Skip to main content
Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Identifying the Environmental Conditions Favouring West Nile Virus Outbreaks in Europe

  • Matteo Marcantonio ,

    Contributed equally to this work with: Matteo Marcantonio, Annapaola Rizzoli

    Affiliation Department of Biodiversity and Molecular Ecology, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy

  • Annapaola Rizzoli ,

    Contributed equally to this work with: Matteo Marcantonio, Annapaola Rizzoli

    Affiliation Department of Biodiversity and Molecular Ecology, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy

  • Markus Metz,

    Affiliation Department of Biodiversity and Molecular Ecology, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy

  • Roberto Rosà,

    Affiliation Department of Biodiversity and Molecular Ecology, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy

  • Giovanni Marini,

    Affiliation Department of Biodiversity and Molecular Ecology, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy

  • Elizabeth Chadwick,

    Affiliation School of Bioscience, Cardiff University, Cardiff, Wales, United Kingdom

  • Markus Neteler

    Affiliation Department of Biodiversity and Molecular Ecology, Research and Innovation Centre, Fondazione Edmund Mach, San Michele all’Adige, Italy


West Nile Virus (WNV) is a globally important mosquito borne virus, with significant implications for human and animal health. The emergence and spread of new lineages, and increased pathogenicity, is the cause of escalating public health concern. Pinpointing the environmental conditions that favour WNV circulation and transmission to humans is challenging, due both to the complexity of its biological cycle, and the under-diagnosis and reporting of epidemiological data. Here, we used remote sensing and GIS to enable collation of multiple types of environmental data over a continental spatial scale, in order to model annual West Nile Fever (WNF) incidence across Europe and neighbouring countries. Multi-model selection and inference were used to gain a consensus from multiple linear mixed models. Climate and landscape were key predictors of WNF outbreaks (specifically, high precipitation in late winter/early spring, high summer temperatures, summer drought, occurrence of irrigated croplands and highly fragmented forests). Identification of the environmental conditions associated with WNF outbreaks is key to enabling public health bodies to properly focus surveillance and mitigation of West Nile virus impact, but more work needs to be done to enable accurate predictions of WNF risk.


West Nile virus (WNV) is a multi-host mosquito borne virus belonging to the Japanese encephalitis (JE) antigenic complex (genus Flavivirus, family Flaviridae) [1]. Although the majority (∼80%) of human WNV infections are sub-clinical and can pass unnoticed, some 20% of patients experience flu-like symptoms known as West Nile fever (WNF), while approximately 1% develop a severe, and potentially fatal, neuro-invasive disease [2]. While clinical trials for human vaccines are underway [3] prevention currently depends on organized, sustained vector (mosquito) control campaigns and risk communication [48].

Sporadic cases of WNV have been documented in Europe and Africa since it was first identified in Uganda in 1937 [9], but until the 1990s it was considered a low risk for humans and domestic animals. Since then however, WNV has spread rapidly across all populated continents and it is now the most widespread arthropod borne virus in the world [10,11]. In Europe, human cases of WNF have been notified in almost all Eastern, Central, and Southern European countries [12] with hotspots in Italy since 2008 [13], Greece since 2010 [14] and continuous transmission in Russia and Romania since 1996 [15]. The number of WNF cases and the impacts on public health are, so far, limited in Europe relative to other vector borne infection (in 2013, 783 cases of WNF were reported by ECDC in Europe and neighbouring countries, as compared to 45,854 Lyme borreliosis reported by the World Health Organization between 2010 and 2013). However, both escalating case load and increased pathogenicity (e.g. substitution of the NY99 genotype with the more pathogenic WN02 in the USA [16]) are contributing to increased risk. Financial costs associated with the prevention of virus transmission to humans through blood and tissue transplantation are mounting (Blood safety regulation; see [17]).

WNV is maintained in enzootic cycles involving several species of birds, and mosquitoes belonging principally to the Culex pipiens complex [7,1821]. Humans and horses are accidental and dead end hosts since they do not develop a viraemic titre sufficient to infect mosquitoes and amplify the transmission cycle [21]. In Europe, the common house mosquito, Cx. pipiens (Linnaeus, 1758), is considered the principal bridge vector of WNV between birds and mammals (horses and humans), although at least 60 other mosquito species can be found infected with the virus [7,20,21,22]. Culex pipiens occurs in two biological forms, Cx. pipiens pipiens and Cx. pipiens molestus, which can hybridize. Both behaviour and host preference vary between forms, with major implications for risk of transmission to humans depending on their relative abundance [23,24]. WNV ecology in the Old World is complex and several aspects of the WNV transmission cycle are as yet poorly quantified. The co-circulation of at least five lineages with variable pathogenicity and the overlap of new introductions with endemic circulation, render the quantification of the parameters necessary to develop transmission models challenging [10,25,26].

Although favourable environmental conditions for virus transmission seem to occur extensively in the Old World and a widespread circulation of the virus has been demonstrated by serological screening of wildlife and sentinel animals, clinical emergence in humans tends to be unpredictable, sporadic and clustered [12]. The occurrence of spatially and temporally localised hot-spots in emergence is likely to reflect the coincidence of circulating virus strain with favourable environmental (biotic and abiotic) conditions which modulate the interaction between virus, mosquito, and hosts, consequently leading to locally altered pathogen amplification, transmission and disease risks [27,28]. Variation in land use, climate, habitat structure, animal community, human socio-economic status or behaviour can all significantly affect the risk of infection—for example, via impacts on the spatial and temporal distribution of the competent reservoir host assemblages and their immune status, as well as on the local abundance and genetic population structure of mosquito vectors and their vectorial capacity [2933].

At the same time, the availability of simultaneous information on the infection pattern in vectors and birds is lacking at a wide spatial scale, while the cost of integrated surveillance, and the economic and social disparities which affect several EU countries, limit the capacity of high-level institutions to collect detailed and standardised ecological and epidemiological data [35]. Identification of the areas of potential emergence, and predicting temporal and spatial variation in WNV risk therefore remains challenging [34].

