3

I'm currently working on Species Distribution Models and particularly with the framework developed by Ian Renner under the ppmlasso package in R. I have some questions about this framework which have been design in order to be used by practitioner without need of specialist guidance (cf. Renner et al. 2015 + Supplementary material).

If we take the data used by Renner et al. 2015:

library(spatstat)
library(ppmlasso)
load("Eucalyptus sparsifolia Atlas 2012.RData") #Contains X and Y
load("Quad1000.RData")
sp.xy=data.frame(X,Y)
ppm.form = ~ poly(FC, MNT, MXT, Rain, degree = 2, raw = TRUE)+ poly(sqrt(D.Main), sqrt(D.Urb), degree = 2, raw = TRUE)
scales = c(0.5, 1, 2, 4, 8, 16)
findres(scales, formula = ppm.form, sp.xy = sp.xy, env.grid = quad)

Fitting a regularisation path of point process models

quad.1k = sample.quad(quad, 1)
ppm.fit = ppmlasso(ft.int, sp.xy = sp.xy, env.grid = quad.1k,sp.scale = 1, criterion = "nlgcv")

Question 1:

how to obtain the map of predicted intensity? Because the equivalency function in spatstat "predict" doesn't work with ppmlasso objects, and the predict.ppmlasso function doesn't return "im" object but "large matrix"

Diagnostic of the model

In its main text (figure 3c and d) Renner drawn the "Lurking variable plot for x (and then y) coordinate Pearson residuals" with Monte Carlo simulation envelopes for an area-interaction model (Gibbs model).

Question 2:

How can we make these diagnostic plots from Gibbs models (as Ian Renner did) and with the Monte Carlo simulation envelopes? 

diagnose.ppmlasso(ppm.fit, which = "smooth", type = "Pearson", labcex = 1)
# OK
diagnose.ppmlasso(ppm.fit, which = "x", type = "Pearson", compute.sd = TRUE) #doesn't work cause it's designed for ppm and not for Gibbs model
diagnose.ppmlasso(ppm.fit, which = "y", type = "Pearson", compute.sd = TRUE)
#doesn't work cause it's designed for ppm and not for Gibbs model

Question 3:

Is there a mean to acceed to the response curves, linking the intensity with each of the environmental variables implemented in the model?

Thank you very much for helping

Clem H
  • 31
  • 1
  • This is a nicely written question which deserves feedback, but you may clarify a small detail: Where is the data from Renner et al. available? Other than that I think the question is so specific to the Renner et al. methodology that you may try to email Ian Renner directly and ask if he could find time to answer any of this (preferably here on the site). – Ege Rubak Jun 28 '17 at 09:47
  • Thank you for your answer. Data from Renner are available at : https://github.com/wrbrooks/confounding/find/master (and searching for "Eucalyptus sparsifolia Atlas 2012.RData" and "Quad1000.RData" in the list). I'm trying to obtain the information from him but i'm short in time to manage this work that is why I'm using at the same time stakeoverflow. I thought that other users could be interested in this purpose and if Renner give me an answer I would publish it here. – Clem H Jun 28 '17 at 09:55
  • @ClemH The object `ft.int` is not defined in your code – Marco Sandri Jun 28 '17 at 11:50
  • Yes it is a mistake, you have to replace ft.int by ppm.form, sorry about that ! – Clem H Jun 28 '17 at 13:42
  • This is a question about the `ppmlasso` package. Unfortunately there is no `ppmlasso` tag in stackoverflow yet. – Adrian Baddeley Aug 10 '20 at 04:03

0 Answers0