Risk assessment of West Nile fever in equines in Europe based on an ecological niche model

Abstract

West Nile fever is a zoonotic viral disease transmitted by Culex mosquitoes, capable of causing severe neurological illness in humans, equines, and birds. Over the past dozen years, Europe has experienced numerous WNF outbreaks of varying severity, leading to significant public health and economic impacts. Consequently, WNF has emerged as a major concern for both public health and veterinary sectors. In this study, we employed spatiotemporal analysis techniques alongside an optimized ecological niche model to investigate the spatiotemporal distribution patterns of historical WNF outbreaks, predict high-risk areas for equine WNF occurrence in Europe, and identify associated risk factors. The spatiotemporal analysis revealed a distinct seasonal pattern, with most outbreaks occurring in September. Between 2010 and 2023, the overall spread direction of equine WNF outbreaks in Europe trended from southeast to northwest, forming seven statistically significant spatiotemporal clusters. Ecological niche modeling indicated that southern Portugal, southern and eastern coastal regions of Spain, southwestern and southeastern regions of France, most parts of Italy, northeastern Germany, western Poland, southwestern coastal regions of Croatia, southwestern coastal regions of Bosnia and Herzegovina, southwestern coastal regions of Montenegro, western coastal regions of Albania, most parts of Greece, southwestern regions of Bulgaria, and Tekirdağ Province in Turkey are high-risk areas for WNF occurrence in equines in Europe. Isothermality, mean temperature of the warmest quarter, mean precipitation of the warmest quarter, elevation, and NDVI are important variables affecting WNF occurrence in equine. This study provides valuable insights to support targeted West Nile virus surveillance and optimized resource allocation across Europe.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12917-026-05298-x.

Introduction

West Nile fever (WNF) is a zoonotic disease caused by the West Nile virus (WNV), a member of the Flavivirus genus (Flaviviridae family) that is closely related to the Japanese encephalitis and dengue viruses [1]. Susceptible animals include equines, geese, ducks, and various bird species beyond poultry [2]. In nature, WNV is maintained in a bird-mosquito-bird cycle and can spill over incidentally into equines (EWNF) and humans (HWNF) [3]. Both equines and humans are considered dead‑end hosts, as they typically do not develop sufficient viremia to reinfect mosquitoes and sustain the transmission cycle [4]. While most avian infections are asymptomatic, certain species, such as American crows, exhibit high mortality [5]. Infected birds may exhibit neurological signs (e.g., ataxia, paralysis) as well as non‑neurological symptoms (e.g., depression, weight loss) [6]. Among mammals, equines are particularly susceptible; approximately 10% of infected individuals develop neurological manifestations such as weakness, ataxia, and limb paralysis, with reported mortality rates ranging from 22% to 44% [7–10]. Evidence suggests that WNV can also be transmitted directly among birds via the fecal‑oral route [6, 11]; however, the primary transmission mechanism remains indirect, mediated by mosquitoes. More than 60 mosquito species have been found to be infected with WNV, with Culex species identified as the principal vectors [12]. The global spread of WNV is further complicated by bird migration and the movement of multiple vector species, making its prevention and control an ongoing international challenge.

WNV was first isolated in 1937 from the blood of a febrile woman in the West Nile region of Uganda, from which its name originates [13]. Subsequently, the virus has spread progressively across Africa, the Middle East, Europe, and Asia, causing outbreaks of varying severity [14]. These outbreaks spurred sustained scientific research and improved characterization of WNF transmission patterns [15]. Nevertheless, WNF was not widely regarded as a major public health concern until the 1990s, when it began to cause increasingly frequent and severe disease in both humans and equines [16]. In 1999, WNF was first detected in North America and spread rapidly thereafter [17]. By 2005, the virus had disseminated across the continental United States, Canada, and parts of Latin America [18]. Between 1999 and 2012, 37,008 human cases were reported in the United States, of which 16,196 had neurologic symptoms [19]. The impact on equines was even more pronounced, with over 25,000 cases reported and a case mortality rate as high as 33% [19], underscoring WNF as a serious veterinary and public health threat. Following the introduction of WNV lineage 2 into Hungary in 2004 [20], this lineage progressively expanded across Europe, reaching Romania, Greece, Austria, Serbia, and Italy over the ensuing years [21, 22]. Since 2010, Europe has experienced a marked increase in WNF outbreak frequency and scale [21]. The disease pattern has shifted from sporadic, localized events to recurrent annual outbreaks, accompanied by a continuous geographic expansion [23]. Affected countries now include Austria, Bosnia and Herzegovina, Bulgaria, Croatia, the Former Yugoslav Republic of Macedonia, France, Germany, Greece, Italy, Portugal, Romania, Slovenia, and Spain. (WOAH, https://wahis.woah.org/).

