Modelling the spatial distribution of the nuisance mosquito species Anopheles plumbeus (Diptera: Culicidae) in the Netherlands
© Ibañez-Justicia and Cianci; licensee BioMed Central. 2015
Received: 26 January 2015
Accepted: 16 April 2015
Published: 1 May 2015
Landscape modifications, urbanization or changes of use of rural-agricultural areas can create more favourable conditions for certain mosquito species and therefore indirectly cause nuisance problems for humans. This could potentially result in mosquito-borne disease outbreaks when the nuisance is caused by mosquito species that can transmit pathogens. Anopheles plumbeus is a nuisance mosquito species and a potential malaria vector. It is one of the most frequently observed species in the Netherlands. Information on the distribution of this species is essential for risk assessments. The purpose of the study was to investigate the potential spatial distribution of An. plumbeus in the Netherlands.
Random forest models were used to link the occurrence and the abundance of An. plumbeus with environmental features and to produce distribution maps in the Netherlands. Mosquito data were collected using a cross-sectional study design in the Netherlands, from April to October 2010–2013. The environmental data were obtained from satellite imagery and weather stations. Statistical measures (accuracy for the occurrence model and mean squared error for the abundance model) were used to evaluate the models performance. The models were externally validated.
The maps show that forested areas (centre of the Netherlands) and the east of the country were predicted as suitable for An. plumbeus. In particular high suitability and high abundance was predicted in the south-eastern provinces Limburg and North Brabant. Elevation, precipitation, day and night temperature and vegetation indices were important predictors for calculating the probability of occurrence for An. plumbeus. The probability of occurrence, vegetation indices and precipitation were important for predicting its abundance. The AUC value was 0.73 and the error in the validation was 0.29; the mean squared error value was 0.12.
The areas identified by the model as suitable and with high abundance of An. plumbeus, are consistent with the areas from which nuisance was reported. Our results can be helpful in the assessment of vector-borne disease risk.
Mosquitoes (Diptera:Culicidae) are known to be vectors of a large number of pathogens around the globe and are considered as prime candidates for transmitting (re-)emerging vector-borne diseases (VBDs) in Europe . The increased mobility of humans, that has also increased the mobility of livestock and pathogens, as well as environmental modifications and climate changes can contribute to the (re-)emergence of vector-borne diseases . Furthermore, mosquito bites can cause a considerable nuisance for humans and mammals. Severe nuisance can have negative economic consequences (e.g., in tourism, work productivity outdoors, meat and dairy production) . These nuisance situations can eventually lead to autochthonous VBD cases, when in non-endemic areas infectious reservoirs, either humans (travellers, temporary workers) or animals (livestock, migrating animals) come in contact with high density of mosquito vectors.
Anopheles plumbeus is a mosquito species commonly found in forests, where larvae are usually found in water in rot-holes of trees with high salinity and deficiency of oxygen . They can also be found in containers with stagnant rain water and groundwater, such as tyres, rainwater casks and cemetery vases [5,12]. In the last decade, this species has also been associated with abandoned stables where it breeds in the rain water collected in the manure cellars . This species is known to be a particularly aggressive biter, feeding at any time of the day on different mammalian hosts (including humans), and to a lesser degree on birds and reptiles . In June 2006, nuisance caused by An. plumbeus was reported for the first time in the Netherlands, near the city of Nijmegen . Since then, An. plumbeus nuisance have been reported every year in the Netherlands, mostly in proximity to abandoned pig stables (Ibanez-Justicia, unpublished).
An understanding of the spatial extent of potential vector species, their abundance and seasonal activity, is important for estimating levels of risks of VBDs and enabling better targeting for surveillance and control. In order to develop basic reproduction number (R0) models and construct risk maps that indicate the risk for an outbreak after an introduction, abundance data of vectors are an essential parameter [15,16]. Although vector presence and abundance are not the only factors determining whether or not a pathogen can spread in an area, determining the distribution of the vector is an essential step in studying the risk of transmission of a pathogen. Given the nuisance and potential risk for the human health, such information on An. plumbeus is needed. Currently, no information on the potential spatial distribution of this species is available for the Netherlands.
