Purell 11Sep2006 Desktop Garp

8/12/2019 Purell 11Sep2006 Desktop Garp

1/36

The World According to GARP *or

Using presence points to map a species using theGenetic Algorithm for Rule-set Production (GARP)

Melora K. PurellTropical Conservation Biology and Environmental Science

Masters Program

University of Hawaii - Hilo

*with apologies to Mr. John Irving


2/36

Overview

Some niche modeling theory Primer for using Desktop GARP

Practical advice and examples


3/36

Niche Modeling

Where is species X likely to be found?

Applications:Rare/threatened/endangered species

Alien/exotic/invasive species


4/36

Niche Modeling

Climatic or environmentalparametersPoints of known presence

Nichemodel/map


5/36

Environmental Envelopes

Fuzzy EnvelopeCrisp Envelope vs.

Parameters are definedrelative to each other.

Example - The species is found inareas that are:

Less than 500 m, if theannual rainfall is >250 mm

and the slope is < 5%.Greater than 500m, if thereis koa forest present andannual rainfall is > 300 mm.

Exact parameters defined foreach environmentalvariable.

Example - The species is foundin areas that are always:

Above 500 m elevation

Less than 3% slope

Mesic koa and/or ohiaforest


6/36

GARP

GARP is a genetic algorithm that creates ecologicalniche models for species.

These fuzzy envelope models describe environmentalconditions under which the species should be able tomaintain populations.

GARP uses a set of point localities where the species isknown to occur and . . .

A set of geographic layers representing theenvironmental parameters that might limit the species'capabilities to survive.

*from the Desktop GARP website


7/36

Genetic algorithm is a search technique used in

computing to find true or approximate solutions to

optimization and search problems. It involvesmultiple iterations of rule-setting in order to best fit

the samples to the desired outcome.

GARP is creating rules about theenvironment that will make all thepoints fit into the predicted range.


8/36

GARP inputs

1. Create MS Excel file with presencelocations

2. Choose GIS layers to use ensure allare same projection and same spatial

resolution ( ArcMap )3. Clip GIS layers in ArcView using GARP

Datasets extension4. Create a GARP Dataset using Dataset

Manager


9/36

GARP inputs

Data points in Excel

No te:

(decimald egr ees)


10/36

GARP inputs

Environmental Layers

Rasters/gridsAll layers must have

same projectionSpatial resolution andspatial extent dependon your study area

Kohala Mountain

10m resolution

Windward aspect raster

Rainfall grid

Main Hawaiian islands100m resolution


11/36

GARP inputs

Making the grids that GARP likes . . .

1. In ArcView , turn on

Spatial Analyst extensionand add grids to the frame

Also add one polygon asa mask for the entirestudy area


12/36

GARP inputs Making the grids that GARP likes . . .

2. Use the GARP Datasetsextension to clip grids and

resample to the same cellsize.

And after a few more questions. . .


13/36

GARP inputs

Thats the ASCIIraster grid that can

be input into GARP!

Voila!


14/36

GARP inputs

Creating a GARP Dataset using Dataset Manager

1. Put all your ASCII grids into one folder2. Copy one of your grids in the folder and

rename it mask.asc


15/36

GARP inputs Creating a GARP Dataset using Dataset Manager

3. Load your layers by choosing one

grid in your folder

4. Type in some

dataset info thensave into thesame folder

Now you have a Dataset of Environmental Layersto use in the Desktop GARP program.


16/36

Running GARP

Upload Excel File


17/36

Running GARP

Load Dataset


18/36

Running GARP Make decisions about

runs, rules and points


19/36

Choose output formatand file locations

Running GARP


20/36

Running GARP Run the program!


21/36

GARP outputs

Maps as ASCII grids


22/36

GARP outputs

Statistics as Excel file

GA


23/36

GARP outputsSome key stats. . .

The total number of cells that werenot covered by the rules

GARP


24/36


The % of the land area that waspredicted as presence

GARP


25/36


Error: the percentage of presence points thatdid not fit in the predicted presence area

M ki ti l


26/36

Making an optimal map

Choose 20 runs that have the best fit for% predicted area presence (commission)and the lowest omission error, as well asno non-predicted areas.

Add those 20 maps together in GIS tocreate an optimal map with cells valuesthat range from 0 to 20.

Areas with high cell values have a highprobability of being accurate.


27/36

Examples of Optimal GARP maps


28/36




29/36


58 presencepoints from field

samplingElevation, Slope,Windward aspectand Rainfall layers

Masters thesisproject andpotential use ininvasive speciesmanagement



30/36




31/36


550 presence

points frommultiple sources

Elevation, Slope,Windward aspectand Rainfall layers

Cost-benefitanalysis forbiological control

Thing to Consider with GARP


32/36


Presence Points

must be spatially independent .

need to represent the full extent of the speciesacross the range of the environmental variables:

- very difficult with rare species

- invasive species that are still spreading - field sampling vs. museum records - output range



33/36


GIS layers

must include all available layers that mayinfluence species distribution

must be continuous/numerical variables no

classes, categories, or ranksmust be reliable : garbage in, garbage out!



34/36


Some Important Strengths

GARP works with whatever GIS layers youhave the output spatial resolution andextent depend upon the GIS layers you use

The output from GARP is spatially complete

It is easy to run 100s of GARP models, sooutputs can be averaged, eliminatingpotential machine learning errors

Wang and Wang, 2006



35/36


One Major Weakness

The GARP rule sets are true black boxes. Theuser can infer the rules from the output, but notin sufficient detail to make strong statementsabout ecological mechanisms that lead to thepredicted distributions.

With great appreciation to . . .


36/36

With great appreciation to . . .My Spatial nalysis gurus

Lisa Canale, ITER program manager

Cynthia Phillips, GIS specialist

Sun Park, PhD., UH-Hilo, Dept. of Geography My graduate committee :

Becky Ostertag, PhD., UH-Hilo, Dept. of Biology

Julie Denslow, PhD., USDA Forest Service

Rosemary Sherriff, PhD., UH-Hilo, Dept. of Geography

Sheldon Furutani, PhD., UH-Hilo, CAFNRM

Research assistants/hiking companions :

Shenandoah Marr, Randee Tubal, Steve Souder

and. . all those that helped me through the Desktop GARP list-serve

Thank You!

Purell 11Sep2006 Desktop Garp

Documents