Current treatment for WNF remains supportive, with no specific antiviral therapy available [24]. The most effective preventive measures consist of vaccination campaigns and vector control strategies [16]. Since 2008, the EU has had a vaccine available for the prevention of WNV infection in equines [25]. Nevertheless, vaccination entails substantial costs, and achieving full coverage would impose a considerable economic burden [26]. Given the increasing frequency of WNF outbreaks in Europe and the high costs associated with prevention and control, there is a clear need to strengthen proactive, risk‑based surveillance. In the era of big data, spatiotemporal analysis techniques offer a powerful tool to guide and optimize such targeted monitoring efforts.

Spatiotemporal analysis techniques are powerful tools for analyzing the characteristics of the temporal and spatial distribution of infectious diseases and contribute to a better understanding of disease transmission patterns [27]. For example, spatial epidemiology can be applied to describe and analyze the geographical patterns of diseases, elucidating the relationships among pathogens, hosts, and environmental factors [28]. Time‑series models are valuable for forecasting future disease trends and seasonal fluctuations [28]. Ecological niche modeling, grounded in ecological theory, examines the associations between climatic/environmental variables and disease occurrence, enabling the prediction of potential pathogen distribution and the identification of key risk factors [29]. In this study, we applied time-series decomposition to analyze the trends and seasonality of WNF outbreaks in equines in Europe. The linear directional mean was used to quantify the overall direction of outbreak spread. Spatiotemporal clustering analysis was conducted to detect significant space‑time clusters of equine WNF outbreaks. Furthermore, an ecological niche model was constructed using bioclimatic, geographic, landscape, and livestock variables to identify high‑risk areas and determine risk factors for equine WNF occurrence in Europe. The findings of this study aim to support policymakers in designing targeted surveillance and prevention strategies for WNF.

Materials and methods

Collection of outbreak data

WNF occurrence data were obtained from the Global Animal Disease Information System (EMPRES-i+, https://empres-i.apps.fao.org/) and the World Animal Health Information System (WAHIS, https://wahis.woah.org/). Data collection spanned from January 1, 2010, to December 31, 2023. The collected information includes the start date of the outbreak, location, latitude and longitude coordinates, affected animal species, and the number of cases. The raw data were cleaned to remove records with incomplete information and duplicates, resulting in a final dataset comprising 602 unique WNF outbreak records (Supplementary Document 1).

Time series decomposition

Time series decomposition was employed to examine the temporal characteristics of disease events, specifically their seasonal patterns and long-term trends [30]. The WNF outbreak data were aggregated monthly to construct a time series of WNF events. A STL (Seasonal and Trend decomposition using Loess) time series decomposition model was utilized to decompose the WNF time series. STL operates on an additive principle, and its expression is . , , , and respectively represent the model output, trend component, seasonal component, and residual. Time series decomposition was conducted in R 4.2.3 using the fpp2 package [31].

Linear directional mean

The linear directional mean is generally used to measure the average direction of a set of linear factors and can be utilized to analyze the macro-level spread trend and direction of epidemic spread [32]. The output of the linear directional mean is a new linear factor. Its direction represents the average direction of all input linear factors, and its length denotes the average length of all input linear factors. In this study, the linear directional mean was employed to characterize the overall directional tendency of WNF outbreaks in equines in Europe. Firstly, the center outbreak of the WNF was calculated for each year (the average center of the WNF outbreaks each year). Next, these center outbreaks were connected pairwise, transforming them into line factors. Finally, the linear directional mean of all segments was computed to summarize the predominant direction of spread. The linear directional mean analysis was performed in ArcGIS 10.2, and the principles are detailed at https://desktop.arcgis.com/en/arcmap/latest/tools/spatial-statistics-toolbox/h-how-linear-directional-mean-spatial-statistics-w.htm.

