A wind density model to quantify the airborne spread of Culicoides

Available online at www.sciencedirect.com
Preventive Veterinary Medicine 87 (2008) 162–181
www.elsevier.com/locate/prevetmed
A wind density model to quantify the airborne spread of
Culicoides species during north-western Europe
bluetongue epidemic, 2006
Guy Hendrickx a,1,*, Marius Gilbert b, Christoph Staubach c,
Armin Elbers d, Koen Mintiens e, Guillaume Gerbier f, Els Ducheyne a,1
a
Avia-GIS, Risschotlei 33, Zoersel B-2980, Belgium
Lutte Biologique et Ecologie Spatiale, Université Libre de Bruxelles, Avenue D. Roosevelt 50,
Brussel B-1050, Belgique
c
Friedrich-Loeffler Institut, Bundesforschungsinstitut für Tiergesundheit, Seestrasse 55, Wusterhausen 16868, Germany
d
Wageningen University and Research Centre, Central Institute for Animal Disease Control,
Department of Virology, P.O. Box 2004, Lelystad NL-8203 AA, The Netherlands
e
Veterinary and Agrochemical Research Centre, Co-ordination Centre for Veterinary Diagnostics,
Groeselenberg 99, Brussels B-1180, Belgium
f
Centre de Coopération Internationale en Recherche Agronomique pour le Développement,
Campus international de Baillarguet TA 30 E, 34398 Montpellier Cedex 5, France
b
Abstract
Increased transport and trade as well as climate shifts play an important role in the introduction,
establishment and spread of new pathogens. Arguably, the introduction of bluetongue virus (BTV) serotype
8 in Benelux, Germany and France in 2006 is such an example. After its establishment in receptive local
vector and host populations the continued spread of such a disease in a suitable environment will mainly
depend on movement of infected vectors and animals. In this paper we explore how wind models can
contribute to explain the spread of BTV in a temperate eco-climatic setting. Based on previous work in
Greece and Bulgaria filtered wind density maps were computed using data from the European Centre for
Medium-Range Weather Forecasts (ECMWF). Six hourly forward wind trajectories were computed at
pressure levels of 850 hPa for each infected farm as from the recorded onset of symptoms. The trajectories
were filtered to remove wind events that do not contribute to possible spread of the vector. The suitable wind
events were rastered and aggregated on a weekly basis to obtain weekly wind density maps. Next to this,
cumulated wind density maps were also calculated to assess the overall impact of wind dispersal of vectors.
A strong positive correlation was established between wind density data and the horizontal asymmetrical
spread pattern of the 2006 BTV8 epidemic. It was shown that short (<5 km), medium (5–31 km) and long
* Corresponding author. Tel.: +32 474 319 571; fax: +32 3 458 2979.
E-mail address: ghendrickx@avia-gis.be (G. Hendrickx).
1
These authors contributed equally to this paper.
0167-5877/$ – see front matter # 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.prevetmed.2008.06.009
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
163
(>31 km) distance spread had a different impact on disease spread. Computed wind densities were linked to
the medium/long-distance spread whilst short range spread was mainly driven by active Culicoides flight.
Whilst previous work in the Mediterranean basin showed that wind driven spread of Culicoides over sea
occurred over distances of up to 700 km, this phenomenon was not observed over land. Long-distance
spread over land followed a hopping pattern, i.e. with intermediary stops and establishment of local virus
circulation clusters at distances of 35–85 km. Despite suitable wind densities, no long range spread was
recorded over distances of 300–400 km. Factors preventing spread Eastwards to the UK and Northwards to
Denmark during the 2006 epidemic are discussed. Towards the east both elevation and terrain roughness,
causing air turbulences and drop down of Culicoides, were major factors restricting spread. It is concluded
that the proposed approach opens new avenues for understanding the spread of vector-borne viruses in
Europe. Future developments should take into consideration both physical and biological factors affecting
spread.
# 2008 Elsevier B.V. All rights reserved.
Keywords: Culicoides; Europe; Bluetongue; Wind density model; Long-distance spread
1. Introduction
Whilst micro-organisms causing disease are increasingly being moved around the world by
rising rates of trade and travel, the dispersal of insect vectors also has been implicated in the
introduction of vector-borne pathogens into new areas. The impact of this potential route of
introduction must not be underestimated. Ritchie and Rochester (2001) showed for example that
Japanese Encephalitis was introduced into Australia by wind-blown Culex spp., whilst Baker
et al. (1990) demonstrated the reinvasion of cleared land by wind-borne Simulium spp. spread by
monsoonal winds during the rainy season in West Africa. Therefore, analysing the process of
arrival, establishment and consequent spread of these vectors is very important.
Bluetongue is an arboviral disease, part of the former OIE List A disease group, that causes
high mortality not only in certain breeds of sheep but also in other domestic and wild ruminants.
It is caused by an orbivirus that is transmitted between ruminant hosts by the bites of Culicoides
biting midges (Diptera: Ceratopogonidae). Having occasionally occurred in Europe mostly at
latitudes below 408N until 1998, in recent years the disease has invaded Southern and
Mediterranean Europe (Baylis, 2002; Mellor and Wittmann, 2002). Invasion of this area has first
been linked to the northward spread of its main African vector: Culicoides imicola and at the
margin of this invading vector’s range by indigenous European Culicoides species. Such areas
where European Culicoides species (of which the C. obsoletus-complex, the C. pulicariscomplex and C. Newsteadi are the most abundant) are suspected of transmitting the disease
include Italy (Goffredo and Meiswinkel, 2004; Caracappa et al., 2003; De Liberato et al., 2005;
Torina et al., 2004), Bulgaria (Purse et al., 2006), Greece (Patakakis, 2004) and the Balkan
countries (Panagiotatos, 2004). More recently (Thiry et al., 2006) transmission over large areas
(see other papers in this special issue) of a non-Mediterranean bluetongue virus serotype (BTV8)
based solely on indigenous Culicoides species at a latitude above 508N in the Benelux, Germany
and France has been documented in great detail. Currently BTV8 has further spread in the United
Kingdom and Denmark, and also reached Switzerland.
