Watch videos with subtitles in your language, upload your videos, create your own subtitles! Click here to learn more on "how to Dotsub"

ENM Choosing Algorithms 1

0 (0 Likes / 0 Dislikes)
Good morning everybody! we are going to move on now to so we've reached the point, or we are almost ready another twenty minutes for me and we are going to start learning some models and talking about running BIOCLIM models so those of you who haven't had a run of any of these models before XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX having run your first models, they can be simple bioclim models and for reasons that will become clear before we get down I'm going to go through a few general considerations that I think are important things to be thinking about at this stage one important thing ultimately is what model do we select we are going to talk through a few different methods this week we are going to talk you through BIOCLIM we are going to give you some general introduction to XXXXXX really from a practical perspective how you have to start running thoose models and huge amount of tutorials and tutorials available to really feed to progress the really comfortable using these methods but we are going to set you on your way with that we are going to talk through XXXXXX as well and within the open model of the frame work you are going to kind of hopefully get on way with the tools and models to run three different methods but from those of you who are already starting to look to the XXXX table see that there are tons of methods out there there are many different algorithms that arnd of problem e being used to address the same kind of problem and the question is which methods should we select we are going to give you some reasons why we selected three We have some XXXXX with them they have been shown to be theoritically good methods that have been widely applied and they are used XXXXXXXXXXXXXXXXXX within a couple of days here you are going to be able to actually XXXXXXX but there are some alternatives that stand alone you know, model selection. we wanted to do something that you can get away start running a model the methods that has a XXXXXXXXXX XXXXXXXXXXX so what are the considerations you should be thinking of? we just emphasize in this in the process remember yesterday we spoke about species occurence, about environment data we are now going to focus on this integrity of what the actual algorithm that we are going to use to do this association, or the corelation between the presence records or the presence absence records and the environmental data but I ought to emphasize that the natural algorithm used is one half of the much broader XXXX process it tends to emphasize how often the largest part of it XXXXX XXXXXXXXX of these models is actually reading the data and understanding the data making decisions about what XXXX XXXX XXXXX frame work is the actual algorithm you use is just a one part of this much broader modellin g process and among the very important decisions are outside the actual algorithm you use when someone comes along and says "Oh yeah I'm doing some ecological niche modelling XXXXXX XXXX models or I run BIOCLIM models that's great. that's just really one relatively in some respect slow XXX XXXX XXXXX How you selected your data, how you make decisions about what the background XXX XX or you dealing in bias or how you are dealing in bias all these other points and I would like to you will find questions to identify XXXXXX had the data so that the algorithm, run the algorithm, you got to predict the XXXXXXXXXXX you got to interpret the actual algorithm the XXX model or whatever algorithm you are using is just one part of a much broader modelling process so let's look at two hung up on what the actual algorithm is it's just a table that lists sum by no means all of the approaches that have been applying on the range from approaches like trying ten below the BIOCLIM model actually meant to be XXXXX XXXXXXX XXXX from here XXXXXXX having a table trying to refer to what XXXXXXXXXXX kind of method this is the generic approach like statistical approach regression tree or generalized attitude model a classification tree. this is kind of a generic approach this is the kind of approach you might see listed in software and model for a software for example classification trees and generalized linear models and a bunch of other approaches being implemented in an approachable to the XXXXX has named XXXXXXX My thesis was working on the main species and had it's XXXX an artificial new network this a table that kind of trying to help you get away with what's in the XXXXXX but there are some generic approaches XXXXXXXXXXXXX network doesn't need to be run within this software kind of tool that we were using called species it could be run in any number of different frameworks but this is just a kind of way to help you understand so you got the methods here, the models or the software are XXX XXX you might see here and the really crucially there is this issue what is the data type that a particular model takes. so depending on your data and some decision you need to make about for example either you could use absence data XXXXXXXX is it reliable best XXXXX in many ways is what kind of model you might use so for example simple XXXXXX for those XXXXXXX or how to know this distance might have been another example of this this is just kind of species space approach, XXXXX I'm going to go through it in just a minute BIOCLIM domain might be the software of all the tools that are used and they are presence only methods, they will only require presence data so Town just stops on this because some of the discussions we just had about movement and disposal XXXXX how that selection of your study area can affect your model predictions there are quite some models that simply because applied to because they are not taking background into account at all I am going to clarify that in a minute quite now it's only our presence data and you don't have XXXXXXXXX about what your study reasons should be there might be a good set of tools used they are also simple, straight forward and easy to interpret another set of tools and examples might be XXXXX XXXXX XXXXX and further niche structure analysis thoughce we refer to it's presence XXXXXX approaches now I'm going to XXXXX that in a minute what exactly that refers to another sef approaches say XXXXX network or somebody's approaches like XXXXXX