Evaluation of the impacts of climate
change on disease vectors through
ecological niche modelling
B.M. Carvalho
1,2,3
*, E.F. Rangel
2
and M.M. Vale
1
1
Laboratório de Vertebrados, Instituto de Biologia, Universidade Federal do
Rio de Janeiro, Rio de Janeiro, Brazil:
2
Laboratório Interdisciplinar de
Vigilância Entomológica em Diptera e Hemiptera, Instituto Oswaldo Cruz,
Fundação Oswaldo Cruz, Rio de Janeiro, Brazil:
3
Pós-Graduação em Ecologia
e Evolução, Universidade do Estado do Rio de Janeiro, Rio de Janeiro, Brazil
Abstract
Vector-borne diseases are exceptionally sensitive to climate change. Predicting
vector occurrence in specific regions is a challenge that disease control programs
must meet in order to plan and execute control interventions and climate change
adaptation measures. Recently, an increasing number of scientific articles have ap-
plied ecological niche modelling (ENM) to study medically important insects and
ticks. With a myriad of available methods, it is challenging to interpret their results.
Here we review the future projections of disease vectors produced by ENM, and as-
sess their trends and limitations. Tropical regions are currently occupied by many
vector species; but future projections indicate poleward expansions of suitable cli-
mates for their occurrence and, therefore, entomological surveillance must be con-
tinuously done in areas projected to become suitable. The most commonly applied
methods were the maximum entropy algorithm, generalized linear models, the gen-
etic algorithm for rule set prediction, and discriminant analysis. Lack of consideration
of the full-known current distribution of the target species on models with future pro-
jections has led to questionable predictions. We conclude that there is no ideal ‘gold
standard’ method to model vector distributions; researchers are encouraged to test
different methods for the same data. Such practice is becoming common in the
field of ENM, but still lags behind in studies of disease vectors.
Keywords: vector-borne diseases, spatial distribution, ensemble modelling, Aedes,
Anopheles, Lutzomyia, Phlebotomus, Culicoides, Triatoma, Ixodes
(Accepted 14 November 2016; First published online 15 December 2016)
Introduction
Climate change is happeningmore quickly and strongly than
predicted, and the anthropic influence in this process is now
clear (IPCC, 2014). Projections from several greenhouse gas
emission scenarios agree on an increase of the mean earth sur-
face temperature by the end of the 21st century, with continents
heating more than oceans and high latitude regions heating
more than the tropics. Longer and more frequent heat waves
will probably occur, as well as more intense precipitation events
in several regions (IPCC, 2014). Increased floods, droughts, fires,
heat waves and air pollutants will directlyimpact human health.
Indirect impacts on human health will arise from ecological dis-
turbances and social responses to disruptions to agriculture, and
to water and food supplies. Vector-borne diseases will also in-
crease, compounded by human migrations towards endemic
areas (Woodward et al., 2014).
Vector-borne diseases are exceptionally sensitive to climate
change because they emerge from complex transmission
*Author for correspondence
Phone: +55 21 2562 1375
E-mail: brunomc@ioc.fiocruz.br
Bulletin of Entomological Research (2017) 107, 419–430 doi:10.1017/S0007485316001097
© Cambridge University Press 2016
https://doi.org/10.1017/S0007485316001097 Published online by Cambridge University Press
cycles involving several species of pathogens, vectors and
hosts (Parham et al., 2015). Most disease vectors are arthro-
pods, including insects and ticks. Climate change should,
therefore, cause changes in disease distribution, density, sea-
sonality and prevalence, and might prompt adaptation of vec-
tors and hosts to new transmission cycles (Kovats et al., 2001;
Brooks & Hoberg, 2007; Rosenthal, 2009; Mills et al., 2010).
The ecology of arthropod vectors should be impacted by
climate change at three levels of biological organization: (i)
at the individual level – being ectothermic organisms, vectors’
metabolism varies with daily fluctuations in temperature,
which may affect physiological traits related to vector compe-
tence (Paaijmans et al., 2013) such as muscle activity (Harrison
& Roberts, 2000) and biting rates, although this latter influence
is not entirely clear (Rogers & Randolph, 2006; Ready, 2013);
(ii) at the population level – changes in climate should influ-
ence abundance, density, seasonality, survival rates, gener-
ation time, fecundity and dispersion ability, allowing vectors
to colonize new habitats more efficiently (Mills et al., 2010;
Stange & Ayres, 2010; Eisen et al., 2014); (iii) at the community
level – parasite–vector interactions can be influenced by tem-
perature (Hlavacova et al., 2013), and new species of vectors or
hosts can adapt to existing transmission cycles (Kovats et al.,
2001; Rosenthal, 2009; Parham et al., 2015).