In this study, we modelled the potential spatial distribution, expressed in occurrence (predicted probability of presence) and abundance of An. plumbeus in the Netherlands, based on data collected during the National Mosquito Survey and environmental data. The occurrence was modelled to predict the environmental suitability of the species using a random classification forest model. The abundance was modelled using a random regression forest model with the aim to identify areas where mosquito peaks could be expected. Random (classification and regression) forest models allow external validation through a bootstrapping procedure. The occurrence model was validated also with an external dataset. The resulting maps are in agreement with the reported nuisance for this species and the predictions show a good matching with an external dataset used to validate the model.
Species distribution modelling links the occurrence or the abundance of species with environmental data and estimates the similarity of the conditions at any site based on the conditions at the locations of known occurrence/abundance of a species. Here we describe the mosquito data collection, the environmental data used and the statistical methods applied in this study.
Mosquito data used for the modelling were obtained from the national mosquitoes survey that was carried out from April to October 2010–2013 by the Dutch Centre for Monitoring of Vectors. Mosquitoes were captured using CO2 baited Mosquito Magnet Liberty Plus MM3100 (Woodstream® Co., Lititz, USA). Traps were randomly distributed in the country following a cross-sectional study design, with the following constraint: 40% of the traps were placed in urban areas, 40% in rural-agricultural areas and 20% in natural areas . Urban and agricultural areas were sampled more intensively, because of the potential higher human and veterinary health risk in those areas due to higher exposure.
Data consisted of mosquito abundance data, sampled at 778 locations. For this study the abundance data were reclassified into data of presence (when at least one mosquito was found in the trap) and absence (when no mosquitoes were found in the trap). Each of the locations was sampled only once and each trap was active for one week. The content of the traps was collected weekly and sent to the CMV laboratory. In the laboratory, mosquitoes were morphologically identified using, among others, the Culicidae key specifically designed for rapid field-identification of Dutch adult Culicidae (modified key after Snow , Schaffner et al. , Verdonschot , Becker et al. ). Twenty-seven mosquito species were found in the National Mosquito Survey and An. plumbeus was the 7th mosquito species most commonly found in the Netherlands. This species was active in the whole period of the survey, from April until October . When a presence and an absence point were in the same square kilometre only the presence point was used because presences inform about the places that are environmental suitable for a species, but absences do not necessarily indicate the opposite . This reduced the number of locations used in the analysis from 778 to 766.
Surveys used for the validation
Total nr locations
Target longitudinal sampling
Larval sampling, manual aspirator, BG-Sentinel trap
MM-Liberty Plus trap
MM-Liberty Plus trap
Target longitudinal sampling
MM-Liberty Plus trap
Target longitudinal sampling
CDC light trap, manual aspirator
2010, 2011, 2013, 2014
Check at locations of reported nuisance
Larval sampling, manual aspirator
Target longitudinal sampling
MM-Liberty Plus trap, CDC light trap
Fourier components from temporal Fourier analysis of an imagery time series
Fourier mean for entire time series
Amplitude of annual cycle
Amplitude of bi-annual cycle
Amplitude of tri-annual cycle
Phase of annual cycle
Phase of bi-annual cycle
Phase of tri-annual cycle
Proportion of total variance due to annual cycle
Proportion of total variance due to bi-annual cycle
Proportion of total variance due to tri-annual and cycle
Proportion of total variance due to all three cycles
Environmental predictor variables
Middle Infra-red (MIR)
Day-time land surface temperature (DLST)
Night-time land surface temperature (NLST)
Enhanced vegetation index (EVI)
Normalised difference vegetation index(NDVI)
Digital elevation model (DEM)
Gridded Population of the World
Human population density
European Environment Agency
Corine land cover
Three distribution modelling techniques suitable for occurrence data were applied: non-linear discriminant analysis , random classification forest  and generalised linear model . For each model, the accuracy was assessed using (i) sensitivity, i.e. the ability of a model to correctly identify known positive sites; (ii) specificity, i.e. the ability of a model to correctly identify known negative sites; (iii) the area under the curve, (AUC) that can be roughly interpreted as the probability that a model will correctly predict positive and negative sites . Of the three techniques, random forest provided the best accuracy and therefore the results of this model are presented.
