At the same time, more efforts should be spent on new in-depth ecological investigations, possibly through (i) dedicated data collection, (ii) model approaches capable of accounting for species interactions, and (iii) including the assessment of the role of the environment in the maintenance and transmission of AIVs among wild birds. This could be achieved through the planning and implementation of specific environmental sampling activities that would integrate the existing surveillance plans for AI, with a specific focus on abiotic matrices [101], which might represent an additional meaningful source of information for the AIVs early detection. All together, the integration of new evidences would allow for generating novel hypotheses on the most likely combinations of species and environmental conditions that could facilitate the transmission chain of infection from wild to domestic populations.

5. Conclusions

Wild birds are of fundamental importance in the spread and transmission of AIVs, but their role in the epidemiology of AI is still elusive. Clarifying the interface between wild and domestic birds would be crucial to foster site-specific and ecologically informed risk mitigation strategies for AI outbreaks and safeguarding public health [1]. The results of this study allowed for the identification of groups of wild bird species whose distributions were most strongly associated with the geographic location of the 49 primary outbreaks that occurred during the HPAI H5N8 epidemic of 2017–2018 in Italy, also classifying them in order of importance.

The present study provided interesting results in this regard. In fact, the species belonging to the orders considered reservoirs (i.e., Anseriformes and Charadriiformes), although positively associated with the onset of outbreaks, were of lesser importance in explaining the phenomenon compared to other wild birds, particularly Pelecaniformes, Gruiformes, and Galliformes.

Author Contributions

Conceptualization and methodology: P. Mulatti, D. Fornasiero, and L. Martelli. Software: L. Martelli and D. Fornasiero. Formal analysis: L. Martelli., D. Fornasiero, and J. A. Martínez-Lanfranco. Investigation: Francesco Scarton and A. Spada. Data curation: Francesco Scarton, A. Spada, D. Fornasiero, and L. Martelli. Writing–original draft preparation: L. Martelli. Writing–review and editing: D. Fornasiero, P. Mulatti, Francesca Scolamacchia, Francesco Scarton, A. Spada, J. A. Martínez-Lanfranco, and G. Manca. Visualization: L. Martelli. Supervision: P. Mulatti. All authors have read and agreed to the published version of the manuscript.

References

[1] J. H. Verhagen, R. A. M. Fouchier, N. Lewis, "Highly Pathogenic Avian Influenza Viruses at the Wild-Domestic Bird Interface in Europe: Future Directions for Research and Surveillance," Viruses, vol. 13 no. 2,DOI: 10.3390/v13020212, 2021.

[2] D. E. Swayne, D. L. Suarez, "Highly Pathogenic Avian Influenza," Revue Scientifique et Technique de l’OIE, vol. 19 no. 2, pp. 463-482, DOI: 10.20506/rst.19.2.1230, 2000.

[3] C. Narrod, J. Zinsstag, M. Tiongco, "A One Health Framework for Estimating the Economic Costs of Zoonotic Diseases on Society," EcoHealth, vol. 9 no. 2, pp. 150-162, DOI: 10.1007/s10393-012-0747-9, 2012.

[4] T. Kuiken, R. A. M. Fouchier, M. P. G. Koopmans, "Being Ready for the Next Influenza Pandemic?," The Lancet Infectious Diseases, vol. 23 no. 4, pp. 398-399, DOI: 10.1016/S1473-3099(23)00117-2, 2023.

[5] M. Leguia, A. Garcia-Glaessner, B. Muñoz-Saavedra, "Highly Pathogenic Avian Influenza A (H5N1) in Marine Mammals and Seabirds in Peru," Nature Communications, vol. 14 no. 1,DOI: 10.1038/s41467-023-41182-0, 2023.

[6] J. M. Fair, N. Al-Hmoud, M. Alrwashdeh, "Transboundary Determinants of Avian Zoonotic Infectious Diseases: Challenges for Strengthening Research Capacity and Connecting Surveillance Networks," Frontiers in Microbiology, vol. 15,DOI: 10.3389/fmicb.2024.1341842, 2024.

