Dear Dr. Fitzpatrick,

I am working with the gdm package and encountered an issue when trying to run PCA on raster data. Initially, I attempted:

terra::prcomp(transRasts, maxcell = 5e5)

but received the errors:

Error: 'prcomp' is not an exported object from 'namespace:terra'

Then I ran simply prcomp() on the SpatRaster object (transRast) like in the tutorial, and I got the error below:

Error in svd(x, nu = 0, nv = k) : infinite or missing values in 'x'
In addition: Warning message:
In prcomp.default(transRasts, maxcell = 5e+05) :
extra argument ‘maxcell’ will be disregarded

I realized this might be due to the class of the object I was passing — prcomp expects a numeric matrix or data frame, not a SpatRaster.

To work around this, I extracted only the PC axis bands from the raster (excluding xCoord and yCoord), converted them into a matrix with values(), cleaned NAs, and then ran stats::prcomp successfully:

names(transRasts)

bands <- c("PC1","PC2","PC3","PC5") # exclude XCoord and YCoord to perform PCA

rts <- transRasts[[bands]]

M <- values(rts, mat = TRUE)
M <- M[complete.cases(M), , drop = FALSE]

pca_model <- stats::prcomp(M, center = TRUE, scale. = TRUE)

pcaRast <- terra::predict(transRasts, pca_model, index = 1:3)

This works, and I got the final dissimilarity map. However, I wanted to confirm whether this approach makes sense in the context of gdm. In particular, by excluding the XCoord and YCoord layers from the PCA, am I ignoring geographic distance effects in the dissimilarity map? Or is it correct to leave coordinates out of the PCA?

Are there any alternative to circumvent the error with prcomp? I tried other pca functions, but then they do not work with terra::predict() or they do not accept SpatRaster**

Any clarification or guidance you can provide would be greatly appreciated.

Best regards,