Multiscale spatial drivers of WNV human risk in Europe: climate acts globally, equines locally

Background

West Nile virus (WNV) risk in Europe is shaped by climate and equine co-circulation, but the spatial scale at which these determinants operate has not been empirically characterised. Standard models impose uniform effects, masking local heterogeneity relevant for One Health surveillance design.

Methods

Human WNV incidence was modelled across 239 NUTS3 regions in five European countries (2010–2022) using Multiscale Geographically Weighted Regression (MGWR), which estimates an optimal spatial bandwidth independently per predictor. Determinants included summer and spring temperature, minimum winter temperature, warm-season precipitation, equine WNV co-circulation, and GDP per capita. MGWR was compared against OLS (R²=0.43) and standard Geographically Weighted Regression (GWR, R²adj=0.69, AICc=296).

Results

MGWR substantially improved fit (R²adj=0.75, AICc=242, ΔAIC=54 vs GWR). Climate determinants operated at near-global spatial scales (summer temperature k = 237 neighbours; winter minimum k = 234), consistent with continental vector ecology. Equine co-circulation was a strictly local predictor (k = 53 neighbours), with significant associations in 38.5% of regions concentrated in southeastern Europe (Greece median β = 0.39, Italy β = 0.17, Germany β < 0). The equine signal was independent of climate (ΔAIC=19.9) and amplified by warm temperature anomalies, identifying regions where integrated One Health data most improves risk prediction.

Conclusions

WNV risk determinants operate at fundamentally different spatial scales: climate establishes continental transmission potential while equine co-circulation signals local viral activity. Strengthening equine surveillance in warm southeastern European regions maximises predictive return of One Health integration, particularly as climate change expands WNV suitability northward.

Key messages

• WNV determinants act at different scales: climate sets continental risk, while equine cases signal local viral activity. MGWR modeling reveals that integrated surveillance is a strictly local tool.

• The equine signal is strongest in warm, southeastern Europe. Prioritizing One Health data in these hotspots maximizes predictive returns as climate change shifts WNV transmission suitability northward.