From the complete data set of the BTV-8 infected area in 2006, we estimated the reproduction number between herds, R
, during the BTV-8 epidemic in 2006 to be around 4 in the second half of the summer. This part of that summer was slightly less warm than usual followed by a warm September. The reproduction ratio declined below 1 in the fall. This pattern was also consistent for the three countries that were analysed separately. We are the first to quantify these parameters for such an infection, so these results cannot be compared to earlier publications. The differences found between the countries are surprisingly small, which means that the essential features driving the epidemic were comparable throughout the affected region. The reproduction number between herds was of the same order as that found for major dreaded livestock infections, like FMD [18, 24].
This epidemic was able to hit most ruminant herds in a large area, which is unprecedented for notifiable diseases in Western Europe in recent decades. This huge impact was a result of a lack of available effective control measures in 2006. In FMD and CSF epidemics, isolation of infected herds immediately after detection in the form of a stand-still or transport ban, leads to a major reduction of transmission, bringing R
close to or below 1 depending on local conditions. [22, 25]. However, for a vector-borne disease, national veterinary authorities in the affected area assumed a ban on animal transport to be less effective and therefore these restrictions were not applied in the rigorous way that is common for other notifiable diseases like FMD or CSF.
From the sensitivity analysis, we learned that the R
estimates are not very sensitive to assumptions on the duration of the infectious period of a herd. It remains in the same range in all cases. This analysis also visualises the effect of temperature on transmission. Temperature is the only really obvious connection explaining the decline of the epidemic in the autumn, which fits with predictions from earlier theoretical modelling studies . The decline of R
during the season cannot be explained by increasing the efficacy of the control measures, because none were newly introduced during the period when the transmission declined in the autumn. The temperature is thought to affect many aspects of the vector-host transmission system, like the intrinsic incubation period in the vector, the biting rate and lifespan of the vector, and with some more delay also the density of the vector population. All these aspects lead to reduced transmission at lower temperatures.
By studying the results for extremely short infectious periods for a herd, we were able to visualise the impact of temperature on such a system. The number calculated in this way may led to a less precise quantification of reproduction number between herds, but it surely is a good method for visualising the impact of temperature on transmission. We found that a few degrees temperature decline in the range between 15 and 20°C can lead to a reduction of transmission up to a factor 10. This is a substantial difference in transmission as a result of a rather common temperature range difference in Western Europe. Thus, the transmission appears to be extremely sensitive to this aspect, as was previously expected but never proven [10, 11]. It also supports findings that the infection has more difficulty invading Scandinavia, where summer temperatures are slightly lower.
We also determined a spatial transmission kernel of the infection between herds and found that although for The Netherlands the shape of the spatial transmission kernel for BTV-8 is quite similar to those found in earlier studies on AI and FMD [21, 22], the spatial scale at which transmission for BTV-8 took place, was much higher. Where transmission of FMD is mostly restricted to a few kilometers, BTV-8 easily spreads over about 15 km ranges. This difference probably follows from two typical features of the BTV-8 epidemic. First of all, unlike FMD and AI, there were no attempts to control this infection with a strict ban on animal transport, which is a common regulation for FMD and AI epidemics. Secondly, the vector can easily move around several kilometers, and more if helped by wind . Thus, longer distance transmission of this infection is to be expected. The effect of transmission via the wind, leading to an asymmetrical spatial spread is not included in the model. During the summer and fall of 2006, the wind direction was quite variable, leading to a very diffuse pattern throughout the infectious period. A specific analysis of the effect of wind in the 2006 BTV-8 epidemic can be found in Hendrickx et al. .
In the analysis over separate countries, by not allowing transmission over the border, we forced long distance transmission into the analysis. These extra long distance transmissions are not needed in a joint analysis of the complete infected area. Thus, the latter gives a more reliable estimate of the long distance transmission averaged over the duration of the outbreak and all the countries involved.
