ENM GARP 1
0 (0 Likes / 0 Dislikes)
This is the GARP XXXXX
that's for genetic XXXX and rules of prodution
although I have changed the name to roots of prediction
and this was developed in the eighties or nineties by XXXXXXX from Australia
and this is a XXXXXXX procedure that wasn't XXXXX for different XXXXXXXX
and it was implemented for niche modellinmg by XXXXXXXX
but the general idea of genetic algorithms is that there is XXXXXX methods or disXXXXX and error methods
for solutions of very complex problems for which deterministic methods like
statistical methods are inefficient in the sense that
could be large large months of data for analysis
they are called genetic algorithms because they inspire in the process of natural selection
in the sense that they evaluate a model
XXX puts XXXX on the original models and then select the most fit one
and discard the ones that are not performing well.
we'll see, all the steps that I XXXXX producing a niche model in regard XXXXXXXX
one of the XXXXXX that I have seen in the development of this field is that more and more soft work are available
which is good
but because XXXX become very easy to use by general users
then comes a black box.. do you know what's inside the XXXXXX?
this is the case for GARP. It was very popular like five ten years ago
and many people started to use it, but most of the people were not knowledgeable about how it works
so they weren't very efficient users of this software
the steps of XXXXXXX. let's sayyou have a set of hundred data for one animal
the first step is that usually for XXXXXXX into 2 sets
One for XXXXX the role model, the joining set
and the other set is for external XXXXXX of the model
so you are not going to use that set of points in any part of the process. we'll have varient or conjugative XXXXXX
from the training data, that's the second step, XXXXX the system with samples of these data
let's say you kept XXXXX 50 of the 100 points
if you sample these fifty points to 1250 presence data and it takes from the rest of the study area
where you don't have XXXXX, another 1250 pixels known as pseudo absences
it's assuming that in these areas you don't have to XXXXXX on species
it's an absence of. if you have a full set of 2500 points of presences and absences
it doesn't matter how many points you have in your initial set
your own points, you end up with 2500 points of presence and absence
then split this whole set of 2500 in half and use one half for training model
and the other half for internal testing of the model
so the first step for creating the niche model is that these internal drawing points are used to characterize
the conditions in which you can find these presences and absences
in the form of rules that is called in GARP which are these if then statements
and it produces these rules on the four different methods
three forms of XXXXX XXXX such as lightning
and XXXXXXX of a logistic regression
so in a way GARP is a meta algorithm
because it uses all the algorithms like these ones
to produce niche model and then this is a genetic algorithm to modify these original rules
so it produces a set of tens, I don't know, it could be one of these rules
with the joining points
then it uses the internal XXXXXnts X points XXXXXX data sets
to see how well these rules predict XXXXXXX points
here the confusion matrix, (we'll see what is a confusion matrix)
In a sense what it does is to see how many parts of presence and absence XXXXXX predicted by this XXXXXX
Basic process course on percentage points will predict presences and absences
it keeps like a qualification for ps great for each on of the groups
and builds a set of XXXX or XXXXX of these rules
sorted in terms of the performance, the rules are performed best at the top
and so forth.
then the entrance to the genetic algorithms. the genetic algorithm XXXXXX is to input variation to these rules
see for example if we have here the rule that says "XXXXXX between 10 and 20 XXXX will be this"
and precipitation is between 250 and 200, and the elevation is between 1000 and 2000
then, this species is present
that's a rule. That's climatic XXXXXX
the genetical XXXX is changed in XXXXXX bodies XXXXXXXXXXXXXXX
there are different rules for changing the XXXXX XXXXX
certainly by deletions, deletions by plant location etc
but the idea is to input some XXXXX into these rules
and then what it does is that it XXXXXXXXX the performance of this new rule, this XXXXXX
if the XXXXXXX rule performs well than the modern
then it substitutes this rule in the rules and XXXXXXX
if it doesn't you just scrap that this isn't a rule
so if you do this process of rule generation or evaluation of substitutional relation
for many number of interactions as you said as elusive
or until each converges, which means that it doesn't improve the whole performance of the XXXXXXXX
if you study the process you'll see that intiation of the process you will improve and improve your prediction capacity
well, there is a point in which you don't improve in this prediction capacity
it stops at that point
and it ends up with a set of these "if" "then" rules
that characterizes the conditions for the species presence and absence
then with this set of rules goes to the geographic space and asksto each one of the peices
if it needs conditions XXXXX by the rules
and gives the communication of presence or absence to each one of the XXXXXXX depending on the conditions of the biomes or XXXXXXXXX
so the final call of GARP more than in process is in binary math
in math that says species is present or species is absent
in contrast to what XXXXXXXXXX produces this is only a binary
well, there are 2 implementations of GARP
this is the best of GARP, this is a stand alone software
that we used through the last few years
and there is an implementation as you saw XXXXXXXXXXXXX
actually nothing more than there are 4 implementations of GARP
it's well known to use the version implemented in novel model XXXXXXX
well, there is something else that's important for you to know
XXXXXXXXXX run processes in their own generation of niche model since the very beginning of the XXXXXXXXXXX
then the generation of the rules also is running process
every time you run a model in GARP with subtle XXXX of them
you get a somehow different math
so that could become a problem because you save which of this bunch of maps
or which one of the 100 maps is the good one
which one is not good
so XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX selecting the models that are more informative
based on the predictive capacity
the way this implementation works is based on the evaluation of the errors
of the model.
when you produce a model there are 2 ways that the model can run
when it fails to predict the known presence which is XXXXXXXXX falls outside the prediction
that's called a XXXXXXXXXXXXX
and the other way you can face is when it over-predicts or
it fails to predict the absences
when the model says that this species should be present and XXXXXXX absent
it does form the commissioner and that's evaluating these confusion makers
we use this information to evaluate the quality of the models