A random classification forest model consists of an ensemble of trees. To create a reliable model, it is generally considered necessary to have the same number of presence and absence points as input. This is because having a different number will create a bias in the model prediction towards the more prevalent category (presence or absence) . For this reason, a ‘balanced’ subset of the data, i.e., a dataset with the same number of presences and absences, was selected. The output produced by the model is an environmental suitability indicator, expressed as a value between 0 (low suitability) and 1 (high suitability). The predictions are visualised in a map with colours ranging from red (high suitability) to blue (low suitability). A list of the most important variables used in the model is given based on the mean decrease in Gini index [31,34]. Random forest allows external validation through a bootstrapping procedure: for each tree, a random subset of the full dataset is sampled with replacement. The model validation is carried out for each tree using the points not used from the full dataset. This validation method is referred to as external, because the model is validated using data that are not used to build the tree. The comparison of the observed and predicted results enables us to calculate accuracy statistics, such as sensitivity and specificity. These measures are calculated for each tree and then averaged to give the overall values.
The predictions produced by the random classification forest were also externally validated against 45 observations from other surveys (Table 1) that reported only the presence of An. plumbeus. Comparing the observations obtained with the other surveys and the predictions made by the model using National Mosquito Survey data, the error rate was calculated as the proportion of incorrectly predicted pixels to the total number of points used in the validation.
The abundance of the species was modelled using a random regression forest model. The abundance data were transformed according to the formula log 10 (abundance + 1) . Because the aim was to identify areas where mosquito peaks could be expected, only the data collected in months in which peaks were observed were selected (June-September). The predicted environmental suitability obtained with the occurrence model described above, was included as one of the predictor variables for modelling the abundance of the species, as it is frequently done in this type of analysis [36-40]. The predicted abundance is interpreted as the expected maximum number of mosquitoes caught in a trap in a certain pixel. The predictions are visualised in a map with colours ranging from light green (low abundance) to dark green (high abundance). The importance of the predictors was assessed using the Increase in Node Purity (INP). The difference between observed and predicted values was expressed as the mean squared error. The analysis has been performed with the software Vecmap demo version . The maps have been produced with Quantum GIS .
Results and discussion
The probability of occurrence (environmental suitability) and the abundance of An. plumbeus have been predicted using mosquito field data and environmental data. The estimated environmental suitability and abundance are shown in maps. The important environmental variables used in the models and the accuracy of the models are discussed. The fact that out of three different modelling techniques for occurrence data random forest model was selected based upon its higher classification accuracy is consistent with earlier findings; random forest has been reported to outperform other traditional modelling techniques [43-45].
List of the top 10 most important variables in the occurrence model
CMORPH precipitation, phase of bi-annual cycle
CMORPH precipitation, phase of annual cycle
Worldclim precipitation, phase of annual cycle
Worldclim precipitation, proportion of total variance due to annual cycle
MIR, phase of annual cycle
NTLS temperature, minimum value
DTLS temperature, amplitude of annual cycle
CMORPH precipitation, maximum value
Fair accuracy was obtained with the model (AUC = 0.73), which showed a better ability in identifying suitable environments (sensitivity 0.71) than unsuitable environments (specificity 0.66). The accuracy is improved as compared with a first attempt of predicting the environmental suitability for An. plumbeus in the Netherlands, where the environmental suitability was extrapolated from Belgium to the Netherlands (sensitivity = 0.50, specificity = 0.49) . The error rate calculated to compare the predicted values to data of other surveys was low (0.29); 71% of the pixels were correctly predicted, meaning that the model could make good predictions in non-sampled areas. However, this is only a partial validation because it considers only presence points and does not give information about the performance of the model in predicting absence points.
