ArticlePDF Available

Abstract and Figures

In this paper we present a novel methodology applied in Spain to model spatial abundance patterns of potential vectors of disease at a medium spatial resolution of 5 x 5 km using a countrywide database with abundance data for five Culicoides species, random regression Forest modelling and a spatial dataset of ground measured and remotely sensed eco-climatic and environmental predictor variables. First the probability of occurrence was computed. In a second step a direct regression between the probability of occurrence and trap abundance was established to verify the linearity of the relationship. Finally the probability of occurrence was used in combination with the set of predictor variables to model abundance. In each case the variable importance of the predictors was used to biologically interpret results and to compare both model outputs, and model performance was assessed using four different accuracy measures. Results are shown for C. imicola, C. newsteadii, C. pulicaris group, C. punctatus and C. obsoletus group. In each case the probability of occurrence is a good predictor of abundance at the used spatial resolution of 5 x 5 km. In addition, the C. imicola and C. obsoletus group are highly driven by summer rainfall. The spatial pattern is inverse between the two species, indicating that the lower and upper thresholds are different. C. pulicaris group is mainly driven by temperature. The patterns for C. newsteadii and C. punctatus are less clear. It is concluded that the proposed methodology can be used as an input to transmission-infection-recovery (TIR) models and R0 models. The methodology will become available to the general public as part of the VECMAP™ software.
Content may be subject to copyright.
Introduction
Bluetongue, a vector-borne arboviral (Orbivirus;
Reoviridae) infectious disease listed by the World
Organization for Animal Health (OIE), is transmitted
by Culicoides biting midges spp. (Diptera:
Ceratopogonidae). Twenty-four different bluetongue
virus (BTV) serotypes are currently known worldwide.
In European outbreaks, five serotypes have been iden-
tified in the Mediterranean biome (BTV1, 2, 4, 9, 16)
and four in the temperate biome (BTV1 and 8 plus
two alleged vaccine strains BTV6 and BTV11).
Depending on the serotype, BTV causes high morbid-
ity and mortality in certain breeds of sheep and other
domestic and wild ruminants (Elbers et al., 2008; Le
Gal et al., 2008; Allepuz et al., 2010). The disease has
a high economic impact on livestock, e.g. the total eco-
nomic losses of the recent BTV8 epidemic in the
Netherlands amounted to 32 million in 2006 and
€164-175 million in 2007 (Velthuis et al., 2010).
In the European Mediterranean biome the BTV
spread that started in the late 1990s was largely, but
not solely, related to the originally tropical vector
C. imicola. At its distribution margins BTV was also
related to indigenous European Culicoides species,
most notably species of the C. obsoletus complex
(Purse et al., 2008). Whether the presence of C. imico-
la in the Mediterranean biome is related to a recent
invasion followed by the incursion of BTV serotypes
or it has already been present for a longer period of
time is still an open question (Conte et al., 2009), but
recent work has shown that the theory of a recent
introduction is highly unlikely (Mardulin et al., 2013).
Until 2006, bluetongue remained limited to the
areas around the Mediterranean, i.e. the
Mediterranean biome. However, from August 2006, in
the absence of C. imicola, an unprecedented introduc-
tion, establishment and spread in the temperate biome
of the non-Mediterranean BTV8 serotype, solely based
Geospatial Health 8(1), 2013, pp. 241-254
Corresponding author:
Els Ducheyne
Avia-GIS
Risschotlei 33, 2980 Zoersel, Belgium
Tel. +32 3 458 2979
E-mail: educheyne@avia-gis.be
Abundance modelling of invasive and indigenous Culicoides
species in Spain
Els Ducheyne
1
, Miguel A. Miranda Chueca
2
, Javier Lucientes
3
, Carlos Calvete
4
, Rosa
Estrada
3
, Gert-Jan Boender
5
, Els Goossens
1
, Eva M. De Clercq
1
, Guy Hendrickx
1
1
Avia-GIS, Zoersel, Belgium;
2
Laboratory of Zoology and Emerging Diseases, University of the Balearic
Islands, Mallorca, Spain;
3
Unidad de Sanidad y Produccion Animal, Centro de Investigacion y Tecnologia,
Agroalimentaria, Zaragoza, Spain;
4
Departamento de Patalogia Animal, Universidad de Zaragoza, Zaragoza,
Spain;
5
Central Veterinary Institute, Wageningen University, Lelystad, The Netherlands
Abstract. In this paper we present a novel methodology applied in Spain to model spatial abundance patterns of potential vec-
tors of disease at a medium spatial resolution of 5 x 5 km using a countrywide database with abundance data for five
Culicoides species, random regression Forest modelling and a spatial dataset of ground measured and remotely sensed eco-cli-
matic and environmental predictor variables. First the probability of occurrence was computed. In a second step a direct
regression between the probability of occurrence and trap abundance was established to verify the linearity of the relation-
ship. Finally the probability of occurrence was used in combination with the set of predictor variables to model abundance.
In each case the variable importance of the predictors was used to biologically interpret results and to compare both model
outputs, and model performance was assessed using four different accuracy measures. Results are shown for C. imicola,
C. newsteadii, C. pulicaris group, C. punctatus and C. obsoletus group. In each case the probability of occurrence is a good
predictor of abundance at the used spatial resolution of 5 x 5 km. In addition, the C. imicola and C. obsoletus group are high-
ly driven by summer rainfall. The spatial pattern is inverse between the two species, indicating that the lower and upper
thresholds are different. C. pulicaris group is mainly driven by temperature. The patterns for C. newsteadii and C. punctatus
are less clear. It is concluded that the proposed methodology can be used as an input to transmission-infection-recovery (TIR)
models and R
0
models. The methodology will become available to the general public as part of the VECMAP
TM
software.
Keywords: spatial abundance modelling, medium resolution, Culicoides, random Forests, Spain.
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
on indigenous Culicoides species, have been observed
north of 50° N in Benelux, Germany and France
(Saegerman et al., 2008). In the following 2 years, this
serotype spread very rapidly in the temperate part of
Europe, including the temperate tip of Sweden and
Norway. The “restriction zone” imposed by the
European Union (EU) now covered an area up to
approximately 2.3 million km
2
, i.e. 43% of the total
European territory.
Whilst BTV1 had been restricted to the European
eastern Mediterranean biome until 2006, a newly
introduced BTV1 strain originating from Morocco
invaded the C. imicola range in 2007 in Spain (OIE,
2007a) and in Portugal (OIE, 2007b). From the latter
country it spread to the northern part of Spain and
also southern France (OIE, 2007c) into areas where
C. imicola is absent and where indigenous European
midge species of the C. obsoletus complex and C. new-
steadii are responsible for transmission of the BTV.
The potential further expansion northwards was
stopped in 2009 by a massive, countrywide vaccina-
tion campaign in France. In addition, BTV8 also
spread southwards into the European Mediterranean
biome, invaded Spain and also appeared in Italy, cre-
ating a large geographical overlap between both
serotypes in Europe.
Several authors have modelled the probability of the
occurrence of Culicoides species in the Mediterranean
basin, relating the observed presence/absence or abun-
dance of the midges to meteorological and environ-
mental variables mainly derived from satellite imagery.
These relationships are established using either statis-
tical techniques such as non-linear discriminant analy-
sis as used by Tatem et al. (2003) in Portugal, logistic
regression (Calvete et al., 2008) and more recently,
data-mining techniques such as random Forests (Peters
et al., 2011).
In order to develop transmission-infection-recovery
(TIR) models (Szmaragd et al., 2009, 2010) or basic
reproduction number (R
0
) models (Hartemink et al.,
2009), abundance data of Culicoides spp. are an
essential parameter. Some abundance models yield
abundance classes as output (Tatem et al., 2003),
while other studies established a direct linear relation-
ship between probability of occurrence and abundance
measured in the traps (Calvete et al., 2008; Guis et al.,
2011). The first group of methods could be used as
input for TIR models or for R
0
models (e.g. Hartemink
et al., 2011) but because the output is categorical map
outputs that indicate risk in indirect manner; the sec-
ond category assumes that there is a linear relationship
between probability of occurrence and abundance, a
relationship that remains to be validated for
Culicoides. In this paper, we go one step further and
present a new methodology to estimate abundance
data using random Forests and a wide set of meteoro-
logical and environmental data sets. Outputs are gen-
erated as a continuous raster at a spatial resolution of
5 x 5 km. First, we estimate the probability of occur-
rence, in a second step a direct regression between the
probability of occurrence and trap abundance is estab-
lished to verify the linearity of the relationship; final-
ly, probability of occurrence is used in combination
with an additional set of predictor variables to esti-
mate the abundance.
Material and methods
Entomological data
Data collected on mainland Spain and the Balearic
Islands under the Spanish Bluetongue National
Surveillance Programme in 2007 (Calvete et al., 2008)
were used in this study. Culicoides spp. specimens
were caught using ultraviolet light traps, fitted with a
suction fan and a collection vessel containing ethanol
and ethylene glycol in water to preserve the samples.
The traps were positioned outside selected farms with
a minimum of 10 large ruminants and not further than
30 m away from livestock. The traps were operational
for one night per week in each farm. Monthly aggre-
gated data showing the maximum catch per farm were
available for this study. A hand-held global position-
ing system receiver recorded the coordinates of the
sample locations (Fig. 1). The red dots indicate traps
where specimens were found, while the green dots rep-
resent traps where no specimens were caught. The
abundance of a species was calculated and reported as
log
10
(n+1), where n is equal to the number of individ-
uals caught in a trap. The radius of the red circle in the
figure is a measure of the abundance.
