A quick note for reference. Timeseries image or grid data can be fourier decomposed to create a new image dataset of the timeseries. This is based on work by Rogers at Oxford University. The components extracted from a timeseries are the amount (mean), variability (annual amplitude) and timing (annual phase). Various classification methods can then be used on the data e.g discriminant analysis, decision trees etc.
Fourier in 4D?