Hi Andrea,

I think modeling single species distribution can be useful either for rare (endemic, threatened) or common species depending on what is the question you want to answer. You will find two examples following the links bellow.

Cheers

Luiz

http://dx.doi.org/10.1371/journal.pone.0031246

http://onlinelibrary.wiley.com/doi/10.1002/ece3.1094/full

I agree with Luiz. We have been working for several years on producing a single species model for great crested newt (Triturus cristatus) a European Protected Species at the edge of its range. The model has proved useful for habitat creation to ensure the preservation of genetically distinct populations. 

Hi andrea

You can read in this link (https://www.researchgate.net/profile/Joao_Giovanelli/publication/225767203_Giovanelli_JGR_Haddad_CFB_Alexandrino_J_Predicting_the_potential_distribution_of_the_alien_invasive_American_bullfrog_Lithobates_catesbeianus_in_Brazil_Biological_Invasion/links/540df5800cf2f2b29a3a73ad.pdf)

ABout your problem in MaxEnt, DIVA- GIS

Regards

Lubis .Muhammad Zainuddin , M.Sc

Article Predicting the potential distribution of the alien invasive ...

Article Environmental gradients explaining prevalence and intensity ...

Article Assisted migration of plants: Changes in latitudes, changes ...

Hello Andrea, I recommend you this paper:

Garland & Adolph 1994 Limitations on inferring adaptation:

http://scholarship.claremont.edu/hmc_fac_pub/279/

This paper talks about the limitations when you work with only two species, from adaptation and a statistical perspective. It is neccesary working with three or more species.

My best wishes.

Angélica Arenas UT

Agree with Luiz, It depent of what you wanna to do and the focus of your research

Thank you all, I agree completely with Ricardo, David and Luiz. Depends of my research question, and I will work on that. Best wishes!

Hi, sometimes many revisors or journals want to take a the best paper, just for have a marvelous reaserch but sometimes is necessary have to work with one species that maybe is necesary or playing with subspecies for answer a big question about the conservation or future conservation for one or more subspecies into your species, i hope that u understand my point, good work!

Agree with all guys.. Depends of the question of your research. Good luck.

Hi Andrea, I have a similar situation as I am working on an endemic and threatened plant species, I intend to map its known distribution and use the climatic data to characterise its habitat and then predict other unknown/ potential locations where it can be found. Any idea on how I could proceed? I think there are certain restrictions when you use Maxent- you need only presence data, other models use both presence and absence data. 

Hi Andrea,

As it was mentioned before, it is useful to model endemic or threatened species, but also can be useful for invasive species. 

Here is an example:

Article Combining a Climatic Niche Model of an Invasive Fungus with ...

 Hi Andrea (All),

In asking your question about your research aims, you should firstly clarify whether you are trying to model the present distribution of your species, or its potential distribution.  In the first case you could use a Species Distribution Model (e.g. DIVA, Bioclim, MAxEnt), whereas, in the second case you should prefer a Niche model (e.g., CLIMEX).  There is a subtle, but important difference between these types of models and their methods.  Unfortunately, the literature is very undisciplined and confused.  Perhaps it is easiest to think of it this way.  If you have a dataset and you want to create a descriptive model you can use one or several correlative statistical models or methods (e.g. regression, GLM, Mixed effects models, etc).  The "art" here is to create a model that is not "over-fitted", nor too general.  The warning with such models is that any confidence you have developed in the model (e.g. by considering goodness of fit statistics) should be discounted if you apply the model to a novel set of conditions (eg future climate scenario or, for an invasive species, into a new continent).  If you want a model that you can apply more generally, you need to capture the responses of your state variable to environmental variables in your model.  In the case of an species where we want to know where it can persist, we can start by considering how its population responds to temperature and soil moisture in terms of potential for growth and survival.  So long as you capture the right mechanisms at the right level of scale, you should be reasonably well-placed to apply your model into situations where the climate is novel.  You still need to check the results.

