Form and Generation of Bioclim and GARP Rules

The idea of BIOCLIM is to find a single rule that identifies all areas with a similar climate to the locations of the species. To do this, the basic BIOCLIM algorithm finds the climatic range of the points for each climatic variable. The rule, consisting of ranges of climate for all climate variables, then encloses all points, within statistically defined limits.

For example, the following rule was formed from the ranges of climate variables that enclose 90% of the data points as determined by calculating the mean and the standard deviation of the points.


IF	TANN=(23,29]degC AND TMNCM=(10,16]degC AND TMXWM=(35,38]degC 
	AND TSPAN=(19,27]degC AND TCLQ=(21,23]degC AND TWMQ=(29,30]degC 
	AND TWETQ=(24,32]degC AND TDRYQ=(19,26]degC AND RANN=(609,1420]mm 
	AND RWETM=(156,319]mm AND RDRY=(1,1]mm AND RCV=(101,123]mm2 
	AND RWETQ=(460,874]mm AND RDRYQ=(0,9]mm AND RCLQ=(1,16]mm 
	AND RWMQ=(272,532]mm AND TMEL=(17,263]masl AND TMXEL=(40,303]masl 
	AND TMNEL=(4,230]masl AND TREL=(0,105]masl AND LONG=(128,136]deg 
	AND LAT=(-12,-15]deg
THEN	SP=PRESENT

Central assumptions are used in Bioclim are:

the distribution of the species is determined by climate,
the distribution of the climatic variables is standard normal, and
all variables with restricted ranges influence the species of interest.

With all modelling systems if the assumptions of a method are not satisfied then the results will be unreliable, or simply quite wrong. While Bioclim has been shown to give satisfactory results for many species, there was a percieved need to develop a system with less restrictive assumptions. Thus GARP (Genetic Algorithm for Rule-set Production) was developed with aims to develop models with:

all forms of environmental data, including categorical soils data,
arbitrary distributions, including bi-modal (two-peaked),
concise rules using few variables.

The solution proposed by GARP is to produce a set of rules predicting presence and absence, each one statistically significant at increasing the probability of presence or absence of a species. An example of a rule set is given below:

IF	TCLQ=(6,19]degC
THEN	SP=ABSENT

IF	GEO=(28,241]c AND SRT=(3,4]c AND TMNEL=(-19,308]masl AND LAT=(-13,-39]deg
THEN	SP=ABSENT

IF	RWMQ=(107,1176]mm
THEN	SP=PRESENT

IF	GEO=(6,244]c AND TMNEL=(285,1480]masl
THEN	SP=ABSENT

The production of a set of rules raises the problem of conflict. For example, at a given point, one rule might predict the presence of a species and another the absence. In these cases GARP predicts using the rule with the highest expected accuracy.

GARP has a number of other features for increasing the rigour, reliability, and flexability when modelling species distributions. In the context of this application GARP can be seen as value-adding to the BIOCLIM method by increasing the accuracy and suggesting causal factors. This can be seen by comparing the predictive accuracy and the rules generated by the two methods.