Culicoides biting midges can be passively dispersed over long distances by prevailing winds
(Sellers, 1992; Sellers and Maarouf, 1989, 1991; Braverman and Chechik, 1996) leading to rapid
spread of the diseases that they carry. When characterizing the movements of insect vectors,
Reynolds et al. (2006) distinguish between two types of movement behaviour—‘vegetative’
164
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
short-distance movements that are directed towards resources (required for growth and
reproduction) and straightened-out, generally longer distance migratory movements. This pattern
of spread is referred to in literature as a stratified dispersal pattern (Hengeveld, 1989). For
Culicoides in particular, Sellers (1992) similarly observed two types of flight (1) short-distance
flights that occur in any direction (both up- and downwind) and at low or zero wind speeds and (2)
long-distance dispersal, up to several 100 km, that occurs at wind speeds greater than the unaided
flight speed of Culicoides and where midges are carried by the wind. The modern view is that
these long-distance movements may not be accidental on the part of the insect but could be
actively initiated and maintained (Dingle, 1996 in Reynolds et al., 2006). For Culicoides, there
are suggestions that these movements may be terminated (i.e. the midge lands) either actively (by
the insect ceasing to move its wings and descending), or because of wind drops (Sellers, 1992), or
due to terrain topography (Bishop et al., 2000, 2005). For this dispersal process to be of
epidemiological importance, once landed, the midge must survive long enough to replicate the
virus to a transmissible level and to bite a susceptible ruminant host. The probability of the latter
is influenced by the local habitat, the weather conditions and the presence of hosts at destination.
The association between incursions and wind events were qualitatively assessed by Sellers
and Pedgley (1985) and Sellers et al. (1978). They examined wind events in relation to the
incursion of bluetongue in Cyprus, Turkey and Portugal and matched them with dominant wind
events. Alba et al. (2004) investigated the possibility of introduction of infected midges on the
Balearic Islands from Sardinia during the 2000 outbreaks using a formal trajectory analysis. They
calculated backward wind trajectories for a limited number of days, those when the climatic
conditions (humidity, temperature and wind) were considered to be highly appropriate for flight
activity of the midges. Using this approach, they could match the Balearic outbreak with previous
outbreaks in Sardinia. This showed that infected Culicoides could have been transported by wind
from Sardinia to the Balearic Islands, thereby causing the recorded bluetongue outbreaks. This
was further extended by Ducheyne et al. (2007) in Greece and Bulgaria, where forward and
backward wind trajectory analysis was used to describe two epidemics involving several
serotypes. In this paper, this methodology was further developed using data from 2006 BTV8
epidemic in the Benelux, France and Germany.
2. Material and methods
2.1. Epidemiological data
The Animal Disease Notification System (ADNS) database was used as basic epidemiological
database. Using information gathered at national level dates of earliest reporting were included
and in this study the earliest known date available was used, this being either (a) the date of
confirmation of laboratory diagnosis, (b) the date of suspicion by a veterinarian (often the date
samples were sent for analysis to the lab) or (c) the date of start of symptoms as recorded during a
post hoc epidemiological questionnaire conducted by local veterinary authorities. Whilst the first
case was officially recorded on 17 August 2006, first symptoms were traced back for one farm as
early as 6 June 2006. After this very early and still controversial first case, farms started
consistently recording symptoms of the disease from week 34 onwards, i.e. from 21 August 2006.
Every farm with one or more positive case(s) was georeferenced at the farmhouse level. Farms
with wrongly assigned coordinates, i.e. outside the range of the epidemic, were corrected after
checking with the original national databases. It remains unclear from the ADNS database
whether recurrent cases within one farm are truly new cases or if they are sequels from previously
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
165
reported cases. Still, in the following analysis these few recurrent cases were included as separate
cases. Case density was determined from the point data using kernel density estimation with a
radius of 20 km and an output cell size of 1 km 1 km. Because kernel density interpolation is a
distance-based operation, the case point data layer was reprojected to Lambert Conformal Conic
projection with the following parameters: standard parallel 1 = 35, standard parallel 2 = 65,
longitudinal central meridian = 10, latitude of Projection Origin = 52, false easting = 4,000,000 m and false northing = 2,800,000 m. Such a projection is needed to take the
earth curvature into consideration when calculating distances.
Next to case data, cattle data collated by the Friedrich Loeffner Institute per Local
Adminstrative Unit (LAU 2, Eurostat) was used to correct case density for cattle density. For
some LAUs no cattle data is available. The cattle density was rastered with the same output
resolution as the case kernel density data layer (1 km 1 km). The cattle density was reprojected
to the Lambert Conformal Conic projection with the same parameters as the case density data.
2.2. Wind density maps
The European Centre for Medium-Range Weather Forecasts (www.ecmwf.int) compiles raw
data from various sources (ground- and satellite-measured) and reanalyses these data using
atmospheric models. From the ECMWF archive more than 60 different parameters, including
relative humidity, temperature and wind-related parameters are archived every 6 h at 23 different
pressure levels. In addition 91 vertical model levels are also available.
For this study ECMWF data with corner coordinates 588N 148E upper left corner and 348N
178W lower right corner were used. The east and north components of the wind vectors, the U and
V components, respectively, from 1 June 2006 until 30 November 2006 at four different pressure
levels, 700, 850, 925 and 1000 hPa, i.e. approximate altitudes of 3000, 1450, 760 and 100 m,
respectively, were extracted at a spatial resolution of 0.258 0.258 latitude and longitude. The
spatial resolution of the original vector data was improved as described below.
Reynolds et al. (2006) suggest that nocturnal migrants might be concentrated at 200–400 m
above ground which agrees with the very few field findings so far for Culicoides (Chapman et al.,
2004, Reynolds and Carpenter, unpublished data). On the other hand, Alba et al. (2004) found
that pressure levels at higher altitudes (900–1000 hPa) matched well with the dispersal of
Culicoides from Sardinia to the Balearic Islands, even though this is not supported with empirical
evidence. Ducheyne et al. (2007) showed that forward trajectories at 850 hPa (approx. 1450 m
altitude) match with historical outbreak patterns during 2 epidemic years of BT in Greece and
Bulgaria involving multiple serotypes.
