BIOMOD2 script
From Csegro:
"We used an ensemble of four different techniques: generalized linear models (GLM), generalized boosted regression models (GBM; Friedman et al., 2000), random forest (RF, Breiman 2001) and maximum entropy modeling." Us probably not maxent
"Models were calibrated with 70% of the data and evaluated with the remaining 30% of the data. The procedure was replicated 25 times per species with random training and evaluation datasets, adding up to 100 models per species (4 techniques x 25 runs each). We compared four different evaluation metrics: Roc AUC (Fielding and Bell 1997), TSS (Allouche et al. 2006), KAPPA (Allouche et al. 2006) and the “presence-only” evaluator Boyce index (Boyce et al. 2002, Hirzel et al. 2006) to assess model performance. The final ensemble model was constructed from selected models with the evaluation metric TSS larger than 0.7, by taking a weighted average proportional to the TSS values. "
myBiomodData <- BIOMOD_FormatingData(resp.var = myResp,
expl.var = myExpl,
resp.xy = myRespXY,
resp.name = myRespName)
myBiomodOption <- BIOMOD_ModelingOptions()
myBiomodModelOut <- BIOMOD_Modeling(
myBiomodData,
models = c('SRE','CTA','RF','MARS','FDA'),
models.options = myBiomodOption,
NbRunEval=3,
DataSplit=80,
Prevalence=0.5,
VarImport=3,
models.eval.meth = c('TSS','ROC'),
SaveObj = TRUE,
rescal.all.models = TRUE,
do.full.models = FALSE,
modeling.id = paste(myRespName,"FirstModeling",sep=""))
myBiomodModelEval <- get_evaluations(myBiomodModelOut)
get_variables_importance(myBiomodModelOut)
myBiomodEM <- BIOMOD_EnsembleModeling(
modeling.output = myBiomodModelOut,
chosen.models = 'all',
em.by='all',
eval.metric = c('TSS'),
eval.metric.quality.threshold = c(0.7),
prob.mean = T,
prob.cv = T,
prob.ci = T,
prob.ci.alpha = 0.05,
prob.median = T,
committee.averaging = T,
prob.mean.weight = T,
prob.mean.weight.decay = 'proportional' )