Bias scale inversion with sampling effort

[Jennifer Smits]

Hi All

I know there is much debate over bias and how to deal with it in Maxent. But i have a slightly different problem and would love some advice. I am in the unique situation where i know my sampling effort, it ranges from sites being revisited 22 times to 1. It is a very wide-spread and common species with a fairly large range in distribution, red kangaroo Macropus rufus. And unfortunately areas with high sampling effort are often coincident with areas with highly suitable habitat and larger numbers. 

Applying a straight sampling intensity or a relative sampling intensity bias file (1 to 10) destroys the output of the model. It seems to  up-weigh the importance of points in the lower sampled areas un-proportionately -- the map output shows higher presence probability values where less sampling effort has occurred which in reality is incorrect. I would hope that in accounting for bias you are removing the effect and thus the probability would be evened out? It also causes the model to predict outside of a sensible range (eg. arid specialist red kangaroos occurring tropical and alpine areas).

I have tried it all - bias files, background masks etc. I have tried adding a bias file to the covariates/environmental layers as well. If i apply bias as a simple background mask or a stepped euclidean distance from points the result is pleasing BUT totally ignores that i have sampling intensity data.

I would like to try inverting the bias scale and wondered whether this was a silly idea or feasible. By this i mean areas where there is low or no sampling intensity will be 10 and areas where there is high sampling intensity as 1. Why? Because where i have high sampling intensity, I have greater confidence that the habitat is suitable and where i have low sampling intensity I have less confidence.

Comments? Thoughts? Ideas?

Thanks in advance

Jenny - Australian Wildlife Services