Hello Beverley,

Several methods can be used. E.g. a zero-inflated binomial GLM or a hurdle model can deal with such data.

HTH,

Ákos

Hi Ákos Bede-Fazekas, thanks very much for your suggestions. I will read-up more on these two methods. Thanks again!

I recommend you see : Species distributions models: A synthetic revision

The species distribution models can be generated, in principle, with any appropriate statistical classifier for the type of variable modeled, dichotomous if we have presence / absence data and continuous if they are abundance data. The most comprehensive review to date of the available methods is Elith et al. (2006), where the advantages and problems of each method are mentioned. The different techniques can be basically classified into three groups: Discriminant techniques They are those that need presence and absence data to build the classifier. They can in turn be subdivided into: (1) classification trees —CART, classification and regression trees— (Breiman et al. 1984) and a host of derived techniques, such as random forest (Breiman 2001), boosted regression trees —BRT— ( Friedman 2001, Elith et al. 2008) or mixture discriminant analysis —MDA— (Hastie & Tibshirani 1996); (2) ordering techniques, such as canonical correspondence analysis —CCA— (Guisan et al. 1999); (3) neural networks —ANN, assisted neural network— (Pearson et al. 2002, Venables & Ripley 2002); (4) multivariate adaptive regression splines —MARS— (Friedman 1991); (5) generalized linear models —GLM, generalized linear models—, among which logistic regression can be considered the pioneer method (Mccullagh & Nelder 1989), which has been followed by techniques such as support vector machines —SVM— (Vapnik 1995) or generalized regression analysis and spatial prediction or GRASP (Lehmann et al. 2003); (6) generalized additive models —GAM, generalized additive models— (Hastie & Tibshirani 1986), GROSS (Hastie et al. 1994); (7) quantile regression (Vaz et al. 2008); (8) Bayesian statistics (Termansen et al. 2006, Latimer et al. 2009); and (9) maximum entropy, among which Maxent stands out (Phillips et al. 2006, Phillips & Dudík 2008). Maxent generates its own absences, called "background" so it is not necessary to enter absence data in the program. From some of these methods, "multi-response" extensions have emerged, which use as absences those places where the organism under study has not been found, but others related —ecologically or phylogenetically— with it (Elith & Leathwick 2007). Descriptive techniques These techniques, which only require presence data, were the first used in modeling: (1) deductive, such as expert opinion, (2) geographic envelopes or convex hull (Worton 1995); (3) environmental envelopes, such as BIOCLIM (Busby 1986, 1991), ANUCLIM (Houlder et al. 1999), BIOMAP (Nix and Switzer 1991), HABITAT (Walker & Cocks 1991) or FLORAMAP (Jones & Gladkov 1999); (4) mathematical distance methods, such as DOMAIN, which uses the Gower distance (Walker & Cocks 1991, Carpenter et al. 1993) or Biomapper, ENFA, LIVES and MADIFA, which are based on the Mahalanobis distance (Hirzel et al. 2002, Calenge et al. 2008); and, finally, (5) Expectation — Maximization Algorithm —EM— (Ward et al. 2008). Mixed techniques These methods employ several rules, some of them descriptive and others discriminating, while generating their own pseudo-absences; Desktop-GARP (Stockwell & Peters 1999) and OM-GARP (Elith et al. 2006) are the best and most widely known. From the individual models obtained with different methods, "consensus models" can be generated, in which the final model indicates the degree of coincidence between various models (Araújo & New 2007, Marmion et al. 2009); BIOMOD (Thuiller et al. 2009) is a tool specifically programmed for the generation of consensus models.