outliers in SDM

[Andrew Jacobson]

Hello,

I have seen very little discussing the importance of outlier points in SDMs. And to be clear, I am not talking geographic outliers but outliers in environmental space, detected during exploratory data analysis. Is it important to remove outliers? Having seen so little mention of outlier removal from the SDM literature, it seems as if this is not a common practice. 

thanks

[Jennifer Pannell]

I believe you're right that it's not often mentioned in detail in the literature, but it's definitely an important step to thoroughly check your data points. Geographic and environmental outliers are both suspicious and should be looked into (though obviously, geographic outliers may be perfectly fine, but just one record on a whole continent may be an indication that it's not a true point). Personally, if they're from GBIF I check the details of the original record and if possible go back to the primary source and see if it's a valid point or is cultivated, etc. If you're working with large-ish plants it can also be possible to use tools like google streetview to look into it (depending on where the record is from, of course). Other factors such as irrigation can also throw out your data points. I'd imagine/hope most studies do this (or something like it) but just mention it under "data cleaning", as outliers in environmental space can make a huge difference to the model outcome.

Cheers,

Jen