As Culicoides cannot survive when temperature is consistently lower than 7 8C in
combination with a relative humidity lower than 30% (Baylis et al., 2004; Wittmann, 2002),
suitable pressure levels for the occurrence of Culicoides were determined from general
temperature and relative humidity monthly means as derived from ECMWF data. From an
inspection of monthly isotherms and iso-humidity lines for an average year (not shown here) it
can be concluded that at the pressure level 700 hPa (approximately 3000 m altitude), temperature
was consistently lower than 275 K (2 8C) throughout the year and the relative humidity was
between 48 and 70% (Uppala et al., 2005). The combination of such a low temperature and low
humidity levels makes it unlikely that Culicoides will be transported at this pressure level (i.e.
altitude). The temperature and relative humidity at the other pressure levels was within the
optimal range for Culicoides species for at least 4 months, respectively, 288–300 K (15–27 8C)
and 65–80%. This is supported by Reynolds et al. (2006) who mention that migrating insects are
166
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
largely confined to the first 2 km of the atmosphere, with higher flights (c. 3 km) only occurring
under certain circumstances. This range of flight altitudes largely coincides with the atmospheric
(or planetary) boundary layer (ABL) (Drake and Farrow, 1988 in Reynolds et al., 2006) within
which the air is directly affected by the sun’s heating of the Earth’s surface, and by the friction
due to surface roughness. Previous results (Ducheyne et al., 2007) showed good fit between wind
trajectories and subsequent outbreaks at 850 hPa. In addition little difference between the
trajectories at the pressure levels 850 and 925 hPa (respectively, approximate altitudes of 1450
and 760 m) up to a distance of 250 km were shown. Therefore wind trajectories at pressure levels
of 850 hPa (approximately 1450 m), reflecting the likely upper limit of Culicoides presence in
the lower troposphere, have been used in this study.
The trajectories were calculated according to a 2D wind trajectory algorithm first developed by
Codina (1999) and later adapted by Ducheyne et al. (2007) in Grid Analysis and Display System
(GrADS http://www.iges.org/grads/) starting from the earliest date of reporting (6 June) until 30
November. Model outputs are a series of spatially bilinearly interpolated trajectory nodes for each
time step (i.e. every 6 h) from each infected farm, starting for each at the earliest known date of BT
occurrence. Wittmann and Baylis (2000) indicated that midges exhibit active movement
irrespective of wind direction when wind speed is below 3 m/s and that they stay sheltered at wind
speeds above 11 m/s. Therefore trajectories with a wind speed outside this range were filtered out
prior to further analysis. Finally, all individual trajectories were converted to raster data with a
spatial resolution of 0.058 latitude/longitude and summed to obtain weekly wind cumulated density
maps. Weeks are numbered according to 2006 calendar weeks following ISO 8601 standard.
2.3. Short-, medium- and long-distance spread
Before any relationship between spread distance and wind events can be established, new
cases resulting from a local or short-distance spread have to be separated from those resulting
from medium and long-distance spread. To this end, the distance between each new cases
occurring in week t + 1 and the nearest case from week t was determined.
Histograms based on these distances were then used to discriminate between the short-,
medium- and long-distance spread. It was assumed that a short-distance spread occurs at a
distance below which 50% of the new cases occur, whilst a medium-distance spread occurs
between the 50% threshold and the 95% threshold and a long-distance spread above the 95%
threshold. Based on the histograms, the distances for short-, medium- and long-distance spread
were thus retrieved and per week these distances were used to categorize the new cases into one
of the three spread categories.
2.4. Correlation between wind density maps and cases
The wind density maps were reprojected to the same Lambert Conformal Conic projection as
for the case density data corrected for livestock density. Wind data was overlayed with the
corrected kernel case density data layer 4 weeks later. Such a time lag was chosen to account for
the time needed between introduction of infected midges and the reporting of first symptoms and
is discussed in greater detail in Section 3. A sea mask was used for further analysis. Both the
corrected case and wind density maps were exported using an ASCII raster format and all
background (no data) values were removed from the data set. The wind density was binned using
following bin sizes: 0–1, 1–2, 2–4, 4–8, 8–16, 16–32, 32–64, 64–128, 128–256, 256–512, 512–
1024, 1024–2048, 2048–4096. Per bin the mean wind density and the mean corrected case
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
167
density were determined. The mean wind density and mean corrected case density were plotted
against each other, and a linear trend line was computed.
2.5. Circular case distribution
In order to quantify the observed asymmetry in case distribution, the circular case distribution
was plotted using rose diagrams. For each case the angle between a case at week t + 1 and the
nearest case of week t was calculated in radians. In order to correct for the difference between
short-, medium- and long-distance spread, the case distribution was weighted according to the
distance. This to ensure that long-distance events have more weight than short-distance ones
because these long-distance events will influence the actual spread pattern more strongly than the
short-distance ones even though they occur less frequently.
The case data were represented using these rose diagrams because of their inherent circular
nature. Statistics developed to compare the direction of circular data assume that the data to be
compared follow a Von Mises distribution, which is the circular normal distribution. However,
preliminary tests indicated that the distribution of new cases deviated significantly from the bestfit Von Mises distribution ( p > 0.05). It was thus necessary to perform a Monte Carlo simulation
method to estimate if the observed circular distribution of cases could be obtained by chance.
3. Results
3.1. Epidemiological curve
In Fig. 1, the weekly distribution of new cases according to the earliest recorded date is
plotted. Note that this curve may differ from epidemiological curves shown elsewhere because
the earliest known date instead of the date of confirmation has been used (see above).
Fig. 1. Epidemiological curve. Number of newly recorded BTV8 outbreaks in Benelux, France and Germany, per week
using earliest date of inclusion in data base (see text for more details).
168
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
Fig. 2. Weekly aggregated wind density model outputs. The maps show weekly aggregated wind events for each
0.058 0.058 pixel for any given week from week 28 (10–16 July), to week 47 (20–26 November). As given in the legend
below colors range from brown/orange to yellow, green and blue and show the number of recorded potentially infective
wind events originating from any previously infected farm for each week shown.
3.2. Weekly aggregated wind density maps
The weekly aggregated wind density maps depicted in Fig. 2 enable to analyze wind events
that occurred in any given week from week 28 (10–16 July), to week 47 (20–26 November). To
account for border effects model outputs within a range of 250 km from the original model output
margins, as defined by the limits of the used ECMWF data, have been excluded.