Trapped Culicoides spp. were identified as described
by Calvete et al. (2008) resulting in abundance data
for the following species: C. imicola, C. pulicaris,
C. punctatus, C. newsteadii and the Obsoletus group
containing C. obsoletus, C. scoticus, C. montanus,
C. dewulfi and C. chiopterus.
Meteorological and environmental data
Variables describing environmental conditions for
Culicoides were selected based on expert knowledge
and a literature review (Conte et al., 2007; Calvete et
al., 2008). This study included a total of 74 variables
242
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
Fig. 1. Location and abundance of five different Culicoides species in Spain (2007).
belonging to the following categories: land cover,
ground water-related variables, precipitation and land
surface temperature (LST).
The land cover information was derived from the
CORINE dataset (JRC, 2005). The percentages of the
three main land cover classes relevant for Culicoides,
i.e. urban, agriculture and natural, within a 5 x 5 km
pixel were computed. The percentage cover was
determined for land cover classes particularly
favourable for Culicoides, i.e. pastures and forest.
Within the forest class, cover percentage was also
computed for three forest types: broadleaved, conifer-
ous and mixed forest. Human population pressure on
the landscape was assessed by the number of inhabi-
tants/km
2
(GWPv3.0 compiled as described by Wint
(2005)). Other environmental data layers included the
water capacity of the topsoil (JRC, 2009), distance to
waterways (GfK geoMarketing, 2009) and the
GTOPO30 elevation (USGS, 1996). The mean total
yearly precipitation and the mean monthly mean pre-
cipitation were obtained from the WORLDCLIM
dataset (Hijmans et al., 2005). Data from the MODIS
sensor (http://modis.gsfc.nasa.gov) were used to
derive additional variables such as the LST during
daytime and at night-time as well as two vegetation
indices, i.e. the enhanced vegetation index (EVI) and
243
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
Species
Positive
traps
Negative
traps
Tot al
C. imicola
C. newsteadii
C. pulicaris
C. punctatus
C. obsoletus
96
78
112
74
122
62
80
48
84
36
158
158
158
158
158
Table 1. Number of traps and catches of Culicoides spp. in
Spain.
Species
Positive
traps
Negative
traps
Tot al
C. imicola
C. newsteadii
C. pulicaris
C. punctatus
C. obsoletus
62
78
48
74
36
62
78
48
74
36
124
154
96
148
72
Table 2. Number of traps and catches of used to model the
occurrence probability of the various Culicoides spp. in Spain.
the normalized difference vegetation index (NDVI)
(Gao et al., 2000). Time series covering the years
2004, 2005, 2006, 2007 and 2008 were used to
detect yearly trends. After Fourier transformation, the
first three harmonics were included as predictor vari-
ables (Scharlemann et al., 2008). These harmonics
describe seasonal cycles, e.g. in temperature and veg-
etation variables, by approximating the temporal sig-
nal with cosinusiodal waves. The amplitude of the
cosinusoidal wave represents the magnitude and the
phase the timing of when the maximum amplitude is
reached. For real-life temperatures, for example, the
amplitude is the maximum difference between the
lowest and the highest recorded temperature, whilst
the phase indicates when the peak (maximum of the
highest temperature) is achieved. The first harmonics
thus represent the major seasonal differences in tem-
perature; subsequent harmonics describe secondary
and tertiary seasonal phenomena. Finally, the number
of days with a mean temperature above (i) 0°C; (ii)
5 °C; and (iii) 12.5°C were derived from the MODIS
LST time series for the year 2007 for which data on
midges were available.
All data layers were clipped to the extent of the
study area i.e. mainland Spain and Portugal.
Geographical information systems (GIS) manipula-
tions were performed using ArcGIS, 9.3 (ESRI,
2009).
Modelling framework
Random Forest modelling
Models were generated using the random Forest
(RF) approach (Breiman, 2001). This is a robust
ensemble learning technique, which can be applied
either to model probability maps using a random clas-
sification forest or abundance maps through a random
regression forest. The technique consistently outper-
forms traditional modelling techniques such as logistic
regression (Cutler et al., 2007; Peters et al., 2007).
Random classification forests have been used to assess
if temperature and precipitation affect the minimum
infection rate of Culex species for the West Nile virus
in Illinois (Ruiz et al., 2010) and to model the current
spatial distribution of Aedes albopictus in Europe
using a wide set of predictor variables (ECDC, 2009).
RF allows both internal and external validation
through a bootstrapping procedure. For each classifi-
cation or regression tree, the full data set is boot-
strapped, i.e. a number of data points are sampled
from the complete data set with replacement. From
the bootstrapped sample approximately one third of
the data are excluded. This set of the excluded data is
referred to as the “out-of-bag” (OOB) dataset for the
tree; each tree will have a different OOB dataset. Since
these datasets are not used to build the tree, they con-
stitute an independent validation dataset for the tree in
absence of autocorrelation.
To measure the classification error of the random
classification forest, the OOB data for each tree are
classified and the classification error is computed. The
error values for all trees in the forest are averaged to
give the overall classification error. In case of random
regression forests, the error is expressed as the mean
squared error between the predicted values for the
OOB data and the observed data.
Probability of occurrence maps
For each species, the abundance data were classified
into presence and absence classes. If a site were nega-
tive over the entire year, it was classified into the
absent class; all other sites were classified into the
present class. In order to maximize model accuracy
(McPherson et al., 2004), equal numbers of presence
and absence sites were randomly selected.
The probability modelling was based on a balanced
set of presence and absence observations at the sample
sites. The presence/absence traps were selected at ran-
dom from the entire dataset of observations points.
The number of traps used for each species can be
found in Table 2. The performance of the probability
model was assessed using four accuracy measures: per-
244
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
centage of correctly classified instances (PCC), sensi-
tivity, specificity and “area under the receiver operat-
ing curve” (AUCOC). The AUCOC can be roughly
interpreted as the probability that a model will cor-
rectly distinguish a true presence and a true absence at
random. For example, a value of 0.8 for the AUCOC
means that for 80% of the time a random selection
from the positive group (presence) will have a score
greater than a random selection from the negative
class (absence) (Fielding and Bell, 1997). Predictor
variable importance is assessed through the measure-
ment of the decline in performance if the model is run
without the variable. This performance decline is
expressed as the mean decrease in GINI index
(Breiman, 2001). The GINI index is a measure of
homogeneity from 0 (homogeneous) to 1 (heteroge-
neous) versus the contribution of each variable. The
changes in GINI are summed for each variable and
normalised at the end of the calculation.
Abundance maps
Abundance data were log
10
transformed according
to the following formula log
10
(n+1), where n is the
maximum number of individuals caught for the site in
question. The species data for all sites were used in the
abundance modelling.
In many cases, the predicted probability of occur-
rence is a surrogate of habitat suitability, and is there-
fore used as an indirect predictor of species abundance
(Osborne et al., 2001; Boyce et al., 2002; Gibson et
al., 2004; Chefaoui et al., 2005; Calvete et al., 2008).
To test this hypothesis, in a first step, the Pearson cor-
relation coefficient was established between the pre-
dicted probability of occurrence and the observed
abundance data. In a second step, the probability of
occurrence was added as a predictor variable for mod-
elling the abundance of a species using RF. Accuracy
was assessed quantitatively through the calculation of
the mean squared error (MSE) and the coefficient of
determination (cor) between the log
10
-transformed
observed and predicted abundances. The importance
of the predictors for the abundance modelling was
assessed using the “increase in node purity” (INP).
This measure shows how much the impurity, i.e. a
measure for inaccuracy, increases when that variable is
omitted from the model. Important variables have a
high value. Statistical analysis and modelling were per-
formed in the R2.10.1 statistical language environ-
ment (R Development Core Team, 2006) using of the
R-package “rgdal”, version 0.6-25, and
“randomForest”, version 4.5-34.
Results
Observed presence and abundance data
For this Spanish study area, data of 158 traps were
used. For each species, the number of positive traps,
negative traps as well as the total number of traps can
be found in Table 1. The observed data show three dis-
tinct patterns for Culicoides species in Spain (Fig. 1).
C. imicola was mainly present in the drier central
and south-western part of continental Spain and most-
ly absent from the northern more humid part. In addi-
tion some specimens were also caught along the Ebro
Valley and along the north-eastern Mediterranean
coast. The species was also present in the Balearic
Islands.
C. newsteadii and C. punctatus both have a compa-
rable distribution pattern along a north-east/south-
west axis with distinct areas of absence north and
south of this axis. Some more positive sites were found
for C. newsteadii in the southern part of the country
and for C. punctatus in the North.
Finally, C. pulicaris and the C. obsoletus group are
the most widespread Culicoides groups in Spain sug-
gesting they may adapt to a wider range of eco-cli-
matic circumstances than the other species. The distri-
bution pattern of the C. obsoletus group was the
inverse of the pattern observed with C. imicola.
Though C. obsoletus was present in most of the area
covered by C. imicola, low densities were recorded,
whilst no C. imicola were found in northern Spain,
where the highest densities of C. obsoletus have been
recorded.
Probability of occurrence maps
Fig. 2 depicts the probability of occurrence for the
Culicoides species in Spain. C. imicola is predicted to
occur in southwest Spain. It has a crisp distinction
between the high and low probability zones, which
closely resembles the observed distribution pattern.
This is reflected in high accuracy indices: PCC = 0.81
and AUCOC = 0.88 (Table 3). C. newsteadii and
C. punctatus exhibit a similar predicted probability of
occurrence. The gradient between the higher and
lower probability areas is smooth. A large zone shows
a medium probability. The accuracy measures are sim-
ilar for both species: the specificity is fair, 0.78 and
0.72 for C. newsteadii and C. punctatus, respectively)
but the sensitivity is lower, indicating that more sites
are falsely classified as being present. This is even
more pronounced for the distribution of C. pulicaris
245
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
Fig. 2. The predicted probability of occurrence of C. imicola, C. newsteadii, C. pulicaris group, C. punctatus and C. obsoletus group
in Spain.
and C. obsoletus, which are generally more ubiquitous
resulting in a lower PCC = 0.51 and AUCOC = 0.59
(Table 3).