The current study aimed to identify environmental factors associated with WNF occurrence in humans across the Old World. We analysed the association between WNF incidence (derived from WNF number of cases as reported annually by the European Centre for Disease Prevention and Control, ECDC) and a wide range of potential predictors including climate, land use, indices of water, vegetation, conservation status, landscape fragmentation and human population density. By identifying key environmental drivers of WNF, we aim to lay the foundations for the development of statistical models able to predict WNF risk at a continental scale.


Epidemiological data

Data used were provided by the European Centre for Disease Prevention and Control (ECDC), [36], compiled from weekly WNF case reports from 146 areas defined at the Nomenclature of Territorial Units level 3 (NUTS3)/Global Administrative Unit Layers level 1 (GAUL1), originating from 16 different countries across western Asia, Europe and northern Africa (Fig. 1). For nation specific details of data collection see [19] and references therein. Weekly data were pooled, to provide annual totals for 2010 to 2012. Population data were obtained for each area using online national statistical databases (Table 1), so that the number of cases per 100,000 inhabitants (hereafter referred to as the 'incidence') of WNF could be calculated per head of population, per year. Areas with no reports of WNF were excluded from analysis, as it was not possible to discriminate between true negatives, and areas where reporting or diagnosis were inadequate.

Fig 1. Spatial variation in cumulative WNF incidence (cases per100,000 population) between 2010 and 2012 is indicated in colour, within areas delineated using NUTS3/GAUL1 administrative boundaries.

Areas with no reported cases of WNF are shown in grey, and delineated using Country boundaries. Peak incidences are reported in red, these being in Volgograd Oblast, North Eastern Greece and Central Tunisia.

Climatic and environmental variables

All climatic and environmental variables were collated as either vector data (protected areas and water bodies) or gridded raster data (all other variables) for the entire study area, and processed using GRASS GIS [37]. Full details of data sources, resolution, and all variables included in statistical analyses are reported in Table 2.

Raw data

Climatic variables were: i) Land Surface Temperature and ii) Precipitation (daily amount of precipitation, and number of days with precipitation).

Environmental variables were: i) Vegetation index, ii) Water index, iii) Land use (land cover classes from GlobCover [38] and ‘Anthropogenic Biomes’ (global ecological patterns created by sustained direct human interactions with ecosystems) from the Anthromes dataset [39], iv) Protected areas, v) Water bodies, and vi) Intensity of light at night (used as a proxy for human population density, [4042]).

Derived data

All variables were summarized within the NUTS3/GAUL1 areas, from which we derived a series of measures described below.

Seasonal averages and anomalies. For datasets incorporating temporal as well as spatial variation (temperature, precipitation, NDVI, NDWI), data from each year and area were aggregated into nine 16 week periods, these being labelled using months instead of weeks, such as 1–4 (January-April), 2–5 (February-May), 3–6 (March-June),…, to 9–12 (September-December) [43]. For each period, we calculated the average (for temperature, NDVI and NDWI) or the cumulative total (precipitation). From these average or cumulative totals we derived anomalies [19] as the difference between the average (or cumulative) value within the given time period, and the average (or cumulative) value for the same time period recorded in the preceding decade, 2001–2010. We then standardized the anomalies by dividing them by the 2001–2010 standard deviation. Standardized anomalies provide more information about the magnitude of the anomalies because influences of dispersion have been removed. This was repeated for each year (2010–2012). Where referred to in the text, each variable (Temp, PrecTot, PrecDays, NDVI, NDWI) is prefixed with Ano. or Av. for standardised anomaly or average, and the relevant period denoted in subscript (e.g. Ano.Temp2–5 = weekly anomaly temperature during months 2–5).

Landscape. For each land use class (GlobCover and Anthromes), protected areas, and water bodies, percentage cover was calculated within each NUTS3/GAUL1 area. Pielou’s evenness index of diversity [44], and the number of patches of homogeneous land use, were calculated using GlobCover data. For night light intensities, the mean and variance were calculated within each NUTS3/GAUL1 area, and are used as proxies for the average population density, and the fragmentation of the human population, respectively.

Statistical analyses

We investigated the association between incidence of WNF and a range of environmental predictors, measured across Europe and adjacent countries, using available epidemiological data from 2010–2012. For all analyses, we used as the response variable the log-transformed annual incidence of WNF, for each NUTS3/GAUL1 area, respectively. All analyses were performed using the 'R' language and environment for statistical computing [45].

Preliminary analyses

In order to select the most appropriate time window for each temporally variable predictor, linear mixed-effects models (LMMs) were fitted to the response variable, for each time window in turn, for both the average and the standardised anomaly of each of the temporally variable environmental predictors. Year and area (NUTS3/GAUL1) were included as random variables. The inclusion of random factors in the model is needed to take under control the clustering of data in different areas and years [46]. The best model was selected for each predictor using An Information Criterion (AICc), with small sample bias adjustment [47].

LMMs were fitted to test the association between WNF incidence and land use. Two models were created, testing in turn for associations with land cover (GlobCover data), and anthropogenic biomes (Anthromes data). Using a process of multi-model inference [4749], we compared all possible models using the R package ‘MuMIn’ [50]. The best models were selected using a threshold of ΔAICc <2 [47].

The full model

Following preliminary stages of variable selection, we developed further LMMs including all remaining environmental variables (Table 2), with year and area included as random variables. As previously, all possible models were compared using multi-model selection. The consensus set of best models were selected using a threshold of ΔAICc <2 [47], and differences in AICc (ΔAICc) between consecutively ranked models were used to calculate weights and relative evidence ratios for each variable. All variables included in the best models were ranked according to their importance, and the relative evidence weight for each model term was calculated (this being the sum of the IC weights of those models in which the term is included, [49]). These data were used to calculate the model-averaged estimates of the coefficients and their standard error. Although we acknowledge that utilising a process of multi-model selection and inference can lead to the testing of spurious models (that we tried to filter out pre-selecting the variables used in the full model), it is an extremely powerful approach that enables us to present a consensus of landscape and weather predictors from multiple models, rather than only a single ‘best’ model, while also considering model selection uncertainty [47]. After testing the averaged model for multicollinearity using the Variance Inflation Factor (VIF) [51], we compared WNF incidence fitted values versus WNF incidence observed values to assess the averaged model goodness of fit.