From the time series follows that the epidemic and consequently the risk of dispersal starts
with a very slow built-up phase. From week 23 (not shown) until week 31 infected wind events
possibly spreading the disease are limited due to the low number of cases.
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
169
Fig. 3. Cumulative wind density model outputs. The map shows for each 0.058 0.058 pixel the cumulative number of
potentially infective wind events originating from infected farms since their inclusion in the epidemiological database
until week 43 (23–29 October 2006). Grid colors have the same legend as in Fig. 2.
During the entire duration of the epidemic, except for week 30, infected wind events are
blowing in a eastwards direction mainly toward Germany and Denmark. In addition, infected
wind events are also oriented in a northwest/west direction in week 30, 32, 34, 36 and 38
potentially towards France and the UK. Spread by wind to the north or south is limited across the
time series, except for strong southerly wind events toward France in week 43 and 44.
3.3. Cumulated weekly wind density maps
In addition to the weekly aggregated wind density maps depicted in Fig. 2 cumulative wind
densities for the entire duration of the epidemic until any given week are shown in Fig. 3. The
cumulated weekly wind density maps enable to assess the overall impact of infected wind events
170
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
at any moment in time. This is in contrast to the weekly aggregated wind density maps that focus
on the impact of single events.
In Fig. 3, is observed that the potential area of spread covers all countries where cases are
found, i.e. Benelux, Germany and France, but also that South-East Anglia has experienced
infected wind events. In addition, Denmark may also have been under the influence of infected
wind events during most of the epidemic. In contrast to this, it follows that even when all infected
wind events are cumulated (week 47), there is no direct spread over the Alps. Under no
circumstance, wind trajectories across the mountain range have been observed.
In Fig. 4, observed cases are superimposed on the cumulated weekly density map of week 43
with a time lag of 4 weeks. A time lag of 4 weeks between wind density model outputs and
recorded symptoms of disease has been used to account for the biological processes and the chain
of events to take place: spread-infected midges, arrival-infected midges, feeding on susceptible
hosts, development of viraemia in susceptible hosts (source of infection for new midges and
radial short-distance spread), development of symptoms in susceptible hosts, detection/reporting
of symptoms by farmer/veterinarian. Results show that both the cases and the cumulated wind
density maps show a similar horizontal spread pattern across the time series.
3.4. Short-, medium- and long-distance spread
The cumulated frequency of the distance of new cases to the nearest case reported in the
previous week is given in Fig. 5. It follows a typical leptokurtic dispersal kernel, with a very high
number of new cases found very close to cases reported in the previous week, and a few cases
being found at high distances from any previously detected case. For example in week 34 the first
Fig. 4. Observed cases and cumulated wind density. Cumulated wind density at week 43 overlayed with observed cases at
week 47. A time lag of 4 weeks is used here to account for the chain of events prior to reporting of clinical symptoms after
arrival of infected midges. Grid colors have the same legend as in Fig. 2. The pattern of cases and wind is matching.
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
171
Fig. 5. Defining thresholds for low, medium and long range spread. Cumulated frequency of the distance of new cases to
the nearest case reported in the previous week: 50% of the cases occur within 5 km of previous case, 95% within 31 km of
previous case and the remaining 5% at a distance more than 31 km of previous cases.
case is found in Antwerp, Belgium at a distance of 40 km to the nearest case in the previous week.
The same happens with the first case in what will be the Ghent cluster: the nearest previous case is
at a distance of 75 km.
Half (50%) of the new weekly cases were distributed within 5 km of the closest case reported
in the previous week, whilst 95% of new cases were distributed within 31 km. These two distance
thresholds, respectively, 5 and 31 km, were used to define short-distance spread (<5 km),
medium-distance spread (>5 km and < 31 km), and long-distance spread events (>31 km).
3.5. Correlation between wind density maps and cases
From a correlation analysis of the data shown in Fig. 3 it is shown that case density, corrected
for livestock density, increases linearly with wind density (Fig. 6). No cases are observed at low
wind densities events. The Pearson’s correlation between the case density and the wind density
Fig. 6. Relationship between wind density and case density. Correlation between cumulated wind density during week 43
and case density corrected for livestock density in week 47 (see Fig. 4).
172
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
was 0.9319, which is statistically significant at 99% confidence level ( p = 0.00) The relationship
between wind density (x) and the corrected case density (y) as estimated by a general linear
model was determined as: y = 2.3336x 29.202.
3.6. Circular distribution of cases
In Fig. 7, the orientation of new cases in relation to the nearest case reported in the previous
week are presented as rose diagrams and weighted by distance, i.e. more weight is given to cases
Fig. 7. Rose diagrams of the orientation of new cases. Circular distribution of new cases to the nearest case reported in the
previous week, weighted by the distance and expressed in radians.
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
173
at medium and long distance (>5 km). From this follows that the orientation of new cases is not
uniform when distance is taken into account. All rose diagrams shown are significantly different
from a uniform distribution ( p < 0.001, Rao’s spacing test of uniformity). Interestingly, the
primary directions are not identical in time. For example, weeks 34 and 38 have different
directional patterns, whilst week 42 is fairly uniformly distributed. An interesting question will
be to explore if these directional patterns correlate with patterns of known environmental
correlates, in particular the direction of the main winds. However this would require higher
resolution wind data than is currently available.
To further illustrate the potential use of such rose diagram the radial distribution of next weeks
cases, as compared to the previous week for the entire duration of the epidemic per distance class
is given in Fig. 8. The results show an increased asymmetry of distribution patterns with
Fig. 8. Rose diagrams of new cases per distance category. Circular distribution of cases to the nearest case for the entire
duration of the epidemic and grouped by distance class. Values are given in radians. Asymmetry of observed spread
patterns increases with distance.
174
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
increasing distance. Whilst at short distances, the distribution of the cases is omnidirectional, the
opposite is found at larger distances where the direction towards east and west is more
pronounced.
4. Discussion
4.1. Wind density model
To our knowledge the developed cumulative wind density modeling approach is novel. The
sole other ongoing effort on modeling spread of Culicoides by wind is based on plume dispersal
models (Gloster et al., 2007a,b), originally developed for the nuclear industry (post-Chernobyl
disaster) and applied with some success to estimate the spread of FMD by wind (Gloster et al.,
2003).
