Identifying insect outbreaks: a comparison of a blind-source separation method with host vs non-host analyses.
Lionel Humbert, Daniel Kneeshaw.
The identification of past insect outbreaks is often determined using a comparison of host/non-host tree ring growth chronologies. Yet this may be a problem when non-hosts are either affected by the outbreaking insect or when the growth of host and non-host trees does not respond similarly to the same climatic factors. We investigate the use of a blind source separation method to identify past outbreaks. This method, used in neurology and called independent components analysis (ICA), directly identifies disturbance patterns. We analysed the tree-ring data from papers dealing with insect outbreaks. These papers focus on western spruce budworm, pandora moth and Douglas-fir tussock moth outbreaks. We compared the results of the original analyses, conducted using the host/non-host approach, with results from ICA. We detected the outbreaks identified in the original papers. However, the start and end dates for the outbreaks were different in 75 per cent of the ICA analyses. On the other hand, we were able to detect growth reduction in non-host Ponderosa pine chronologies as well as increased growth during outbreak periods. Since conventional methods may be less robust when the growth of non-host trees is affected, the ICA may provide a powerful new method to identify outbreaks in such situation.