Thanks for your response!
Ideally I would use NUTS for sampling. However I just don’t know if it’s feasible. In this instance where I have 3600 (60x60) predictions to make; using Metropolis sampling this takes approximately 30 hours to run. I understand that NUTS is faster at finding effective samples, but in terms of pure run time, Metropolis appears to be significantly faster (I haven’t run this through all the way with NUTS, but I think it could take anywhere up to and beyond 5 times as long). Furthermore I eventually plan to run this on a higher resolution dataset where I have >100,000 predictions to make (I am predicting fields of geo-spatial data).
In any case, I will check the Metropolis result and see how it performs. If it turns out I need to switch to NUTS I think the best thing for me to do will be to divide the data into swaths and run them in parallel on my University’s computer servers. Not ideal, but hey.
Thanks again.