In order to clarify effect size, predictions were made from the best models for each significant predictor variable in turn. All variables but one were fixed at their median values, and predictions made across the full range of the selected variable. For example, to test the association between temperature and WNF incidence, in a model where temperature, precipitation and NDWI were all significant predictors, precipitation and NDWI were added to the model as constants (fixed at their median measured value) while values for temperature were allowed to vary within their observed range.


WNF incidence

The highest WNF incidences during the 2010–2012 period were reported in Volgograd Oblast, North Eastern Greece and Central Tunisia (Fig. 1). Annual incidence across the entire study area varied from 1.31 cases per 100,000 people in 2011 to 2.66 in 2012 (incidence in 2010 reached an intermediate value of 1.68; Fig. 2). In 2010, the highest incidences were reported in eastern Europe, western Asia (Volgograd Oblast) and Greece. In 2011, the highest incidence was reported in Greece, but overall, incidence was low. In 2012 the average incidence was the highest in Greece and Tunisia.

Fig 2. Barplot reporting the total number of cases (grey) and incidence (cases per 100,000 population, red) of WNF in each year (2010, 2011, 2012).

Preliminary analyses

To predict incidence of WNF, the optimum periods over which to measure temperature, precipitation, NDWI and NDVI were late-winter early spring (Av.PrecDays2–5 and Ano.Temp3–6) and summer (Av.NDWI4–7, Ano.NDVI4–7, Av.Temp6–9), respectively. The anthropogenic biomes selected from the initial subset of 14 classes were “Populated Forests” and “Populated Rangelands” (forests or rangelands with scattered, low density human populations), and the land use classes selected from the initial subset of 13 classes were “Mosaic forests” (forests alternated with other land uses such as grasslands), “Irrigated Croplands” and “Mixed Natural Vegetation”.

The full model

A total of 9 models were identified with ΔAICc <2,. Of the 14 explanatory variables considered in the full model, 8 were represented in the consensus set of best models. The Variance Inflation Factor of the full model was less than 2 for all variables, indicating that multicollinearity was not significant. The model-averaged fitted values explained 32% (R2 0.32) of the observed data, moreover the model averaged residuals did not show any evident pattern, being normally distributed around 0. Model-averaged importance of terms and the estimation of their coefficients (Fig. 3; Table 3) showed that the most important predictors affecting WNF incidence were climatic and land use factors.

Fig 3. Summary statistics from the best 9 models (ΔAIC≤2 from the best model) for log-transformed WNF incidence (from 2010 to 2012).

The coefficients have been derived using multi-model averaging. The model term ‘Importance’ is proportional to the number of times that the variable is included in the set of best models and is represented by the colour and size of the bubbles (red/bigger bubble = high importance; blue/smaller bubble = low importance). Where referred to in the figure, each variable (Temp, PrecTot, PrecDays, NDVI, NDWI) is prefixed with Ano. or Av. for standardised anomaly or average, and the relevant period denoted in subscript (e.g. Ano.Temp2–5 = weekly anomaly temperature during months 2–5).

Water index (NDWI) proved a highly significant predictor, negatively correlated with incidence (cases per 100,000 inhabitants) of WNF such that a decrease in spring-early summer vegetation index (Av.NDWI4–7) of 0.10 predicts an increase in incidence of approximately 47 (NDWI theoretically ranges from-1 to 1 but in our study area/years Av.NDWI4–7 ranges between-0.13 and +0.04). Summer average temperatures Av.Temp6–9) were positively correlated with WNF, such that an increase of 1.6°C (as predicted under climate change scenario RCP 2.6; [52] predicts an increase in incidence of 2. Days of precipitation in late winter/early spring (CumulativePrecDays2–5) were also positively correlated with WNF incidence, such that an additional 10 days of precipitation (in the four months considered) predict an increase in incidence of WNF of 29 (where the range within our dataset was between 0 and 58 days of rain in total for this period).

Of the land use variables, the most significant predictors were the percentage of area covered by irrigated croplands, or by populated forests, both of which were positively correlated with incidence of WNF. An increase of 5% in the area cultivated as irrigated croplands predicts an increase in incidence of 113 (in the current study, the average area cultivated in this way varied between 0 and 26% of each NUTS3/GAUL1 area), while an increase of 5% in the area of populated forests predicts an increase of 36 per 100,000 people (in our study, the average area of populated forest varied between 0 and 47%).

Remaining variables were given an evidence weight <0.8, and were not considered significant [53].


West Nile virus is spreading in Europe and neighbouring countries at an increasing rate, with new lineages and variants emerging into new territories. Several factors contribute to the current epidemiological picture, amongst which urbanisation, variation in land use and climate are considered among the most important [19;5457]. To our knowledge the current study is the first attempt to model WNF human incidence at a continental scale in the Old World, including the whole of Europe, northern Africa and western Asia. Despite the large number of studies set in North America, differences in vectors and reservoir host species, in the degree of human exposure and wildlife immune status, suggest that the ecological processes driving WNV ecology in the New World (e.g. [8]) can be only partly considered to apply to the Old World system. In this system, perhaps because of the far longer history of WNV circulation and the recognition of at least five co-circulating lineages with variable pathogenicity, WNV epidemiology and ecology seem to be more complex.

It is well established that climate affects many components of the WNV biological cycle (e.g. [5758]), and that the carrying capacity for both vector and host populations differs with land use [60]. The interplay between biotic and abiotic variables that drive WNV in humans forms a complex and dynamic ecological system, and to adequately describe those dynamics across a large scale, requires a comprehensive set of predictors within a modelling framework. The current study makes use of remote sensing and GIS to enable collation of multiple types of environmental data over a continental spatial scale, at sufficient temporal and spatial resolution to test associations with WNF incidence. The number of studies utilising such tools are increasing [6062] but although a plethora of studies on WNV have been published since the New York outbreak of 1999, only a few authors applied statistical modelling techniques to study spillover potential (i.e. [19,6066]). Only a small subset of these studies use human cases as the response variable and environmental factors as predictors, and most are conducted at a local or regional spatial scale [60,61,64,66,67,68]. Only two studies [19,70] have made an attempt to model WNV outbreaks in humans in the Old World at a regional scale while others modelled WNV circulation in horses [71].

