Automatically partitions data into bins for model training and testing, executes ecological niche models (ENMs) across a range of user-defined settings, and calculates evaluation metrics to help achieve a balance between goodness-of-fit and model complexity.
| Package: | ENMeval |
| Type: | Package |
| Version: | 0.3.0 |
| Date: | 2018-07-23 |
| License: | GNU 3.0 |
The ENMeval package (Muscarella et al. 2014) (1) automatically partitions data into training and testing bins using one of six methods (including several options for spatially independent partitions as well as user-defined bins), (2) executes a series of ENMs using Maxent (Phillips et al. 2006, Phillips et al. 2017) with a variety of user-defined settings (i.e., feature classes and regularization multipliers), conducting k-fold cross validation, and (3) calculates multiple evaluation metrics to aid in selecting model settings that balance model goodness-of-fit and complexity (i.e., "model tuning" or "smoothing").
ENMevaluate is the primary function of the ENMeval package, and multiple other functions highlighted below are called when it is run. The six options for partitioning occurrence data into training and testing (i.e., calibration and evaluation) bins are: n-1 jackknife, random k-fold, user-specified bins, and three explicit methods of masked geographically structured k-fold partitioning (see: get.evaluation.bins). After model training, these bins are used to calculate five metrics of model performance for each combination of settings: model discrimination (AUC of test localities), the difference between training and testing AUC, two different threshold-based omission rates, and the small sample-size corrected version of the Akaike information criterion (AICc), the latter using the unpartitioned dataset. A model prediction (as a raster layer) using the full (unpartitioned) dataset is generated for each combination of feature class and regularization multiplier settings. Similarity of these models in geographic space (i.e., "niche overlap") can be calculated to better understand how model settings change predictions (see calc.niche.overlap). The results of ENMevaluate are returned as an object of class ENMevaluation-class. A basic plotting function (eval.plot) can be used to visualize how evaluation metrics depend on model settings.
- As of version 0.3.0 -
The default ENMevaluate runs the the Maxent algorithm by calling the maxnet package (Phillips et al. 2017) instead of the previous implementation that relied on the 'maxent.jar' Java program called by the dismo package. This is controlled by the new argument, algorithm='maxnet'. A major advantage of this change is that it removes the reliance on Java and the rJava package, which is great but can sometimes cause confusing problems on different computers. Our team has done some fairly extensive testing to ensure this implementation gives the expected results but the maxnet implementation is relatively new (at the time of writing this) and we encourage users to scrutinize their results.
There are some differences between the 'maxnet' and 'maxent.jar' algorithms that may lead to slight numeric differences in the results (at least when hinge feature classes are used). See Phillips et al. (2017) and Phillips (2017) for more details. Additionally, the 'maxnet' algorithm does not provide information on variable importance (from the var.importance() function) because of differences in the underlying models. Users can still choose to use the 'maxent.jar' implementation by setting algorithm='maxent.jar' in the ENMevaluate function (also see note below).
- As of version 0.2.0 -
ENMevaluate includes an option for parallel computing. Setting parallel = TRUE can significantly speed up processing time, particularly for large analyses. For very small analyses, it may actually take longer than running with parallel = FALSE.
Hijmans, R. J., Phillips, S., Leathwick, J. and Elith, J. 2011. dismo package for R. Available online at: https://cran.r-project.org/package=dismo.
Muscarella, R., Galante, P. J., Soley-Guardia, M., Boria, R. A., Kass, J. M., Uriarte, M., and Anderson, R. P. 2014. ENMeval: An R package for conducting spatially independent evaluations and estimating optimal model complexity for Maxent ecological niche models. Methods in Ecology and Evolution, 5: 1198-1205.
Phillips, S. J. 2017. maxnet package for R. Available online at: https://CRAN.R-project.org/package=maxnet.
Phillips, S. J., Anderson, R. P., Dud<U+00ED>k, M., Schapire, R. E. and Blair, M. E. 2017. Opening the black box: an open-source release of Maxent. Ecography, 40: 887<U+2013>893.
Phillips, S. J., Anderson, R. P., and Schapire, R. E. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190: 231-259.
Phillips, S. J., and Dudik, M. 2008. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31: 161-175.
maxnet in the maxnet package
maxent in the dismo package