[R-sig-eco] Time series

[Chris Mcowen]

Dear List,
 
I am working on some fisheries data investigating key correlates of catch from 50 global sites ( huge areas) in relation to a suite of environmental anthropogenic and oceanographic variables.  
 
The data is non-linear and shows high correlation between predictors, therefore I have been using regression trees to get at underlying relationships.
 
I am trying to get at the key drivers of variability within and between areas i.e. what causes catch within area 1 to fluctuate and what causes catch within area 2 to fluctuate etc. The approach have taken is to cluster the areas based on the composition of the catch, under the logic that variability of catch within areas of similar species will be driven by similar variables. I have identified 7 clusters that are significantly (bootstrapped and cross validated ) different.
 
I have data from 1997-2010 (so I guess I will be looking at a multivariate time series analysis) – is it possible to tease out the drivers of variance within each cluster i.e. catch in cluster 1 is driven by temperature where as catch in cluster 2 is driven by chlorophyll a? I was thinking of doing a random forest analysis and extract variable importance measures  on the average of the time series for each cluster but that would only tell me what causes catch to vary between sites within the cluster – which would be strange as they should be similar due to the clustering.
 
Thanks In advance
 
Chris
 
 
Chris Mcowen
Postdoctoral Scientist, Nippon Foundation Nereus Senior Fellow
Marine Assessment and Decision Support Programme
 
UNEP World Conservation Monitoring Centre
219 Huntingdon Road
Cambridge CB3 0DL
United Kingdom
Switchboard: +44 (0)1223 277 314
Fax: +44 (0)1223 277 136
 
www.unep-wcmc.org