Table 4 summarises the 10 most important vari-
ables driving the model. The occurrence of C. imico-
la is most strongly driven by precipitation. The five
most important factors were found to be precipitation
variables such as the precipitation level of the driest
month and the monthly mean precipitation of the
summer months from June until September. The tim-
ing of the greening of the vegetation (phase 1 of the
NDVI and phase 1 of the EVI) and the number of
days with daytime LTS greater than 12 °C were also
important. Indeed, in 2008, the temperature variables
were found to be one of the 10 most important vari-
ables. Spring precipitation was the 10
th
most impor-
tant factor.
The occurrence of C. newsteadii is determined by
the night-time LTS through the first amplitude (peak
of the annual cycle) and the third phase (the timing of
the peak of the tri-annual cycle), forest cover and the
winter precipitation variables (maximum precipita-
tion, plus the precipitation in December and
November).
The occurrence of the C. pulicaris group turned out
to be mainly determined by temperature: seven out of
246
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
10 variables are temperature-related, both day- and
nigh temperatures. These variables were: the peak of
the bi-annual and tri-annual daytime temperature
cycle, the number of days with night temperatures in
each year greater than 5 °C, the number of days above
the freezing point in 2008 and the peak of the tri-
annual night temperature cycle. Elevation, a proxy for
temperature, was also retained. Additionally, the tim-
ing of the bi-annual (phase 2) and tri-annual vegeta-
tion peak (phase 3) of the EVI was found to drive the
probability of occurrence. Thus, the crucial factors for
the occurrence of C. punctatus are a combination of
different types of variables with the main factors being
temperature (first and third amplitude of night-time
LST, the number of days above 0 °C in 2008, the num-
ber of days above 12 °C in 2007 and the mean yearly
night-time land surface temperature), distance to
water and precipitation. However, elevation and forest
cover both play a role.
Finally, the occurrence of the C. obsoletus group is
clearly determined by temperature, both in the day
and during night: seven out of the 10 variables were
Species PCC Sensitivity Specificity Kappa AUCOC
C. imicola
C. newsteadii
C. pulicaris
C. punctatus
C. obsoletus
0.81
0.75
0.62
0.70
0.51
0.76
0.72
0.59
0.69
0.47
0.85
0.78
0.65
0.72
0.56
0.61
0.50
0.24
0.41
0.03
0.88
0.81
0.58
0.80
0.59
Table 3. Accuracy measures of the probability of occurrence
modelling for Culicoides spp. in Spain.
C. imicola C. newsteadii C. pulicaris group C. punctatus C. obsoletus group
Predictor VARI M P Predictor VARIMP Predictor VARIMP Predictor VA R IM P Predictor VARIMP
Minimum
precipitation
6.19 Nighttime LST,
amplitude 1
2.51 Day LST,
amplitude 2
1.52 Nighttime LST,
amplitude 1
2.82 EVI,
phase 3
1.22
July
precipitation
5.63 Forest cover
(% per pixel)
2.3 Nights in 2007
with LST >5 °C
1.35 Distance to water 2.57 Nights in 2008
with LST >5 °C
0.98
September
precipitation
3.46 Nighttime LST,
phase 3
2.13 EVI,
phase 3
1.33 June
precipitation
2.32 EVI,
amplitude 3
0.96
August
precipitation
3.22 December
precipitation
2.13 EVI,
phase 2
1.25 December
precipitation
2.1 Nighttime LST,
phase 3
0.92
June
precipitation
3.08 Population density 2.05 Elevation (from
the DTM model)
1.08 Nighttime LST,
amplitude-3
1.84 NDVI,
phase 1
0.88
NDVI
Phase 1
2.63 Distance to water 1.86 No. of days in 2008
with LST >0 °C
1.08 No. of days in 2008
with temp. >0 °C
1.8 Nights in 2007
with LST >5 °C
0.86
Daytime LST,
phase 2
2.30 Maximum
precipitation
1.83 No. of nights in
2006 with
LST >5 °C
1.04 Days in 2007 with
LST >12 °C
1.73 Nighttime LST,
mean
0.84
EVI, phase 1 1.80 Total
precipitation
1.83 Daytime LST,
amplitude 3
1.02 Forest cover
(% per pixel)
1.57 Days in 2008
with LST >12 °C
0.80
Days in 2008
with daytime
LST >12 °C
1.66 November
precipitation
1.70 No. of nights in
2008 with
LST >5 °C
1.01 Elevation (from
the DTM model)
1.52 Days in 2006
with LST >12 °C
0.79
May
precipitation
1.56 June
precipitation
1.64 Nighttime LST,
amplitude 3
0.99 Nighttime LST,
mean
1.44 Day LST,
phase 1
0.73
Table 4. The 10 most important predictors indicated by the variable importance (VARIMP) for the probability of occurrence of the
C. imicola, C. newsteadii, C. pulicaris group, C. punctatus and C. obsoletus groups.
DTM: digital terrain model; EVI: enhanced vegetation index; LST: land surface temperature; NDVI: normalized difference vegeta-
tion index.
247
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
found to be temperature-related. The timing of the
annual and tri-annual peak of vegetation was also
important but precipitation does not have an influence
on the distribution modelling for this species group.
Development of abundance models
In Table 5 the correlation is given between the pre-
dicted probability of occurrence and the log
10
(n+1)-
observed abundance. This relationship is shown in
Fig. 3. Whilst all reported correlation coefficients were
highly significant, it is clear that correlation is mainly
achieved because of the good match between negative
traps and zero probabilities of occurrence. When
removing the zero observations, no correlation was
observed (not shown here). A simple linear relation-
ship will thus not allow modelling the abundance of
the different Culicoides species.
The output of abundance RF models is shown in
Fig. 4. C. imicola is most abundant (from >200 indi-
viduals up to 2,200 individuals) in Extremadura (cen-
tral Spain). Andalucía in south-western Spain featured
a medium abundance, while other regions had abun-
dance lower than 100 specimens. Very low densities
are predicted in northern Spain, concurrent with the
observed data. C. newsteadii and C. punctatus were
most abundant on the Mediterranean coast and in
central to northern Spain. The Pyrénées are predicted
to have a very low abundance of this species. The
C. pulicaris group, and also the C. obsoletus group, is
predicted to occur very abundantly in the entire coun-
try with the exception of the driest areas, i.e.
Extremadura and Andalucía. A visual comparison of
the predicted abundance map of C. obsoletus (Fig. 4)
and the observed data (Fig. 1) suggests a considerable
underestimation of abundance in the overlapping area
with C. imicola, whilst higher abundances were cor-
rectly predicted in the northern part of Spain. This is
also the case for C. pulicaris, though to a lesser extent.
Results for species with distinct presence/absence
Species Correlation
a
C. imicola
C. newsteadii
C. pulicaris
C. punctatus
C. obsoletus
0.06
0.28
0.97
0.39
0.53
Table 5. Pearson correlation between the predicted probability
of occurrence and log
10
(n+1)-observed abundance of Culicoides
spp. in Spain.
a
Coefficient of determination.
Fig. 3. The relation between the log
10
(n+1) observed abundance and the predicted probability of occurrence of Culicoides spp. in Spain.
248
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
Fig. 4. Predicted abundance of Culicoides spp. in Spain (abundance is represented as log
10
(n+1) where n = number of individuals).
areas, i.e. C. imicola, C. newsteadii and C. punctatus,
showed a more crisp distinction between high and low
abundance areas.
The interpretation of predictions obtained outside
the study area (i.e. Spain) reflects a similar pattern. In
southern France, the absence of C. imicola and the
presence of C. obsoletus group and C. pulicaris were
correctly predicted as compared to the results from the
Epizone-Dynvect project (Balenghien et al., 2010).
Results obtained in Portugal, however, were less clear:
C. imicola was correctly predicted, C. obsoletus to a
lesser extent so, while the discriminating power
between absence and presence zones was less for
C. pulicaris. Results obtained with C. newsteadii and
C. punctatus in these countries could not be compared
since no data were available to us.
In Table 6 the MSE of the abundance model for each
species is given. In case of C. imicola, the MSE on the
log
10
(n+1) data amounts to 0.73. This is even lower
for the other species, which is also reflected in the
determination coefficient. Fig. 5 shows the relation-
ship between the observed and predicted abundance
data.
Table 7 lists the 10 most important predictor vari-
ables together with their variable importance for the
abundance modelling of the different Culicoides
249
species. For all species the most important predictor
variables determining the abundance is the probability
of occurrence. This is concurrent with the previously
identified correlation between the predicted probabili-
ty and the observed log
10
(n+1)-transformed abun-
dance data. The probability can therefore be consid-
ered a necessary condition for the abundance, and the
additional predictor variables will define the level of
abundance within the presence zone. For C. imicola,
C. newsteadii and C. pulicaris these predictor vari-
ables are similar to those found in the probability
mapping. Abundance of C. imicola is mostly driven by
summer precipitation, abundance of C. newsteadii by
temperature and precipitation and C. pulicaris by
number of days below the 5 °C threshold and number
of days with temperatures above freezing. C. puncta-
tus abundance seems to be determined more by tem-
perature variables (five variables), albeit different vari-
ables than be found in the probability of occurrence
model. Additionally, precipitation in June, population
density and the peak of the annual vegetation cycle
influence the abundance. C. obsoletus abundance is
relative to summer precipitation as well as the mean
yearly daytime temperature and the peak of annual
cycle of the temperature at night. Summer precipita-
tion did not influence the probability of occurrence.
