A new tool for fundamental niche modelling
Renato De Giovanni
Centro de Referência em Informação Ambiental, CrIA
openModeller
• Definition
• History
• Motivation and features
• Design
• Interfaces and additional tools
• Algorithms
• Future plans
Definition
openModeller is an open source C++ library completely
dedicated to static spatial distribution modelling.
Applications
Biology: Fundamental niche modelling.
Geology ?
Demography ?
Others ?
openModeller’s history
apr 2003: Initial design of a new modelling environment at CRIA as
a natural consequence of previous experiences with
other tools (DesktopGarp).
oct 2003: First working prototype as part of the speciesLink
project (Fapesp).
dec 2003: Released all source code (sourceforge).
feb 2004: Partnership with BDWorld (CSM / GRID component).
apr 2004: Partnership with University of Kansas (GARP / BTRA).
jan 2005: Released first graphical user interface (Tim Sutton &
Peter Brewer).
may 2005: Basis of a new thematic project funded by Fapesp (4y).
Main Motivation
Facilitate and speed up modelling tasks, offering at the same time
a homogeneous environment to carry out experiments with
different algorithms.
Main features
•
•
•
•
•
•
•
•
Platform independent.
Enables the existence of multiple interfaces on top of it.
Accepts different formats of georeferenced maps.
Accepts different coordinate systems and projections for each
map and for the whole set of occurrence points.
Accepts different cell sizes and extents for each map.
Allows the different algorithms to use exactly the same input and
the same working environment, therefore enabling fair
comparison between all results.
Isolates algorithm logic from other issues related to maps,
georeferencing, input and output formats, etc.
Offers a collaborative and transparent environment for all
interested developers.
Architecture overview
pluggable
algorithms
interfaces
Console
API
API
SOAP
server
Bioclim
open
Modeller
SWIG
wrapper
GARP
CSM
others...
drivers
(GDAL, proj4, etc)
others...
points
(diff. coord systems)
maps
(diff. formats)
Interfaces and additional tools
• Command line / Console suite
– om_console
– om_viewer (X11)
– om_niche (X11)
• SWIG wrapper
– Python
• SOAP interface (prototype server and sample client)
• Web interface
• Graphical User Interface (Linux, Windows, Mac OS)
Console interface
>> om_console request.txt
WKT Coord System =
Species file =
Species =
Map =
Mask =
Output map =
Output mask =
Output format =
Output file =
Algorithm =
Parameter =
Console interface
Console interface
Tool for visualizing maps
>> om_viewer -r request.txt
Tool for visualizing models
>> om_niche request.txt
Web Interface
Web Interface
Graphical User Interface
Graphical User Interface
Graphical User Interface
Development of algorithms
• Metadata definitions (name, version, author, description,
bibliographic references, parameters).
• Method to initialize the algorithm.
• Method to generate the model.
• Method to calculate the probability of occurrence given a
certain vector of environmental values.
Algorithms: Building models
Sampler gives the algorithm vectors
of environmental values from a set of
occurrence points:
Ex: [20˚, 115 mm], [22˚, 100 mm]
open
Modeller
Algorithm
API
Algorithm uses the values to
build a distribution model and stores
an internal representation of it.
Algorithms: Generating distribution maps
For each cell of the resulting map, openModeller
asks the probability of presence sending the
vector of environmental values as a parameter.
Ex: probability for [30˚, 90 mm] ?
open
Modeller
Algorithm
Algorithm answers with a probability of presence.
Ex: prob = F( [30˚, 90 mm] ) = 0.8
Algorithms
• Bioclim
• Climate Space Model (Broken Stick cutoff method)
• GARP (incl. best subset procedures)
• Distance algorithms
– Distance to average
– Minimum distance
Algorithms - Bioclim
• Assumes normal distribution for each environmental variable.
• Envelopes are represented by the interval [m - c*s, m + c*s],
where 'm' is the mean; 'c' is the cutoff parameter; and 's' is the
standard deviation.
• Besides the envelope, each environmental variable has
additional upper and lower limits taken from the maximum and
minimum values related to the set of occurrence points.
• Points are classified as: suitable, marginal or unsuitable.
fig. 1: cutoff = 0.674
fig. 2: cutoff = 0.99
Algorithms - GARP
• Genetic Algorithm for Rule-set Production: models are
represented by a set of rules generated by a genetic algorithm.
• Non-deterministic: produces a different model each time the
algorithm is run.
fig. 1: model 1
fig. 2: model 2
fig. 3: model 3
Algorithms – GARP with Best subsets procedure
• Runs several GARP models and chooses the best ones
according to omission and commission erros.
• Resulting model is the overlapping of models that were selected
in the previous step.
fig. 1: sample model
Algorithms – distance to average
• Normalizes environmental values and parameter.
• Calculates the mean point in environmental space considering
all presence points.
• Probabily of presence is proportional to the Euclidean distance
from the average point (linear decay).
• Parameter determines the maximum accepted distance.
fig. 1: parameter = 0.1
fig. 2: parameter = 0.3
Algorithms – Minimum distance
• Normalizes environmental values and parameter.
• Probabily of presence is proportional to the Euclidean distance
from the closest point (linear decay).
• Parameter determines the maximum accepted distance.
fig. 1: parameter = 0.05
fig. 2: parameter = 0.1
Use case – Byrsonima subterranea Brad. & Markgr.
= original point
= 4 new points
Scope issues & known limitations
•
Works only with static models – dynamic modelling is currently
outside the scope of this tool.
•
None of the algorithms can handle categorical maps (although the
library is already prepared to deal with them).
•
None of the algorithms can handle absence points (except GARP), and
none of the high level interfaces is prepared to receive absence points
as an additional parametrer.
•
Produces only bi-dimensional maps – not prepared to produce models
in three dimensions (especially considering aquatic environments).
•
Still not sufficiently documented!
•
Still not sufficiently tested!
Future plans
• Implementation of other algorithms: neural nets, cellular
automata, GLM, GAM, GRASP, Domain…
• Development of new components to help on pre-processing
and post-analysis.
• Finalize Web and SOAP interfaces.
• Develop SWIG interfaces for other programming languages.
• Improve documentation.
• Implementation of a new and advanced graphical user
interface.
New version of the graphical interface
Institutions & People
Mauro Muñoz
Renato De Giovanni
Tim Sutton
Peter Brewer
Ricardo S. Pereira
Kevin Ruland
Jens Oberender
Thank you
http:// openmodeller . sf . net
renato (at) cria . org . br
Descargar

Slide 1