Still, the shape of the spatial kernel we found in analysing the overall area is very different from the Gaussian kernel for BTV transmission estimated from the same epidemic by Szmaragd et al. [19, 27]. This is because their analysis, which uses the same case dataset, only incorporates reported farms, and ignores the background density of farms present in the area. To regard their study result as a transmission kernel, a uniform density of background farms have to be assumed. Since the background of uninfected and undetected farms are not at all uniformly distributed, this assumption leads to an overestimation of the number of farms in low density regions and consequentially to an underestimation of the number of farms in high density regions. In a kernel estimation for case data, this leads to an underestimation of the probability of long distance transmission and thus to a too much localised kernel estimate.
In Belgium, we found a spatial transmission pattern in 2006 that diverges substantially from that in The Netherlands and Germany. We observed a very heavy tail of long distance transmission in Belgium, suggesting a restricted influence of the spatial component in transmission in Belgium. The kernels for The Netherlands, Germany and the kernel for the whole area show a much more localised spatial transmission.
A separate analysis for the early (with animal movement restrictions) and late period (no animal movement restrictions) of the epidemic in Belgium showed a major difference between the spatial transmission kernels before and after August 24th, 2006. This suggests that unrestricted animal transport has a major impact on the spatial scale of transmission of this infection. It leads to a transmission pattern that is similar to random transmission over space in the area where such transports are allowed freely.
All other countries maintained restrictions on animal transports from the more heavily infected areas, to protect the zone that was not yet affected (heavily), thus reducing long distance transmissions and maintaining a pattern of gradual spread over space. The scale of the spatial transmission kernel fits with the size of the restriction areas (i.e. average transmission distance of 15 km). According to Szmaragd et al.  for the Gaussian kernel, movement restrictions had only a small effect on spread, whereas for fat tailed kernels, such restrictions lead to greatly reduced spread. We observed a large effect of the movement restrictions on the spatial spread, which was consistent with a fat tailed kernel and inconsistent with the Gaussian kernel selected by Szmaragd et al. .
Therefore, our results suggest that further restriction of animal movements could have helped in reducing long distance transmission. Such an intervention did not affect the short distance transmission and did not change the intensity (prevalence) of the epidemic in affected areas, but slowed down the spatial progress (speed of the wave front) of the epidemic. This can be of major importance in protecting neighbouring countries for introduction of the infection.
Finally we add a few remarks on the results to make our conclusions more informative. Firstly, in this study we were able to analyse transmission parameters only on the level of transmission between herds. This follows from the type of data that were collected, i.e. estimated date of infection and location of each infected farm. Within herd information was very limited and will be analysed separately. Although effective transmission within a herd will probably continue later in the season than the transmission between herds, the main spread of the epidemic was already restricted early in the autumn, because transmission between herds had declined below the threshold. Transmission between herds is the essential feature of large scale epidemics in livestock.
Secondly, underreporting and late reporting was probably substantial in this epidemic. If the underreporting was constant throughout the assessed period, this did not influence the amplitude of the spatial kernel, while the ratio between infected and uninfected was consistently underestimated. It will not have influenced our R
results or the shape of the kernel much. This is because the transmission (for both methods applied here) was quantified from the ratio between infectious and newly infected farms which are registered. Thus it missed a similar fraction of farms on both ends, the infectious and the infected end.
Thirdly, the Culicoides densities were unknown at the time of the study. Since then it has become clear that suitable vector species for BTV-8 are prevalent throughout the area [12–16]. Details on the vector abundance and vector competence for the various Culicoides species, was not available in sufficient detail to analyse its influence.
In conclusion we found that
animal transport restrictions can slow down the spatial spread of BTV-8 substantially.
spatial transmission during the BTV-8 epidemic took place mostly within a 20 km range.
at temperatures below 15°C, the transmission of BTV-8 between farms was limited to such a low level that the epidemic was fading out.
the reproduction number of BTV-8 between herds is about 4 in a normal summer in Western Europe.