[7] A. Pohlmann, J. King, A. Fusaro, "Has Epizootic Become Enzootic? Evidence for a Fundamental Change in the Infection Dynamics of Highly Pathogenic Avian Influenza in Europe, 2021," MBio, vol. 13 no. 4,DOI: 10.1128/mbio.00609-22, 2022.

[8] G. Graziosi, C. Lupini, E. Catelli, S. Carnaccini, "Highly Pathogenic Avian Influenza (HPAI) H5 Clade 2.3.4.4b Virus Infection in Birds and Mammals," Animals (Basel), vol. 14 no. 9,DOI: 10.3390/ani14091372, 2024.

[9] World Organisation for Animal Health, "World Animal Health Information System (WAHIS)," 2023. (accessed on 29 August 2024) https://www.oie.int/wahis_2/public/wahid.php/Wahidhome/Home

[10] O. Flavia, K. Sascha, S. Carola, "The Role of Mammals in Avian Influenza: A Review," EFSA Supporting Publications, vol. 21 no. 3,DOI: 10.2903/sp.efsa.2024.EN-8692, 2024.

[11] M. Agüero, I. Monne, A. Sánchez, "Highly Pathogenic Avian Influenza A(H5N1) Virus Infection in Farmed Minks, Spain, October 2022," Eurosurveillance, vol. 28 no. 3,DOI: 10.2807/1560-7917.ES.2023.28.3.2300001, 2023.

[12] C. Adlhoch, A. Fusaro, J. L. Gonzales, "Avian Influenza Overview December 2020 – February 2021," EFSA Journal, vol. 19 no. 3,DOI: 10.2903/j.efsa.2021.6497, 2021.

[13] C. Adlhoch, A. Fusaro, J. L. Gonzales, "Avian Influenza Overview June – September 2022," EFSA Journal, vol. 20 no. 10,DOI: 10.2903/J.EFSA.2022.7597, 2022.

[14] C. Adlhoch, A. Fusaro, J. L. Gonzales, "Avian influenza overview December 2022 – March 2023," EFSA Journal. European Food Safety Authority, vol. 21 no. 3,DOI: 10.2903/j.efsa.2023.7917, 2023.

[15] E. Niqueux, O. Guionie, A. Schmitz, J. Hars, V. Jestin, "Presence of Serum Antibodies to Influenza A Subtypes H5 and N1 in Swans and Ibises in French Wetlands, Irrespective of Highly Pathogenic H5N1 Natural Infection," Avian Diseases, vol. 54 no. s1, pp. 502-508, DOI: 10.1637/8804-040109-ResNote.1, 2010.

[16] European Food Safety Authority (EFSA), I. Aznar, F. Baldinelli, A. Papanikolaou, A. Stoicescu, Y. Van der Stede, "Annual Report on Surveillance for Avian Influenza in Poultry and Wild Birds in Member States of the European Union in 2020," EFSA Journal, vol. 19 no. 12,DOI: 10.2903/j.efsa.2021.6953, 2021.

[17] S. A. Shriner, J. J. Root, "A Review of Avian Influenza A Virus Associations in Synanthropic Birds," Viruses, vol. 12 no. 11,DOI: 10.3390/v12111209, 2020.

[18] R. G. Webster, W. J. Bean, O. T. Gorman, T. M. Chambers, Y. Kawaoka, "Evolution and Ecology of Influenza A Viruses," Microbiological Reviews, vol. 56 no. 1, pp. 152-179, DOI: 10.1128/mr.56.1.152-179.1992, 1992.

[19] N. J. Hill, M. A. Bishop, N. S. Trovão, "Ecological Divergence of Wild Birds Drives Avian Influenza Spillover and Global Spread," PLOS Pathogens, vol. 18 no. 5,DOI: 10.1371/journal.ppat.1010062, 2022.

