I have a dataset of sample concentrations of several ions/elements (Ca 2+, SO4 2-, K +, S, Ca, V, etc.) from an ice core which I am using to detect volcanic eruptions. The raw version of the dataset has a high resolution (many samples per year) for recent years and an increasingly low resolution for older samples. I also have a resampled version of the dataset, where the data points are interpolated at a regular time interval (quarter-year frequency), for my other analyses, as most types of time series analyses assume that the data is collected at regular time intervals.

I would like to perform an EOF (Empirical Orthogonal Function) analysis on the dataset, because a mode that has a high sulfate loading may represent volcanic activity. I read somewhere (sorry, I don't remember where exactly, but I think it was on Cross Validated) that the even spacing of the time series is not required for EOF/ Principal Component Analysis, but I am not sure why.

Now my question is: for my EOF analysis, should I use the raw data (the unevenly-spaced time series) for maximum precision of my results, or should I use the resampled data (linearly interpolated data points at a quarter year frequency) in order to avoid all the problems entailed by an uneven time series? (I am not sure exactly what those are, but I imagine problems like overestimating trends based on recent observations or underestimating trends present in older observations)



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.