Climatic anomalies are considered among the most important drivers of WNV emergence, and ambient temperature can be a determinant of outbreaks [33]. Permissive meteorological conditions are necessary for the persistence of active transmission [19], and empirical studies suggest that temperature is a key factor influencing WNV evolution and dissemination [72]. Indeed, the relationship between temperature and increased WNV incidence in humans has repeatedly been confirmed both at national and multinational scales [70,7374]. In the current study, by using spatially continuous input data, we managed to avoid weak interpolation methods which only employ sparse point data. We found that average summer temperatures between the months of June and September (Av.Temp6–9) are positively correlated with WNF human incidence, in concurrence with [19] who found that the summer temperature anomaly before the upsurge was the main driver. Similarly, [70] found a significant relationship between summer temperature anomalies and WNV risk in Europe. The effect of temperature is likely to be mediated through its impact on the distribution, behaviour and survival of the mosquito vector, via direct impacts on the virus and on its’ extrinsic incubation period in the competent vectors (which is reduced at higher temperatures [72]), and via impacts on host distribution and behaviour. The nature of the association with temperature is therefore complex—for example, although warmer temperatures may have a negative effect on mosquito survival, they increase mosquito biting and development rates, and pathogen replication [72] and can lead to increased human exposure risk through altered human activity patterns [8].

Previously, [19] found no significant correlation between WNF incidence and either precipitation or humidity in Europe. In the present study, however, we observe positive associations between WNF incidence and the total number of days with precipitation in late winter-spring (Av.Rain2–5), and the percentage of irrigated croplands, suggesting a strong link to spring precipitation, and the presence of standing water. Standing water is a prerequisite for larval development of the mosquito vector, without which they cannot complete their biological cycle [75,76]. WNV transmission to humans is inefficient and infrequent [8], and cases of WNF occurs more often when mosquito population density rises (i.e. see [29]). While [19] found no link to precipitation, their data were limited to 2010, and landscape structure was not considered [77]. Furthermore, the authors used point source climatic data (meteorological stations) which may not represent spatial and temporal variations as well as global earth observations [60]. We believe the results of the current study are a more robust representation of the association between human incidence of WNF and spring precipitation/standing water in Europe, but recognise also that the nature of this association is likely to vary across the geographic range of the virus and the precipitation regimes, as reflected by conflicting evidence in literature (e.g. a positive correlation between precipitation and human incidence in eastern US, but a negative correlation in the west [63], in southern Florida [78] and Israel [79]).

Although precipitation in spring is positively associated with WNF incidence, a strong negative association is seen between WNF incidence and summer NDWI (Av.NDWI4–7). [70] also report an association between NWDI and WNV risk in Europe, specifically a positive association with anomalies in early June. NDWI is a proxy for the amount of water in the ecosystem, and low NDWI may be indicative of drought. Recent research suggests that drought may lead to subsequent localised increases in mosquito numbers and disease outbreaks [56,63]. Drought conditions may encourage the aggregation of both hosts (birds) and vectors (mosquitoes) at remaining water bodies, increasing rates of transmission [80], while potentially also influencing their vector competence [25]. Culex pipiens thrives during drought conditions by exploiting larval habitats with high organic matter (a consequence of drying content) and artificial containers not reliant on precipitation [81]. Furthermore, drought conditions may force amplifier bird species into urban and suburban areas where water is more freely available, thereby bringing competent hosts into contact with competent urban vectors [8]. Associations between dry summers and increased WNV outbreaks have been indicated in a number of US studies [82].

Wetlands [61,83,84], agricultural area [68,86], and urban infrastructure [66] have previously been reported to influence vector populations or WNV transmission [85]. In the current study, the detailed resolution of land use classes allows a distinction to be made between irrigated, and rain-fed crop lands. Irrigated croplands were positively associated with WNF, in accordance with research in the US where an increased human incidence of WNF was found to be associated with crop land cover and water catchment depressions in the otherwise semi-arid environment of Texas [87]. Rain-fed crop lands in the current study show no significant association with WNF, possibly because the more variable water supply is less favourable to mosquito populations.

A positive association with urban infrastructure has been repeatedly demonstrated in the US (e.g. [73,88,89], although with some exceptions and east-west differences e.g. [66]. This association is not apparent in the current study, and has not been reported in other European studies, although the largest outbreaks in the Old World, as in the New, have occurred in urban areas (e.g. Bucharest in 1995, Volgograd in 1999; Belgrade 2012; [9092]). This probably reflects significant differences in urban planning and infrastructures design, demonstrating the importance of region specific analyses. Indeed, US suburban areas are often characterized by assemblages of houses surrounded by vegetation (gardens, public parks, trees) which provide the optimal habitat for interaction of synantropic birds, mosquitoes and humans. On the other hand, Old World cities, especially those with a longer history, typically include far less green space. Instead, small buildings surrounded by vegetation are more common in rural areas of the Old World, where WNF cases are common.

In the current study, results indicate a clear positive association of WNF cases with populated forests. ‘Populated forests’ are defined as those with relatively low human population density (1–10 individuals/km2), and are characterised by a mixture of forest, human habitation, farmlands and transitional habitat, and usually occur in agricultural areas [39]. In such areas houses are usually scattered within forest patches which provide refuge, nesting and feeding opportunity for birds, including species which are considered highly competent reservoir hosts for WNV [8]. The occurrence of natural and protected areas was not significantly associated with WNF, but it is likely that increased sprawl of urban settlements causes fragmentation of previously pristine forest systems, increasing contact rates between vectors (mosquitoes), competent reservoir hosts (birds) and dead-end hosts such as humans [64,85], therefore enhancing transmission from the sylvatic cycle [69].

Landscape structure, patchiness and matrix organization have previously been associated with variation in vector population and virus transmission at a local scale [85,86], but although our model suggests a negative association with an index of heterogeneity (the Pielou Index), the model support for this term is low and the result therefore inconclusive.