[20] J. Brown, R. Poulson, D. Carter, "Susceptibility of Avian Species to North American H13 Low Pathogenic Avian Influenza Viruses," Avian Diseases, vol. 56 no. 4s1, pp. 969-975, DOI: 10.1637/10158-040912-Reg.1, 2012.

[21] W. Puryear, K. Sawatzki, N. Hill, "Highly Pathogenic Avian Influenza A(H5N1) Virus Outbreak in New England Seals, United States," Emerging Infectious Diseases, vol. 29 no. 4, pp. 786-791, DOI: 10.3201/eid2904.221538, 2023.

[22] EFSA Panel on Animal Health and Welfare (AHAW), S. More, D. Bicout, "Avian influenza," EFSA Journal, vol. 15 no. 10, pp. 4991-233, DOI: 10.2903/j.efsa.2017.4991, 2017.

[23] D.-H. Lee, K. Bertran, J.-H. Kwon, D. E. Swayne, "Evolution, Global Spread, and Pathogenicity of Highly Pathogenic Avian Influenza H5Nx Clade 2.3.4.4," Journal of Veterinary Science, vol. 18 no. S1, pp. 269-280, DOI: 10.4142/jvs.2017.18.S1.269, 2017.

[24] N. Gaidet, J. Cappelle, J. Y. Takekawa, "Potential Spread of Highly Pathogenic Avian Influenza H5N1 by Wildfowl: Dispersal Ranges and Rates Determined From Large-Scale Satellite Telemetry," Journal of Applied Ecology, vol. 47 no. 5, pp. 1147-1157, DOI: 10.1111/j.1365-2664.2010.01845.x, 2010.

[25] L. A. Reperant, N. S. Fučkar, A. D. M. E. Osterhaus, A. P. Dobson, T. Kuiken, "Spatial and Temporal Association of Outbreaks of H5N1 Influenza Virus Infection in Wild Birds With the 0°C Isotherm," PLOS Pathogens, vol. 6 no. 4,DOI: 10.1371/journal.ppat.1000854, 2010.

[26] G. Wang, D. Zhan, L. Li, "H5N1 Avian Influenza Re-Emergence of Lake Qinghai: Phylogenetic and Antigenic Analyses of the Newly Isolated Viruses and Roles of Migratory Birds in Virus Circulation," Journal of General Virology, vol. 89 no. 3, pp. 697-702, DOI: 10.1099/vir.0.83419-0, 2008.

[27] A. Caron, V. Grosbois, E. Etter, N. Gaidet, M. de Garine-Wichatitsky, "Bridge Hosts for Avian Influenza Viruses at the Wildlife/Domestic Interface: An Eco-Epidemiological Framework Implemented in Southern Africa," Preventive Veterinary Medicine, vol. 117 no. 3-4, pp. 590-600, DOI: 10.1016/j.prevetmed.2014.09.014, 2014.

[28] T. E. Burns, C. Ribble, C. Stephen, "Use of Observed Wild Bird Activity on Poultry Farms and a Literature Review to Target Species as High Priority for Avian Influenza Testing in 2 Regions of Canada," Canadian Veterinary Journal, vol. 53 no. 2, pp. 158-166, 2012.

[29] B. Olsen, V. J. Munster, A. Wallensten, J. Waldenström, A. D. M. E. Osterhaus, R. A. M. Fouchier, "Global Patterns of Influenza A Virus in Wild Birds," Science, vol. 312 no. 5772, pp. 384-388, DOI: 10.1126/science.1122438, 2006.

[30] P. Mulatti, N. Ferrè, S. Marangon, "Spatial Distribution of 2000-2007 Low Pathogenicity Avian Influenza Epidemics in Northern Italy," Pandemic Influenza Virus, pp. 232-247, 2010.

[31] D. Fornasiero, A. Fusaro, B. Zecchin, "Integration of Epidemiological and Genomic Data to Investigate H5N1 HPAI Outbreaks in Northern Italy in 2021–2022," Pathogens, vol. 12 no. 1,DOI: 10.3390/pathogens12010100, 2023.