Knowledge of vectors’ spatial distributi ons is essential to
assess transmission risks in different regions. Predicting
vector occurrence in specific regions is a challenge that
many disease control programs must meet in orde r to
plan and execute control interventions and adaptation mea-
sures more efficiently. With the popularization of GIS (geo-
graphic information systems), increasing avail ability of
species occurrence data, disease infor mation and environ-
mental variables, various methods of spatial analysis and
mathematica l modelli ng have become common in the scien-
tific literature. The methods that correlate these available
data in order to predict species’ distributions are known
as ecological niche models (ENMs) or species distribution
models and h ave been wi dely used in studies of ecology,
biogeography and c onservation (Guisan & Zimmermann,
2000; Guisan & Thuiller, 2005; Elith & Leathwick, 2009
).
Recen tly, an increasing number of scientific artic les have ap-
plied these models to study distribu tions of many medically
important insect and tick species.
Ecological niche models are perhaps the most used meth-
ods to link climatic and environmental conditions to the distri-
bution of species. In an ENM, an algorithm takes as input
occurrence records of the studied species and calculates their
relation with environmental variables, producing a surface of
environmental suitability or probability of occurrence (Guisan
& Zimmermann, 2000; Franklin, 2010; Peterson et al., 2011).
There are two basic approaches to apply an ENM in studies
of vector-borne diseases. The first considers the entire trans-
mission cycle and their ecological relationships as a ‘black
box’, and analyses the geographical distribution of the disease
occurrence, as if it were a single species (e.g. Nieto et al., 2006;
Yé et al., 2007; Williams et al., 2008; Arboleda et al., 2009 ). This
approach indirectly groups all component species of the trans-
mission cycle, as well as their environmental needs and eco-
logical interactions, losing, therefore, important details of the
transmission process. However, the occurrence of the disease
is often the only information available, and this becomes the
only modelling option. The second approach is to model
each species from the transmission cycle individually, and
evaluate areas of co-occurrence afterwards. This approach
offers the opportunity to distinguish different reasons for the
presence or absence of disease transmission in certain loca-
tions. For example, the disease may be absent due to the lack
of the pathogen, an appropriate vector or a reservoir host
(Peterson et al., 2011). Areas with the presence of only vectors
and competent hosts may be treated as vulnerable – a particu-
larly important situation nowadays, when species are artifi-
cially transported by humans and new diseases emerge in
areas where they would not naturally occur (Komar, 2003;
Ready, 2008, 2010; Daszak et al., 2013).
Comparative studies show that most of uncertainty in
ENM comes from using different modelling algorithms
(Buisson et al., 2009; Diniz-Filho et al., 2009; Elith & Graham,
2009). With the wide variety of methods, it is an additional
challenge to interpret and compare the results of studies on
vector distributions, so that they can be effectively used in con-
trol programs. Here we review the future projections of dis-
ease vectors produced by ENMs, and assess trends and
limitations of the methods applied.
Methods
We performed a systematic review of the literature using
four online databases: (i) Web of Science (http://isiwebof-
knowledge.com); (ii) Scopus (http://www.hub.sciverse.
com); (iii) Pubmed (http://www.ncbi.nlm.nih.gov/pubmed);
and (iv) Scientific Electronic Library Online (SciELO) (http://
www.scielo.org). The Web of Science is the most comprehen-
sive database of peer-reviewed articles published in English,
as well as being the most used in systematic reviews
(Falagas et al., 2008; Gavel & Iselid, 2008). However, Scopus
covers a larger number of journals that publish articles in lan-
guages other than English (Falagas et al., 2008; Gavel & Iselid,
2008). PubMed is the most frequently consulted source for in-
formation in the biomedical field (Falagas et al., 2008). The
SciELO database, although less comprehensive, includes
many Latin American journals that are not included in the
other consulted databases.