Increases in the risk of WNF emergence in humans may arise due to temporal extension of the transmission season, increased spatial extent of habitat suitable for hosts and vectors, or increased intensity of virus amplification and circulation among birds and mosquitoes. All of these variables are likely to vary with changes in climate, human population expansion and land use change. A better understanding of the environmental drivers of WNF may ultimately be used to map the spatial variation in risk across the continent. In conjunction with long range meteorological forecasts, environmental data might be used further, to forecast inter annual change. Together, these could be used to target public health actions and so mitigate risk from WNF emergence. At the current time however, a number of weaknesses in the data available mean that predictive modelling is unlikely to be accurate. In particular, the complexity of the viral transmission cycle remains poorly understood in the Old World, while human WNF incidence data are limited by geographic variation in the accuracy of diagnosis, the establishment of surveillance, and the organization of national reporting systems.

Further research is needed, integrating interdisciplinary research across human, veterinary and environmental health (i.e., in line with the ‘One Health’ initiative [93]). In particular, enhanced monitoring of WNV circulation in the environment (via the combined use of sentinel birds and the virological screening of mosquitoes) and the recording of environmental variables [94], and a European shared database collecting geo-localized reports of human WNF, would provide important advances in data quality. This combined with collection of socio-economic data, and enhanced environmental data (for example, inclusion of species specific bird density data or routes of migrations) would help the development of more reliable predictive models.

In conclusion, although further interdisciplinary research is required to develop accurate predictive models of WNF risk, the current study, making use of a multi model inference framework (for a complete overview on the topic please refer to the milestone paper on model selection by [48]), provides a valuable starting point, and successfully identifies and confirms a number of variables which are associated with WNV emergence in Europe, Asia and North Africa.


We kindly thank Dr Bertrand Sudre of ECDC and three anonymous reviewers for precious insights on an earlier version of the manuscript. The PhD Scholarship of Matteo Marcantonio is supported by FIRS|T (FEM International Research School e Trentino). We acknowledge the E-OBS dataset from the EU-FP6 project ENSEMBLES ( and the data providers in the ECA&D project ( We are grateful to the NASA Land Processes Distributed Active Archive Center (LP DAAC) for making the MODIS LST data available.

Author Contributions

Conceived and designed the experiments: M. Marcantonio RR GM. Performed the experiments: M. Marcantonio AR. Analyzed the data: M. Marcantonio GM. Contributed reagents/materials/analysis tools: M. Marcantonio EC M. Metz MN. Wrote the paper: M. Marcantonio AR EC MN RR M. Metz.