[32] P. Mulatti, A. Fusaro, F. Scolamacchia, "Integration of Genetic and Epidemiological Data to Infer H5N8 HPAI Virus Transmission Dynamics During the 2016-2017 Epidemic in Italy," Scientific Reports, vol. 8 no. 1,DOI: 10.1038/s41598-018-36892-1, 2018.

[33] L. Martelli, D. Fornasiero, F. Scarton, "Study of the Interface Between Wild Bird Populations and Poultry and Their Potential Role in the Spread of Avian Influenza," Microorganisms, vol. 11 no. 10,DOI: 10.3390/microorganisms11102601, 2023.

[34] W. T. Harvey, P. Mulatti, A. Fusaro, "Spatiotemporal Reconstruction and Transmission Dynamics During the 2016-17 H5N8 Highly Pathogenic Avian Influenza Epidemic in Italy," Transboundary and Emerging Diseases, vol. 68 no. 1, pp. 37-50, DOI: 10.1111/tbed.13420, 2021.

[35] Commission Delegated Regulation (EU), "Commission Delegated Regulation (EU) 2020/689 of 17 December 2019 Supplementing Regulation (EU) 2016/429 of the European Parliament and of the Council as Regards Rules for Surveillance, Eradication Programmes, and Disease-Free Status for Certain Listed and Emerging Diseases," Official Journal of the European Union, vol. 174, pp. 211-340, 2019.

[36] L. Alexakis, A. Fusaro, T. Kuiken, "Avian Influenza Overview March–June 2024," EFSA Journal, vol. 22 no. 7,DOI: 10.2903/j.efsa.2024.8930, 2024.

[37] Cornell Lab of Ornithology, "eBird Basic Dataset. Version: EBD_relDec-2023," 2024.

[38] M. Mazzucato, G. Marchetti, M. Barbujani, "An Integrated System for the Management of Environmental Data to Support Veterinary Epidemiology," Frontiers in Veterinary Science, vol. 10,DOI: 10.3389/fvets.2023.1069979, 2023.

[39] C. O. Justice, E. Vermote, J. R. G. Townshend, "The Moderate Resolution Imaging Spectroradiometer (MODIS): Land Remote Sensing for Global Change Research," IEEE Transactions on Geoscience and Remote Sensing, vol. 36 no. 4, pp. 1228-1249, DOI: 10.1109/36.701075, 1998.

[40] European Environment Agency, "Corine Land Cover 2018," 2020. accessed on 6 July 2024 https://www.eea.europa.eu/data-and-maps/data/external/corine-land-cover-2018

[41] T. G. Farr, P. A. Rosen, E. Caro, "The Shuttle Radar Topography Mission," Reviews of Geophysics, vol. 45 no. 2,DOI: 10.1029/2005RG000183, 2007.

[42] M. Neteler, "Estimating Daily Land Surface Temperatures in Mountainous Environments by Reconstructed MODIS LST Data," Remote Sensing, vol. 2 no. 1, pp. 333-351, DOI: 10.3390/rs1020333, 2010.

[43] WorldClim, "Bioclimatic Variables," 2024. https://www.worldclim.org/data/bioclim.html

[44] The European Parliament and the Council of the European Union, "Directive 2000/60/EC of the European Parliament and of the Council of 23 October 2000 establishing a Framework for Community Action in the Field of Water Policy," Official Journal of the European Communities, 2000.

[45] B. Naimi, N. A. S. Hamm, T. A. Groen, A. K. Skidmore, A. G. Toxopeus, "Where is Positional Uncertainty a Problem for Species Distribution Modelling?," Ecography, vol. 37 no. 2, pp. 191-203, DOI: 10.1111/j.1600-0587.2013.00205.x, 2014.

[46] L. E. Escobar, A. Lira-Noriega, G. Medina-Vogel, A. Townsend Peterson, "Potential for Spread of the White-Nose Fungus (Pseudogymnoascus Destructans) in the Americas: Use of Maxent and NicheA to Assure Strict Model Transference," Geospatial Health, vol. 9 no. 1, pp. 221-229, DOI: 10.4081/gh.2014.19, 2014.

