GARPSummary
org.ecoinformatics.seek.gis.java_gis.GARPSummary

This actor examinesof the values in individual pixels in an ascii grid file and summarizes those values.

There are three inputs; 'input'is an ASCI grid (output of GARP); 'pointFileName' is the file of testing locations (long, lat) used to evaluate the prediction; and 'ruleSetFileName is the ruleSetFile which is needed later to reproduce the predicted distribution.
The output is a string containing the omission, commission, and the ruleSetFileName, separated by tabs. These should be saved (in a File?) for determination of the 'best' result.

Ricardo Periera, provided the following recipe for calculating omission and commision in an e-mail to Dan Higgins, 10/5/2004

However, those error statistics (omission and commission) could be calculated outside GARP code. Here is the recipe:
1) Get a set of presence data points (species occurrences - x, y coordinates) to test
2) Project the GARP model onto geography (map generated by GarpProjection actor)
3) Overlay the presence points from item #1 onto the map generated on #2. The percentage of those points that fall in a pixed not predicted present is your OMISSION. Say, out of 100 points, only 45 fall on white pixels, the other 55 fall on black ones, your omission is 55% or 0.55.
4) Commission, when we don't have real absence points (our case) is the proportion of area predicted present with regard to the total area of interest, not counting masked pixels. So if 40% of the area is predicted present, your commission error is 40% or 0.40.
5) Then, select those GARP runs that show omission below a certain omission threshold, say 5 or 10%.
6) From those runs selected in #5, sort them by commission error, and then get the 50% of the models that are around the median value for commission. If you got, say, 20 models in item #5, now you have 10 models that make up your best subset of models.
7) Sum up the maps for the best subset of models in item #6, that is your final prediction map for your species.


Author(s): Dan Higgins NCEAS UC Santa Barbara
Version:


pointFileName
'pointFileName' is the file of testing locations (long, lat) used to evaluate the prediction

ruleSetFileName
'ruleSetFileName is the ruleSetFile which is needed later to reproduce the predicted distribution.

omissionValue
The omission value

commissionValue
The commission value

outputRuleSetFileName
output ruleset file name