Fav: Favourability (probability without the effect of sample...

View source: R/Fav.R

FavR Documentation

Favourability (probability without the effect of sample prevalence)

Description

Computes prevalence-independent favourability for a species' presence, based on a presence/(pseudo)absence model object, or on a vector of predicted probability values plus either the modelled binary response variable, the total number of modelled ones and zeros, or the prevalence (proportion of ones) in the modelled binary response (i.e., in the training data).

Usage

Fav(model = NULL, obs = NULL, pred = NULL, n1n0 = NULL, sample.preval = NULL, 
method = "RBV", true.preval = NULL, verbosity = 2)

Arguments

model

a binary-response, presence/(pseudo)absence probability-producing model object of class "glm", "gam", "gbm", "randomForest" or "bart" (computed with keeptrees=TRUE).

obs

alternatively to 'model', a vector of the 1 and 0 values of the binary response variable (e.g. presence-absence of a species) in the model training data. This argument is ignored if 'model' is provided.

pred

alternatively to 'model', a numeric vector, RasterLayer or SpatRaster of predicted presence probability values, produced by a presence/(pseudo)absence modelling method yielding presence probability. This argument is ignored if 'model' is provided.

n1n0

alternatively to 'obs' or 'sample.preval', an integer vector of length 2 providing the total numbers of modelled ones and zeros (in this order) of the binary response variable in the model training data. Ignored if 'obs' or 'model' is provided.

sample.preval

alternatively to 'obs' or 'n1n0', the prevalence (proportion of ones) of the binary response variable in the model training data. Ignored if 'model' is provided.

method

either "RBV" for the original Real, Barbosa & Vargas (2006) procedure, or "AT" if you want to try out the modification proposed by Albert & Thuiller (2008) (but see Details!).

true.preval

the true prevalence (as opposed to sample prevalence), necessary if you want to try the "AT" method (but see Details!).

verbosity

numeric value indicating the amount of messages to display; currently meaningful values are 0, 1, and 2 (the default).

Details

Methods such as Generalized Linear Models, Generalized Additive Models, Random Forests, Boosted Regression Trees / Generalized Boosted Models, Bayesian Additive Regression Trees and several others, are widely used for modelling species' potential distributions using presence/absence data and a set of predictor variables. These models predict presence probability, which incorporates the prevalence (proportion of presences) of the species in the modelled sample. So, predictions for restricted species are always generally low, while predictions for widespread species are always generally higher, regardless of the actual environmental quality. Barbosa (2006) and Real, Barbosa & Vargas (2006) proposed an environmental favourability function which is based on presence probability and cancels out uneven proportions of presences and absences in the modelled data. Favourability thus assesses the extent to which the environmental conditions change the probability of occurrence of a species with respect to its overall prevalence in the study area. Model predictions become, therefore, directly comparable among species with different prevalences. The favourability function is implemented in the fuzzySim package and is also in the SAM (Spatial Analysis in Macroecology) software (Rangel et al. 2010).

Using simulated data, Albert & Thuiller (2008) proposed a modification to the favourability function, but it requires knowing the true prevalence of the species (not just the prevalence in the modelled sample), which is rarely possible in real-world modelling. Besides, this suggestion was based on the misunderstanding that the favourability function was a way to obtain the probability of occurrence when prevalence differs from 50%, which is incorrect (see Acevedo & Real 2012).

To get environmental favourability with either the Real, Barbosa & Vargas ("RBV") or the Albert & Thuiller ("AT") method, you just need to get model predictions of presence probability from your data, together with the proportions of presences and absences in the modelled sample, and then use the 'Fav' function. Input data for this function are either a model object resulting from the glm function, or the vector of presences-absences (1-0) of your species and the corresponding presence probability values, obtained e.g. with predict(mymodel, mydata, type = "response"). Alternatively to the presences-absences, you can provide either the sample prevalence or the numbers of presences and absences in the dataset that was used to generate the presence probabilities. In case you want to use the "AT" method, you also need to provide the true (absolute) prevalence of your species.

Value

If 'model' is provided or if 'pred' is a numeric vector, the function returns a numeric vector of the favourability values. If 'model' is not provided (which would override other arguments) and 'pred' is a RasterLayer or a SpatRaster, the function returns an object of the same class, containing the favourability values.

Author(s)

A. Marcia Barbosa

References

Acevedo P. & Real R. (2012) Favourability: concept, distinctive characteristics and potential usefulness. Naturwissenschaften 99: 515-522

Albert C.H. & Thuiller W. (2008) Favourability functions versus probability of presence: advantages and misuses. Ecography 31: 417-422.

Barbosa A.M.E. (2006) Modelacion de relaciones biogeograficas entre predadores, presas y parasitos: implicaciones para la conservacion de mamiferos en la Peninsula Iberica. PhD Thesis, University of Malaga (Spain).

Rangel T.F.L.V.B, Diniz-Filho J.A.F & Bini L.M. (2010) SAM: a comprehensive application for Spatial Analysis in Macroecology. Ecography 33: 46-50.

Real R., Barbosa A.M. & Vargas J.M. (2006) Obtaining environmental favourability functions from logistic regression. Environmental and Ecological Statistics 13: 237-245.

See Also

multGLM

Examples


# obtain a probability model and its predictions:

data(rotif.env)

names(rotif.env)

mod <- with(rotif.env, glm(Abrigh ~ Area + Altitude + 
AltitudeRange + HabitatDiversity + HumanPopulation, 
family = binomial))

prob <- predict(mod, data = rotif.env, type = "response")


# obtain predicted favourability in different ways:

Fav(model = mod)

Fav(obs = rotif.env$Abrigh, pred = prob)

Fav(pred = mod$fitted.values, n1n0 = c(112, 179))

Fav(pred = mod$fitted.values, sample.preval = 0.38)

fuzzySim documentation built on Oct. 31, 2022, 1:07 a.m.