[47] R. A. Baquero, A. M. Barbosa, D. Ayllón, "Potential Distributions of Invasive Vertebrates in the Iberian Peninsula Under Projected Changes in Climate Extreme Events," Diversity and Distributions, vol. 27 no. 11, pp. 2262-2276, DOI: 10.1111/ddi.13401, 2021.

[48] P. F. Pereira, A. M. Barbosa, C. Godinho, P. A. Salgueiro, R. R. Silva, R. Lourenço, "The Spread of the Red-Billed Leiothrix ( Leiothrix lutea ) in Europe: The Conquest by an Overlooked Invader?," Biological Invasions, vol. 22 no. 2, pp. 709-722, DOI: 10.1007/s10530-019-02123-5, 2020.

[49] M. B. Kursa, W. R. Rudnicki, "Feature Selection With the Boruta Package," Journal of Statistical Software, vol. 36 no. 11,DOI: 10.18637/jss.v036.i11, 2010.

[50] T. A. Lake, R. D. Briscoe Runquist, D. A. Moeller, "Predicting Range Expansion of Invasive Species: Pitfalls and Best Practices for Obtaining Biologically Realistic Projections," Diversity and Distributions, vol. 26 no. 12, pp. 1767-1779, DOI: 10.1111/ddi.13161, 2020.

[51] J. A. Swets, "Measuring the Accuracy of Diagnostic Systems," Science, vol. 240 no. 4857, pp. 1285-1293, DOI: 10.1126/science.3287615, 1988.

[52] A. H. Fielding, J. F. Bell, "A Review of Methods for the Assessment of Prediction Errors in Conservation Presence/Absence Models," Environmental Conservation, vol. 24 no. 1, pp. 38-49, DOI: 10.1017/S0376892997000088, 1997.

[53] O. Allouche, A. Tsoar, R. Kadmon, "Assessing the Accuracy of Species Distribution Models: Prevalence, Kappa and the True Skill Statistic (TSS)," Journal of Applied Ecology, vol. 43 no. 6, pp. 1223-1232, DOI: 10.1111/j.1365-2664.2006.01214.x, 2006.

[54] J. Pearce, S. Ferrier, "Evaluating the Predictive Performance of Habitat Models Developed Using Logistic Regression," Ecological Modelling, vol. 133 no. 3, pp. 225-245, DOI: 10.1016/S0304-3800(00)00322-7, 2000.

[55] M. E. Miller, S. L. Hui, W. M. Tierney, "Validation Techniques for Logistic Regression Models," Statistics in Medicine, vol. 10 no. 8, pp. 1213-1226, DOI: 10.1002/sim.4780100805, 1991.

[56] G. Rapacciuolo, D. B. Roy, S. Gillings, R. Fox, K. Walker, A. Purvis, "Climatic Associations of British Species Distributions Show Good Transferability in Time But Low Predictive Accuracy for Range Change," PLoS ONE, vol. 7 no. 7,DOI: 10.1371/journal.pone.0040212, 2012.

[57] P. Acevedo, R. Real, "Favourability: Concept, Distinctive Characteristics and Potential Usefulness," Naturwissenschaften, vol. 99 no. 7, pp. 515-522, DOI: 10.1007/s00114-012-0926-0, 2012.

[58] R. Real, A. M. Barbosa, J. W. Bull, "Species Distributions, Quantum Theory, and the Enhancement of Biodiversity Measures," Systematic Biology, vol. syw072,DOI: 10.1093/sysbio/syw072, 2016.

[59] R. Real, A. M. Barbosa, J. M. Vargas, "Obtaining Environmental Favourability Functions From Logistic Regression," Environmental and Ecological Statistics, vol. 13 no. 2, pp. 237-245, DOI: 10.1007/s10651-005-0003-3, 2006.

