Hi all,
I would like to simulate the spatial distribution of compositional data, in my case age classes of fish. My data consists of the age composition (pooled into 4 or 5 age classes) of samples of fish at non-random locations. I want to interpolate these data using stochastic geostatistical simulations to get age composition at unsampled grid locations. I am aware from Aitcheson (2005) and others that I first need to transform my data (logratio). I am assuming that after the transformation I still could not simulate the distributions of the age classes separately, and would need to simulate the joint distribution of the age classes, using an approach something like that described by Babak et al (2010), and then backtransform the simulation results to proportions. However, I am not sure at this point whether following the procedure I've described above will constrain the sum of the age class proportions at a simulated node to be 1. Any opinions on whether the approach described will work or not? Any suggestions for an alternate method? Thanks in advance. Paul Walline, Alaska Fisheries Science Center paul.wall...@noaa.gov