Sellers' Revisited: A Big Data Reassessment of Historical Outbreaks of Bluetongue and African Horse Sickness due to the Long-Distance Wind Dispersion of Midges.

The possibility that outbreaks of bluetongue (BT) and African horse sickness (AHS) might occur long-distance wind dispersion (LDWD) of their insect vector ( spp.) was proposed by R. F. Sellers in a series of papers published between 1977 and 1991. These investigated the role of LDWD by means of visual examination of the wind direction of synoptic weather charts. Based on the hypothesis that simple wind direction analysis, which does not allow for wind speed, might have led to spurious conclusions, we reanalyzed six of the outbreak scenarios described in Sellers' papers. For this reanalysis, we used a custom-built Big Data application ("") which couples a user-friendly web-interface with an established atmospheric dispersal model (""), thus enabling more sophisticated modeling than was possible when Sellers undertook his analyzes. For the two AHS outbreaks, there was strong support from our reanalysis of the role of LDWD for that in Spain (1966), and to a lesser degree, for the outbreak in Cyprus (1960). However, for the BT outbreaks, the reassessments were more complex, and for one of these (western Turkey, 1977) we could discount LDWD as the means of direct introduction of the virus. By contrast, while the outbreak in Cyprus (1977) showed LDWD was a possible means of introduction, there is an apparent inconsistency in that the outbreaks were localized while the dispersion events covered much of the island. For Portugal (1956), LDWD from Morocco on the dates suggested by Sellers is very unlikely to have been the pathway for introduction, and for the detection of serotype 2 in Florida (1982), LDWD from Cuba would require an assumption of a lengthy survival time of the midges in the air column. Except for western Turkey, the BT reanalyses show the limitation of LDWD modeling when used by itself, and indicates the need to integrate susceptible host population distribution (and other covariate) data into the modeling process. A further refinement, which will become increasingly important to assess LDWD, will be the use of virus and vector genome sequence data collected from potential source and the incursion sites.