Maxent software for species habitat modeling

Currently available version: 3.2.1 (see new features below).

Use this site to download software based on the maximum-entropy approach for species habitat modeling.  This software takes as input a set of layers or environmental variables (such as elevation, precipitation, etc.), as well as a set of georeferenced occurrence locations, and produces a model of the range of the given species.

Further description of this approach can be found in:

• Steven J. Phillips, Miroslav Dudík, Robert E. Schapire.
  A maximum entropy approach to species distribution modeling.
  In Proceedings of the Twenty-First International Conference on Machine Learning, pages 655-662, 2004.
  pdf

• Steven J. Phillips, Robert P. Anderson, Robert E. Schapire.
  Maximum entropy modeling of species geographic distributions.
  Ecological Modelling, 190:231-259, 2006.
  (datasets used in this paper are available below)
  pdf

Terms of use:   This software may be freely downloaded and used for all educational and research activities.  This software may not be used for any commercial or for-profit purposes.  The software is provided "as-is", and does not come with any warranty or guarantee of any kind.  The software may not be further distributed.  By clicking on the download button below, you agree to these terms.

Please provide your name, institution and email address prior to downloading.

Discussion Group

There is a google discussion group for users of this software at http://groups.google.com/group/Maxent.  Please sign up for this group for discussions and questions about the software, and also to get information about updates, bugs, etc.

Tutorial

A tutorial explaining how to use this software is provided here as a word document.  The data that goes along with this tutorial is available from this zip file.  A Spanish translation of the tutorial, provided by Paolo Ramoni-Perazzi, is available here.

Datasets

Here are the datasets used in the Ecological Modeling paper referenced above:

• coverages.zip - the coverages we used in modeling.
• ipcc.zip - the raw ipcc data, as dowloaded from the IPCC Data Distribution Centre.  Also the annual and monthly variables we used for our study, extracted from the raw ipcc data and converted into world-wide coverages in .asc format.
• samples.zip - the training and testing sample localities we used.

What's new in Version 3.2.1

• Response curves are now in logistic (probability) space, rather than exponent (linear) space, so they're easier to interpret
• There's a second set of response curves and accompanying text, to show response of the species to variables individually rather than marginal responses
• The code that makes clamping pictures has been rewritten. The clamp amount now shows what effect clamping can have on the logistic prediction, rather than the exponent.
• There is support for .bil format
• In pictures of predictions, the legend now goes in the corner with least overlap with non-NODATA
• In .csv files, we detect ";" as separator (and in that case, also "," in place of "."), as made by French Excel
• File browsers now show .CSV files in addition to .csv
• The dontcache and responseCurvesExponent command-line arguments are described in help file.
• There is no longer a complaint about extra columns in SWD input files if those columns are deselected layers
• There is a warning if occurrence data is not in SWD format when background is in SWD format
• Fixed bug in writing _samplePredictions.csv file: training predictions were sorted by prediction value, but their lat/long weren't staying with them
• Fixed bug in .html: was writing wrong feature categories when auto features was overridden
• Fixed bug in project: clamping wasn't always keeping response constant outside the training range, e.g. for quadratic features

New features in Maxent version 3.0

• Variable contributions: The html output has a table that gives the percentage contribution of each variable to the Maxent model. This gives a new way (in addition to the jackknife) of interpreting the influence of each variable on the model.
• Logistic output format: There is a new default output grid format, called "logistic". Logistic values range from 0 to 1, and model the probability of presence of the species, assuming that sampling effort is similar to the effort expended to obtain the training samples. See the help button for more details.
• Clamping picture: Whenever a model is projected onto a different set of environmental variables, a picture is produced showing where clamping has occurred, colored according to the amount of clamping.
• DIVA data types. Compatibility problems have been resolved, so Maxent can now successfully read and write DIVA's .grd format.
• Increased categorical regularization. The default regularization of categorical features was causing over-fitting for some data sets, especially those with larger numbers of categories and training presences. Categorical regularization has therefore been increased, but is still adjustable using the "beta_categorical" flag.
• Better memory usage. Larger grids of environmental data can now be used without running out of memory.
• Faster file handling. By default, any .asc file used during training or projection will have a cached compressed version saved in a subdirectory called "maxent.cache". The cached version is then used to speed up future runs.
• ROC plot axis labels have been fixed.
• Bug fixed in jackknifing: previously, test performance of leave-one-out runs was too low when running with lots of training presences.
• New options to control clamping during projection. The flag "fadebyclamping" causes the prediction at a point to decrease when there is clamping there, with the amount of decrease determined by the amount of clamping. The "dontextrapolate" option causes the prediction to be zero wherever there is clamping.
• More data and details added to maxentResults.csv file.