I am using QGIS 2.18.0 on Windows 10.
I have a set of observations (1,428 points) for a given species (presence only data). These data were generated from a convenience sampling (hunters, field observations by biologists, etc.) and thus they are subjected to strong spatial bias.
Experts tell me that I should build a sampling bias surface in order to account for that bias in my analyses (Maxent).
To do so, I need to select at least one variable which is highly correlated with the density of observations.
I have two such variables in mind, namely 1) a shapefile with all the roads in the study area, and 2) a raster of population density.
The rationale behind using variable 1 (roads) would be that, in general, hunters tend to get their prey near to roads. However, I am having a hard time finding a way around the analysis.
One alternative might be to generate a regular grid, measure the total length of roads and the number of observations within each cell. Then, I could extract those values and perform count regression to see if the association between counts and density of roads is statistically significant. If so, then I can use that layer as sampling bias surface.