[R-sig-Geo] Building a prediction raster when the statistical model was built from sampling units of different sizes

[Nelly Reduan]

Hello,



I would like to build a predictive map of capture success from a GAM. To build the GAM, I used data of capture success (dependent variable in the model) and proportions of land cover types (independent variables in the model) that were estimated within 50 trapping sites (thus, sample size = 50 observations). The size of trapping sites is ranged from 24 km² to 236 km² (mean = 133 km²). I tested whether or not the capture success varied between the trapping sites and the capture success didn't vary between the sites. Is it statistically correct to create a prediction raster map with cells of equal size given that the trapping sites are of different sizes?  If it's correct, should I use raster cells of 5 km x 5 km based on the minimum size of trapping sizes or 16 km x 16 km based on the maximum size or 12 km x km based on the mean size ? If it's not correct, what should I do? Instead of calculating proportions of land cover types within trapping sites, should I calculate the proportions within a buffer around the centroid of trapping sites?



Thank you very much for your time.

Have nice day.

Nell

	[[alternative HTML version deleted]]