Spatiotemporal clustering analysis

Space‑time clustering analysis was performed to detect statistically significant spatiotemporal clusters of disease events, which can reveal localized high‑risk periods and areas [33, 34]. The principle involves constructing a cylindrical scanning window in the time-space dimension, with the time range of disease occurrence as the height and the spatial extent of disease occurrence as the base within the study area. Both the base radius and the height of the cylinder are dynamically adjusted in the iterations until the preset upper limits are reached. In this analysis, the minimum temporal window was set to 1 year, and the maximum temporal window could not exceed 50% of the total study period. The probability model was set as a space-time permutation model, and P-values were calculated using standard Monte Carlo methods. Clusters with a P-value < 0.05 were considered statistically significant. This analysis was completed in SaTScan v10.1.2 [35].

Ecological niche modeling

Filtering of WNF occurrence data

To eliminate spatial autocorrelation between WNF occurrence points, duplicate records were filtered using the “Trim duplicate occurrences” tool in ENMTools software [36]. This ensured that there is at most one WNF occurrence point in each grid of the environmental raster data, preventing overfitting of the model. Following this filtering process, a total of 449 WNF outbreak points remained for modeling.

Variable collection

Based on the ecological mechanisms driving the West Nile virus (WNV) transmission cycle, a set of variables potentially influencing WNF occurrence was assembled [37]. Horse density data were obtained from FAO Livestock Systems (https://www.fao.org/livestock-systems/en/). Nineteen bioclimatic variables, wind speed, and elevation were sourced from the WorldClim website (https://www.worldclim.org/data/worldclim21.html). These bioclimatic variables capture annual, seasonal, and extreme trends in temperature and precipitation, providing a robust basis for modeling the ecological niche and predicting potential geographic distributions. Monthly wind speed values were averaged to produce an annual mean. The Normalized Difference Vegetation Index (NDVI) was downloaded from the National Tibetan Plateau Data Center [38], and the average NDVI was obtained using the maximum value composite method [39, 40]. Land cover data were obtained from European Space Agency GlobCover 2009 products (http://due.esrin.esa.int/page_globcover.php), which include 22 land cover classes. To ensure consistency across all datasets, the spatial resolution of every raster layer was standardized to 2.5 arc-minutes, matching the extent of the study area.

Modeling procedures

The MaxEnt model was employed to predict the risk areas for WNF occurrence in equines in Europe by combining WNF outbreaks with environmental variables. Variable selection was conducted using the SDMtune package in R [41]. Modeling was initially performed using all variables, followed by an examination of the most important variables based on their percentage contribution. Variables exhibiting an absolute value of the Pearson correlation coefficient > 0.7 were classified as highly correlated. A “leave-one-out” Jackknife test was then performed to discard the variable that minimally reduces the model performance among all the correlated variables. A new model was constructed using the remaining variables, and the process was repeated until the correlation between all retained variables was below the threshold. The results of variable selection are shown in Table 1. Model parameter optimization was performed using the ENMeval package in R [42]. In MaxEnt, two key parameters were tuned: the regularization multiplier (RM), which penalizes model complexity to reduce overfitting, and the feature class (FC) combination, which defines the mathematical relationships (e.g., linear L, quadratic Q, hinge H, product P, threshold T) used to fit the response of WNF occurrence to environmental predictors. The RM values were set to range from 0.2 to 4.0 (incremented by 0.2), with six different FC settings (L, H, LQ, LQH, LQHP, LQHPT). This resulted in a total of 126 candidate models with different parameter settings. Model selection was based on the Akaike Information Criterion corrected for small sample sizes (AICc), with the lowest AICc value identifying the best-fitting model [43, 44]. The final model was constructed using the selected variables and the optimal parameter set (RM and FC). All other settings were kept at default, and the model was run with 5‑fold cross‑validation. Model performance was assessed in terms of the Area Under the Receiver Operating Characteristic Curve (AUC) and True Skill Statistics (TSS).

