We have a lot of line-transect samples of bird communities (spanning from 2 to 10 km long). Within transect the data are obviously autocorrelated, we therefore cannot compare transects simply splitting the data to get variability.
How can we cope with this problem, to obtain comparable data for between-transects comparison?
Ciao Guido. Do you want to compare transects in term of bird community indices i.e. species richness or model species density or just presence absence?
I think you could use the lenght of your transect as an offset in poisson GLMs. or the article Spatial Models for Line Transect Sampling by Sharon L. Hedley and Stephen T. Buckland (Journal of Agricultural, Biological, and Environmental Statistics, Vol. 9, No. 2 (Jun., 2004), pp. 181-199) could help you (see link)
By the way, as long as each single transect is your sample unit you should correct your bird community indicies by the length of the transect and be more worried about between transects autocorrelation.
I would be worried about spatial autocorrelation within trasects only if you want to model each presence point inside a transect and compare it with other transects.
I hope it helps.
Enrico
http://link.springer.com/article/10.1198%2F1085711043578
Hi Guido,
just to add this: Within GLMs and other models using R, you can correct for a certain amount of autocorrelation of your response variable by applying a CorARMA correlation structure that has to be further specified, and then watching the remaining autocorrelation by function acf().
Best regards
Armin
Dear Enrico,
I am actually interested in spatial autocorrelation within transects, my question was specifically on this type of problem.
Dear Armin,
many thanks to your answer, it seems very interesting