  1. 1. Murray KO, Walker C, Gould E. The virology, epidemiology, and clinical impact of West Nile virus: a decade of advancements in research since its introduction into the Western Hemisphere. Epidemiol Infect. 2011;139(6):807–17. pmid:21342610
  2. 2. Artsob H, Gubler DJ, Enria DA, Morales MA, Pupo M, Bunning ML, et al. West Nile Virus in the New World: trends in the spread and proliferation of West Nile Virus in the Western Hemisphere. Zoonoses Public Health 2009;56:357–69. pmid:19486320
  3. 3. Iyer AV, Kousoulas KG. A Review of Vaccine Approaches for West Nile Virus. International Journal of Environmental Research and Public Health 2013;10:4200–23. pmid:24025396
  4. 4. Arnold C. West Nile virus bites back. Lancet Neurol 2012;11:1023–4. pmid:23153403
  5. 5. Campbell GL, Marfin AA, Lanciotti RS, Gubler DJ. West Nile virus. Lancet Infect Dis 2002;2:519–29. pmid:12206968
  6. 6. Dauphin G, Zientara S, Zeller H, Murgue B. West Nile: worldwide current situation in animals and humans. Comparative Immunology, Microbiology and Infectious Diseases 2004;27:343–55. pmid:15225984
  7. 7. Martín-Acebes MA. West Nile virus: A re-emerging pathogen revisited. World Journal of Virology 2012;1:51. pmid:24175211
  8. 8. Reisen WK. Ecology of West Nile virus in North America. Viruses 2013;5:2079–105. pmid:24008376
  9. 9. Smithburn KC, Hughes TC, Burke AW, Paul JH. A neurotropic virus isolated from the blood of a native of Uganda. The American Journal of Tropical Medicine and Hygiene 1940;20:471–92.
  10. 10. May FJ, Davis CT, Tesh RB, Barrett ADT. Phylogeography of West Nile Virus: from the Cradle of Evolution in Africa to Eurasia, Australia, and the Americas. J Virol 2011;85:2964–74. pmid:21159871
  11. 11. Ozdenerol E, Taff GN, Akkus C. Exploring the Spatio-Temporal Dynamics of Reservoir Hosts, Vectors, and Human Hosts of West Nile Virus: A Review of the Recent Literature. International Journal of Environmental Research and Public Health 2013;10:5399–432. pmid:24284356
  12. 12. Sambri V, Capobianchi M, Charrel R, Fyodorova M, Gaibani P, Gould E, et al. West Nile virus in Europe: emergence, epidemiology, diagnosis, treatment, and prevention. Clin Microbiol Infect 2013;19:699–704. pmid:23594175
  13. 13. Barzon L, Pacenti M, Franchin E, Pagni S, Lavezzo E, Squarzon L, et al. Large human outbreak of West Nile virus infection in north-eastern Italy in 2012. Viruses 2013;5:2825–39. pmid:24284876
  14. 14. Papa A. West Nile virus infections in humans-Focus on Greece. J Clin Virol 2013;58:351–3. pmid:23562617
  15. 15. Tsai T, Popovici F, Cernescu C, Campbell G, Nedelcu N. West Nile encephalitis epidemic in southeastern Romania. The Lancet 1998;352:767–71. pmid:9737281
  16. 16. Moudy RM, Meola MA, Morin LL, Ebel GD, Kramer LD. A Newly Emergent Genotype of West Nile Virus Is Transmitted Earlier and More Efficiently by Culex Mosquitoes. Am J Trop Med Hyg 2007;77:365–70. pmid:17690414
  17. 17. Semenza JC, Domanović D. Blood supply under threat. Nature Clim Change 2013;3:432–5.
  18. 18. Jimenez-Clavero MA. Animal viral diseases and global change: bluetongue and West Nile fever as paradigms. Front Gene 2012;3:105. pmid:22661980
  19. 19. Paz S, Malkinson D, Green MS, Tsioni G, Papa A, Danis K, et al. Permissive Summer Temperatures of the 2010 European West Nile Fever Upsurge. PLoS ONE 2013;8:e56398. pmid:23431374
  20. 20. Petersen LR, Brault AC, Nasci RS. West Nile virus: review of the literature. JAMA 2013;310:308–15. pmid:23860989
  21. 21. Petersen LR, Marfin AA. West Nile Virus: A Primer for the Clinician. Ann Intern Med 2002;137:173–9. pmid:12160365
  22. 22. Gomes B, Sousa CA, Vicente JL, Pinho L, Calderón I, Arez E, et al. Feeding patterns of molestus and pipiens forms of Culex pipiens (Diptera: Culicidae) in a region of high hybridization. Parasit Vectors 2013;6:93. pmid:23578139
  23. 23. Gomes B, Sousa CA, Novo MT, Freitas FB, Alves R, Côrte-Real AR, et al. Asymmetric introgression between sympatric molestus and pipiens forms of Culex pipiens (Diptera: Culicidae) in the Comporta region, Portugal. BMC Evolutionary Biology 2009;9:262. pmid:19895687
  24. 24. Osório HC, Zé-Zé L, Amaro F, Nunes A, Alves MJ. Sympatric occurrence of Culex pipiens (Diptera, Culicidae) biotypes pipiens, molestus and their hybrids in Portugal, Western Europe: feeding patterns and habitat determinants. Med Vet Entomol 2013.
  25. 25. Bakonyi T, Ferenczi E, Erdélyi K, Kutasi O, Csörgő T, Seidel B, et al. Explosive spread of a neuroinvasive lineage 2 West Nile virus in Central Europe, 2008/2009. Veterinary Microbiology 2013;165:61–70. pmid:23570864
  26. 26. Calzolari M, Monaco F, Montarsi F, Bonilauri P, Ravagnan S, Bellini R, et al. New incursions of West Nile virus lineage 2 in Italy in 2013: the value of the entomological surveillance as early warning system. Vet Ital 2013;49:315–9. pmid:24002939
  27. 27. Dobson A, Foufopoulos J. Emerging infectious pathogens of wildlife. Phil Trans R Soc Lond B 2001;356:1001–12. pmid:11516378
  28. 28. Karesh WB, Dobson A, Lloyd-Smith JO, Lubroth J, Dixon MA, Bennett M, et al. Ecology of zoonoses: natural and unnatural histories. The Lancet 2012;380:1936–45.
  29. 29. Colborn JM, Smith KA, Townsend J, Damian D, Nasci RS, Mutebi J-P. West Nile virus outbreak in Phoenix, Arizona—2010: entomological observations and epidemiological correlations. J Am Mosq Control Assoc 2013;29:123–32. pmid:23923326
  30. 30. Dowling Z, Ladeau SL, Armbruster P, Biehler D, Leisnham PT. Socioeconomic status affects mosquito (Diptera: Culicidae) larval habitat type availability and infestation level. J Med Entomol 2013;50:764–72. pmid:23926774
  31. 31. Morin CW, Comrie AC. Regional and seasonal response of a West Nile virus vector to climate change. Proc Natl Acad Sci USA 2013;110:15620–5. pmid:24019459
  32. 32. Patz JA, Daszak P, Tabor GM, Aguirre AA, Pearl M, Epstein J, et al. Unhealthy Landscapes: Policy Recommendations on Land Use Change and Infectious Disease Emergence. Environ Health Perspect 2004;112:1092–8. pmid:15238283
  33. 33. Paz S, Semenza JC. Environmental drivers of West Nile fever epidemiology in Europe and Western Asia—a review. Int J Environ Res Public Health 2013;10:3543–62. pmid:23939389