An obvious disadvantage of the latter is that these are based on a known density of pollutant at
source which then is carried away in the atmospheric boundary layer and further diluted in the
overall air mass by prevailing winds. Based on complex meteorological models, the fall-out over
the downwind area is then calculated. This concept whilst applicable for air borne transmission of
weightless and inactive particles such as pollutants and viruses may not be directly applicable to
the spread of insects by wind.
In the developed approach, forward wind trajectories originating from infected pixels have
been calculated individually using relatively straightforward algorithms and near-real time
available six hourly meteorological data (ECMWF). Because (1) wind trajectories which are a
priori unsuitable for Culicoides lift-off have been filtered out and (2) only pixels with confirmed
positive herds as starting point were included, it is believed that the outcome is particularly
adapted to create cumulative density maps of potentially infective wind events and thus to
contribute to estimating the risk of medium and long-distance spread of BTV outbreaks. In our
model wind dispersal is not described as a single event but rather as a series of cumulated events.
This may better reflect the risk of spread.
Based on the proof of concept made in this paper the way forward will be to take the best of
both worlds and further investigate how the main advantages of the 3D wind-plume models may
contribute to the 2D wind density models described in this paper. This should nevertheless be
done with caution and will need additional information on the actual presence and densities of
midges at various altitudes and in varying weather conditions. As an example, our 2D wind
density models neither account for the vertical wind shear, nor for the vertical velocity of
diffusion which are included in 3D plume models. Under stable conditions wind shear can
amount to an angular spread of up to 608 in the atmospheric boundary layer in which dispersion
of pollutants released from the ground mainly takes place. It nevertheless is questionable
whether midges still capable of transmitting BTVat drop down are transported in a similar way,
especially since temperatures are consistently too low throughout the year for midge survival at
higher altitudes (700 hPa, 3000 m), most likely causing stop of active wing movement of the
midges and rapid drop down. The latter may contribute to explain the observed high proportion
of spread at distances lower than 31 km and the overall spread at a pace of 15 km/week of the
outbreak front.
Whilst it may be argued that not taking into consideration vertical wind shear de facto yields a
lowest order indication of the potential airborne spread, i.e. underestimation of the risk of
airborne spread, biological facts as the example given above may at least partly counterbalance
this: insects simply are not weightless, non-moving particles as pollutants are.
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
175
4.1.1. Belgium
In Belgium, the initial area of infection was in the province of Liege. This initial area remained
confined until week 34 when the first cases were observed in the province of Antwerp (at a
distance of approximately 40 km). From the weekly aggregated wind density maps (Fig. 2), it can
be deduced that potentially infective wind events may have occurred in week 32 or 30, i.e.,
respectively, 2–4 weeks prior to the case symptoms.
In what will later be know as the Ghent Cluster, the first case is detected in week 36. For this
particular case a similar ‘wind time lag’ is observed as for the Antwerp cases: suitable winds
were found in week 34 and 32 thus with a time lag of 2–4 weeks. Other suitable winds also
occurred in week 30. However, given the low density of infected wind events, the distance
between the case in week 36, and the nearest case in week 30 (150 km), this is considered
unlikely.
This observed hopping pattern fits the threshold set for long-distance spread in Fig. 5.
4.1.2. France
Though in Fig. 3, wind densities south over France may appear as rather high, these are mainly
caused by high density events towards the end of the epidemic in end October/early November
during weeks 43 and 44 (Fig. 2). These events occurred when the epidemic was declining and
may have resulted in new outbreaks in November in France. However during that period viral
development may already have been hampered due to less favorable weather conditions.
Fig. 9. Impact of roughness of terrain on the spread of Culicoides, Cumulated wind density overlayed with all observed
cases. Yellow lines show stop of spread, despite high wind density toward East and Southeast. Elevations above 300 m are
depicted in grey and show barrier role of altitude related to roughness of terrain to withhold spread of Culicoides by wind.
In grey: areas >300 m altitude.
176
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
4.1.3. Germany
An important observation in Fig. 9 is the limited spread east beyond the Koln–Aachen cluster
(yellow lines) despite the consistent trend throughout time of high wind density events east: large
blue area without red outbreak dots (Fig. 4).
This may be related to the increasing altitude (grey areas in Fig. 9) and roughness of terrain
creating turbulences and drop down of midges. This phenomenon has been discussed by
Bishop et al. (2000, 2005) and was also suspected by Ducheyne et al. (2007) in Bulgaria. In
this outbreak it appears as if infected midges take off from the Maastricht cluster, cross the
Ruhr valley eastwards and drop down in large numbers on the opposite side creating a local
cluster which then further grows through local spread. Altitude may also similarly prevent
spread south. Analysis of other confounding factors such as night temperature, lower livestock
densities, land cover differences, etc. which may further enhance this effect are currently been
carried out.
Though at least over a part of the southern ‘spread limit’, altitude and roughness of terrain may
play an important role. Even though the movement of infected midges through valleys must also
be further analyzed, it is very likely that the single most important variable affecting spread south
is the low density of wind events throughout the epidemics.
4.1.4. United Kingdom
An important question raised during the 2006 epidemic was the risk of introduction to the UK.
This question has become very pertinent given the recent outbreaks in the UK (starting 22
September 2007). An analysis of the time series of aggregated weekly wind densities maps given
in Fig. 2 contributes to this assessment. The fact that no outbreaks were observed in UK in 2006
may be related to the following observations:
(a) All weeks where wind trajectories were recorded sweeping over larger parts of southern UK
occurred at an early stage of the epidemic in August during weeks 30, 32 and 34. These events
individually yielded an overall low wind density (mostly <8 and in week 34 locally up to 16–
32) per pixel and, more importantly, the distance to the closest infected herd was more than
400 km (of which more than half over land) making it very unlikely for infected midges to
reach UK, and should they to do so, in sufficient numbers.