Discussion
Aggregation of time series
The Spanish dataset originates from a network of
traps, which have been sampled on a regular basis dur-
ing one year. The available aggregated data (i.e. maxi-
mum catch per month) allow reducing the risk of false
negative trap sites and enable the use of annual maxi-
mum catch figures per trap site as a measure of abun-
dance leading to improved apparent density estimates.
It is important to note that this apparent density is not
the same as the true population density, because the
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
Species MSE Cor
a
Intercept Coefficient
C. imicola
C. newsteadii
C. pulicaris
C. punctatus
C. obsoletus
0.81
0.75
0.62
0.70
0.51
0.76
0.72
0.59
0.69
0.47
0.85
0.78
0.65
0.72
0.56
0.88
0.81
0.58
0.80
0.59
Table 6. The mean of squared error (MSE) and Pearson correla-
tion coefficient with intercept and regression coefficient between
the observed and predicted data for Culicoides spp. in Spain.
a
Coefficient of determination.
Fig. 5. The relation between the observed abundance of Culicoides spp. in Spain and the predicted abundance (represented as log
10
(n+1)
where n = number of individuals).
250
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
C. imicola C. newsteadii C. pulicaris group C. punctatus C. obsoletus group
Predictor VARI M P Predictor VARIMP Predictor VARIMP Predictor VA R IM P Predictor VARIMP
Probability of
occurrence
50.88 Probability of
occurrence
67.80 Probability of
occurrence
39.15 Probability of
occurrence
86.72 Probability of
occurrence
44.62
Minimum
precipitation
29.18 Nighttime LST,
amplitude 1
10.32 Elevation (from
the DTM model)
10.58 Nighttime LST,
amplitude 1
9.39 August
precipitation
7.55
July
precipitation
27.66 Population density 6.94 Nights in 2007
with LST >5 °C
8.67 Probability of
occurrence
4.78 September
precipitation
4.71
September
precipitation
18.61 Distance to water 4.14 No. of days in
2008 with
temp. >0 °C
6.29 Nighttime LST,
phase 1
4.71 Minimum
precipitation
3.90
August
precipitation
15.17 Total
precipitation
4.04 No. of nights in
2006 with
LST >5 °C
4.23 No. of days in
2007 with
temp. > 0°C
3.36 Daytime LST,
amplitude 0
3.86
June
precipitation
9.88 December
precipitation
3.19 No. of nights in
2008 with
LST >5 °C
3.90 Nighttime LST,
phase 3
3.26 July
precipitation
3.51
Daytime LST,
phase 2
5.85 NDVI
phase 3
2.78 Nighttime LST,
amplitude 2
3.07 June
precipitation
3.20 NDVI,
phase 2
3.47
Daytime LST,
phase 1
3.63 Nighttime LST,
phase 1
2.78 Distance to water 2.20 Total
precipitation
2.98 Nighttime LST,
amplitude 1
3.42
May
precipitation
2.87 Nighttime LST,
amplitude 3
2.45 No. of days in
2006 with
temp. >0 °C
2.04 EVI,
amplitude 0
2.75 Total precipitation 3.32
Nighttime LST,
phase 3
2.82 Nighttime LST,
amplitude 3
2.26 EVI,
amplitude 2
2.00 No. of days in
2007 with
LST >12 °C
2.29 EVI,
phase 3
3.29
Table 7. The 10 most important predictors indicated by the variable importance (VARIMP) for the probability of occurrence of the
C. imicola, C. newsteadii, C. pulicaris group, C. punctatus and C. obsoletus groups.
DTM: digital terrain model; EVI: enhanced vegetation index; LST: land surface temperature; NDVI: normalized difference vegeta-
tion index.
apparent density is dependent on a wide range of fac-
tors such as trap type, location type, presence of hosts
nearby and the weather. The data did not allow cor-
recting for these potential biases.
Comparing abundance data
It is possible to directly use abundance data as input
to the model because the abundance was measured
using a standardised protocol (trap type, trap fre-
quency, morphological identification). A wide range
of trapping techniques is in use for the collection of
adult mosquitoes (Kline, 2006). They differ in design
and vary greatly in effectiveness and usefulness
(Campbell, 2003). Comparing the performance of
three trap mechanisms (using the “CO
2
-baited mos-
quito magnet liberty plus trap”, the “BG sentinel
trap” and the “gravid trap”), Versteirt (2012) noted
that the species caught and the abundances per
species differed greatly between the different traps
and thus that abundance values are not comparable.
This leaves obviously the issue that the proposed
methodology cannot be directly applied to continen-
tal-wide mapping if the observed data are not uni-
formly collected and illustrates that it is essential to
design a sampling strategy tailored to the study objec-
tives prior to the field work. When working with his-
torical datasets from different sources and collected in
different ways, methods will have to be designed to
compensate for discrepancies.
251
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
252
Predictors for Culicoides probability of occurrence
and abundance
In this study, Culicoides species with a clear distinc-
tion between areas of presence and absence scored
considerably better in models predicting the probabil-
ity of occurrence than species with a wider distribu-
tion range. This suggests that at a pixel resolution of 5
x 5 km for the latter (i.e. C. pulicaris and the C. obso-
letus group) may not contain enough information to
discriminate between unsuitable regions in Spain.
Additional predictor data will therefore be needed to
improve the discriminating capacity of the models.
This is also reflected when analysing the correlation
between the predicted probability of occurrence and
the observed abundance. While the RF approach does
not allow investigating the direction of influence, it
does permit assessing the importance of each variable.
Future research with techniques such as boosted
regression trees (Elith et al., 2009) may overcome this
disadvantage.
Precipitation, especially summer rainfall (June-
September), was the most influential factor determin-
ing the distribution of C. imicola. Field observations
indicate that the population of C. imicola peaks in the
September-October period and it seems that the sum-
mer rainfall has a direct impact on the population.
This is concurrent with the observations from Calvete
et al. (2008, 2009), who noted that the coefficient of
variation and the total amount of precipitation was
retained in all logistic regression models for C. imico-
la. In contrast, models produced by Wittmann et al.
(2001) did not include any variable related to rainfall.
Our research shows that temperature and tempera-
ture variation, expressed either through Fourier-
transformed variables or through a coefficient of vari-
ation, are also important. Work by Wittmann et al.
(2001), Purse et al. (2004) and Calvete et al. (2008,
2009) confirms this relationship. In the case of the
C. obsoletus group, the main factors were related to
temperature (seven out of the 10 most important fac-
tors) and vegetation. The population of this species
group peaks in summer time. Most of the temperature
variables are related to days colder than a 0 °C (which
has to do with overwintering), 5 °C or 12 °C (tem-
perature when insect activity starts) respectively, but
the first phase of land surface daytime temperature is
also included in the model. This is in agreement with
the findings of Calvete et al. (2008), which indicate
that the mean temperature and the coefficient of vari-
ation were the significant factors (in addition to the
mean NDVI).
Models for C. newsteadii, C. pulicaris and C. punc-
tatus are sparse and little information on the driving
factors for these species is known. From these results
it is concluded that the developed methodology
enables to produce sufficiently accurate abundance
models for Culicoides species in Spain. Whilst it is
clear that the approach may still be improved, it still
provides a good basis for further work.
Using probability of occurrence as a predictor for
abundance
In all the abundance models, the probability of
occurrence is the most important factor. This factor
was added as a predictor given that many authors
indicated that there is a (non)-linear relationship
between probability of occurrence and abundance. In
our results, the linear relationship proved to be too
weak to create abundance maps directly from the
probability of occurrence maps and therefore the
probability of presence was added into the set of pre-
dictors for the abundance model using random regres-
sion forest. While this would introduce correlation in
traditional statistical modelling techniques, and thus
may cause considerable variable inflation, this is
allowed when using the RF technique, which is not
affected by the predictor variables correlation. For all
species, the RF model delineates the zones of proba-
bility of presence (the necessary condition), and with-
in that zone the level of abundance is determined by
the remaining predictor variables.
Conclusion
The methodology for the creation of abundance
models, and its application for multiple Culicoides
species in Spain and Portugal using RF, is shown. The
results indicate that this modelling approach is robust
and that the predictor variables that are retained by
the model are concurrent with existing studies for the
mapping of probability of occurrence. The develop-
ment of abundance models using a continuous output
has not been attempted before and it is shown here
that using a combination of probability of occurrence
maps and a set of dedicated predictor variables, an
accurate output can be obtained and used as input to
TIR or R
0
models.
Acknowledgements
This work was sponsored by the EU network of Excellence,
EPIZONE (contract nr FOOD-CT-2006 016236) and the out-
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
253
come of two internal call projects (IC 6.6 BT EPIDEMIOLGY
and IC 6.7 BT-DYNVECT). Additional funding was provided
by Central Veterinary Institute (The Netherlands) under con-
tract BO-08-010-021.
References
Allepuz A, Garcia-Bocanegra I, Napp S, Casal J, Arenas A, Saez
M, Gonzalez MA, 2010. Monitoring bluetongue disease (BTV-
1) epidemic in southern Spain during 2007. Prev Vet Med 96,
263-271.
Balenghien T, Bodker R, Kiel E, De Deken R, Chirico J, Lucientes
J, Carpenter S, Elbers ARW, Calistri P, Miranda M, Staubach
C, Van der Stede Y, Guis H, 2010. Dynvect’s overview of the
Culicoides surveillance systems in the EU and distribution
maps of key species. In: 4
th
Annual Meeting EPIZONE, Saint-
Malo, France, 7-10 June 2010, Saint-Malo, France.
Boyce MS, Vernier PR, Nielsen SE, Schmiegelow FKA, 2002.
Evaluating resource selection functions. Ecol Model 157, 281-
300.
Breiman L, 2001. Random Forests. Mach Learn 45, 5-32.
Calvete C, Estrada R, Miranda MA, Borrás D, Calvo JH,
Lucientes J, 2008. Modelling the distributions and spatial
coincidence of bluetongue vectors Culicoides imicola and the
Culicoides obsoletus group throughout the Iberian Peninsula.
