In wallaceEcoMod/wallace: A Modular Platform for Reproducible Modeling of Species Niches and Distributions

Component: Partition Occurrence Data

ORIENTATION

Evaluating the accuracy of a predictive model usually requires testing it on independent data; however, as truly independent data are difficult to obtain for most biodiversity datasets, researchers often partition a single dataset into subsets. Occurrence data are partitioned into two categories: 1) those used to make the model (i.e. "training" or "calibration"), and 2) others that are withheld for evaluating it (i.e. "testing" or "evaluation" data) (Guisan and Zimmermann 2000; Peterson et al. 2011). One widely-used technique, k-fold cross-validation, involves partitioning the full dataset into k groups, then iteratively building a model using all groups but one and testing this model on the left-out group. This iterative process results in k models. Evaluation statistics can be averaged over these k models to help in model selection, leading to the production of a full model that includes all the data if desired.

There are myriad particular ways to partition data for niche/distributional models, but they can be split into two main categories that are offered in 1) Module: Non-spatial Partition and 2) Module: Spatial Partition.

REFERENCES

Guisan A., & Zimmermann N.E. (2000). Predictive habitat distribution models in ecology. Ecological Modelling, 135(2-3), 147-186. DOI: 10.1016/S0304-3800(00)00354-9

Peterson, A.T., Soberón J., Pearson, R.G., Anderson, R.P., Martinez-Meyer, E., Nakamura M., & Araújo, M.B. (2011). Evaluating Model Performance and Significance. In: Ecological Niches and Geographic Distributions, Princeton, New Jersey: Monographs in Population Biology, 49. Princeton University Press. DOI: 10.23943/princeton/9780691136868.003.0009

wallaceEcoMod/wallace documentation built on March 24, 2024, 5:15 p.m.