Table 1.

Variables used in the model

Code	Variable Name	Time	Source
bio_2	Mean Diurnal Range	1970–2000	WorldClim version 2.1
bio_3	Isothermality
bio_7	Temperature Annual Range
bio_8	Mean Temperature of Wettest Quarter
bio_10	Mean Temperature of Warmest Quarter
bio_13	Precipitation of Wettest Month
bio_15	Precipitation Seasonality
bio_18	Mean Precipitation of Warmest Quarter
wind	Wind Speed
elev	Elevation
horse	Horse density	2015	FAO livestock systems
landcover	Land cover	2009	European Space Agency
ndvi	Normalized Difference Vegetation Index	1981–2015	National Tibetan Plateau Data Center

Results

Spatial and temporal distribution characteristics

Time‑series decomposition revealed a distinct seasonal pattern in WNF occurrence, with a consistent peak observed in September each year (Fig. 1). The trend component further indicated that WNF outbreaks of varying intensity occurred annually between 2010 and 2023. As illustrated in Fig. 2, the annual centroid of outbreaks shifted over time: initially located in Italy (2010–2011), it moved to Greece (2012–2014), then to France (2015), Portugal (2016), back to Greece (2017–2018), to Germany (2019–2022), and finally to France again in 2023. The overall trend in the spread of WNF outbreaks from 2010 to 2023 is from southeast to northwest. Spatiotemporal clustering analysis identified seven statistically significant spatiotemporal clusters of WNF during the study period (Fig. 3; Table 2). The first spatiotemporal cluster is located in southwestern France, Andorra, and northeastern Spain, including 35 outbreak locations in 2023. The second spatiotemporal cluster covers most of Germany, Poland, the Czech Republic, Slovakia, Austria, and Hungary, including 104 outbreak locations from 2019 to 2022. The third spatiotemporal cluster is located in southeastern France, which is the smallest spatiotemporal cluster, including 30 outbreak locations in 2015. The fourth spatiotemporal cluster is located in southern Italy, including 73 outbreak locations in 2010. The fifth spatiotemporal cluster is located on the island of Sardinia in Italy, including 38 outbreak locations in 2011. The sixth spatiotemporal cluster is located in northeastern Italy, including 38 outbreak locations in 2011. The seventh spatiotemporal cluster is located in southeastern Europe, covering over a dozen countries. It is the largest spatiotemporal cluster, including 149 outbreak locations from 2012 to 2014.

Fig. 1.

Components of the time series of WNF monthly events in equines in Europe from 2010 to 2023. a time series. b trend component. c seasonal component. d remainder component

Fig. 2.

Linear directional mean (LDM) of WNF outbreak in equines in Europe from 2010 to 2023. The star symbol indicates the central outbreaks of WNF each year (the average center of the WNF outbreaks each year). The arrow indicates the direction of outbreaks

Fig. 3.

The result of spatio-temporal cluster analysis for WNF outbreaks in equines in Europe from 2010 to 2023. Red points represent WNF outbreaks, and red circles represent spatio-temporal clusters

Table 2.

The attributed value of spatiotemporal clustering analysis for WNF outbreaks in equines in Europe from 2010 to 2023

Cluster	Center coordinates	Radius (km)	Time frame	Number of cases	Test statistic	Monte Carlo P-value
1	(43.7143 N, 0.1899 E)	281.66	2023	69	74.8223	< 0.05
2	(51.0100 N, 14.8600 E)	488.63	2019–2022	107	70.5801	< 0.05
3	(43.6478 N, 4.6949 W)	34.90	2015	48	66.2707	< 0.05
4	(38.1512 N, 15.3071 E)	429.26	2010	97	63.4131	< 0.05
5	(40.0390 N, 8.4014 E)	116.26	2011	61	46.2984	< 0.05
6	(45.7001 N, 12.6577 E)	69.33	2011	60	45.5157	< 0.05
7	(43.1030 N, 27.8927 E)	907.32	2012–2014	48	32.4711	< 0.05

Model parameter optimization and evaluation

Parameter optimization using ENMeval produced AICc values for all candidate models. As shown in Fig. 4, the optimal MaxEnt model was achieved with a regularization multiplier (RM) of 1.4 and a feature combination (FC) of LQHPT. The final model demonstrated strong predictive performance, with mean AUC and TSS values of 0.947 and 0.831, respectively.