Med Vet Entomol 22, 124-134.
Calvete C, Estrada R, Miranda MA, Borrás D, Calvo JH,
Lucientes J, 2009. Ecological correlates of bluetongue virus in
Spain: predicted spatial occurrence and its relationship with
the observed abundance of potential Culicoides spp. vector.
Vet J 182, 235-243.
Campbell CB, 2003. Evaluation of five mosquito traps and a
horse for West Nile vectors on a north Florida equine facility
(MS Thesis). University of Florida, Gainesville.
Chefaoui RM, Horal J, Lobo JM, 2005. Potential distribution
modelling, niche characterization and conservation status
assessment using GIS tools: a case study of Iberian Copris
species. Biol Conserv 122, 327-338.
Conte A, Gilbert M, Goffredo M, 2009. Eight years of
Culicoides imicola survey in Italy: can we detect range expan-
sion? J Appl Ecol 46, 1332-1339.
Conte A, Goffredo M, Ippoliti C, Meiswinkel R, 2007.
Influence of biotic and abiotic factors on the distribution and
abundance of Culicoides imicola and the Obsoletus complex
in Italy. Vet Parasitol 150, 333-344.
Cutler R, Edwards TC, Bear KH, Cutler A, Hess KT, Gibson J,
Lawler JJ, 2007. Random Forests for classification in ecology.
Ecology 88, 2783-2792.
ECDC, 2009. Development of Aedes albopictus risk maps.
ECDC, Stockholm, 52 p.
Elbers ARW, Backx A, Mintiens K, Gerbier G, Staubach C,
Hendrickx G, van der Spek A, 2008. Field observations during
the bluetongue serotype 8 epidemic in 2006 II. Morbidity and
mortality rate, case fatality and clinical recovery in sheep and
cattle in the Netherlands. Prev Vet Med 87, 31-40.
Elith J, Leathwick JR, Hastie T, 2009. A working guide to
boosted regression trees. J Anim Ecol 77, 802-813.
ESRI, 2009. ArcGIS 9.3 software.
Fielding AH, Bell JF, 1997. A review of methods for the assess-
ment of prediction errors in conservation presence/absence
models. Environ Conserv 24, 38-49.
Gao X, Huete AR, Ni W, Miura T, 2000. Optical-biophysical
relatinships of vegetation spectra without background con-
tamination. Remote Sens Environ 74, 609-620.
GfK, 2009. Available at: http://www.gfk-geomarketing.de/
(accessed on February 2011).
Gibson LA, Wilson BA, Cahill DM, Hill J, 2004. Spatial pre-
diction of rufous bristlebird habitat in a coastal heathland: a
GIS-based approach. J Appl Ecol 41, 213-223.
Guis H, Caminade C, Calvete C, Morse AP, Tran A, Baylis M,
2011. Modelling the effects of past and future climate on the
risk of bluetongue emergence in Europe. J R Soc Interface 9,
339-350.
Hartemink NA, Purse BV, Meiswinkel R, Brown HE, de Koeijer
A, Elbers ARW, Boender GJ, Rogers DJ, Heesterbeek JAP,
2009. Mapping the basic reproduction number (R
0
) for vector-
borne diseases: a case study on bluetongue virus. Epidemics 1,
153-161.
Hartemink NA, Vanwambeke SO, Heesterbeek H, Rogers D,
Morley D, Pesson B, Davies C, Mahamdallie S, Ready P, 2011.
Integrated mapping of establishment risk for emerging vector-
borne infections: a case study of canine leishmaniasis in south-
west France. PLoS One 6, e20817.
Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A, 2005.
WORLDCLIM: Very high resolution interpolated climate sur-
faces for global land areas. Int J Climatol 25, 1965-1978.
JRC, 2005. IMAGE2000 and CLC2000, products and
Methods. Available at: http://ies.jrc.cec.eu.int/ (accessed on
February 2011).
JRC, 2009. European soil database. Available at:
http://eusoils.jrc.ec.europa.eu/ESDB_Archive/ESDB/index.htm
(accessed on February 2011).
Kline DI, Patnaude M, Barnard DR, 2006. Efficacy of four trap
types for detecting and monitoring Culex spp. in North
Central Florida. J Med Entomol 43, 1121-1128.
Le Gal MC, Dufour B, Geoffroy E, Zanella G, Rieffel JN,
Pouilly F, Moutou F, 2008. La fièvre catarrhale ovine due au
sérotype 8 dans les Ardennes françaises en 2007: taux de mor-
bidité, mortalité, létalité et signes cliniques observeés chez les
bovines et les ovins. Ann Med Vet 152, 240-245.
Mardulyn P, Goffredo M, Conte A, Hendrickx G, Meiswinkel
R, Balenghien T, Sghaier S, Lohr Y, Gilbert M, 2013. Climate
change and the spread of vector-borne diseases: using approx-
imate Bayesian computation to compare invasion scenarios for
E. Ducheyne et al. - Geospatial Health 8(1), 2013, pp. 241-254
254
the bluetongue virus vector Culicoides imicola in Italy. Mol
Ecol in press.
McPherson JM, Jetz W, Rogers DJ, 2004. The effect of species’
range sizes on the accuracy of distribution models: ecological
phenomenon or statistical artifact? J Appl Ecol 41, 811-823.
OIE, 2007a. Bluetongue Spain (immediate notification)
26/07/2007. Paris: World Organisation for Animal Health.
OIE, 2007b. Bluetongue Portugal (immediate notification)
21/09/2007. Paris: World Organisation for Animal Health.
OIE, 2007C. Bluetongue France (immediate notification)
23/11/2007. Paris: World Organisation for Animal Health.
Osborne PE, Alonso JC, Bryant RG, 2001. Modelling land-
scape-scale habitat using GIS and remote sensing: a case study
with great bustards. J Appl Ecol 38, 458-471.
Peters J, De Baets B, Van doninck J, Calvete C, Lucientes J, De
Clercq E, Ducheyne E, Verhoest NEC, 2011. Absence reduc-
tion in entolomological studies surveillance data to improve
niche-based distribution models for Culicoides imicola. Prev
Vet Med 100, 15-28.
Peters J, De Baets B, Verhoest N, Samson R, Degroeve S, De
Becker P, Huybrechts W, 2007. Random Forests as a tool for
ecohydrological distribution modelling. Ecol Model 207, 304-
318.
Purse BV, Brown HE, Harrup L, Mertens PPC, Rogers DJ,
2008. Invasion of bluetongue and other orbivirus infections
into Europe: the role of biological and climatic processes. Rev
Sci Tech Oie 27, 427-442.
Purse BV, Tatem AJ, Caracappa S, Rogers DJ, Mellor PS, Baylis
M, Torina A, 2004. Modelling the distributions of Culicoides
bluetongue virus vectors in Sicily in relation to satellite-derived
variables. Med Vet Entomol 18, 90-101.
R Development Core Team, 2006. R stats v2.10.1.
Ruiz MO, Chaves LF, Hamer GL, Sun T, Brown WM, Walker
ED, Haramis L, Goldberg TL, Kitron U, 2010. Local impact
of temperature and precipitation on West Nile virus infection
in Culex species mosquitoes in northeast Illinois, USA. Parasit
Vectors 3.
Saegerman C, Berkvens D, Mellor PS, 2008. Bluetongue epi-
demiology in the European Union. Emerg Infect Dis 14, 539-
544.
Scharlemann JPW, Benz D, Hay S, Purse BV, Tatem A, Wint W,
Rogers DJ, 2008. A novel algorithm for temporal Fourier pro-
cessing MODIS data for ecological and epidemiological appli-
cations. PLoS One 3, e1408.
Szmaragd C, Wilson AJ, Carpenter S, Wood JLN, Mellor PS,
2009. A modeling framework to describe the transmission of
bluetongue virus within and between farms in Great Britain.
PLoS One 4, e7741.
Szmaragd C, Wilson AJ, Carpenter S, Wood JLN, Mellor PS,
2010. The spread of bluetongue virus serotype 8 in Great
Britain and Its Control by Vaccination. PLoS One 5, e9353.
Tatem AJ, Baylis M, Mellor PS, Purse BV, Capela R, Pena I,
Rogers DJ, 2003. Prediction of bluetongue vector distribution
in Europe and North Africa using satellite imagery. Vet
Microbiol 97, 13-29.
USGS, 1996. Available at: GTOP030,http://eros.usgs.gov/#/
Find_Data/Products_and_Data_Available/gtopo30_info
(accessed on February 2011).
Velthuis AGJ, Saatkamp HW, Mourits MCM, de Koeijer AA,
Elbers ARW, 2010. Financial consequences of the Dutch blue-
tongue serotype 8 epidemics of 2006 and 2007. Prev Vet Med
93, 294-304.
Versteirt V, 2012. Taxonomic and functional biodiversity of
indigenous and exotic mosquito species (Culicidae) in Belgium
(PhD Thesis). University of Antwerp, Belgium.
Wint W, 2005. Global World Population v3.0 compiled for
EDEN DMT. Available at: http://ergodd.zoo.ox.ac.uk/eden/
(accessed on February 2011).
Wittmann EJ, Mellor PS, Baylis M, 2001. Using climate data
to map the potential distribution of Culicoides imicola
(Diptera: Ceratopogonidae) in Europe. Rev Sci Tech Oie 20,
731-740.
... Since BTV is an arthropod-borne virus (arbovirus), the distribution of BTV is directly related to the distribution of its vector [10]. Culicoides species distribution is related to different climatological, topographic and anthropogenic factors [22][23][24][25][26]. BTV transmission is highly influenced by temperature. ...
... A review of scientific literature was conducted to choose significant variables to generate predictive models for the Spanish principal vectors [22][23][24][25][42][43][44][45][46][47]. The selected variables can be classified into six different categories: climatic, vegetation indices, host, orography, land cover and soil composition. ...