(b) Later during the epidemic some more dense wind events reached the Eastern tip of East
Anglia during week 36 and to a lesser extend during weeks 37, 38, 41 and 42 at the fringe of
mainly northbound wind events. Recorded wind densities were highest during week 36
reaching a density of up to 35 potentially infective wind events in a few pixels. During that
week the shortest distance to a known outbreak was 380 km, the latter being the first outbreak
recorded in what will become the Ghent cluster in Belgium. During weeks 41 and 42 the
Ghent cluster was already well established, the number of newly recorded cases was at its
peak (Fig. 1) and in Belgium outbreaks were recorded in the coastal area. Nevertheless
recorded wind densities in East Anglia were rare and very low.
(c) After the peak of the epidemic (Fig. 1, weeks 41–43) winds were dominantly blowing from
west to east and no wind events reached UK.
4.1.5. Denmark
Finally we can assess the potential risk of spread towards Denmark, where bluetongue may
have a devastating economical impact. Whilst from the data shown here Denmark may have been
at risk during the 2006 epidemic given the high density of infected wind events coming from
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
177
Germany, an outbreak was probably not found because of the large overland distance (320 km)
between the cases in Germany.
4.2. Short-, medium- and long range spread
Across the sea Culicoides spread has been suggested over distances up to 700 km (Sellers,
1992; Braverman and Chechik, 1996). This is due to the relative smoothness of the sea surface
offering no barrier to horizontal airflows. This effect is further enhanced when sea surface
temperature is lower than air temperature (which is normally the case) due to a stabilizing effect
on the atmospheric boundary layer (Sørensen et al., 2000). Recently wind spread over sea in the
Mediterranean was documented (Ducheyne et al., 2007) for BTV outbreaks in 1999–2001.
Evidence was provided suggesting wind spread of BTV16 over a distance of 750 km over sea
from Israel to Rhodos. This paper, which actually served as a basis for the methodology
developed here, also showed evidence supporting wind spread over land. The latter nevertheless
never occurred over such long distances.
In this paper we show that over land, due to the roughness of terrain, wind spread occurred
mainly in a hopping pattern where large distances may be covered between the start and the end
of the epidemic at a rate of 15 km per day, but with ‘intermediary stops’ mostly within a range of
31 km (Gerbier et al., 2007).
This pattern of disease spread is described in the ecological literature as ‘‘stratified dispersal’’
and is typical of dispersal events resulting from processes taking place at different spatial scales,
short-distance contagion, and long-distance jump spread (Shigesada and Kawasaki, 1997).
A similar pattern is described here and includes:
(a) Local spread which is caused by active movement of midges randomly in all directions and
creates circular clusters of outbreaks (Reynolds et al., 2006) and is characterized by a very
high number of cases reported at very short distance. Here, these clusters reach a cumulated
radius of more than 5 km towards the end of the epidemic and are discussed elsewhere in
special issue (Gerbier et al., 2007).
(b) Medium-distance spread which cannot be related to ‘un-aided’ midge movement and which
regularly occurs within a range of 5–31 km.
(c) Long-distance spread, i.e. where a low, but non-null, number of cases are reported at long
distances (distance >31 km) from the nearest case.
4.3. Circular case distribution
One particular feature of the BTV8 epidemic under discussion is the predominant spread
in eastern and to a lesser in western direction. From the wind density maps presented in
Figs. 2 and 3 and the rose diagrams in Fig. 8 it may be concluded that the density of wind
events contributes to explain at least part of this horizontal spread pattern. Whilst the analysis
in Figs. 2 and 3 is qualitative, the rose diagrams in Fig. 8 shows the observed relation
quantitatively. As said earlier, wind is not the sole variable to explain the disease spread but
to date no other variable has been identified which describes the observed spatial spread of
outbreaks in such a comprehensive way, including no spread southwards and a very limited
spread northwards.
In order to include wind into the circular distribution analysis the following points should be
taken into consideration, all of which are currently being studied:
178
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
(1) The time lags were considered uniform throughout the epidemic but it is likely that these time
lags are longer than one week and variable throughout the epidemic.
(2) Weather conditions (temperature, rainfall) throughout the epidemic and during a single day
influence midge flight and should therefore also be included.
(3) ECMWF data are very useful for trajectory analysis, but due to their coarse resolution this
type of data may not take local effects arising from roughness of terrain sufficiently into
account. Using higher resolution simulated wind data may further improve the developed
methodology towards a much more accurate assessment of the impact of wind on vector
spread.
4.4. Animal movement
Movement of BTV positive animals is widely accepted as being a potential source of BTV
spread and must be taken into consideration in addition to wind. An analysis of animal
movement data is currently in the process of being published (Mintiens et al., submitted for
publication). Interestingly the animal spread patterns are omnidirectional and do not always
overlap with the case density patterns, whilst wind density patterns do. In addition whilst a
large number of animal movements were reported from Belgium, only very few were
recorded in Germany. Though this does not mean that BTV8 was not spread through animal
movement it shows that (1) in many areas where animals originating from outbreak areas
were moved to, no BTV8 spread occurred, and (2) that in large areas were BTV8 spread
occurred no animal movements were recorded. The overlap with wind density is much more
consistent.
5. Conclusions
In this paper we have shown that based on the developed wind models:
(a) Density of wind events contributed to explain at least part of the horizontal asymmetrical
spread pattern of the epidemic. The proposed wind density modeling approach may become
an important risk management tool and could enable the implementation of more cost
efficient monitoring and prevention (e.g. vaccination) activities.
(b) Short (<5 km), medium (5–31 km) and long (>31 km) distance spread had a different
impact on disease spread. Computed wind densities were linked to the medium/longdistance spread. Whilst wind driven spread of midges over sea was shown to reach up to
700 km (Ducheyne et al., 2007), the same was not true here over land. Long-distance
spread over land followed a hopping pattern, i.e. with intermediary stops and
establishment of local virus circulation at distances of 35–85 km (e.g. Liege–Antwerp–
Ghent transect). No long range spread was recorded over distances of 300–400 km (e.g. to
UK and Denmark).
(c) Medium/long-distance spread of the outbreaks followed an asymmetric pattern and this
asymmetry increased with distance. short-distance spread occurred in a symmetrical pattern.
This suggests that the latter is mainly driven by insect individual inherently random
movement, whilst the former is driven by an external factor.
(d) A positive relationship between wind density and case density was established,
nevertheless more work is needed to further fine-tune this and to establish the temporal
relationship between disease spread and wind density patterns more precisely. Current
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
179
data suggest time lags of 2–4 weeks occur between infective wind events and the
reporting of BT symptoms.