  34. 34. Petersen LR, Fischer M. Unpredictable and difficult to control—the adolescence of West Nile virus. N Engl J Med 2012;367:1281–4. pmid:23013055
  35. 35. Mulatti P, Ferguson HM, Bonfanti L, Montarsi F, Capelli G, Marangon S. Determinants of the population growth of the West Nile virus mosquito vector Culex pipiens in a repeatedly affected area in Italy. Parasites & Vectors 2014;7:26.
  36. 36. ECDC 2014. West Nile fever maps. Accessed 01/07/2014. Link:
  37. 37. Neteler M, Bowman MH, Landa M, Metz M. GRASS GIS: A multi-purpose open source GIS. Environmental Modelling & Software 2012;31:124–30.
  38. 38. Patrice B, Pierre D, Carsten B, Leon S, Christelle V, Mireille HUC, et al. GlobCover—Products Description and Validation Report 2008.
  39. 39. Ellis EC, Ramankutty N. Putting people in the map: anthropogenic biomes of the world. Frontiers in Ecology and the Environment 2007;6:439–47.
  40. 40. Elvidge C, Baugh K, Hobson V, Kihn E, Kroehl H, Davis E, et al. Satellite inventory of human settlements using nocturnal radiation emissions: a contribution for the global toolchest. Global Change Biology 1997;3:387–95.
  41. 41. Zhuo L, Ichinose T, Zheng J, Chen J, Shi PJ, Li X. Modelling the population density of China at the pixel level based on DMSP/OLS non-radiance-calibrated night-time light images. International Journal of Remote Sensing 2009;30:1003–18.
  42. 42. Small C, Elvidge CD, Balk D, Montgomery M. Spatial scaling of stable night lights. Remote Sensing of Environment 2011;115:269–80.
  43. 43. Rosà R, Marini G, Bolzoni L, Neteler M, Metz M, Delucchi L, et al. Early warning of West Nile virus mosquito vector: climate and land use models successfully explain phenology and abundance of Culex pipiens mosquitoes in north-western Italy. Parasites and Vectors 2014; 7:269. pmid:24924622
  44. 44. Pielou EC. The measurement of diversity in different types of biological collections. J Theor Biol 1966;13:131–44.
  45. 45. R Development Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing; 2012.
  46. 46. Zuur AF, Ieno EN, Walker N, Saveliev A, Smith G. Mixed Effects Models and Extensions in Ecology with R. 2009.
  47. 47. Burnham KP, Anderson DR. Model selection and multimodel inference: A practical Information-Theoretic Approach. Second. Springer; 2002.
  48. 48. Johnson JB, Omland KS. Model selection in ecology and evolution. Trends in Ecology & Evolution 2004;19:101–8.
  49. 49. Buckland ST, Burnham KP, Augustin NH. Model Selection: An Integral Part of Inference. Biometrics 1997;53:603.
  50. 50. Bartoń K. MuMIn: Multi-model inference; R package version 1.9.13 2013.
  51. 51. Hair JF Jr, Anderson RE, Tatham RL, Black WC. (1995). Multivariate Data Analysis (3rd ed). New York: Macmillan.
  52. 52. IPCC. Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. New York, USA: Cambridge University Press; 2013.
  53. 53. Calcagno V, de Mazancourt C. glmulti: An R Package for Easy Automated Model Selection with (Generalized) Linear Models. Journal of Statistical Software. 2010;34(12):1–29.
  54. 54. Bradley CA, Gibbs SEJ, Altizer S. Urban land use predicts west nile virus exposure in songbirds. Ecol Appl 2008;18:1083–92. pmid:18686573
  55. 55. Bradley CA, Altizer S. Urbanization and the ecology of wildlife diseases. Trends Ecol Evol 2007;22:95–102. pmid:17113678
  56. 56. Chase JM, Knight TM. Drought-induced mosquito outbreaks in wetlands. Ecology Letters 2003;6:1017–24.
  57. 57. Dohm DJ, O’Guinn ML, Turell MJ. Effect of environmental temperature on the ability of Culex pipiens (Diptera: Culicidae) to transmit West Nile virus. J Med Entomol 2002;39:221–5. pmid:11931261
  58. 58. Kilpatrick AM. Globalization, Land Use, and the Invasion of West Nile Virus. Science 2011;334:323–7. pmid:22021850
  59. 59. Valiakos G, Papaspyropoulos K, Giannakopoulos A, Birtsas P, Tsiodras S, Hutchings MR, et al. Use of Wild Bird Surveillance, Human Case Data and GIS Spatial Analysis for Predicting Spatial Distributions of West Nile Virus in Greece. PLoS ONE 2014;9:e96935. pmid:24806216
  60. 60. Chuang T-W, Wimberly MC. Remote Sensing of Climatic Anomalies and West Nile Virus Incidence in the Northern Great Plains of the United States. PLoS ONE 2012;7:e46882. pmid:23071656
  61. 61. Young SG, Tullis JA, Cothren J. A remote sensing and GIS-assisted landscape epidemiology approach to West Nile virus. Applied Geography 2013;45:241–9.
  62. 62. Brown H, Duik-Wasser M, Andreadis T, Fish D. Remotely-sensed vegetation indices identify mosquito clusters of West Nile virus vectors in an urban landscape in the northeastern United States. Vector Borne Zoonotic Dis 2008;8:197–206. pmid:18452400
  63. 63. Landesman WJ, Allan BF, Langerhans RB, Knight TM, Chase JM. Inter-annual associations between precipitation and human incidence of West Nile virus in the United States. Vector Borne Zoonotic Dis 2007;7:337–43. pmid:17867908
  64. 64. Chuang T-W, Hockett CW, Kightlinger L, Wimberly MC. Landscape-level spatial patterns of West Nile virus risk in the northern Great Plains. Am J Trop Med Hyg 2012;86:724–31. pmid:22492161
  65. 65. Allan BF, Langerhans RB, Ryberg WA, Landesman WJ, Griffin NW, Katz RS, et al. Ecological correlates of risk and incidence of West Nile virus in the United States. Oecologia 2009;158:699–708. pmid:18941794
  66. 66. Bowden SE, Magori K, Drake JM. Regional Differences in the Association Between Land Cover and West Nile Virus Disease Incidence in Humans in the United States. Am J Trop Med Hyg 2011;84:234–8. pmid:21292890
  67. 67. Harrigan RJ, Thomassen HA, Buermann W, Cummings RF, Kahn ME, Smith TB. Economic Conditions Predict Prevalence of West Nile Virus. PLoS ONE 2010;5:e15437. pmid:21103053
  68. 68. LaBeaud AD, Gorman A-M, Koonce J, Kippes C, McLeod J, Lynch J, et al. Rapid GIS-based profiling of West Nile virus transmission: defining environmental factors associated with an urban-suburban outbreak in Northeast Ohio, USA. Geospat Health 2008;2:215–25. pmid:18686270
  69. 69. Kilpatrick AM, Daszak P, Jones MJ, Marra PP, Kramer LD. Host heterogeneity dominates West Nile virus transmission. Proc Biol Sci 2006;273:2327–33. pmid:16928635