... We first generated monthly probability of presence maps, henceforth occurrence maps, for C. imicola and the Obsoletus complex species, to include those as a variable for abundance maps [24,26]. For occurrence maps, Culicoides spp. ...
Article
Full-text available
Bluetongue virus (BTV) causes a disease that is endemic in Spain and its two major biological vector species, C. imicola and the Obsoletus complex species, differ greatly in their ecology and distribution. Understanding the seasonality of BTV transmission in risk areas is key to improving surveillance and control programs, as well as to better understand the pathogen transmission networks between wildlife and livestock. Here, monthly risk transmission maps were generated using risk categories based on well-known BTV R0 equations and predicted abundances of the two most relevant vectors in Spain. Previously, Culicoides spp. predicted abundances in mainland Spain and the Balearic Islands were obtained using remote sensing data and random forest machine learning algorithm. Risk transmission maps were externally assessed with the estimated date of infection of BTV-1 and BTV-4 historical outbreaks. Our results highlight the differences in risk transmission during April-October, June-August being the period with higher R0 values. Likewise, a natural barrier has been identified between northern and central-southern areas at risk that may hamper BTV spread between them. Our results can be relevant to implement risk-based interventions for the prevention, control and surveillance of BTV and other diseases shared between livestock and wildlife host populations.
... Other species, namely Culicoides newsteadi Austen and Culicoides pulicaris (L.), have also been incriminated as vectors following the isolation of Bluetongue virus from field specimens (Vanbinst et al. 2009, Goffredo et al. 2015, Foxi et al. 2016. Culicoides pulicaris species is less prevalent and is not closely linked to livestock than the others (Harrup et al. 2013(Harrup et al. , 2016; however, this species shares a similar overall distribution to the Obsoletus complex in Spain , Ducheyne et al. 2013, Talavera et al. 2015. Conversely, C. newsteadi has a high population density in Spain and is distributed predominantly in the south (Ortega et al. 1999, Del Río et al. 2013, Ducheyne et al. 2013. ...
... Culicoides pulicaris species is less prevalent and is not closely linked to livestock than the others (Harrup et al. 2013(Harrup et al. , 2016; however, this species shares a similar overall distribution to the Obsoletus complex in Spain , Ducheyne et al. 2013, Talavera et al. 2015. Conversely, C. newsteadi has a high population density in Spain and is distributed predominantly in the south (Ortega et al. 1999, Del Río et al. 2013, Ducheyne et al. 2013. ...
... As well as C. imicola, the species C. newsteadi has also been found to prefer lower altitudes (Torina et al. 2004), and sites with warmer average temperatures (Ortega et al. 1999, Purse et al. 2004. Culicoides pulicaris has shown a preference for the soil of open pastures (Harrup et al. 2013, Zimmer et al. 2014b, and this species appears in the same sites as Obsoletus complex (Ortega et al. 1998, Ducheyne et al. 2013, and also in coastal zones (Ramilo et al. 2017). Obsoletus complex and C. pulicaris have been found to overwinter in stables during winter (Kameke et al. 2017), therefore, C. pulicaris is expected to be related to the same variables as the Obsoletus complex, such as lower temperatures and areas with livestock (Purse et al. 2004, Sanders et al. 2011, Searle et al. 2012. ...
Article
Full-text available
Bluetongue is a viral disease affecting wild and domestic ruminants transmitted by several species of biting midges Culicoides Latreille. The phenology of these insects were analyzed in relation to potential environmental drivers. Data from 329 sites in Spain were analyzed using Bayesian Generalized Linear Mixed Model (GLMM) approaches. The effects of environmental factors on adult female seasonality were contrasted. Obsoletus complex species (Diptera: Ceratopogonidae) were the most prevalent across sites, followed by Culicoides newsteadi Austen (Diptera: Ceratopogonidae). Activity of female Obsoletus complex species was longest in sites at low elevation, with warmer spring average temperatures and precipitation, as well as in sites with high abundance of cattle. The length of the Culicoides imicola Kieffer (Diptera: Ceratopogonidae) female adult season was also longest in sites at low elevation with higher coverage of broad-leaved vegetation. Long adult seasons of C. newsteadi were found in sites with warmer autumns and higher precipitation, high abundance of sheep. Culicoides pulicaris (Linnaeus) (Diptera: Ceratopogonidae) had longer adult periods in sites with a greater number of accumulated degree days over 10°C during winter. These results demonstrate the eco-climatic and seasonal differences among these four taxa in Spain, which may contribute to determining sites with suitable environmental circumstances for each particular species to inform assessments of the risk of Bluetongue virus outbreaks in this region.
... Using entomological data collected on farms and environmental variables obtained from satellite imagery, it is possible to model and map the abundance of vectors. Culicoides abundance maps for Europe can be found either at a national [15][16][17] or a continental scale for C. imicola [18,19] and for the Obsoletus ensemble [20]. The Culicoides maps available at a continental scale for Europe are usually created with abundance data collected within a limited area of the mapped region. ...
... We hypothesised that Culicoides abundance may be predicted for a large area of Europe using a RF approach and climatic and environmental predictors. These predictors have proven effective in previous Culicoides studies [15,23,25,26]. The entomological dataset covers nine countries and represents the largest entomological dataset aggregated to date comprising 595 sampled livestock farms with 30,626 trap collections and 8,539,420 recorded specimens. ...
... The models were not able to predict the highest range of observed abundance, making relatively similar predictions throughout the range of the observed abundance. Nevertheless, our resulting maps displayed a regional C. imicola abundance similar to previous studies that modelled C. imicola abundance in Spain [15]. Our models were able to recognise environmental factors on a regional scale, which allowed us to estimate the abundance distribution of C. imicola quite accurately, as our maps are comparable to those presented in other studies in Spain [14,24,50]. ...
Article
Full-text available
Background: Culicoides biting midges transmit viruses resulting in disease in ruminants and equids such as bluetongue, Schmallenberg disease and African horse sickness. In the past decades, these diseases have led to important economic losses for farmers in Europe. Vector abundance is a key factor in determining the risk of vector-borne disease spread and it is, therefore, important to predict the abundance of Culicoides species involved in the transmission of these pathogens. The objectives of this study were to model and map the monthly abundances of Culicoides in Europe. Methods: We obtained entomological data from 904 farms in nine European countries (Spain, France, Germany, Switzerland, Austria, Poland, Denmark, Sweden and Norway) from 2007 to 2013. Using environmental and climatic predictors from satellite imagery and the machine learning technique Random Forests, we predicted the monthly average abundance at a 1 km2 resolution. We used independent test sets for validation and to assess model performance. Results: The predictive power of the resulting models varied according to month and the Culicoides species/ensembles predicted. Model performance was lower for winter months. Performance was higher for the Obsoletus ensemble, followed by the Pulicaris ensemble, while the model for Culicoides imicola showed a poor performance. Distribution and abundance patterns corresponded well with the known distributions in Europe. The Random Forests model approach was able to distinguish differences in abundance between countries but was not able to predict vector abundance at individual farm level. Conclusions: The models and maps presented here represent an initial attempt to capture large scale geographical and temporal variations in Culicoides abundance. The models are a first step towards producing abundance inputs for R0 modelling of Culicoides-borne infections at a continental scale.
... Culicoides-borne viruses can be introduced through vectors from Africa (Allepuz et al., 2010;Fernández-Carrion et al., 2018). Annual rainfall ranges from 170 to 1000 mm (516 mm on average), whilst the average temperature is around 17ºC ( (Calvete et al., 2006;Ducheyne et al., 2013;Pérez et al., 2012;Talavera et al., 2015). These studies have highlighted that C. imicola, which is the main competent vector for both BTV and SBV in the Mediterranean Basin (Nolan et al., 2008;Versteirt et al., 2017;Cuéllar et al., 2018;Pages et al., 2018), is the most widely distributed species in southwestern Spain. ...
Thesis
The transmission of pathogens between domestic and wild hosts greatly impacts animal and public health, biodiversity conservation and socio-economic contexts. The present PhD thesis deals with the study of the epidemiology of shared pathogens relevant for veterinary medicine in the Iberian Peninsula, with a special focus on the wildlife-livestock interface. The related epidemiological background, as well as the current research perspectives and knowledge blanks, are reviewed throughout the introduction section. In the first chapter of this thesis (Chapter 1), an innovative blood extraction method is proposed as an alternative to conventional sampling techniques in wild ruminants, representing a relevant step forward to better perform high quality sampling for disease surveillance and epidemiological wildlife studies. Chapter 2 describes the first Schmallenberg disease outbreak in Spain and assesses the local spread of the causative virus and associated risk factors in livestock. In Chapter 3, nation-wide studies evaluate the role played by wild ruminant species in the maintenance and transmission of this emerging pathogen (Chapter 3.1), as well as of pestiviruses (Chapter 3.2), a group of viruses endemic to livestock in Iberia. Finally, in Chapter 4, a finer approach to the wildlife-livestock interface is developed in Doñana National Park to provide in-depth information on the interspecies transmission of pathogens, which can follow different routes and pathways: pathogens directly transmitted through close or non-close interactions (Chapter 4.1) and pathogens indirectly transmitted through vectors (Chapter 4.2). Diverse serological methods, alone or combined with anatomical, pathological, and molecular tools, were used in each section. The blood sampling method developed in this PhD thesis could be systematically used in wild ruminant species for wildlife disease surveillance at international level favouring more accurate data comparisons. Likewise, epidemiological findings provide additional information on the spatio-temporal dynamics of both emerging and endemic shared pathogens and the epidemiological role played by wild ruminant species in mainland Spain. Overall, our findings revealed the usefulness of shared disease monitoring to better drive and prioritise control strategies in specific wildlife-livestock interfaces.