(e) UK has been at risk of introduction during the 2006 epidemics, but:
(1) Wind events with the highest potential risk of introduction, and pointing to East Anglia,
occurred mainly at an earlier stage of the epidemic, when infected herd clusters were too
distant from the Channel coast line;
(2) At a later stage infected herd clusters were well established within the 31 km medium
distance of spread range from the coast, but potentially infective wind events over UK
showed a very low density.
(3) It is interesting to note that bluetongue virus currently is circulating in UK in the area
highlighted by the wind models.
(f) Terrain roughness may be an important factor preventing spread of infected midges and it will
be important to assess this impact to predict risk of further spread in Europe. Whilst this
blocking effect can be deduced from epidemiological data it is not predicted by the model as
is, highlighting the need for further development such as including vertical aspects (3D) and
higher resolution wind data.
(g) Independently modeled spread based on animal movement data showed that whilst some of
the spread may be related to animal movement this did not account for explaining the
asymmetrical spread of the epidemic.
From the above it is clear that a strong proof of concept has been presented in this paper. The
proposed wind density model should be further enhanced to take into consideration both physical
vertical and biological factors affecting spread and evolve from 2D to 3D models. Both aspects
are currently being addressed by the authors and an airborne trapping device is currently under
development to assess midge presence and densities at varying altitudes and in varying weather
conditions.
This proof of concept opens new avenues for understanding the spread of vector-borne viruses
in Europe. Future development should take into consideration both physical and biological
factors. This will be of great value especially since pathogens and insect vectors of disease
increasingly move or are being moved around the global village due to rising rates of trade and
travel. The latter being further enhanced by environmental changes thus paving the way for a
series of similar events in the near future.
Acknowledgements
This research was sponsored by the European Food Safety Authority (EFSA), Parma, Italy,
contract CT/EFSA/SCAD/2006/01, and the Belgian Federal Authority on Public Health, through
a sub-contract to the Veterinary and Agrochemical Research Centre (VAR), Ukkel, Brussels. The
constructive comments of two anonymous reviewers is greatly acknowledged.
Conflict of interest
None of the authors (Guy Hendrickx, Marius Gilbert, Christoph Staubach, Armin Elbers,
Koen Mintiens, Guillaume Gerbier, Els Ducheyne) has a financial or personal relationship with
other people or organisations that could inappropriately influence or bias the paper entitled ‘‘A
wind density model to quantify the airborne spread of Culicoides species during north-western
Europe bluetongue epidemic, 2006’’.
180
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
References
Alba, A., Casal, J., Domingo, M., 2004. Possible introduction of bluetongue into the Balearic Islands, Spain, in 2000, via
air streams. Vet. Rec. 155, 460–461.
Baker, R.H.A., Guillet, P., Sékétéli, A., Poudiougo, P., Boakye, D., Wilson, M.D., Bissan, Y., 1990. Progress in controlling
the reinvasion of windborne vectors into the western area of the Onchocerciasis Control Program in West Africa. Phil.
Trans. R. Soc. Lond. B 328, 731–750.
Baylis, M., 2002. The re-emergence of bluetongue. Vet. J. 154, 5–6.
Baylis, M., O’Connel, L., Purse, B.V., 2004. Modelling the distribution of bluetongue vectors. Vet. Ital. 40 (3), 176–181.
Bishop, A.L., Barchia, I.M., Spohr, 2000. Models for the dispersal in Australia of the arbovirus vector, Culicoides
brevitarsis Kieffer (Diptera: Ceratopogonidae). Prev. Vet. Med. 47, 243–254 2000.
Bishop, A.L., Spohr, L.J., Barchia, I.M.M., 2005. Effects of altitude, distance and waves of movement of Culicoides
brevitarsis Kieffer (Diptera: Cerotopogonidae). Prev. Vet. Med. 65, 135–145.
Braverman, Y., Chechik, F., 1996. Air streams and the introduction of animal diseases borne on Culicoides (Diptera:
Ceratopogonidae) into Israel. Rev. Sci. Technol. OIE 15, 1037–1052.
Caracappa, S., Torina, A., Guercio, A., Vitale, F., Calabro, A., Purpari, G., Ferrantelli, V., Vitale, M., Mellor, P.S., 2003.
Identification of a novel bluetongue virus vector species of Culicoides in Sicily. Vet. Rec. 153 (3), 71–74.
Chapman, J.W., Reynolds, D.R., Smith, A.D., Woiwod, I.P., 2004. An aerial netting study of insects migrating at high
altitude over England. Bull. Entomol. Res. 94 (2), 123–136.
Codina, B., 1999, Contribució a l’estudi del rentat de l’atmosfera per la pluja (in Catalan), Ph.D., Dept. of Astronomy and
Meteorology, University of Barcelona. ISBN: 84-475-2062-5.
De Liberato, C., Scavia, G., Lorenzetti, R., Scaramozzino, P., Amaddeo, D., Cardeti, G., Scicluna, M., Ferrari, G.,
Autorino, G.L., 2005. Identification of Culicoides obsoletus (Diptera: Ceratopogonidae) as a vector of bluetongue
virus in central Italy. Vet. Rec. 156 (10), 301–304.
Ducheyne, E., De Deken, R., Bécu, S., Codina, B., Nomikou, K., Mangana-Vougiaki, O., Georgiev, G., Purse, B.V.,
Hendrickx, G., 2007. Quantifying the dispersal of Culicoides species in Greece and Bulgaria by wind. Geospatial
Health 1, 177–189.
Gloster, J., Champion, H.J., Sørensen, J.H., Mikkelsen, T., Ryall, D.B., Astrup, P., Alexandersen, S., Donaldson, A.I.,
2003. Airborne transmission of foot-and-mouth disease virus from Burnside Farm, Heddon-on-the-Wall, Northumberland, during the 2001 epidemic in the United Kingdom. Vet. Rec. 152 (17), 525–533 (Erratum in: Vet. Rec., 152
(20), 628).
Gloster, J., Mellor, P.S., Burgin, L., Sanders, S., Carpenter, S., 2007a. Will bluetongue come on the wind to the United
Kingdom in 2007? Vet. Rec. 160, 422–426.
