SDMPlay (version 2.0)

compute.brt: Compute BRT (Boosted Regression Trees) model

Description

Compute species distribution models with Boosted Regression Trees

Usage

compute.brt(x, proj.predictors, tc = 2, lr = 0.001, bf = 0.75,
           n.trees = 50, step.size = n.trees, n.folds= 10, fold.vector = NULL)

Arguments

x

SDMtab object or dataframe that contains id, longitude, latitude and values of environmental descriptors at corresponding locations

proj.predictors

RasterStack of environmental descriptors on which the model will be projected

tc

Integer. Tree complexity. Sets the complexity of individual trees

lr

Learning rate. Sets the weight applied to individual trees

bf

Bag fraction. Sets the proportion of observations used in selecting variables

n.trees

Number of initial trees to fit. Set at 50 by default

step.size

Number of trees to add at each cycle, set equal to n.trees by default

n.folds

Number of subsets into which the initial dataset (x) is divided for model evaluation procedures (cross-validation). Set to 10 by default.

fold.vector

Vector indicating the fold group to which each data belongs to.

Value

• model$algorithm "brt" string character

• model$data x dataframe that was used to implement the model

• model$response Parameters returned by the model object: list of 41, see gbm.step for more info

• model$raster.prediction Raster layer that predicts the potential species distribution

• model$eval.stats List of elements to evaluate the model: AUC, maxSSS, COR, pCOR, TSS, ntrees, residuals

Details

The function realises a BRT model according to the gbm.step function provided by Elith et al.(2008). See the publication for further information about setting decisions. The map produced provides species presence probability on the projected area.

References

Elith J, J Leathwick & T Hastie (2008) A working guide to boosted regression trees. Journal of Animal Ecology, 77(4): 802-813.

Barbet Massin M, F Jiguet, C Albert & W Thuiller (2012) Selecting pseudo absences for species distribution models: how, where and how many? Methods in Ecology and Evolution, 3(2): 327-338.

See Also

gbm.step

Examples

# NOT RUN {
#Download the presence data
data('ctenocidaris.nutrix')
occ <- ctenocidaris.nutrix
# select longitude and latitude coordinates among all the information
occ <- ctenocidaris.nutrix[,c('decimal.Longitude','decimal.Latitude')]

#Download some environmental predictors
data(predictors2005_2012)
envi <- predictors2005_2012
envi

#Create a SDMtab matrix
SDMtable_ctenocidaris <- SDMPlay:::SDMtab(xydata=occ,
                                         predictors=predictors2005_2012,
                                         unique.data=FALSE,
                                         same=TRUE)

#Run the model
model <- SDMPlay:::compute.brt(x=SDMtable_ctenocidaris, proj.predictors=envi,lr=0.0005)

#Plot the partial dependance plots
dismo::gbm.plot(model$response)

#Get the contribution of each variable to the model
model$response$contributions

#Get the interaction between variables
dismo::gbm.interactions(model$response)

#Plot some interactions
int <- dismo::gbm.interactions(model$response)
dismo::gbm.perspec(model$response,int$rank.list[1,1],int$rank.list[1,3])

#Plot the map prediction
library(grDevices) # add nice colors
palet.col <- colorRampPalette(c('deepskyblue','green','yellow', 'red'))( 80 )
raster::plot(model$raster.prediction, col=palet.col,
            main="Prediction map of Ctenocidaris nutrix distribution")
data('worldmap')
#add data
points(worldmap, type="l")
points(occ, col='black',pch=16)

REMARK: see more examples in the vignette tutorials
# }