Fig. 4.

The AICc values for all maxent candidate models. Each point represents a combination of parameters

Important variables and response curves of the model

As detailed in Table 3; Fig. 5, the percent contribution, permutation importance, and Jackknife test results identified several key environmental predictors. Isothermality (bio3), mean temperature of the warmest quarter (bio10), and elevation (elev) demonstrated high values in both percent contribution and permutation importance. Variables yielding high training and test gains when used in isolation included isothermality, mean temperature of the warmest quarter, mean precipitation of the warmest quarter (bio18), elevation, and the Normalized Difference Vegetation Index (NDVI). Based on a synthesis of these metrics, the following variables were selected as important determinants of WNF occurrence: isothermality, mean temperature of the warmest quarter, mean precipitation of the warmest quarter, elevation, and NDVI. The response curves in Fig. 6 illustrate the relationships between these predictors and the estimated probability of WNF occurrence. The occurrence of WNF increases and then levels off as bio3 increases (Fig. 6a). For mean precipitation of warmest quarter, elevation, and mean temperature of warmest quarter, the occurrence of WNF is roughly negatively correlated with mean precipitation of the warmest quarter and elevation, while it is roughly positively correlated with mean temperature of the warmest quarter (Fig. 6b and d). The response curve of NDVI reveals that the occurrence of WNF first increases and then decreases as NDVI increases, with a preference for moderate NDVI values (Fig. 6e).

Table 3.

Percent contribution and permutation importance of variables in maxent model

Variable	Percent contribution (%)	Permutation importance (%)
bio_10	44.9	30.5
bio_3	20	33.3
elev	16.7	14.5
bio_15	5.6	7.1
wind	4.1	2.2
bio_8	2.6	3.3
bio_7	1.5	1.1
bio_2	1.2	1.5
horse	1	0.6
ndvi	1	2.5
landcover	0.7	1.1
bio_18	0.5	2.2
bio_13	0.2	0.2

Fig. 5.

Results of the jackknife test for the variable importance of the model. Cyan depicts the gain of the model constructed with other variables besides this one, blue depicts the gain of the model constructed with this variable only, and red depicts the gain of the model constructed with all variables. a train gain. b test gain

Fig. 6.

Response curves for important variables of the model. The curves show the mean response of the 5-fold cross-validation (red) and the mean +/- one standard deviation (blue). a bio_3. b bio_10. c bio_18. d elev. e ndvi

Suitability map of WNF occurrence in equines in Europe

Based on the natural breaks classification method, the suitability map for WNF occurrence was categorized into three risk levels: low, medium, and high (Fig. 7). The high-risk areas for WNF occurrence in equines in Europe include southern Portugal, southern and eastern coastal regions of Spain, southwestern and southeastern regions of France, most parts of Italy, northeastern Germany, western Poland, southwestern coastal regions of Croatia, southwestern coastal regions of Bosnia and Herzegovina, southwestern coastal regions of Montenegro, western coastal regions of Albania, most parts of Greece, southwestern regions of Bulgaria, and Tekirdağ Province in Turkey.

Fig. 7.