and XXXX mini models, XXX aptitude models it refers to presence-absence approaches, that is they try to take into account presence data and absence data I'm going to try explain that in the next few slides there are a whole set of techniques if you have presence data and good reliable absence data that you have some confidence in then that might be another way of steering in towards a certain class of algorithm to your modelling. I'm going to emphasize here, some of these tools and XXXXXX XXXXXX XXX more about it later some of these tools are invented in our XXXXXX thos of you who are very comfortable with using "R" that is growing environment between these models and I'm going to mention some of the work that XXXXX has done with the XXXXXX package in "R" that makes a lot of these analysis relatively straight forward if you are OK with the "R" environment so some of these tools like regression trees and XXXXXXXXXX regression XX and whole bunch of others infact most XXXXXX actually now implemented within XXXX framework so that's what I wanted to mention but we actually are going to play without this, this week that's a challenege beyond what we are going to do in the next few days OK, now there is a whole bunch of approaches that have been applied and categorized in different ways and one XXXX XXX what to do it is XXXXXXX practical way to do it is what will eight requirements for average models OK let's first take a general consideration about emphasize is this idea of what the data requirement for a species is so XXXXXXX fell obvious OK, you might have presence and absence data so the blacks stars give a representing of presences XXXX XXXXX XXXX XXXXXXXXXX and the red crosses here represent absence data they are hypothetical in this particular instance but just take XXXX XXXXXX equally you might know we have presence data, XXXXXX we emphasized yesterday that's very often the case when using these XXXXX or keeping the data very often you only have records of where you found the species rather than records of where you didn't another class of approaches that we use is referred to pseudo-absences this is where you will only have presence data but you will apply essentially presence-absence algorithm so an algorithm that tries to contrast presences or categorize presences and absences and approach has been used with some success basically random XXXXX sample a whole bunch of points and from study region and referred to those as pseudo-absences so your challenge then becomes using algorithms to classify your presences from your group of pseudo-absences you are making assumption that it's going to include some presences but mostly absences there is a whole body of literature on the pros and cons of doing it up to the detail now I just want to emphasize that there is that class of models as well I mentioned that these absence records are pseudo-absences commonly taken. randomly there are some good ways of thinking that you can use say presence only approaches to the model sensibly assessed where those absences should be so there might be randomly some examples of particular environments that you think species doesn't occupy OK, that's just a whole class of approaches to selecting or using algorithms that use presence and absence data absences by sampling these pseudo-absences XXXXXXX approaches we've talked about is essentially presence and pseudo-absence type contrast to that a very subtly different approach for the important and theoritical distinction you'll see in literature presence background approaches, this is where you'll only have presence data but the algorithm that you use is theoritically set up the thinking behind this is to contrast your presence data with the environments that are available to the species that's just kind of trying to visualize here by, say in Madagascar it's a climate surface presented here but the point is, what we try to do is contrast these presence records with all the environments that are available to the species let's come back to the discussion we had of course what that background is what those environments are, XXXX find by your study XXXXX it's the same selection of pseudo-absences so infact this is the way XXXXX XXXXX jump over a presence background approach infact how this is done is we sample an awful lot of records from this background say 10000 records from background we shall XXXXXX XXXXX a lot like XXXXX pseudo-absences that's almost the same, but there is an important theoritical distinction in terms of when you are using a presence pseudo-absence you tend to be thinking XXXX distinguished by presence is from my pseudo-absences my presences for my absences, the present background approaches are taking more theoriticle approaches say I'm trying to contrast my presences from the environments that are available for the species so there is a theoritical way of thinking about this but there's an important practical distinction which is the presence background approaches your presence records where you have got XXXX records is also a part of your background. so you don't have two sets of data that are mutually explicit you have your presence records and you have your background records that might include some of your presence records OK, they are not reallyy XXXX records that the localities XXXX observe your records with pseudo absences your sampling from the study region that is not what you are going to take in pseudo-absence that includes you know that is absolutely what are the localities where you found your species it's subtle distinction and in practise those two data sets might look very similar your pseudo-absences in your back ground might be very similar but your background data has the opportunity to include some of your presence records but the important thing really is the stereotical distinction between XXXXXX presences and absences where I might try to contrast my presences from the environments that are suitable and you know these approaches can take very very different ways of thinking about the XXXXX statistically or XXXXX in context but there's a theoritical way of thinking behind it that contrasts these presence - pseudo absence approaches from the presence background approaches .

Video Details

Duration: 15 minutes and 23 seconds
Country: United States
Language: English
Views: 36
Posted by: townpeterson on Jul 12, 2013

ENM in ICIPE

Caption and Translate

    Sign In/Register for Dotsub above to caption this video.