... Precipitation may influence distribution through an impact on the availability of breeding sites. Ducheyne et al. (2013) demonstrated that precipitation, especially summer rain-fall (June-September) was the most influential factor determining C. imicola distribution in Spain. This is in agreement with the observations made by Calvete et al. (2008Calvete et al. ( , 2009) who noted that the coefficient of variation and the total amount of precipitation significantly influenced the presence of C. imicola (Calvete et al. 2008(Calvete et al. , 2009. ...
Article
The increasing threat of vector-borne diseases (VBDs) represents a great challenge to those who manage public and animal health. Determining the spatial distribution of arthropod vector species is an essential step in studying the risk of transmission of a vector-borne pathogen (VBP) and in estimating risk levels of VBD. Risk maps allow better targeting surveillance and help in designing control measures. We aimed to study the geographical distribution of Culicoides imicola, the main competent vector of Bluetongue virus (BTV) in sheep in Tunisia. Fifty-three records covering the whole distribution range of C.imicola in Tunisia were obtained during a 2-year field entomological survey (August 2017 – January 2018 and August 2018 – January 2019). The ecological niche of C. imicola is described using ecological-niche factor analysis (ENFA) and Mahalanobis distances factor analysis (MADIFA). An environmental suitability map (ESM) was developed by MaxEnt software to map the optimal habitat under the current climate background. The MaxEnt model was highly accurate with a statistically significant area under curve (AUC) value of 0.941. The location of the potential distribution of C. imicola is predicted in specified regions of Tunisia. Our findings can be applied in various ways such as surveillance and control program of BTV in Tunisia.
... Precipitation may influence distribution through an impact on the availability of breeding sites. Ducheyne et al. (2013) demonstrated that precipitation, especially summer rain-fall (June-September) was the most influential factor determining C. imicola distribution in Spain. This is in agreement with the observations made by Calvete et al. (2008Calvete et al. ( , 2009) who noted that the coefficient of variation and the total amount of precipitation significantly influenced the presence of C. imicola (Calvete et al. 2008(Calvete et al. , 2009. ...
... Precipitation may influence distribution through an impact on the availability of breeding sites. Ducheyne et al. (2013) demonstrated that precipitation, especially summer rain-fall (June-September) was the most influential factor determining C. imicola distribution in Spain. This is in agreement with the observations made by Calvete et al. (2008Calvete et al. ( , 2009) who noted that the coefficient of variation and the total amount of precipitation significantly influenced the presence of C. imicola (Calvete et al. 2008(Calvete et al. , 2009. ...
Article
Full-text available
The increasing threat of vector-borne diseases (VBDs) represents a great challenge to those who manage public and animal health. Determining the spatial distribution of arthropod vector species is an essential step in studying the risk of transmission of a vector-borne pathogen (VBP) and in estimating risk levels of VBD. Risk maps allow better targeting surveillance and help in designing control measures. We aimed to study the geographical distribution of Culicoides imicola, the main competent vector of Bluetongue virus (BTV) in sheep in Tunisia. Fifty-three records covering the whole distribution range of C.imicola in Tunisia were obtained during a 2-year field entomological survey (August 2017 – January 2018 and August 2018 – January 2019). The ecological niche of C. imicola is described using ecological-niche factor analysis (ENFA) and Mahalanobis distances factor analysis (MADIFA). An environmental suitability map (ESM) was developed by MaxEnt software to map the optimal habitat under the current climate background. The MaxEnt model was highly accurate with a statistically significant area under curve (AUC) value of 0.941. The location of the potential distribution of C. imicola is predicted in specified regions of Tunisia. Our findings can be applied in various ways such as surveillance and control program of BTV in Tunisia.
Article
A cross‐sectional study was carried out to assess the circulation of bluetongue virus (BTV) and Schmallenberg virus (SBV) within the wild and domestic ungulate host community in Doñana National Park (southwestern Spain). A total of 440 animals, including 138 cattle, 102 red deer (Cervus elaphus), 101 fallow deer (Dama dama) and 99 wild boar (Sus scrofa) were sampled in 2015 during the seasonal peak of Culicoides spp. (summer‐autumn). Serum and spleen samples were analysed to detect exposure (using commercial blocking ELISAs) and infection (by RT‐PCR), respectively, to BTV and SBV. Cattle were not tested by BTV‐bELISA because all were previously vaccinated against BTV serotypes 1 and 4. High BTV seroprevalences were found in red deer (97.0%) and fallow deer (64.7%). Antibodies against SBV were detected in 37.0% of cattle, 16.8% of red deer, 23.5% of fallow deer and 2.0% of wild boar. Thirty‐eight of the 203 deer (18.7%; 17 red deer and 21 fallow deer) were co‐exposed to both viral agents. BTV‐4 RNA was confirmed in four red deer and two fallow deer. SBV RNA was found in two fallow deer. Co‐infections were not detected. Our results indicate high exposure, widespread distribution, and active circulation of BTV and SBV in the ruminant community in the study area. We provide additional evidence for the potential role of wild cervids as reservoirs of these Culicoides‐borne viruses in two different epidemiological scenarios: with vaccination (BTV) and without vaccination (SBV) of sympatric livestock. This study highlights the importance of wildlife surveillance, particularly of cervid species, for the proper execution of control programmes of Culicoides‐borne diseases in extensively reared livestock. This article is protected by copyright. All rights reserved
Article
Full-text available
Bluetongue virus (BTV) serotype 8 has been circulating in Europe since a major outbreak occurred in 2006, causing economic losses to livestock farms. The unpredictability of the biting activity of midges that transmit BTV implies difficulty in computing accurate transmission models. This study uniquely integrates field collections of midges at a range of European latitudes (in Sweden, The Netherlands, and Italy), with a multi-scale modelling approach. We inferred the environmental factors that influence the dynamics of midge catching, and then directly linked predicted midge catches to BTV transmission dynamics. Catch predictions were linked to the observed prevalence amongst sentinel cattle during the 2007 BTV outbreak in The Netherlands using a dynamic transmission model. We were able to directly infer a scaling parameter between daily midge catch predictions and the true biting rate per cow per day. Compared to biting rate per cow per day the scaling parameter was around 50% of 24 h midge catches with traps. Extending the estimated biting rate across Europe, for different seasons and years, indicated that whilst intensity of transmission is expected to vary widely from herd to herd, around 95% of naïve herds in western Europe have been at risk of sustained transmission over the last 15 years.
Article
Full-text available
Bluetongue virus (BTV) is an arbovirus of ruminants that has been circulating in Europe continuously for more than two decades and has become endemic in some countries such as Spain. Spain is ideal for BTV epidemiological studies since BTV outbreaks from different sources and serotypes have occurred continuously there since 2000; BTV-1 has been reported there from 2007 to 2017. Here we develop a model for BTV-1 endemic scenario to estimate the risk of an area becoming endemic, as well as to identify the most influential factors for BTV-1 persistence. We created abundance maps at 1-km² spatial resolution for the main vectors in Spain, Culicoides imicola and Obsoletus and Pulicaris complexes, by combining environmental satellite data with occurrence models and a random forest machine learning algorithm. The endemic model included vector abundance and host-related variables (farm density). The three most relevant variables in the endemic model were the abundance of C. imicola and Obsoletus complex and density of goat farms (AUC 0.86); this model suggests that BTV-1 is more likely to become endemic in central and southwestern regions of Spain. It only requires host- and vector-related variables to identify areas at greater risk of becoming endemic for bluetongue. Our results highlight the importance of suitable Culicoides spp. prediction maps for bluetongue epidemiological studies and decision-making about control and eradication measures.
Chapter
Full-text available
Bluetongue (BT) is a reportable disease of considerable socioeconomic concern and of major importance in the international trade of animals and animal products. Before 1998, BT was considered an exotic disease in Europe. From 1998 through 2005, at least 6 BT virus strains belonging to 5 serotypes (BTV-1, BTV-2, BTV-4, BTV-9, and BTV-16) were continuously present in the Mediterranean Basin. Since August 2006, BTV-8 has caused a severe epizootic of BT in northern Europe. The widespread recrudescence and extension of BTV-8 infections in northern Europe during 2007 suggest that requirements for BTV establishment may now be fulfilled in this area. In addition, the radial extension of BTV-8 across Europe increases the risk for an encounter between this serotype and others, particularly those that occur in the Mediterranean Basin, where vector activity continues for more of the year. This condition increases the risk for reassortment of individual BTV gene segments.
Technical Report
Full-text available
The aim of the work was to (1) produce a map that shows the precise current distribution of Aedes albopictus in Europe; and (2) map the risk for establishment of Aedes albopictus in Europe, in the event of its introduction. From this, it can be concluded that the temperate strains of Aedes albopictus are here to stay — and that they will spread. In addition, new populations may become established in other parts of Europe. Surveillance of the introduction and spread of this vector, in particular in areas at risk, is important in order to be prepared for the mosquito’s role in the transmission of diseases. In order to prepare the maps presented in this technical report, a huge amount of work was carried out by collecting data from many entomologists all over Europe, resulting in a database containing an enormous amount of valuable information. Regular revision of this database would provide the possibility to easily update the Aedes albopictus distribution maps, providing relevant information for public health specialists. The distribution maps show the areas for which no mosquito surveillance data are available, while the risk maps show which areas are suitable for the mosquito’s establishment in case it is introduced. This information should support entomologists and public health experts in their efforts to ensure preparedness for mosquito-borne diseases. Additionally needed information includes the surveillance of the introduction, possible establishment and spread of other potential disease vectors, such as Aedes aegypti. There are still many unknowns with regard to European Aedes albopictus strains for different pathogens, and further research into this topic is required. More research is also needed on the development of adapted control measures to eradicate newly established mosquito colonies and to prevent the further spread of the mosquito. Finally, in order to improve our understanding of mosquito introduction pathways and the evolution and adaptation of mosquito populations, further research is needed on the genetic homogeneity or heterogeneity of European mosquito populations.