Gloster, J., Mellor, P.S., Manning, A.J., Webster, H.N., Hort, M.C., 2007b. Assessing the risk of windborne spread of
bluetongue in the 2006 outbreak of disease in northern Europe. Vet. Rec. 160, 54–56.
Goffredo, M., Meiswinkel, R., 2004. Entomological surveillance of bluetongue in Italy: methods of capture, catch
analysis and identification of Culicoides biting midges. Vet. Ital. 40 (3), 260–266.
Gerbier, G., Staubach, S., Elbers, A., Mintiens, K., 2007. Time space distribution of virus spread between herds. Prev. Vet.
Med. submitted.
Hengeveld, R., 1989. Dynamics of biological invasions. Chapman and Hall, London, 160 p.
Mellor, P.S., Wittmann, E.J., 2002. Bluetongue virus in the Mediterranean Basin 1998–2001. Vet. J. 164 (1), 20–37.
Mintiens, K., Méroc, E., Faes, C., Cortiñas Abrahantes, J., Hendrickx, G., Staubach, C., Gerbier, G., Elbers, A.R.W, Aerts,
M., De Clercq, K., Deluyker, H., submitted for publication. Impact of human interventions on the spread of
Bluetongue virus sero-type 8 during the 2006 epidemic in north-western Europe. Prev. Vet. Med.
Panagiotatos, D.E., 2004. Regional overview of bluetongue viruses, vectors, surveillance and unique features in Eastern
Europe between 1998–2003. Vet. Ital. 40 (3), 61–73.
Patakakis, M.J., 2004, Culicoides imicola in Greece. In: MacLachlan, N.J., Pearson, J.E. (Eds.), In Bluetongue, vol. 40, Pt
1, Proceedings, pp. 232–34.
Purse, B.V., Nedelchev, N., Georgiev, G., Veleva, E., Boorman, J., Denison, E., Veronesi, E., Carpenter, S., Baylis, M.,
Mellor, P.S., 2006. Spatial and temporal distribution of bluetongue and its Culicoides vectors in Bulgaria. Med. Vet.
Entomol. 20, 335–344.
Reynolds, D.R., Chapman, J.W., Harrington, R., 2006. The migration of insect vectors of plant and animal viruses. Adv.
Vir. Res. 67, 453–517.
Ritchie, S.A., Rochester, W., 2001. Wind-blown mosquitoes and introduction of Japanese encephalitis into Australia.
Emer. Infect. Dis. 7 (5), 900–903.
G. Hendrickx et al. / Preventive Veterinary Medicine 87 (2008) 162–181
181
Sellers, R.F., Maarouf, A.R., 1991. Possible introduction of epizootic hemorrhagic disease of deer virus (serotype 2) and
bluetongue virus (serotype 11) into British Columbia in 1987 and 1988 by infected Culicoides carried on the wind.
Can. J. Vet. Res. 55 (4), 367–370.
Sellers, R.F., 1992. Weather, Culicoides, and the distribution and spread of bluetongue and African horse sickness viruses.
In: Walton, T.E., Osburn, B.I. (Eds.), Bluetongue, African Horse Sickness, and Related Orbiviruses. CRC, Boca
Raton, FL, p. 1042.
Sellers, R.F., Maarouf, A.R., 1989. Trajectory analysis and bluetongue virus serotype 2 in Florida 1982. Can. J. Vet. Res.
53 (1), 100–102.
Sellers, R.F., Pedgley, D.E., 1985. Possible windborne spread to western Turkey of bluetongue virus in 1977 and of
Akabane virus in 1979. J. Hyg. (Lond.) 95 (1), 149–158.
Sellers, R.F., Pedgley, D.E., Tucker, M.R., 1978. Possible windborne spread of bluetongue to Portugal. June–July 1956. J.
Hyg. (Lond.) 81 (2), 189–196.
Shigesada, N., Kawasaki, K., 1997. Biological invasions: theory and practice. Oxford University Press, Oxford, UK,
205 p.
Sørensen, J.H., Mackay, D.K., Jensen, C.O., Donaldson, A.I., 2000. An integrated model to predict the atmospheric spread
of foot-and-mouth disease virus. Epidemiol. Infect. 124 (3), 577–590.
Thiry, E., Saegerman, C., Guyot, H., Kirten, P., Losson, B., Rollin, F., Bodmer, M., Czaplicki, G., Toussaint, J.F., De
Clerq, K., Dochy, J.M., Dufey, J., Gilleman, J.L., 2006. Bluetongue in northern Europe. Vet. Rec. 159, 327.
Torina, A., Caracappa, S., Mellor, P.S., Baylis, M., Purse, B.V., 2004. Spatial distribution of bluetongue virus and its
Culicoides vectors in Sicily. Med. Vet. Entomol. 18 (2), 81–89.
Uppala, S.M., Kållberg, P.W., Simmons, A.J., Andrae, U., da Costa Bechtold, V., Fiorino, M., Gibson, J.K., Haseler, J.,
Hernandez, A., Kelly, G.A., Li, X., Onogi, K., Saarinen, S., Sokka, N., Allan, R.P., Andersson, E., Arpe, K.,
Balmaseda, M.A., Beljaars, A.C.M., van de Berg, L., Bidlot, J., Bormann, N., Caires, S., Chevallier, F., Dethof, A.,
Dragosavac, M., Fisher, M., Fuentes, M., Hagemann, S., Hólm, E., Hoskins, B.J., Isaksen, L., Janssen, P.A.E.M.,
Jenne, R., McNally, A.P., Mahfouf, J.-F., Morcrette, J.-J., Rayner, N.A., Saunders, R.W., Simon, P., Sterl, A.,
Trenberth, K.E., Untch, A., Vasiljevic, D., Viterbo, P., Woollen, J., 2005. The ERA-40 re-analysis. Quart. J. R.
Meteorol. Soc. 131, 2961–3012.
Wittmann, E., 2002, Temperature and the transmission of arboviruses by Culicoides biting midges, Ph.D., University of
Bristol.
Wittmann, E.J., Baylis, M., 2000. Climate change: Effects on Culicoides-transmitted viruses and implications for the UK.
Vet. J. 160 (2), 107–117.