I'm looking at 30 years of growing season Landsat 5 NDVI for small areas (three sites, each with 9 pixels). For some years there are only one or two available images due to cloud cover, other years have more. I do not believe I have enough information to do any sort of interpolation (correct me if I'm wrong), so as I see it I should either take the annual max pixel value per site, or an annual average of the max pixel value for each date with available imagery per site.
For example:
Site A (Max pixel value per available date)
May 19, 1984: 0.67
June 3, 1984: 0.71
August 4, 1984: 0.65
If I take the max per year I would get a value of 0.71, if I took the average of the available dates I would get 0.68.
Either way will not be perfect given the inconsistency between years (e.g. one year might have images available for June, July, and august, while another year might only have one image for May). I'm not sure which method is more appropriate/justifiable as I have not found literature that deals with this specific issue.
