Handling Spatial Autocorrelation and Multicolinearity when modelling altitude of forestline
Earlier I asked this question on GRASS ML , but maybe it is more suited here:
Zofie and me are trying to mode altitude of forest line in Fensocandia using GRASS and R.
We have a couple of 100k pixels that we assume to represent forest line and now want to model their altitude with explanatory variables (latitude, terrain, temperature, precipitation and the like). In a next step we want to use projected data (climate scenarios) in the model in order to predict possible effects of climate change. But now we are a bit unsure about what modeling technique to use.
Our data to be modeled is zero-inflated (from 0 - ~1150m) with a significant amount of spatial autocorrelation. And also the explanatory variables are spatially auto-correlated and have a lot of collinearity.
We have been looking at (amongst others):
Regression kriging, but we have doubts that R will be able to handle the amount of data (even on a highmem server), pluss that we are unsure if we can replace current with future climate data in such a model
GLS, but also here we face problems with excessive resource consumption in R for e.g. accounting for spatial autocorrelation and on a subsample residuals still showed spatial autocorrelation.
Can anyone recommend other types of models or R packages we should look at, that can handle spatial autocorrelation and multicolinearity?
We would be glad for any hint also on where to look for more information (articles, textbooks…)?