[60] N. Ahmed, C. Atzberger, W. Zewdie, "Species Distribution Modelling Performance and its Implication for Sentinel-2-Based Prediction of Invasive Prosopis juliflora in Lower Awash River Basin, Ethiopia," Ecological Processes, vol. 10 no. 1,DOI: 10.1186/s13717-021-00285-6, 2021.

[61] T. A. Lima, R. Beuchle, A. Langner, R. C. Grecchi, V. C. Griess, F. Achard, "Comparing Sentinel-2 MSI and Landsat 8 OLI Imagery for Monitoring Selective Logging in the Brazilian Amazon," Remote Sensing, vol. 11 no. 8,DOI: 10.3390/rs11080961, 2019.

[62] M. B. Lyons, D. A. Keith, S. R. Phinn, T. J. Mason, J. Elith, "A Comparison of Resampling Methods for Remote Sensing Classification and Accuracy Assessment," Remote Sensing of Environment, vol. 208 no. September 2017, pp. 145-153, DOI: 10.1016/j.rse.2018.02.026, 2018.

[63] J. Elith, H. Graham, C. P. Anderson, "Novel Methods Improve Prediction of Species’ Distributions From Occurrence Data," Ecography, vol. 29 no. 2, pp. 129-151, DOI: 10.1111/j.2006.0906-7590.04596.x, 2006.

[64] C. F. Dormann, J. Elith, S. Bacher, "Collinearity: A Review of Methods to Deal With it and a Simulation Study Evaluating Their Performance," Ecography, vol. 36 no. 1, pp. 27-46, DOI: 10.1111/j.1600-0587.2012.07348.x, 2013.

[65] M. Greenacre, R. Primicerio, Multivariate Analysis of Ecological Data, 2013.

[66] R Core Team, "R: A Language and Environment for Statistical Computing," 2022. https://www.r-project.org/

[67] Posit team, "RStudio: Integrated Development Environment for R," 2024. Posit Software http://www.posit.co/

[68] R. J. Hijmans, "Raster: Geographic Data Analysis and Modeling (R Package Version 3.6-14)," 2023. https://cran.r-project.org/package=raster

[69] R. Hijmans, S. Phillips, J. Leathwick, J. Elith, "Dismo: Species Distribution Modeling," 2023. R package version 1.3-14 https://cran.r-project.org/package=dismo

[70] R. J. Hijmans, "Terra: Spatial Data Analysis (R Package Version 1.7-3)," 2023. https://cran.r-project.org/package=terra

[71] A. M. Barbosa, "FuzzySim: Applying Fuzzy Logic to Binary Similarity Indices in Ecology," Methods in Ecology and Evolution, vol. 6 no. 7, pp. 853-858, DOI: 10.1111/2041-210X.12372, 2015.

[72] R. Bivand, T. Keitt, B. Rowlingson, "Rgdal: Bindings for the “Geospatial” Data Abstraction Library (R Package Version 1.6-7)," 2023. https://cran.r-project.org/package=rgdal

[73] R. S. Bivand, E. Pebesma, V. Gómez-Rubio, Applied Spatial Data Analysis With R,DOI: 10.1007/978-1-4614-7618-4, 2013.

[74] E. J. Pebesma, R. Bivand, "Classes and Methods for Spatial Data in R," R News, vol. 5 no. 2, 2005. https://cran.r-project.org/doc/Rnews/

[75] C. Carlson, "Embarcadero: Species Distribution Models With BART, Version 1.2.0.1003," 2023.

[76] V. Dorie, "Dbarts: Discrete Bayesian Additive Regression Trees Sampler (R Package Version 0.9-23)," 2023. https://cran.r-project.org/package=dbarts

[77] B. Naimi, M. B. Araújo, "Sdm: A Reproducible and Extensible R Platform for Species Distribution Modelling," Ecography, vol. 39 no. 4, pp. 368-375, DOI: 10.1111/ecog.01881, 2016.