Searches were conducted in March 2015, through different
combinations of the following key words: ‘ecologic* niche
model*’‘species distribution model*’, ‘climat* model* ’, ‘vec-
tor’, ‘disease’. The initial results (N = 572) were limited to arti-
cles published until 2014 that applied ENMs to predict areas of
occurrence or environmental suitability of arthropods vectors
of diseases. Articles that used models to explain the relation-
ship of the vectors with environmental variables, without pre-
dictive mapping, were excluded from the analysis. Studies
with models based only on the occurrence of disease or risk
maps generated without vector information were also dis-
carded. After removing duplicates and refining selections,
146 articles were reviewed (Table S1).
The articl es were d escribed under the following categor-
ies: vector species and main associated disease; study area;
types of biological data; types of environmental data; ap-
plied meth od; inclusion of future projections (Table S1).
Studies including future projections were analysed in great-
er detail in relation to biological dat a (number of reco rds,
data source), environmental data (number of variables,
approxim ate spatial resolution), methods (algorithm em-
ployed, use of ensemble models based on different algo-
rithms) and future projections (years, general circulation
model, climate change scenario) (Table S2). The main results
of th e future projections were summarized by vector group
and f urther described.
B.M. Carvalho et al.
420
https://doi.org/10.1017/S0007485316001097 Published online by Cambridge University Press
Results and discussion
Application of different modelling methods
Seventeen different modelling methods were used to pre-
dict vector distributions. The most common was the max-
imum entropy algorithm (MaxEnt, 43 articles) (Phillips et al.,
2006), followed by generalized linear models (GLM, 34)
(Guisan et al., 2002), the genetic algorithm for rule set predic-
tion (GARP, 25) (Stockwell, 1999), and discriminant analysis
(12) (Rogers et al., 1996)(fig. 1, Table S1). Other methods
were less frequently applied, such as CLIMEX (5) (Sutherst
& Maywald, 1985), ENFA (3) (Hirzel et al., 2002), BRT (2)
(Elith et al., 2008), BIOCLIM (1) (Booth et al., 2014) and
Random Forests (1) (Breiman, 2001)(fig. 1, Table S1). For com-
prehensive descriptions of ENM algorithms, see Franklin
(2010) and Peterson et al. (2011).
The predictive performance of MaxEnt has exceeded other
algorithms in several comparative studies (Elith et al., 2006;
Foley et al., 2009, 2010; Larson et al., 2010; Arboleda et al.,
2012). In addition, its popularity can probably be explained
by the fact that it is implemented in free software with a user-
friendly interface, good documentation and many options for
parameterization. Generalized linear models were the second-
most frequent method because they offer more flexibility than
machine learning algorithms (e.g. MaxEnt and GARP), thus
improving model fit and ecological interpretations of para-
meters (Franklin, 2010). Also noteworthy is the use of
CLIMEX, a mechanistic (process-based) algorithm.
Mechanistic models are based on vector’s biological processes,
such as duration of life cycle, biting rates, dispersal ability,
temperature limits for larvae development, etc. The inclusion
of this type of data improves the biological meaning of models,
but they require solid empirical knowledge about the vectors’
physiology, which makes parameterization a challenge
(Kearney & Porter, 2009; Dormann et al., 2012; Fischer et al.,
2014).
Models produced by different algorithms may have dis-
similar, even contrasting outputs (Dormann et al., 2008;
Diniz et al., 2009; Elith & Graham, 2009; Li & Wang, 2013).
Independent evaluations have often been unable to identify
a single recommended algorithm for all circumstances (Elith
et al., 2006; Elith & Graham, 2009; Li & Wang, 2013; Qiao
et al., 2015). An alternative to avoid the choice of a particular
method is to test models produced by a set of algorithms (Qiao
et al., 2015) and combine their results as an ensemble model
(Araújo & New, 2007; Marmion et al., 2009). With a set of mod-
els produced by a number of algorithms, uncertainty can be
properly quantified, thus improving the study’s result
(Pearson et al., 2006 ; Owens et al., 2013; Qiao et al., 2015). The
use of multiple algorithms was present in over 70% of general
ENM studies published recently (Guillera-Arroita et al., 2015),
but it was under-represented in ENM of disease vectors for the
same period (approximately 10%). This represents a signifi-
cant delay in disease vector studies in relation to what is cur-
rently being published.