Observed abundance used in the model
List of the top 10 most important variables in the abundance model
Worldclim precipitation, phase of annual cycle
Worldclim precipitation, proportion of total variance due to bi-annual cycle
NDVI, amplitude of annual cycle
Worldclim precipitation, amplitude of bi-annual cycle
MIR, amplitude of annual cycle
NTLS temperature, phase of bi-annual cycle
Worldclim precipitation, total variance
CMORPH precipitation, phase of bi-annual cycle
The aim of this study was to investigate the potential spatial distribution of An. plumbeus in the Netherlands. Using random (classification and regression) forest models, we identified areas with high environmental suitability and high abundance of this species in south-eastern provinces of Limburg and Brabant. These areas coincide with the areas where in recent years most nuisances have been reported. The predictions of the occurrence model were accurate and matched the external dataset used for validation. The abundance model predictions also matched the observation.
The output of species distribution modelling method can be used as an input for risk assessment of establishment and spread of vector-borne diseases [47,48]. Understanding and depicting the potential spatial distribution of mosquito species with modelling techniques is of increasing importance, especially for nuisance mosquito species that can cause economic implications or impact on human health.
The authors would like to thank Nienke Hartemink and Marieta Braks for a critical reading of the manuscript and useful comments. The authors would also like to thank Els Ducheyne and Wesley Tack for helpful discussion and suggestions. This study was funded by EU grant FP7-261504 EDENext and is catalogued by the EDENext Steering Committee as EDENext311 (http://www.edenext.eu). The contents of this publication are the sole responsibility of the authors and do not necessarily reflect the views of the European Commission.
- Versteirt V, De Clercq E, Dekoninck W, Damiens D, Ayrinhac A, Jacobs F, et al. Mosquito Vectors of Disease: Spatial Biodiversity, Drivers of Change, and Risk. Final Report. Brussels: Belgian Science Policy; 2009.Google Scholar
- Randolph SE, Rogers DJ. The arrival, establishment and spread of exotic diseases: patterns and predictions. Nat Rev Microbiol. 2010;8:361–71.PubMedView ArticleGoogle Scholar
- Connelly CR, Carlson DB. Florida Coordinating Council on mosquito control. Florida mosquito control: The state of the mission as defined by mosquito controllers, regulators, and environmental managers. Vero Beach, Florida: University of Florida, Institute of Food and Agricultural; 2009.Google Scholar
- Ibañez-Justicia A, Stroo A, Dik M, Beeuwkes J, Scholte EJ: National Mosquito (Diptera, Culicidae) Survey in the Netherlands 2010–2013. J Med Entomol 2015. doi: 10.1093/jme/tju058.Google Scholar
- Schaffner F, Thiéry I, Kaufmann C, Zettor A, Lengeler C, Mathis A, Bourgouin C. Anopheles plumbeus (Diptera: Culicidae) in Europe: a mere nuisance mosquito or potential malaria vector? Malar J. 2012;11:393.PubMed CentralPubMedView ArticleGoogle Scholar
- Krüger A, Rech A, Su X-Z, Tannich E. Two cases of autochthonous Plasmodium falciparum malaria in Germany with evidence for local transmission by indigenous Anopheles plumbeus. Trop Med Int Heal. 2001;6:983–5.View ArticleGoogle Scholar
- Shute PG, Maryon M. Malaria in England past, present and future. J R Soc Promot Health. 1974;94:23–9.View ArticleGoogle Scholar
