I never hit the limit. You can check our papers, if you are interested. Up to about hundred should go. You can reduce the number of predictor variable a posteriori by stepwise backward elimination, usually to two or three predictive variables. Alternatively, you can a priori reduce the number of variable in various ways. All described in our range of papers.

Why not, these variables seem relevant for modelling the actual (not potential) distribution of tigers. Of course you will need to use raster data but you can convert the point data to rasters showing the distance to these points. For the prey species, what's sometimes done is to model their distribution first, then you can use the resulting suitability maps of these species as an input in your model.

Thank you respected Hein Van Gils and Tobias Fremout for your valuable suggestion. With this suggestion, now, I can proceed with developing the proposal. Your help is much appreciated.

Sincerely

Kamal Ghimire

Dear Kamal Ghimire

You are now thinking properly. There is no limit to the number of variables that you can enter into your SDM. For example, some variables may be suitable for plants and not for animals, birds or marine organisms. Each organism has its own Ecological Niche that govern a distribution. Your skill in knowing these conditions in reality first before testing them inside the maxent, and thus you can judge the results of Model whether they are correct or not. I recommend to read this article: http://abc.museucienciesjournals.cat/files/ABC_39-1_pp_99-114.pdf

All the best

Thank you, Karim Omar and Joaquín Duque Lazo for your valuable feedback. They are indeed very helpful for my research.

Well, some limitations might be in case of small sample size. Although there is no strict standard, the rule of thumb is that a single variable often needs ten data. Using an unbalanced number of variables might cause overfitting problems. So, yes, you can first include the highest number of variables but then try to reduce them down to a reasonable and ucorrelated set.

You can use different environmental variables, but the ideal method for habitat suitability modeling is to use variables that are more important to the species under study.

Hello Kamal Ghimire,

Good approach, normally we include in Bioclimate variables and topographic layers in SDM, and such data is depending on Spp types (Plant, Trees or wildlife spp) you should add forest cover layer, humans settlement layer, water ponds or stream layers, and most important tiger main prey such as wild boar, spotted deer, sambhar deer, swap dear presence location layer. tiger only shift or adopt their territory, when tiger prey shifted or a number of tiger in increased in a limited area. According to me only a few(2/3) bioclimates variables required because tiger can survive dry and wet region, the tiger can survive easily in 1 degree to 45-degree celsius temp. altitude layer maybe you can add, normally tiger found in Shivalik region within (200 -1200m alt) but in Uttarakhand tiger capture Askot Wildlife Sanctuary around 3600 m.

hope my thought will help u.

For variants selection, I suggest an ecological based technique routine. The SDMs prediction for species spreading is really a habitat prediction rather than a model itself. So we should input the real variants which influence the spreading of an animal most(not all). Maybe the prey behavior shift the spatial pattern of your target species, may it does not do as we expected. We rely on the model selection. Of course, settlement, rivers and so on may also play their roles, but we can not make a decision before a model assessment. In general, models service the ecological research, rather than others.

Well, no model is understood to be perfect. More biologically relevant parameters/variables you add, closer you will be to the prediction precision.

it is ok to use a combination of bioclimatic and other bio physical variables. However, will need to do collinearity test to remove highly correlated variables. collinearity test is to remove the environmental biases and improving the model performance. One of the easiest ways of doing this is through VIF analysis, or there are few other methods.

depend on sample size. large number of variables make model complex Which results huge biases

There is no limit on variable selection but the selection of variables must be defined by the necessity or potential influence on the species distribution. Going by your thread where you mentioned some variables which are point data like settlement, prey occurrence, ranger station, etc. MaxEnt accepts only one .CSV file that is the species occurrence data. So, to my understanding, you can't use these point variables in MaxEnt. If at all you want to use these variables then you must convert them all into a raster data format to be used as an environmental layer. Land use / Land cover is recommended and it is mostly used in species distribution models.

It's always good to include variables based on a priori set of assumptions. SDM is not only for creating distribution maps but investigating species niche as well. Having assumptions could lead to some results about species niche. Second, I would consider correlations among variables. Highly correlated variables can have adverse impacts on your results. I appreciate data science but I always try to be an ecologist first and use data science as a tool in my research: Defining hypotheses, testing them ( Through Maxent, etc..), and then interpreting results.

Yes, there is a limitation. All models need to meet a balance between preciseness and parsimonious. More variables and fewer data points, more overfitting problem. You may want to look into this article for more information about the impact of too many variables on the model when there are few records of species

Article Investigating niches and distribution of a rare species in a...