Modelling the potential spatial distribution of mosquito species using three different techniques

Abstract

BackgroundModels for the spatial distribution of vector species are important tools in the assessment of the risk of establishment and subsequent spread of vector-borne diseases. The aims of this study are to define the environmental conditions suitable for several mosquito species through species distribution modelling techniques, and to compare the results produced with the different techniques.MethodsThree different modelling techniques, i.e., non-linear discriminant analysis, random forest and generalised linear model, were used to investigate the environmental suitability in the Netherlands for three indigenous mosquito species (Culiseta annulata, Anopheles claviger and Ochlerotatus punctor). Results obtained with the three statistical models were compared with regard to: (i) environmental suitability maps, (ii) environmental variables associated with occurrence, (iii) model evaluation.ResultsThe models indicated that precipitation, temperature and population density were associated with the occurrence of Cs. annulata and An. claviger, whereas land surface temperature and vegetation indices were associated with the presence of Oc. punctor. The maps produced with the three different modelling techniques showed consistent spatial patterns for each species, but differences in the ranges of the predictions. Non-linear discriminant analysis had lower predictions than other methods. The model with the best classification skills for all the species was the random forest model, with specificity values ranging from 0.89 to 0.91, and sensitivity values ranging from 0.64 to 0.95.ConclusionsWe mapped the environmental suitability for three mosquito species with three different modelling techniques. For each species, the maps showed consistent spatial patterns, but the level of predicted environmental suitability differed; NLDA gave lower predicted probabilities of presence than the other two methods. The variables selected as important in the models were in agreement with the existing knowledge about these species. All model predictions had a satisfactory to excellent accuracy; best accuracy was obtained with random forest. The insights obtained can be used to gain more knowledge on vector and non-vector mosquito species. The output of this type of distribution modelling methods can, for example, be used as input for epidemiological models of vector-borne diseases.

Modelling the monthly abundance of Culicoides biting midges in nine European countries using Random Forests machine learning

“…The overall prediction for that observation is the average of all individual tree's predictions in the forest [38]. The RF technique has previously been used to model the geographical distribution and/or abundance of vectors such as mosquitoes [23], biting midges [17,39] and parasites (Fasciola hepatica) [40]. The advantages of decision trees include their robustness against outliers and their capability to identify complex interactions, including non-linear relationships between the response and predictor variables.…”

“…Machine learning techniques are algorithms that, like classical statistical models, can be used to predict an outcome using predictor variables. The machine learning technique Random Forests (RF) has been proven to outperform classical approaches for species distribution modelling such as generalized linear models (GLM) and logistic regression (LR) [22][23][24]. We hypothesised that Culicoides abundance may be predicted for a large area of Europe using a RF approach and climatic and environmental predictors.…”

“…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.…”

Modelling and mapping heavy metal and nitrogen concentrations in moss in 2010 throughout Europe by applying Random Forests models

“…Areas of RF application are, amongst others, astronomy, autopsy, transport planning, medicine, and environmental sciences (Fawagreh et al 2014). Examples for the latter category were given by Cianci et al (2015), Evans and Cushman (2009), Howard et al (2014) and Magness et al (2008) predicting species, Deloncle et al (2007) predicting weather regimes, Pal (2005) classifying forests and Thums et al (2008) marine species, Rothwell et al (2008) for Cd, Hg, and Pb concentrations in moss by using CART (Breiman et al 1984). Contrary to CART, RF in conjunction with the Geographic Information System (GIS) used here were additionally applied for regression mapping, i.e.…”

Caracterización edafoclimática de la zona productora de lima ácida Tahití (Citrus latifolia Tanaka) en el Tolima (Colombia), afectada por una fisiopatía

“…Esta herramienta ha sido empleada en estudios sobre temas como el efecto del cambio climático en el sistema cultivo (Evangelista, Young, & Burnett, 2013;Läderach, Martinez-Valle, Schroth, & Castro, 2013;Schroth, Läderach, Cuero, Neilson, & Bunn, 2015), humedad del suelo (Topete-Ángel et al, 2014), distribución de plagas (Holt, Salkeld, Fritz, Tucker, & Gong, 2009) y distribución de especies de interés (Castro-Díez, Godoy, Saldaña, & Richardson, 2011;Cianci, Hartemink, & Ibáñez-Justicia, 2015;Kumar & Stohlgren, 2009).…”