- Sokolova MI, Snow KR. Malaria vectors in European Russia. Eur Mosq Bull. 2002;12:1–6.Google Scholar
- Schaffner F, Angel G, Geoffroy B, Hervy JP, Rhaiem A, Brunhes J. The mosquitoes of Europe, an identification and training programme. Montpellier: Fr IRD; 2001.Google Scholar
- Takken W, Kager PA, Verhave JP. Will malaria return to North-west Europe? In: Emerg pests vector-borne Dis. Wageningen: Wageningen Academic Publishers; 2007. p. 23–34.View ArticleGoogle Scholar
- Bradshaw WE, Holzapfel CM. Fitness and habitat segregation of British tree-hole mosquitoes. Ecol Entomol. 1991;16:133–44.View ArticleGoogle Scholar
- Marshall JF. The British mosquitoes. London: Brit. Mus. Nat Hist; 1938. p. 341.Google Scholar
- Dekoninck W, Hendrickx F, Van Bortel W, Versteirt V, Coosemans M, Damiens D, et al. Human-induced expanded distribution of Anopheles plumbeus, experimental vector of West Nile virus and a potential vector of human malaria in Belgium. J Med Entomol. 2011;48:924–8.PubMedView ArticleGoogle Scholar
- Bueno Marí R, Jiménez Peydró R. Study of the malariogenic potential of Eastern Spain. Trop Biomed. 2012;29(1):39–50.PubMedGoogle Scholar
- Hartemink NA, Purse BV, Meiswinkel R, Brown HE, De Koeijer A, Elbers ARW, et al. Mapping the basic reproduction number (R0) for vector-borne diseases: A case study on bluetongue virus. Epidemics. 2009;1:153–61.PubMedView ArticleGoogle Scholar
- Hartemink N, Vanwambeke SO, Heesterbeek H, Rogers D, Morley D, Pesson B, et al. Integrated Mapping of Establishment Risk for Emerging Vector-Borne Infections: A Case Study of Canine Leishmaniasis in Southwest France. PLoS One. 2011;6:e20817.PubMed CentralPubMedView ArticleGoogle Scholar
- Snow KR. Mosquitoes. Naturalist’ Handbooks 14. Slough, England: Richmond Publishing Co. Ltd; 1990. p. 66.Google Scholar
- Verdonschot P. Family Culicidae. In: Beuk PLT, editor. Checklist of the Diptera in the Netherlands. Utrecht: KNNV Publishers; 2002. p. 98.Google Scholar
- Becker N, Petric D, Zgomba M, Boase C, Madon M, Dahl C, et al. Mosquitoes and Their Control. Heidelberg, Dordrecht, New York: Springer; 2010. p. 577.View ArticleGoogle Scholar
- Lobo JM, Jiménez-Valverde A, Hortal J. The uncertain nature of absences and their importance in species distribution modelling. Ecography (Cop). 2010;33:103–14.View ArticleGoogle Scholar
- European Centre for Disease Prevention and Control (ECDC). Environmental Risk Mapping: Aedes albopictus in Europe. Stockholm: ECDC; 2013. Available from: http://www.ecdc.europa.eu/en/publications/Publications/climate-change-environmental-risk-mapping-aedes.pdf.Google Scholar
- Justice C, Townshend J, Vermote E, Sohlberg R, Descloitres J, Roy D, et al. Preliminary land surface products from the NASA Moderate Resolution Imaging Spectroradiometer (MODIS). In: Geosci Remote Sens Symp 2000 Proceedings IGARSS 2000 IEEE 2000 Int. Volume 3. 2000. p. 1157–62.Google Scholar
- MODIS. [modis.gsfc.nasa.gov].Google Scholar
- Rogers DJ, Williams BG. Tsetse distribution in Africa: seeing the wood and the trees. 1994.Google Scholar
- Rogers DJ. Satellites, space, time and the African trypanosomiases. Adv Parasitol. 2000;47:129–71.PubMedGoogle Scholar
- Scharlemann JPW, Benz D, Hay SI, Purse BV, Tatem AJ, Wint GRW, et al. Global data for ecology and epidemiology: a novel algorithm for temporal Fourier processing MODIS data. PLoS One. 2008;3:e1408.PubMed CentralPubMedView ArticleGoogle Scholar
- WorldClim - Global Climate Data [www.worldclim.org].