Article
Full-text available
Summary 1. To develop a conservation management plan for a species, knowledge of its distribu- tion and spatial arrangement of preferred habitat is essential. This is a difficult task, especially when the species of concern is in low abundance. In south-western Victoria, Australia, populations of the rare rufous bristlebird Dasyornis broadbenti are threatened by fragmentation of suitable habitat. In order to improve the conservation status of this species, critical habitat requirements must be identified and a system of corridors must be established to link known populations. A predictive spatial model of rufous bristlebird habitat was developed in order to identify critical areas requiring preservation, such as corridors for dispersal. 2. Habitat models generated using generalized linear modelling techniques can assist in delineating the specific habitat requirements of a species. Coupled with geographic information system (GIS) technology, these models can be extrapolated to produce maps displaying the spatial configuration of suitable habitat. 3. Models were generated using logistic regression, with bristlebird presence or absence as the dependent variable and landscape variables, extracted from both GIS data layers and multispectral digital imagery, as the predictors. A multimodel inference approach based on Akaike's information criterion was used and the resulting model was applied in a GIS to extrapolate predicted likelihood of occurrence across the entire area of concern. The predictive performance of the selected model was evaluated using the receiver operating characteristic (ROC) technique. A hierarchical partitioning protocol was used to identify the predictor variables most likely to influence variation in the dependent variable. Probability of species presence was used as an index of habitat suitability. 4. Negative associations between rufous bristlebird presence and increasing elevation, 'distance to creek', 'distance to coast' and sun index were evident, suggesting a preference for areas relatively low in altitude, in close proximity to the coastal fringe and drainage lines, and receiving less direct sunlight. A positive association with increasing habitat complexity also suggested that this species prefers areas containing high vertical density of vegetation. 5. The predictive performance of the selected model was shown to be high (area under the curve 0·97), indicating a good fit of the model to the data. Hierarchical partitioning analysis showed that all the variables considered had significant independent contribu- tions towards explaining the variation in the dependent variable. The proportion of the total study area that was predicted as suitable habitat for the rufous bristlebird (using probability of occurrence at a ≥ 0·5 level) was 16%. 6. Synthesis and applications. The spatial model clearly delineated areas predicted as highly suitable rufous bristlebird habitat, with evidence of potential corridors linking coastal and inland populations via gullies. Conservation of this species will depend on management actions that protect the critical habitats identified in the model. A multi- scale approach to the modelling process is recommended whereby a spatially explicit model is first generated using landscape variables extracted from a GIS, and a second
Article
Full-text available
Cited By (since 1996): 2, Export Date: 26 April 2012, Source: Scopus
Article
Bluetongue virus serotype 8 is a midge-transmitted, non infectious disease which first appeared in Northern Europe in 2006. The animal health divisional office of the Ardennes (08) department in France designed a form which vets working in this department could use to report their observations. Using data obtained via this form, morbidity, mortality and case fatality rates were calculated and symptoms recorded. Morbidity rates were 2.6% for cattle and 3.8% for sheep. Mortality rates were 0.18% for cattle and 1.8% for sheep. Case fatality rates were 6.9% for cattle and 48.3% for sheep. Vets reported similar clinical signs in both cattle and sheep, although they did appear to vary in the latter according to the flock's official disease status.
Article
Random forests are a combination of tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the forest. The generalization error for forests converges a.s. to a limit as the number of trees in the forest becomes large. The generalization error of a forest of tree classifiers depends on the strength of the individual trees in the forest and the correlation between them. Using a random selection of features to split each node yields error rates that compare favorably to Adaboost (Y. Freund & R. Schapire, Machine Learning: Proceedings of the Thirteenth International conference, ***, 148–156), but are more robust with respect to noise. Internal estimates monitor error, strength, and correlation and these are used to show the response to increasing the number of features used in the splitting. Internal estimates are also used to measure variable importance. These ideas are also applicable to regression.
Article
Data collected in the Netherlands during the Bluetongue serotype 8 (BTV-8) epidemic indicated that in outbreak cattle herds, predominantly dairy and nursing cows were clinically affected and not young stock, beef cattle, beef calves, or breeding animals. In outbreak sheep flocks, mainly ewes and – if present – rams, were clinically affected and not the lambs. Median morbidity rate in outbreak herds was 1.85 per 100 sheep-month at risk and 0.32 per 100 cattle-month at risk for sheep and cattle, respectively. The mean proportion of BT-affected animals in outbreak herds that recovered from clinical disease was approximately eight times higher for cattle compared to sheep in the Netherlands. Median mortality rate in outbreak herds was 0.5 per 100 sheep-month at risk of dying and 0 per 100 cattle-month at risk of dying for sheep and cattle, respectively. Median recovery time of both sheep and cattle that recovered from clinical disease in outbreak herds was 14 days. Median case fatality was 50% in sheep outbreak flocks and 0% in outbreak cattle herds. It is concluded that morbidity and mortality in outbreak cattle herds was very limited during the BTV-8 epidemic in the Netherlands in 2006. In outbreak sheep flocks, morbidity was limited, with exceptions for a few flocks. However, almost 50% of the clinically sick sheep died in outbreak sheep herds.
Article
For a better evaluation of the accuracy of VIs in estimating biophysical parameters, a "true" VI value attributed only to the vegetation signal and free of any contamination is needed. In this article, pure vegetation spectra were extracted from a set of open and closed canopies by unmixing the green vegetation signal from the background component. Canopy model-simulation and reflectances derived from graph-based linear extrapolation were used to unmix and derive a "true" vegetation signal, equivalent to a perfect absorber (free boundary) canopy background reflectance condition. Optical-biophysical relationships were then derived for a variety of canopy structures with differences in foliage clumping, horizontal heterogeneity, and leaf type. A 3-dimensional canopy radiative transfer model and a hybrid geometric optical-radiative transfer model (GORT) were used to simulate the directional-hemispherical reflectances from agricultural, grassland, and forested canopies (cereal and broadleaf crop, grass, needleleaf, and broadleaf forest). The relationships of the extracted red and near-infrared reflectances and derived vegetation indices (VIs) to various biophysical parameters (leaf area index, fraction of absorbed photosynthetically active radiation, and percent ground cover) were examined for the pure vegetation spectra. The results showed normalized difference vegetation index (NDVI) relationships with biophysical parameters to become more asymptotic over the pure vegetation conditions. The extraction of pure vegetation signals had little effect on the soil-adjusted vegetation index (SAVI), which had values equivalent to those obtained with the presence of a background signal. NDVI values were fairly uniform across the different canopy types, whereas the SAVI values had pronounced differences among canopy types, particularly between the broadleaf and cereal/needleleaf structural types. These results were useful not only in selecting suitable vegetation indices to characterize specific canopy biophysical parameters, but also in understanding a "true" VI behavior, free of background noise. (C) 2000 Elsevier Science Inc.
Article
Bluetongue (BT) is a commonly cited example of a disease with a distribution believed to have recently expanded in response to global warming. The BT virus is transmitted to ruminants by biting midges of the genus Culicoides, and it has been hypothesized that the emergence of BT in Mediterranean Europe during the last two decades is a consequence of the recent colonization of the region by Culicoides imicola and linked to climate change. To better understand the mechanism responsible for the northward spread of BT, we tested the hypothesis of a recent colonization of Italy by C. imicola, by obtaining samples from more than 60 localities across Italy, Corsica, Southern France, and Northern Africa (the hypothesized source point for the recent invasion of C. imicola), and by genotyping them with 10 newly identified microsatellite loci. The patterns of genetic variation within and among the sampled populations were characterized and used in a rigorous Applied Bayesian Computation framework to compare three competing historical hypotheses related to the arrival and establishment of C. imicola in Italy. The hypothesis of an ancient presence of the insect vector was strongly favoured by this analysis, with an associated P ≥ 99%, suggesting that causes other than the northward range expansion of C. imicola may have supported the emergence of BT in southern Europe. Overall, this study illustrates the potential of molecular genetic markers for exploring the assumed link between climate change and the spread of diseases.
Article
1. Conservation scientists and resource managers increasingly employ empirical distribution models to aid decision-making. However, such models are not equally reliable for all species, and range size can affect their performance. We examined to what extent this effect reflects statistical artefacts arising from the influence of range size on the sample size and sampling prevalence (proportion of samples representing species presence) of data used to train and test models. 2. Our analyses used both simulated data and empirical distribution models for 32 bird species endemic to South Africa, Lesotho and Swaziland. Models were built with either logistic regression or non-linear discriminant analysis, and assessed with four measures of model accuracy: sensitivity, specificity, Cohen's kappa and the area under the curve (AUC) of receiver-operating characteristic (ROC) plots. Environmental indices derived from Fourier-processed satellite imagery served as predictors. 3. We first followed conventional modelling practice to illustrate how range size might influence model performance, when sampling prevalence reflects species' natural prevalences. We then demonstrated that this influence is primarily artefactual. Statistical artefacts can arise during model assessment, because Cohen's kappa responds systematically to changes in prevalence. AUC, in contrast, is largely unaffected, and thus a more reliable measure of model performance. Statistical artefacts also arise during model fitting. Both logistic regression and discriminant analysis are sensitive to the sample size and sampling prevalence of training data. Both perform best when sample size is large and prevalence intermediate. 4. Synthesis and applications. Species' ecological characteristics may influence the performance of distribution models. Statistical artefacts, however, can confound results in comparative studies seeking to identify these characteristics. To mitigate artefactual effects, we recommend careful reporting of sampling prevalence, AUC as the measure of accuracy, and fixed, intermediate levels of sampling prevalence in comparative studies.