[78] QGIS Development Team, "QGIS Geographic Information System," 2009. Open Source Geospatial Foundation http://qgis.org

[79] K. Soda, Y. Tomioka, T. Usui, "Susceptibility of Herons (family: Ardeidae) to Clade 2.3.2.1 H5N1 Subtype High Pathogenicity Avian Influenza Virus," Avian Pathology, vol. 51 no. 2, pp. 146-153, DOI: 10.1080/03079457.2021.2022599, 2022.

[80] G. Wang, T. Zhang, X. Li, "Serological Evidence of H7, H5 and H9 Avian Influenza Virus Co-Infection Among Herons in a City Park in Jiangxi, China," Scientific Reports, vol. 4 no. 1,DOI: 10.1038/srep06345, 2014.

[81] B. M. Ahmed, M. M. El-Shazly, A. A. El-Sanousi, Y. I. Youssef, "Isolation of H5 Highly Pathogenic Avian Influenza Virus From Cattle Egret ( Bubulcus ibis ) Near Affected Broiler Chicken Flocks in Egypt," Journal of Virological Sciences, vol. 7 no. 2, pp. 62-70, 2017.

[82] J. L. Siembieda, C. K. Johnson, C. Cardona, "Influenza A Viruses in Wild Birds of the Pacific Flyway, 2005-2008," Vector-Borne and Zoonotic Diseases, vol. 10 no. 8, pp. 793-800, DOI: 10.1089/vbz.2009.0095, 2010.

[83] H. D. I. Wu, R. S. Lin, W. H. Hwang, "Integrating Citizen Scientist Data Into the Surveillance System for Avian Influenza Virus, Taiwan," Emerging Infectious Diseases, vol. 29 no. 1, pp. 45-53, DOI: 10.3201/eid2901.220659, 2023.

[84] K. Tarasiuk, A. Kycko, M. Knitter, "Pathogenicity of Highly Pathogenic Avian Influenza H5N8 Subtype for Herring Gulls ( Larus argentatus ): Impact of Homo- and Heterosubtypic Immunity on the Outcome of Infection," Veterinary Research, vol. 53 no. 1,DOI: 10.1186/s13567-022-01125-x, 2022.

[85] G. M. McPhail, S. M. Collins, T. V. Burt, "Geographic, Ecological, and Temporal Patterns of Seabird Mortality During the 2022 HPAI H5N1 Outbreak on the Island of Newfoundland," BioRxiv,DOI: 10.1101/2024.01.17.575746, 2024.

[86] F. McDuie, E. L. Matchett, D. J. Prosser, "Pathways for Avian Influenza Virus Spread: GPS Reveals Wild Waterfowl in Commercial Livestock Facilities and Connectivity With the Natural Wetland Landscape," Transboundary and Emerging Diseases, vol. 69 no. 5, pp. 2898-2912, DOI: 10.1111/tbed.14445, 2022.

[87] A. Gray, "Releasing Pheasants Could Pose a Risk of Transmitting HPAI to Some Wild Bird Species," Veterinary Record, vol. 192 no. 3,DOI: 10.1002/vetr.00100073, 2023.

[88] N. Tammiranta, M. Isomursu, A. Fusaro, "Highly Pathogenic Avian Influenza A (H5N1) Virus Infections in Wild Carnivores Connected to Mass Mortalities of Pheasants in Finland," Infection, Genetics and Evolution, vol. 111,DOI: 10.1016/j.meegid.2023.105423, 2023.

[89] Y. Liang, C. K. Hjulsager, A. H. Seekings, "Pathogenesis and Infection Dynamics of High Pathogenicity Avian Influenza Virus (HPAIV) H5N6 (Clade 2.3.4.4b) in Pheasants and Onward Transmission to Chickens," Virology, vol. 577 no. September, pp. 138-148, DOI: 10.1016/j.virol.2022.10.009, 2022.

[90] K. Bertran, R. Dolz, N. Majó, "Pathobiology of Avian Influenza Virus Infection in Minor Gallinaceous Species: A Review," Avian Pathology, vol. 43 no. 1,DOI: 10.1080/03079457.2013.876529, 2014.

