ecospat.CCV.communityEvaluation.prob

0th

Percentile

Evaluates community predictions directly on the probabilities (i.e., threshold independent)

This function generates a number of community evaluation metrics directly based on the probability returned by the individual models. Instead of thresholding the predictions (ecospat.CCV.communityEvaluation.bin this function directly uses the probability and compares its outcome to null models or average expectations.)

Keywords
~kwd1 , ~kwd2
Usage
ecospat.CCV.communityEvaluation.prob(
      ccv.modeling.data,
      community.metrics=c('SR.deviation','community.AUC','probabilistic.Sorensen'),
      se.th=0.01,
      parallel = TRUE,
      cpus = 4)
Arguments
ccv.modeling.data

a 'ccv.modeling.data' object returned by ecospat.CCV.modeling

community.metrics

a selection of community metrics to calculate ('SR.deviation', 'community.AUC', 'probabilistic.Sorensen', 'probabilistic.Jaccard', 'probabilistic.Simpson')

se.th

the desired precission for the community metrics (standard error of the mean)

parallel

should parallel computing be allowed (TRUE/FALSE)

cpus

number of cpus to use in parallel computing

Value

DataSplitTable

a matrix with TRUE/FALSE for each model run (TRUE=Calibration point, FALSE=Evaluation point)

CommunityEvaluationMetrics.CalibrationSites

a 3-dimensional array containing the community evaluation metrics for the calibartion sites of each run (NA means that the site was used for evaluation)

CommunityEvaluationMetrics.EvaluationSites

a 3-dimensional array containing the community evaluation metrics for the evaluation sites of each run (NA means that the site was used for calibaration)

Note

If the community evaluation metric 'SR.deviation' is selected the returned tables will have the following columns:

  • SR.obs = observed species richness,

  • SR.mean = the predicted species richness (based on the probabilities assuming poission binomial distribution),

  • SR.dev = the deviation of observed and predicted species richness,

  • SR.sd = the standard deviation of the predicted species richness (based on the probabilities assuming poission binomial distribution),

  • SR.prob = the probability that the observed species richness falls within the predicted species richness (based on the probabilities assuming poission binomial distribution),

  • SR.imp.05 = improvement of species richness prediction over null-model 0.5,

  • SR.imp.average.SR = improvement of species richness prediction over null-model average.SR and

  • SR.imp.prevalence = improvement of species richness prediction over null-model prevalence.

If the community evalation metric community.AUC is selected the returned tables will have the following colums:

  • Community.AUC = The AUC of ROC of a given site (in this case the ROC plot is community sensitiviy [percentage species predicted corretly present] vs 1 - community specificity [percentage of species predicted correctly absent])

If any of the other community evaluation metrics ('probabilistic.Sorensen', 'probabilistic.Jaccard', 'probabilistic.Simpson') is selected the returned tables will have the follwing colums:

  • METRIC.mean = The average Sorensen/Jaccard/Simpson based on a number of random draws of the probabilities.

  • METRIC.sd = The standard deviation of Sorensen/Jaccard/Simpson based on a number of random draws of the probabilities.

  • METRIC.CI = The 95% confidence intervall of the average Sorensen/Jaccard/Simpson based on the standard deviation and number of draws. Should normally be <= se.th.

  • nb.it = number of draws used to estimate all the parameters. The draws stop as soon as the desired precission (se.th) is reached or the limit of allowed iterations (default=10'000).

  • composition.imp.05 = improvement of species compostion prediction over the null-model 0.5.

  • composition.imp.average.SR = improvement of the species composition prediction over the null-model average.SR.

  • composition.imp.prevalence = improvement of the species composition prediction over the null-model prevalence.

For detailed descriptions of the null models see Scherrer et al. .....

See Also

ecospat.CCV.createDataSplitTable; ecospat.CCV.communityEvaluation.bin;

Aliases
  • ecospat.CCV.communityEvaluation.prob
Examples
# NOT RUN {
#Loading species occurence data and remove empty communities
data(ecospat.testData)
testData <- ecospat.testData[,c(24,34,43,45,48,53,55:58,60:63,65:66,68:71)]
sp.data <- testData[which(rowSums(testData)>0), sort(colnames(testData))]

#Loading environmental data
env.data <- ecospat.testData[which(rowSums(testData)>0),4:8]

#Coordinates for all sites
xy <- ecospat.testData[which(rowSums(testData)>0),2:3]

#Running all the models for all species
myCCV.Models <- ecospat.CCV.modeling(sp.data = sp.data,
                                     env.data = env.data,
                                     xy = xy,
                                     NbRunEval = 5,
                                     minNbPredictors = 10,
                                     VarImport = 3)
                                     
#Calculating the probabilistic community metrics
myCCV.communityEvaluation.prob <- ecospat.CCV.communityEvaluation.prob(
      ccv.modeling.data = myCCV.Models,
      community.metrics = c('SR.deviation','community.AUC','probabilistic.Sorensen'),
      se.th = 0.02, 
      parallel = FALSE,
      cpus = 4)
# }
Documentation reproduced from package ecospat, version 3.0, License: GPL

Community examples

Looks like there are no examples yet.