The single versus multiple species question has arisen partly because it is a convenient way for some journals to deflect single species modelling papers.  Personally, I view multiple species modelling efforts with deep scepticism.  Having developed many single species models I know how difficult it is to craft a reliable model.  I know that when someone has run many species into an automated ensemble modelling system they cannot have paid suitable attention to the quality of the models.  There are many sources of error and misguidance that modellers are faced with.  Distribution data (e.g. from GBIF) contains many types of errors - some obvious, and some subtle.  Apart from typographical geocoding errors, there can be records attributed to the centre of administrative areas such as countries and states or provinces.  Records can represent permanent populations, ephemeral populations, and in extreme cases a laboratory in which the species was reared.  A modeller needs a keen sense for the smell of garbage and good tools to provide feedback.  These types of errors have resulted in embarrassing results for even highly experienced modellers.  We mostly rely on climatic data to provide us with covariates.  This data too contains many errors in the raw data and in the splined (gridded surfaces that we love).  There is also the potential for spatial/altitudinal mismatch.  Particularly in areas of high topographic relief it is possible for distribution records to be mismatched with climate data.  Wherever possible, modellers should check that the altitude of the record is similar to that of the nearest climate station.  Add taxonomic uncertainty and structural model uncertainty to all of these sources of error and perhaps you will understand why I am deeply sceptical of multispecies ensembles.

I will take this opportunity to challenge the myth that Prishnee asked about.  MaxEnt is not a presence-only model.  It requires presence data and a defined background, and so technically it is a presence-background model.  The background defines the geographical area (and hence covariate space) over which the model will attempt to discriminate between regions that are dense with presence points from those areas where there are no presence points.  Unfortunately the model results are highly sensitive to how you define the background, the sampling biases that are present in the distribution data, the method by which you constrain the model behaviour outside of the covariate training domain (extrapolation), and the selection of the covariates that it can use to explain the distribution pattern.  Whilst the proponents of such systems claim that it is an objective method, that's a long list of subjective options available to the modeller.

In the spirit of full disclosure, I started my modelling career using Bioclim and GARP, before trying out CLIMEX and I have used MaxEnt, GRASP, and a number of other correlative modelling packages.  I am now developing the CLIMEX package.

Good luck with your modelling.

 Thank you for your question which has arisen an interesting discussion. I personally prefer to read about results on  SDMs\ENMs of few species in a restricted area which are supposedly better known and so directly verified by the same authors . For instance, when I read about models based on GBIF data I am always suspicious (also because  the authors have maybe never seen the species). As already stated, it depends on what you want to say, I have personally published with only one species but I had to work more on finding new and interesting methods\conclusions rather than showing simple predictions. If you are interested, please check my page in researchgate.

All the best

Mauro

Hi Dareen, I really appreciate that observation, when I was checking papers about this topic, modelling the present distribution or  potential distribution seemed the same concept, but you clarify those ideas. Hi Mauro, about GBIF data, well, I prefer to use records of  regional collections, specifically voucher specimens, I think Human observation, for example, is not a good data (there is a lot like that inside GBIF), maybe this is the most important thing, I will check your avalaible information.

Thank you very much!

About Gbif data, no dataset should be used as downloaded, it is necessary to be critical and sufficiently familiar with your species to know how to validate records. You can always eliminate the Human observation or fossil records if you think they are not useful for your work. You can always contact the data provider for more information in case of doubt. But it is really a pity to ignore this valuable source of information just like that.

Good luck with your work!  Marta

You are right Marta, is not a question of ignoring information, is just in my case my co-workers know all the localities of my species, so I prefer to ignore human observations, for example, keeping infomation of our regional collections, thank you and good luck too!

Between all this forum/ conversation I would like to ask to everyone.. if anyone has worked with modelling in the past? or has any contacts to talk with? I am interested on this area too.

Marta, you are correct that you shouldn't use data in an unquestioning manner.  These days everyone seems in too much of a hurry to get the work done so they can get it published.  Ironically, with the advent of databases such as GBIF we are now becoming rich in data, and consequently we are finding it more difficult to identify erroneous or misleading outliers.  Sadly, some of the most popular modelling tools provide little or no feedback to the modeller that there is a problem with any points.  Some errors are inconsequential to the model.  The balance is to find tools that highlight those outliers that do affect the model or analysis so that they can be scrutinised carefully.

As for using GBIF data - I would use them with great trepidation, if at all. The geographic information is full of gaps and the taxonomy and thus nomenclature lags behind seriously. Well, it is up to us researchers to fill the gaps and update the data, right!? Unfortunately we all seem to have so much more urgent things to do that there is no time to do everything we should do.

Hi Andrea, here a link to a paper were we predicting the potencial distribution of a invasive mammal, perhaps could be interesting to you

www.researchgate.net/publication/287973398_Taxonomic_identity_of_invasive_rabbits_in_Cuba_First_record_of_Eastern_Cottontail_Sylvilagus_floridanus_Mammalia_Lagomorpha?ev=prf_pub

Article Taxonomic identity of invasive rabbits in Cuba: First record...