MaxEnt
Description
This page explains how to use the MaxEnt Algorithm on the Statistical Manager with the i-Marine portal.. The algorithm is hosted by the D4ScienceAn e-Infrastructure operated by the D4Science.org initiative. e-InfrastructureAn operational combination of digital technologies (hardware and software), resources (data and services), communications (protocols, access rights and networks), and the people and organizational structures needed to support research efforts and collaboration in the large. that supports i-Marine. It is a Maximum-Entropy model for species habitat modeling, based on the implementation by Shapire et al. v 3.3.3k at Princeton University.
In this adaptation the software accepts a table following the Species Product Discovery service model of i-Marine and a set of environmental layers in various formats (NetCDF, WFSWeb Feature Service, WCSWeb Coverage Service, ASC, GeoTiff) via direct links or GeoExplorer UUIDs.
The user can also set the bounding box and the spatial resolution (in decimal degrees) of the training and the projection. The application will adapt the layers to that resolution if this is higher than the native one.
The output is made up of the following components:
- a thumbnail map of the projected model,
- the ROC curve,
- the Omission/Commission chart,
- a table containing the raw assigned values,
- a threshold to transform the table into a 0-1 probability distribution,
- a report of the importance of the used layers in the model,
- ASCII representations of the input layers to check their alignment.
Starting from this output, other processes of the Statistical Manager can be later applied to the raw values, for example to produce a GIS map (e.g. the "Statistical Manager Points to Map" process). Eventually, results can be shared with other participants to the e-InfrastructureAn operational combination of digital technologies (hardware and software), resources (data and services), communications (protocols, access rights and networks), and the people and organizational structures needed to support research efforts and collaboration in the large. using the i-Marine workspace.
Demo Video
Here is a demonstration of the usage of the MaxEnt algorithm on the Statistical Manager: http://goo.gl/TYYnTO
Inputs
The inputs of the MaxEnt procedure are the following:
Parameter Name | Description | Example |
---|---|---|
SpeciesName | The name of the species to model and the occurrence records refer to. If the name is not important this can be a generic string. | Latimeria chalumnae |
MaxIterations | The number of learning iterations of the MaxEnt algorithm | 1000 |
DefaultPrevalence | A priori probability of presence at ordinary occurrence points. Ref. Santika 2010 | 0.5 |
OccurrencesTable | A geospatial table containing occurrence records, following the template of the Species Products Discovery datasets. See section below for more details. | LatimeriaPointsTable |
LongitudeColumn | The table column containing longitude values | decimallongitude |
LatitudeColumn | The table column containing latitude values | decimallatitude |
Z | Value of Z. Default is 0, that means environmental layers processing will be at surface level or at the first avaliable Z value in the layer | 0 |
TimeIndex | Time Index. The default is the first time indexed in the input environmental datasets | 0 |
XResolution | Model projection resolution on the X axis in decimal degrees | 1 |
YResolution | Model projection resolution on the Y axis in decimal degrees | 1 |
Layers | The list of environmental layers to use for enriching the points. Each entry is a layer Title or UUID or HTTP link. See section below for further details. | https://dl.dropboxusercontent.com/u/12809149/wind1.tif |
SPD Input Format
The algorithm needs a table to be uploaded on the Statistical Manager. To use the upload facilities, refer to the Statistical Manager Tutorial page. The uploaded table should follow the Species Product Discovery (SPD) template and can be generate by the SPD service:
Field name | Format |
---|---|
institutioncode | string |
collectioncode | string |
catalognumber | string |
dataset | string |
dataprovider | string |
datasource | string |
scientificnameauthorship | string |
identifiedby | string |
credits | string |
recordedby | string |
eventdate | timestamp without time zone |
modified | timestamp without time zone |
scientificname | string |
kingdom | string |
family | string |
locality | string |
country | string |
citation | string |
decimallatitude | double precision |
decimallongitude | double precision |
coordinateuncertaintyinmeters | string |
maxdepth | double precision |
mindepth | double precision |
basisofrecord | string |
The fields could be also empty, except for the decimallatitude and decimallongitude fields. This allows to apply MaxEnt also to other domains than species distributions modelling. Note that the template closely follows the Darwin Core format.
Feeding the algorithm with Input Maps
In the layers box users can insert links to maps that will be used to associate environmental values to species occurrence records. The + button allows to insert a new layer.
Input Examples
Input from i-Marine GeoExplorer
using the i-Marine GeoExplorer application:
- search for an environmental layer (e.g. temperature)
- click on one of the layers found by the search
- in the "Summary Layer Info" panel on the right scroll down and select the Metadata UUID string (e.g. cd048cb5-dbb6-414b-a3b9-1f3ac512fbff)
- paste the UUID in the layers box in MaxEnt
Input from a WFSWeb Feature Service link
MaxEnt can import WFSWeb Feature Service links residing either on a GeoServer or on a MapServer. The server must be able to produce maps details in json format. In this case you can insert the direct WFSWeb Feature Service link in the layers box, without specifying the bounding box.
Please, use EPSG:4326 as projection.
Input from a WCSWeb Coverage Service link
You can input a direct WCSWeb Coverage Service link.
E.g.:
Please, use EPSG:4326 as projection.
Input from a NetCDF-GRID file
You can input the OpenDAP link to a NetCDF file, only if this contains one single dimension layer.
E.g.:
ASC ESRI-GRID files
You can input a direct http link to an ESRI-GRID file, even using local common publishing tools (e.g. dropbox).
E.g.:
http://thredds.research-infrastructures.eu/thredds/fileServer/public/netcdf/ph.asc
https://dl.dropboxusercontent.com/u/12809149/layer1.asc
GeoTiffs
Http links to GeoTiff files are allowed, even using local common publishing tools (e.g. dropbox). E.g.:
https://dl.dropboxusercontent.com/u/12809149/wind1.tif
Contacts
For questions and bug alerts use this form.