Sampling Bias Question
Maribeth Mitri
Dear MaxEnt community:
I am hoping to get some advice on the right approach to account for an unequal sampling effort/sampling bias in my project. Any guidance would be greatly appreciated.
To provide some context – I currently have a file with about 1200 occurrence points for both Canada and the United States. These points represent the centroid of either a public health unit (Canada) or county (United States) for species occurrence. We had to use the centroid approach as a result of not having more specific coordinates for species location. In cases where public health units/counties are geographically small, the occurrence point will be more accurate when compared to larger public health units/counties. Our geographic extent is currently all of Canada and the United States as this covers the entire survey area – we will likely restrict areas of extreme climate moving forward (i.e., Northern Canada) as we don’t want MaxEnt to predict background samples in these areas. As a result of geographically large public health units/counties in Western Canada/United States, when we plot our occurrence data, it demonstrates higher sampling density in areas of geographically small public health units/counties (not necessarily more species presence). The concern is that by choosing centroids of these administrative borders, we have a sampling bias associated with the size of the county/public health unit. What would be the best method for accounting for this specific bias (restricting background samples, Gaussian Kernel Density, buffer approach, spatial filtering, etc.)? I'm not sure which one would be best suited for the situation.
Thanks in advance!
Dalla Lana School of Public Health
University of Toronto