Risk map of WNF occurrence in equines in Europe. The light yellow color depicts low risk, the gold color depicts moderate risk, and the red color depicts high risk

Discussion

WNV is an important pathogen affecting humans and equines. Once diagnosed, it must be reported to the World Health Organization or the World Organization for Animal Health [45]. WNV is now widely distributed, occurring on every continent except Antarctica, including Africa, the Middle East, Asia, Europe, Australia, and the Americas [46]. The re-emergence of WNF in equines and humans underscores the substantial threat posed by this disease. The introduction of new viral lineages has contributed to persistent WNF outbreaks across Europe over the past dozen years [21–23]. Although vaccines are available, their economic impact remains a considerable concern. In this context, effective and strategic surveillance is essential. Currently, most European Union countries rely primarily on passive surveillance for WNV in equines; only Spain, Croatia, Romania, and Greece implement active equine monitoring programs [47]. Passive surveillance likely detects only a fraction of outbreaks, as most WNV‑infected equines remain asymptomatic. Therefore, there is a clear need to strengthen active, risk‑based surveillance of WNV in equines across Europe. Spatiotemporal analytical techniques can extract valuable insights from epidemiological data to guide pathogen monitoring and optimize the allocation of prevention and control resources. In this study, we applied such techniques, along with an optimized ecological niche model, to characterize the spatiotemporal distribution of historical WNF outbreaks, predict high-risk areas for equine WNF in Europe, and identify associated risk factors. Our time‑series decomposition revealed a distinct seasonal pattern, with most outbreaks peaking in September. This seasonality aligns largely with the population dynamics of mosquito vectors, which typically reach their highest abundance in late summer and early autumn [48]. During these peak vector seasons, intensified vector control measures are critical to reduce spillover risk to equines and humans. The results of the linear directional mean suggest that the overall spread direction of WNF outbreaks in equines from 2010 to 2023 was from southeast to northwest. The sequential connection of annual outbreak centroids visually reconstructs the dynamic spread of the disease. The underlying drivers of this directional pattern—which may include bird migration routes and vector dispersal—warrant further investigation. To identify localized space-time aggregation, we performed spatiotemporal cluster analysis, which detected seven statistically significant clusters during the study period. The most likely spatiotemporal clustering is located in southwestern France, Andorra, and northeastern Spain. Although it does not contain many outbreak locations, it exhibits the highest value of the test statistic, indicating a high degree of clustering of WNF outbreaks. Notably, France and Italy have each hosted multiple distinct spatiotemporal clusters within their territories, suggesting that these countries represent persistent focal regions requiring sustained surveillance attention.

To identify high-risk areas and risk factors for WNF occurrence in equines in Europe, we constructed an ecological niche model using bioclimatic, geographic, landscape, and livestock variables. MaxEnt was selected for its strong predictive performance, distribution maps produced by MaxEnt are comparable to those from ensemble methods [49], and its ability to handle small sample sizes and complex nonlinear relationships between predictors and species occurrence [50, 51]. However, default MaxEnt settings may not be optimal for all species and can lead to overfitting [52, 53]. We therefore optimized the regularization multiplier (RM) and feature classes (FC) based on AICc. The final model exhibited excellent predictive performance, and the resulting suitability map accurately captured historical outbreaks while also highlighting previously unidentified potential risk areas. Based on four different metrics (percent contribution, permutation importance, training gain, and test gain from Jackknife tests), we identified important variables influencing WNF occurrence in equines. The response curves indicate that WNF tends to prefer higher mean temperatures of the warmest quarter and lower mean precipitation of the warmest quarter. This finding is generally consistent with previous research results by Myer and Hahn [54, 55]. Higher temperatures can shorten the developmental time of Culex mosquitoes, accelerate their reproductive rate, and increase blood meal intake [56]. Additionally, higher temperatures enhance the infectivity of WNV, leading to faster replication rates within the vector and thus accelerating transmission efficiency [57–59]. Heavy rainfall can disrupt the habitats of Culex mosquitoes, leading to the washing away of larvae and pupae [60, 61]. Another explanation is that low precipitation leads to fewer water sources, causing birds and mosquitoes to gather around remaining water sources, thereby increasing the probability of WNV transmission [62]. The model results reveal that areas with lower elevation have higher suitability for WNF occurrence. Mosquito species and abundance appear to be high at lower elevations [63]. Indeed, lower elevation is associated with higher temperatures, which indirectly supports the preference of mosquitoes for warmer conditions. The response curve of NDVI shows that suitability for WNF occurrence is highest when NDVI values are moderate. Although corvids can inhabit various habitats within urban areas, they typically prefer areas with grass and trees [55]. The study by Bargaoui et al. suggests that higher NDVI values are associated with higher seroprevalence of WNF in horses [64]. NDVI may represent the habitat of birds, leading to this association with WNF. The ecological niche model suggests that high-risk areas of WNF occurrence in equines in Europe are primarily concentrated in central and southern coastal regions of Europe. Countries involved include Portugal, Spain, France, Italy, Germany, Poland, Croatia, Bosnia and Herzegovina, Montenegro, Albania, Greece, Bulgaria, and Turkey. The climate and environmental conditions in high-risk areas are highly conducive to WNF outbreaks in equines. Our predictions for the Iberian Peninsula are more conservative than those of García-Carrasco et al. [65], which may reflect methodological differences. In many European countries, authorities are well aware of high-risk areas and have already implemented surveillance measures aimed at preventing human infections, which helps prevent WNV spillover [66–70]. Effective WNV control requires a One Health approach that integrates human, animal, and environmental health. The transmission cycle of WNV involves mosquitoes, birds, equines, and humans, necessitating coordinated efforts across multiple sectors. Effective vector control strategies, informed by the spatiotemporal and ecological niche analyses presented in this study, require robust collaboration between public health authorities, veterinary services, and environmental agencies. Such collaboration can facilitate integrated surveillance, timely data sharing on human and equine cases and mosquito populations, and the implementation of targeted vector management interventions in high-risk areas. By fostering cross-sectoral partnerships, a One Health framework can significantly enhance the capacity to monitor, prevent, and control WNF outbreaks, thereby mitigating the risk of spillover into human populations.