[91] C. Abolnik, "A Current Review of Avian Influenza in Pigeons and Doves (Columbidae)," Veterinary Microbiology, vol. 170 no. 3-4, pp. 181-196, DOI: 10.1016/j.vetmic.2014.02.042, 2014.

[92] S. Jeong, J.-H. Kwon, S.-H. Lee, "Subclinical Infection and Transmission of Clade 2.3.4.4 H5N6 Highly Pathogenic Avian Influenza Virus in Mandarin Duck ( Aix galericulata ) and Domestic Pigeon (Columbia Livia Domestica)," Viruses, vol. 13 no. 6,DOI: 10.3390/v13061069, 2021.

[93] M. Peters, J. King, P. Wohlsein, C. Grund, T. Harder, "Genuine Lethal Infection of a Wood Pigeon ( Columba palumbus ) With High Pathogenicity Avian Influenza H5N1, Clade 2.3.4.4b, in Germany, 2022," Veterinary Microbiology, vol. 270,DOI: 10.1016/j.vetmic.2022.109461, 2022.

[94] V. Caliendo, L. Leijten, L. Begeman, "Enterotropism of Highly Pathogenic Avian Influenza Virus H5N8 From the 2016/2017 Epidemic in Some Wild Bird Species," Veterinary Research, vol. 51 no. 1,DOI: 10.1186/s13567-020-00841-6, 2020.

[95] Y. K. Kwon, S. J. Joh, M. C. Kim, "The Susceptibility of Magpies to a Highly Pathogenic Avian Influenza Virus Subtype H5N1," Poultry Science, vol. 89 no. 6, pp. 1156-1161, DOI: 10.3382/ps.2009-00549, 2010.

[96] F. C. Velkers, S. J. Blokhuis, E. J. B. Veldhuis Kroeze, S. A. Burt, "The Role of Rodents in Avian Influenza Outbreaks in Poultry Farms: A Review," Veterinary Quarterly, vol. 37 no. 1, pp. 182-194, DOI: 10.1080/01652176.2017.1325537, 2017.

[97] T. Usui, Y. Uno, K. Tanaka, T. Tanikawa, T. Yamaguchi, "Susceptibility of Synanthropic Rodents ( Mus musculus , Rattus norvegicus and Rattus rattus ) to H5N1 Subtype High Pathogenicity Avian Influenza Viruses," Pathogens, vol. 13 no. 9,DOI: 10.3390/pathogens13090764, 2024.

[98] Y. Si, W. F. de Boer, P. Gong, "Different Environmental Drivers of Highly Pathogenic Avian Influenza H5N1 Outbreaks in Poultry and Wild Birds," PLoS ONE, vol. 8 no. 1,DOI: 10.1371/journal.pone.0053362, 2013.

[99] W. Chen, X. Zhang, W. Zhao, L. Yang, Z. Wang, H. Bi, "Environmental Factors and Spatiotemporal Distribution Characteristics of the Global Outbreaks of the Highly Pathogenic Avian Influenza H5N1," Environmental Science and Pollution Research, vol. 29 no. 29, pp. 44175-44185, DOI: 10.1007/s11356-022-19016-1, 2022.

[100] S. H. Vickers, J. Raghwani, A. C. Banyard, I. H. Brown, G. Fournie, S. C. Hill, "Utilising Citizen Science Data to Rapidly Assess Potential Wild Bridging Hosts and Reservoirs of Infection: Avian Influenza Outbreaks in Great Britain," bioRxiv,DOI: 10.1101/2024.03.28.587127, 2024.

[101] M. Coombe, S. Iwasawa, K. A. Byers, "A Systematic Review and Narrative Synthesis Of the Use Of Environmental Samples For the Surveillance Of Avian Influenza Viruses In Wild Waterbirds," Journal of Wildlife Diseases, vol. 57 no. 1,DOI: 10.7589/JWD-D-20-00082, 2021.