  70. 70. Tran A, Sudre B, Paz S, Rossi M, Desbrosse A, Chevalier V, et al. Environmental predictors of West Nile fever risk in Europe. International Journal of Health Geographics 2014;13:26. pmid:24986363
  71. 71. Pradier S, Leblond A, Durand B. Land cover, landscape structure, and West Nile virus circulation in southern France. Vector-Borne Zoonotic Dis 2008;8:253–63. pmid:18429693
  72. 72. Kilpatrick AM, Meola MA, Moudy RM, Kramer LD. Temperature, Viral Genetics, and the Transmission of West Nile Virus by Culex pipiens Mosquitoes. PLoS Pathog 2008;4:e1000092. pmid:18584026
  73. 73. Pecoraro HL, Day HL, Reineke R, Stevens N, Withey JC, Marzluff JM, et al. Climatic and landscape correlates for potential West Nile virus mosquito vectors in the Seattle region. Journal of Vector Ecology 2007;32:22–8. pmid:17633422
  74. 74. Soverow JE, Wellenius GA, Fisman DN, Mittleman MA. Infectious Disease in a Warming World: How Weather Influenced West Nile Virus in the United States (2001–2005). Environ Health Perspect 2009;117:1049–52. pmid:19654911
  75. 75. Clements AN. The Biology of Mosquitoes: Development, nutrition, and reproduction. Chapman & Hall; 1992.
  76. 76. Becker N. Mosquitoes and their control. Berlin; London: Springer; 2010.
  77. 77. Pérez-Rodríguez A, Fernández-González S, de la Hera I, Pérez-Tris J. Finding the appropriate variables to model the distribution of vector-borne parasites with different environmental preferences: climate is not enough. Global Change Biology 2013:19(11):3245–53. pmid:23606561
  78. 78. Shaman J, Day JF, Stieglitz M. Drought-induced amplification and epidemic transmission of West Nile virus in southern Florida. J Med Entomol 2005;42:134–41. pmid:15799522
  79. 79. Aharonson-Raz K, Lichter-Peled A, Tal S, Gelman B, Cohen D, Klement E, et al. Spatial and temporal distribution of west Nile virus in horses in Israel (1997–2013)-From endemic to epidemics. PLoS ONE. 2014;9(11).
  80. 80. Schöning JM, Cerny N, Prohaska S, Wittenbrink MM, Smith NH, Bloemberg G, et al. Surveillance of Bovine Tuberculosis and Risk Estimation of a Future Reservoir Formation in Wildlife in Switzerland and Liechtenstein. PLoS ONE 2013;8:e54253. pmid:23349839
  81. 81. Shaman J, Day JF, Komar N. Hydrologic Conditions Describe West Nile Virus Risk in Colorado. International Journal of Environmental Research and Public Health 2010;7:494–508. pmid:20616987
  82. 82. Epstein PR. Climate change and emerging infectious diseases. Microbes Infect 2001;3:747–54. pmid:11489423
  83. 83. Johnson BJ, Munafo K, Shappell L, Tsipoura N, Robson M, Ehrenfeld J, et al. The roles of mosquito and bird communities on the prevalence of West Nile virus in urban wetland and residential habitats. Urban Ecosyst 2012;15:513–31. pmid:25484570
  84. 84. Liu H, Weng Q, Gaines D. Spatio-temporal analysis of the relationship between WNV dissemination and environmental variables in Indianapolis, USA. International Journal of Health Geographics 2008;7:66. pmid:19094221
  85. 85. Gardner AM, Lampman RL, Muturi EJ. Land Use Patterns and the Risk of West Nile Virus Transmission in Central Illinois. Vector-Borne and Zoonotic Diseases 2014;14:338–45. pmid:24746038
  86. 86. Deichmeister JM, Telang A. Abundance of West Nile Virus Mosquito Vectors in Relation to Climate and Landscape Variables. Journal of Vector Ecology 2011;36:75–85. pmid:21635644
  87. 87. Warner RD, Kimbrough RC, Alexander JL, Rush Pierce J Jr, Ward T, Martinelli LP. Human west nile virus neuroinvasive disease in Texas, 2003 epidemic: regional differences. Ann Epidemiol 2006;16:749–55. pmid:16978879
  88. 88. Brown HE, Childs JE, Diuk-Wasser MA, Fish D. Ecologic Factors Associated with West Nile Virus Transmission, Northeastern United States. Emerg Infect Dis 2008;14:1539–45. pmid:18826816
  89. 89. Gomez A, Kilpatrick AM, Kramer LD, Dupuis AP, Maffei JG, Goetz SJ, et al. Land Use and West Nile Virus Seroprevalence in Wild Mammals. Emerg Infect Dis 2008;14:962–5. pmid:18507915
  90. 90. Platonov AE, Fedorova MV, Karan LS, Shopenskaya TA, Platonova OV, Zhuravlev VI. Epidemiology of West Nile infection in Volgograd, Russia, in relation to climate change and mosquito (Diptera: Culicidae) bionomics. Parasitol Res 2008;103 Suppl 1:S45–53. pmid:19030885
  91. 91. Popović N, Milošević B, Urošević A, Poluga J, Lavadinović L, Nedelijković J, et al. Outbreak of west Nile virus infection among humans in Serbia, August to October 2012. Eurosurveillance 2013;18.
  92. 92. Reiter P. West Nile virus in Europe: understanding the present to gauge the future. Eurosurveillance 2010.
  93. 93. Coker R, Rushton J, Mounier-Jack S, Karimuribo E, Lutumba P, Kambarage D, et al. Towards a conceptual framework to support one-health research for policy on emerging zoonoses. The Lancet Infectious Diseases 2011;11:326–31. pmid:21376670
  94. 94. Kwan JL, Park BK, Carpenter TE, Ngo V Civen R, Reisen WK. Comparison of enzootic risk measures for predicting West Nile disease, Los Angeles, California, USA, 2004–2010. Emerg Infect Dis. 2012;18(8):1298–306. pmid:22840314
  95. 95. Metz M, Rocchini D, Neteler M. Surface Temperatures at the Continental Scale: Tracking Changes with Remote Sensing at Unprecedented Detail. Remote Sensing 2014;6:3822–40.
  96. 96. Roerink GJ, Menenti M, Verhoef W. Reconstructing cloudfree NDVI composites using Fourier analysis of time series. International Journal of Remote Sensing 2000;21:1911–7.
  97. 97. Haylock MR, Hofstra N, Tank AMGK, Klok EJ, Jones PD, New M. A European daily high-resolution gridded data set of surface temperature and precipitation for 1950–2006. Journal of Geophysical Research—Atmospheres 2008;113:D20119.
  98. 98. Eisen L, Wong D, Shelus V, Eisen RJ. What is the Risk for Exposure to Vector-Borne Pathogens in United States National Parks? Journal of Medical Entomology 2013;50:221–30. pmid:23540107
  99. 99. Gibbs SEJ, Wimberly MC, Madden M, Masour J, Yabsley MJ, Stallknecht DE. Factors Affecting the Geographic Distribution of West Nile Virus in Georgia, USA: 2002–2004. Vector-Borne and Zoonotic Diseases 2006;6:73–82. pmid:16584329
  100. 100. Mack RN, Simberloff D, Mark Lonsdale W, Evans H, Clout M, Bazzaz FA. Biotic invasions: causes, epidemiology, global consequences, and control. Ecological Applications 2000;10:689–710.