A good example of the multiple algorithm approach was a
comparison between models produced by BIOCLIM,
DOMAIN (Carpenter et al., 1993), GARP, GLM (logistic regres-
sion) and MaxEnt to identify areas of high density of Aedes
mosquitoes in Bermuda (Khatchikian et al., 2011). The results
varied between the different algorithms, but since GLM and
MaxEnt performed better, both were used to predict risk
areas of mosquito infestations (Khatchikian et al., 2011).
Another example was a study of the distribution patterns of
natural breeding sites of A. aegypti in Colombia, where models
produced by GARP had fewer omission errors than those pro-
duced by MaxEnt (Arboleda et al., 2012). Models produced by
MaxEnt performed better in certain regions, although areas
predicted as suitable by the two algorithms coincided closely.
The two algorithms were combined into an ensemble model,
where coincident areas were considered suitable with greater
confidence. The combination of methods improved the detec-
tion of natural breeding sites, allowing the optimization of ef-
fort and financial investment in dengue control programs in
the region (Arboleda et al.
, 2012).
Future projections of vector distributions
Over 700 vector species were studied in the 146 reviewed
papers, including mostly mosquitoes (63 articles) and sand
flies (29), followed by works on kissing bugs (18), biting
midges (17), ticks (14), tsetse flies (3), fleas (1) and water
bugs (1) (Table S1). The geographic extent of the reviewed
studies varied from local to global (Table S1). The 31 studies
with future ENM projections mostly point to expansions in re-
sponse to climate change scenarios, accompanied by poleward
shifts (table 1). This trend is being observed for several taxo-
nomic groups, where long-term field studies demonstrate re-
cent species’ movements towards higher latitudes and higher
altitudes in response to climate change (Hickling et al., 2006;
Stange & Ayres, 2010; Chen et al., 2011). There is, however, a
noteworthy methodological issue in about half of the re-
viewed studies (table 1). When projecting into future scen-
arios, models should be trained with the full-known
distribution of the species. If only a subset of the realized
niche is used, future predictions may underestimate environ-
mentally suitable areas and quantifications of range changes
become questionable (Pearson & Dawson, 2003; Guisan &
Thuiller, 2005; Araújo & Peterson, 2012). An additional source
of uncertainty in future forecasts is the extrapolation of models
into climatic conditions that do not presently exist (Fitzpatrick
& Hargrove, 2009). Some ENM algorithms have standard
ways of controlling extrapolation, such as MaxEnt, by limiting
output values to the range of environmental variables under
which the model was calibrated (Phillips et al., 2006).
Alternatively, out-of-range values can be masked directly in
model predictions (Owens et al., 2013; Carvalho et al., 2015).
Aedes aegypti and Aedes albopictus (Diptera: Culicidae)
The main vector of dengue, A. aegypti, is currently distrib-
uted throughout most tropical regions of the world.
Fig. 1. Methods applied in the literature of ecological niche
modelling of arthropod vectors of diseases.
Impacts of climate change on disease vectors 421
https://doi.org/10.1017/S0007485316001097 Published online by Cambridge University Press
Table 1. Overview of the future projections of the distributions of arthropod vectors of diseases.
Species Main disease Study area
Full current
distribution
of species Algorithm
Year of
projections
Difference of pro-
jected area
General range shift
directions Reference
Diptera: Culicidae
Aedes aegypti Dengue Australia No GARP 2030, 2050 Expansion Central, south Beebe et al. (2009)
Aedes aegypti Dengue Global Yes Alpha-shapes 2010–2040 Expansion and
contraction
Several directions Capinha et al. (2014)
Aedes aegypti Dengue Brazil No MaxEnt 2050 Contraction South Cardoso-Leite et al.
