To account for sample bias you can use the "target group" approach: presences of other species of the same taxonomic group are used to identify the sites that have been surveyed. More info in Philips, S.J., Dudik, M. (2008). Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography 31, 161-175.

Hi, we have controlled the effect with the HFP (human foot print) and successfully validated predictions with to passerine species. Watch this work in PlosOne...

Article Effect of Climate Change on Mediterranean Winter Ranges of T...

Check this https://www.researchgate.net/publication/277252871_Estimating_species_diversity_and_distribution_in_the_era_of_Big_Data_To_what_extent_can_we_trust_public_databases

Article Estimating species diversity and distribution in the era of ...

Along with geographic sampling bias, another common problem is "not enough records from across the full domain of environmental conditions suitable for the average individual of a species".  While this is a challenge, it's not necessarily systematic bias.  I think there is good reason to be concerned that GBIF records are not numerous enough to sufficiently estimate the environmental space for many plants and animals.  

I've not used GBIF records for modeling, but just for fun I went and searched the GBIF database for English Ivy Hedera helix, an ubiquitous and cosmopolitan invasive in N America.  There were only 670 records from the USA, but they ranged from southern Alabama and the panhandle of Florida to the Pacific Northwest and Hawaii.  

That's a pretty huge chunk of environmental space and it seems pretty clear that this plant is far more widely distributed than the GBIF records indicate.  I would be skeptical of a niche estimate in this case

Can you find additional occurrence data for your species?  Museum records?

Hello, thank you all for answering.

Jessica and Jose Luis, I had never thought in this way to cope with bias, I really apprecciate your comments.

Alfredo, I´m reading the paper.

Jason:

I'm pretty sure that I can find more information about my species, but I´m really constrained with time to earn my master degree.

I think that kind of data gaps are ubiquitous whether in museum records or data portals such as GBIF, due to bias inside sampling of biodiversity ( spatial bias to accessible areas, incomplete sampling across an area, incomplete detectability of individuals and other issues) 

I will speak to the question more generally. Each data set and each taxon and each region will have particular sorts of bias. For example, some regions are access limited, so that a major agent of bias is that of the environments that the roads and trails and rivers cross. Other regions have big citizen scientist legions ready to go out and collect data… In a case like that human presence or proximity to cities might be more important. Very frequently, one political entity is good at reporting, where as another is not, and understanding those patterns will help you quite a bit to make decisions about what is oversampled and what is undersampled. The problem, of course, is that under occurrence in a region could also reflect lower-quality conditions for the species, or even absence of the species. So my suggestion would be something quite a bit more holistic: you have to understand your particular dataset to be able to respond appropriately to the specific biases that affect your particular situation. Hope this helps ATP

Hi Carlos,

If you use MaxEnt software to model your GBIF records you can input a 'bias layer' into the model that corrects for geographical bias e.g. samples collected on transects or roads. See here for how to make one: https://scottrinnan.wordpress.com/2015/08/31/how-to-construct-a-bias-file-with-r-for-use-in-maxent-modeling/

Best of luck, Mark

Hi all!

 In collaboration with Prof Tellería and others, we tested some approaches to assess for bias in MAXENT  with bird ring data. We used field samples to evaluate our models and found that "target group background" approach failed in those areas where presences fron sister groups used as background were not available, since MAXENT were not "informed" about the "behavior" of the species for many environmental values (easy to see with MESS analysis). Here you have a panel communication about this issue (in Spanish): 

Poster ¿Modelizando aves o anilladores? El sesgo de los datos de an...

Hello,

Not sure anyone has recommeded this paper yet: http://www.sciencedirect.com/science/article/pii/S1574954113001155

Cheers,

Frederico