- National Weather Service Climate Prediction Center [www.cpc.ncep.noaa.gov].
- Socioeconomic Data and Applications Center [sedac.ciesin.columbia.edu].Google Scholar
- European Environment Agency [www.eea.europa.eu].
- Breiman L. Random forests. Mach Learn. 2001;45:5–32.View ArticleGoogle Scholar
- McCullagh P, Nelder JA. Generalized Linear Models. Volume 37. London: Chapman & Hall; 1989.View ArticleGoogle Scholar
- Fielding AH, Bell JF. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ Conserv. 1997;24:38–49.View ArticleGoogle Scholar
- Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification and Regression Trees. Belmont, California, USA: Wadsworth international group; 1984.Google Scholar
- Avia-GIS. Abundance Modeling of Mosquito and Biting Midge Species in the Netherlands. 2012.Google Scholar
- Osborne PE, Alonso JC, Bryant RG. Modelling landscape-scale habitat use using GIS and remote sensing: a case study with great bustards. J Appl Ecol. 2001;38:458–71.View ArticleGoogle Scholar
- Boyce MS, Vernier PR, Nielsen SE, Schmiegelow FKA. Evaluating resource selection functions. Ecol Modell. 2002;157:281–300.View ArticleGoogle Scholar
- Gibson LA, Wilson BA, Cahill DM, Hill J. Spatial prediction of rufous bristlebird habitat in a coastal heathland: a GIS-based approach. J Appl Ecol. 2004;41:213–23.View ArticleGoogle Scholar
- Chefaoui RM, Hortal J, Lobo JM. Potential distribution modelling, niche characterization and conservation status assessment using GIS tools: a case study of Iberian < i > Copris</i > species. Biol Conserv. 2005;122:327–38.View ArticleGoogle Scholar
- Calvete C, Estrada R, Miranda MA, Borras D, Calvo JH, Lucientes J. Modelling the distributions and spatial coincidence of bluetongue vectors Culicoides imicola and the Culicoides obsoletus group throughout the Iberian peninsula. Med Vet Entomol. 2008;22:124–34.PubMedView ArticleGoogle Scholar
- Avia-GIS. VecmapTM. A one-stop-shop for vector mapping, Zoersel, Belgium. http://www.avia-gis.com/vecmap. 2014.
- QGIS Development Team. QGIS Geographic Information System. 2012.Google Scholar
- Cutler DR, Edwards Jr TC, Beard KH, Cutler A, Hess KT, Gibson J, et al. Random forests for classification in ecology. Ecology. 2007;88:2783–92.PubMedView ArticleGoogle Scholar
- Peters J, Baets BD, Verhoest NEC, Samson R, Degroeve S, Becker PD, et al. Random forests as a tool for ecohydrological distribution modelling. Ecol Modell. 2007;207:304–18.View ArticleGoogle Scholar
- Cianci D, Hartemink N, Ibáñez-Justicia A. Modelling the potential spatial distribution of mosquito species in the Netherlands using three different techniques. Int J Health Geogr. 2015;14:10. doi: 10.1186/s12942-015-0001-0.PubMed CentralPubMedView ArticleGoogle Scholar
- Townroe S, Callaghan A. British container breeding mosquitoes: the impact of urbanisation and climate change on community composition and phenology. PLoS One. 2014;9:e95325.PubMed CentralPubMedView ArticleGoogle Scholar
- Miller J. Species distribution modeling. Geogr Compass. 2010;4:490–509.View ArticleGoogle Scholar
- Fischer EAJ, Boender G-J, Nodelijk G, de Koeijer AA, Van Roermund HJ. The transmission potential of Rift Valley fever virus among livestock in the Netherlands: a modelling study. Vet Res. 2013;44:58.PubMed CentralPubMedView ArticleGoogle Scholar
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.