POCplot: Presence-only smoothed calibration plot

In peterbat1/fitMaxnet: Fit MaxEnt Niche Models Using maxnet

Presence-only smoothed calibration plot

Description

Produce a smoothed calibration plot for a presence-only model by the method of Phillips and Elith

Usage

POCplot(
  pred,
  back,
  linearize = FALSE,
  capUpperValues = TRUE,
  title = "Calibration plot",
  filename = NULL
)

Arguments

pred	Numeric. Vector of model predictions at occurrence points.
back	Numeric. Vector of model predictions at background points.
linearize	Logical. Should the logistic transform be applied to model predictions? The default is FALSE; see Details.
capUpperValues	Logical. Should computed values in smoother fit be clamped to 1?
title	Character. Title to appear in the plot. Default is a generic "Calibration plot".
filename	Character. Full path to a file into which the calibration plot will be written. Default (NULL) will cause the plot to appear on the standard graphics device.

Details

This is an adaptation of code published by Phillips and Elith (2010. POC plots: calibrating species distribution models with presence-only data. Ecology 91:2476–2484).

The original script set the default value of the parameter linearize to TRUE which applies a logistic transformation to the y-values in the POCplot. In development trials, this frequently produced numerical failures when values approaching 1 where transformed. It is perhaps intended to transform 'raw' values supplied by fitted ENMs to the logistic scale. However, the most frequent output from MaxEnt and MaxEnt-like models is already transformed onto a logistic or complementary log-log ('cloglog') scale. Inadvertent double application of a linearising transformation will cause numeric failures and severely distorted calibration plots. So, the default for linearize is set to FALSE.

Value

A named list with the follow elements:

predd

Data frame of results from the smoothing process; columns are: x = predicted probability of presence, y = (modelled) probability of presence, se = std error of the y-value at each x-value

mse

A numeric object giving the Mean Squared Error between the line of equality between probability of presence (y) and predicted probability of presence (x) and the y-values

peterbat1/fitMaxnet documentation built on Sept. 17, 2024, 10:50 p.m.