2

I'm attempting to predict elk elevational selection using elk location points (Elk) relative to an NED elevational raster (Elev) using maxent in R. Both are projected in NAD83 using lat/long, and I've cropped the raster to get around memory issues in R.

I've divided the Elk dataset into 5 groups, 4 of which are now the training group (ElkTrain) and 1 I've reserved as the test group (ElkTest). I also created my own background data (NonElk) with its own training and test data (NonElkTrain, NonElkTest). I'm running into this error (with and without using my background data) and I can't find anyone discussing this relative to maxent or when using only one dataframe:

> Max <- maxent(x=Elev, p=ElkTrain, a=NonElkTrain)

or

> Max <- maxent(x=Elev, p=ElkTrain, a=NULL, nbg=5000, factors=NULL, removeDuplicates=TRUE)

Error in match.names(clabs, names(xi)) : names do not match previous names

In addition: Warning message:

In .local(x, p, ...) : 1 (0.02%) of the presence points have NA predictor values

Since I'm only using one dataframe (ElkTrain), what names aren't matching?

Heikki
  • 2,214
  • 19
  • 34
S. Smythe
  • 23
  • 5

1 Answers1

1

That is a bug that occurs when you make a maxent model with only one predictor. You can see it with the data from the example in ?maxent

library(dismo)
# example data
predictors <- stack(list.files(path=paste(system.file(package="dismo"), '/ex', sep=''), pattern='grd', full.names=TRUE ))
occ <- read.table(paste(system.file(package="dismo"), '/ex/bradypus.csv', sep=''), header=TRUE, sep=',')[,-1]
bg <- randomPoints(predictors, 1000)

# this works
me <- maxent(x=predictors[[1:2]], p =occ)

# fails
me <- maxent(x=predictors[[1]], p =occ)
#Error in match.names(clabs, names(xi)) : 
#  names do not match previous names

This is because with a single layer, the matrix is dropped (the cause of many R bugs...), illustrated here:

extract(predictors[[1:2]], occtrain[1:2,])
#     bio1 bio12
#[1,]  263  1639
#[2,]  263  1639

extract(predictors[[1]], occtrain[1:2,])
#[1] 263 263

I will fix that. But here is are two work-arounds.

= Either make a single layer RasterStack (as suggested by you); the simplest approach:

prd <- stack(predictors[[1]])
me <- maxent(x=prd, p =occ)

= Or make a data.frame with extracted raster values for presence and background points:

abs <- cbind(pa=0, bio1=extract(predictors[[1]], bg))
prs <- cbind(pa=1, bio1=extract(predictors[[1]], occ))

and use these data to build the maxent model

x <- data.frame(rbind(prs, abs))
m <- maxent(x[,2,drop=F], p=x[,1,drop=F] )

p <- predict(predictors, m)
plot(p)
Robert Hijmans
  • 40,301
  • 4
  • 55
  • 63