The development version of the ‘ modEvA‘ R package, currently available on R-Forge, computes the Boyce index and a range of other metrics that evaluate model predictions against presence/absence or presence/background data, including the area under the ROC or under the precision-recall curve ( AUC), sensitivity, specificity, TSS, Cohen’s kappa, Hosmer-Lemeshow goodness-of-fit, Miller calibration statistics, and several others. Whether or not the computation requires an explicit identification of the absences, or just presence points and a map of the entire study area, simply depends on how each method or metric is implemented in each particular software package. So, they need the same type of data that presence-absence methods need. “non-presence”) pixels, and can’t produce proper predictions without non-presence pixels - you can also try for yourself. The same applies to presence-background modelling methods like Maxent and ENFA, which use both the presence and the remaining (i.e. both presences and (pseudo)absences, or the Boyce index cannot be computed: try using a raster map with values only in pixels with presence and see what happens. We need both pixels with and pixels without presence records, i.e. the complementary proportion) must be… non-presences, which are equally necessary for the computation of this index (as for any other model evaluation metric). But, if there’s a proportion of presences in each class, then the rest of the class (i.e. It measures the predicted-to-expected ratio of presences in each class of predicted values, i.e., whether classes / regions with higher predicted values have indeed higher proportions of presences than expected by chance. 2002) is often described as a presence-only metric for evaluating the predictions of species distribution (or ecological niche, or habitat suitability) models (e.g. Thanks are due to Arpat Ozgul and François Guilhaumon, who helped me through the steep part of the R learning curve (to me it was actually more like an overhang). Comments and suggestions are very welcome, although a quick response cannot be guaranteed! Leave a comment here if you use these functions in publications - this way I can cite you in the blog, track how people are using the tools, and justify the time and effort dedicated to them. P lease note that these tools are experimental (like R, they come with absolutely no warranty) and occasionally edited with corrections or improvements, so preferably check back for the latest version before you use a function. Most of them are now being included in R packages - please cite the appropriate package (mentioned in the blog post featuring each function) if you use these tools, which are released under a Creative Commons Attribution license. This website provides a series of free software tools, mostly written in R, that can be used for various tasks related to the analysis of species’ distribution and ecology (although many of them can be applied to other purposes).
0 Comments
Leave a Reply. |