This study has several limitations. All epidemiological data used for the study were obtained from FAO and WOAH. Consequently, underreporting or incomplete information from member countries could introduce bias, meaning our findings may not fully reflect the true epidemiology of WNF. Additionally, the covariates used in our study are static, which may lead to biases in the modeling results. Although the bioclimatic variables capture seasonal trends, they represent long-term averages (1970–2000) and may not precisely match the temperature and precipitation dynamics during specific outbreak periods. Additionally, the ecological niche modeling in the study did not incorporate bird migration factors due to the lack of relevant data. Infected birds are considered to play a significant role in the long-distance transmission of the virus [6]. While a land-use variable (potentially acting as a proxy for bird habitat) was included, it showed low importance in the final model. Future studies should consider developing ecological niche models for WNF of birds when sufficient data become available. Integrating risk areas and factors for both birds and equine WNF could yield more comprehensive insights for disease prevention and control.

Conclusion

WNF outbreaks exhibit clear seasonality, with the peak typically observed in September. Between 2010 and 2023, the overall spread direction of WNF outbreaks in equines in Europe has been from southeast to northwest, showing seven statistically significant spatiotemporal clusters. High-risk areas for WNF occurrence in equines in Europe include southern Portugal, southern and eastern coastal regions of Spain, southwestern and southeastern regions of France, most parts of Italy, northeastern Germany, western Poland, southwestern coastal regions of Croatia, southwestern coastal regions of Bosnia and Herzegovina, southwestern coastal regions of Montenegro, western coastal regions of Albania, most parts of Greece, southwestern regions of Bulgaria, and Tekirdağ Province in Turkey. Isothermality, mean temperature of the warmest quarter, mean precipitation of the warmest quarter, elevation, and NDVI are important variables affecting WNF occurrence in equines. This study can provide useful information for the prevention and control of WNF in equines in Europe.

Authors’ contributions

Conceptualization: Yue Wang, Qi An. Data curation: Yue Wang. Formal analysis: Qi An. Investigation: Yue Wang. Methodology: Qi An. Visualization: Yue Wang. Writing – original draft: Yue Wang. Writing – review & editing: Qi An, Yifei Liu, Ting Yang, Zhuo Sun.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

The datasets generated and/or analyzed during the current study are available in the article’s Supplementary Document.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.