(2014)
Aedes aegypti Dengue Global Yes CLIMEX 2030, 2070 Contraction, discrete
expansion
Several directions Khormi and Kumar
(2014)
Aedes albopictus Arboviruses Trentino, Italia No GLM (logistic) 2050 Expansion East, west Roiz et al. (2011)
Aedes albopictus Arboviruses Europe Yes MaxEnt 2040, 2070,
2100
Expansion North, east, west Fischer et al. (2011c)
Aedes albopictus Arboviruses Australia, global Yes MaxEnt,
CLIMEX
2030, 2050 Discrete expansion Central Hill et al. (2014)
Aedes stictus Arboviruses Sweden No Other 2020, 2050,
2080
Expansion North Schäfer and Lundström
(2009)
Anopheles
arabiensis
Malaria Sudan and North
of Egypt
No MaxEnt 2050 Expansion Not given Fuller et al. (2012)
Anopheles
arabiensis
Malaria Africa Yes LOBAG-OC 2050 Contraction East, southeast Drake and Beier (2014)
Anopheles gambiae
and Anopheles
arabiensis
Malaria Africa Yes CLIMEX Not given Expansion South, east Tonnang et al. (2010)
Anopheles gambiae
and Anopheles
arabiensis
Malaria Africa Yes GARP 2055 Expansion South, east Peterson (2009)
Anopheles gambiae
and Anopheles
arabiensis
Malaria Africa Yes CLIMEX Not given Expansion South, east Tonnang et al. (2014)
Diptera: Psychodidae
Lutzomyia antophora
and Lutzomyia
diabolica
Leishmaniasis North America
and Mexico
Yes MaxEnt 2020, 2050,
2080
Expansion North, northeast González et al. (2010)
Lutzomyia longipalpis
and Lutzomyia
evansi
Leishmaniasis Colombia No MaxEnt 2020, 2050,
2080
Expansion or
contraction (at
different
scenarios)
North González et al. (2014)
Lutzomyia spp.
(three species)
Leishmaniasis South America Yes GARP 2055 Expansion South, southeast Peterson and Shaw
(2003)
Phlebotominae
(28 species)
Leishmaniasis North and Central
Americas
No GARP 2020, 2050,
2080
Expansion in 97% of
species,
contraction in 3%
Northwest (64% of
species), northeast
(35%), southeast
(0,6%)
Moo-Llanes et al. (2013)
Phlebotomus papatasi Leishmaniasis Southeast Asia No Discriminant
Analysis
Not given Expansion Not given Cross and Hyams (1996)
Phlebotomus
perniciosus
Leishmaniasis Bavaria, Germany No MaxEnt 2040 Expansion Not given Fischer et al. (2011b)
B.M. Carvalho et al.422
https://doi.org/10.1017/S0007485316001097 Published online by Cambridge University Press
Table 1. (Cont.)
Species Main disease Study area Full current
distribution
of species
Algorithm Year of
projections
Difference of pro-
jected area
General range shift
directions
Reference
Phlebotomus ariasi
and Phlebotomus
perniciosus
Leishmaniasis Madrid, Spain No GLM (negative
binomial)
2040, 2070,
2100
Expansion Not given Gálvez et al. (2011)
Phlebotomus spp.
(five species)
Leishmaniasis Southern
Germany
No MaxEnt 2040 Expansion Central, northwest Haeberlein et al. (2013)
Phlebotomus spp.
(five species)
Leishmaniasis Central Europe No MaxEnt 2040, 2070,
2100
Expansion Mostly east Fischer et al. (2011a)
Diptera:
Ceratopogonidae
Culicoides imicola Bluetongue Spain No GLM (negative
binomial)
2040 Stability Not given Acevedo et al. (2010)
Culicoides imicola Bluetongue Europe No GLM (logistic) Not given Expansion North Wittmann et al. (2001)
Culicoides imicola Bluetongue Global Yes CLIMEX 2030, 2070 Expansion and
contraction
Mostly north Guichard et al. (2014)
Hemiptera: Reduviidae
Triatoma gerstaeckeri
and Triatoma
sanguisuga
Chagas
disease
Mexico and USA Yes MaxEnt 2050 Expansion North, northeast Garza et al. (2014)
Triatoma brasiliensis
species complex
Chagas
disease
Northeast Brazil Yes MaxEnt, GARP 2020, 2050 Stability Not given Costa et al. (2014)
Acari: Ixodida
Ixodes ricinus Lyme disease Europe No GARP 2050 Expansion and
contraction
North Boeckmann and Joyner
(2014)
Ixodes ricinus Lyme disease Europe and Asia Yes MaxEnt 2050, 2080 Expansion North, east Porretta
et al. (2013)
Ixodes scapularis Lyme disease USA/Mexico
border
Yes MaxEnt 2050 Expansion Northeast Feria-Arroyo et al. (2014)
Ixodidae (six species) Lyme disease Mediterranean
region
Yes ENFA Not given Not given Not given Estrada-Peña and
Venzal (2007)
Impacts of climate change on disease vectors 423
https://doi.org/10.1017/S0007485316001097 Published online by Cambridge University Press