Sampling in space is a task that requires plenty of thought, as demonstrated by the previous sections. Sampling in space and time (i.e. monitoring) requires even more thought as there are even more options. Generally, if one wants to sample repeatedly then the focus will be (at least partly) on trends through time. It is commonly established in the survey literature that the uncertainty around temporal signals is reduced by repeatedly visiting the same sites (e.g. Urquhart and Kincaid, 1999). This comes at a cost though – less sites are sampled and therefore the sample may not be as representative of the population as it could be. Extreme cases in marine sampling are when the sampling gear actively alters the population size (through extractive sampling) or its habitat (for example removal of epibenthic structure). In these cases, repeatedly sampling the same sites will not reflect the trends in the population.
Intuition tells us that, unless sampling is destructive, then you should revisit at least some of the sites. This is due to the reduction in variation in the temporal signal (the site-to-site variability is removed). The proportion of sites to be revisited, and the pattern of revisits (e.g. rotating panel, fixed panel, and so on – see McDonald, 2003), will depend upon the temporal (and spatial) variability of the biota under consideration (see Perkins et al., 2017, and references therein). A power analysis of some repeat visitation patterns is provided in Urquhart and Kincaid (1999).
In a marine environment, it is often not possible to revisit the same site exactly., in most cases we recommend attempting to revisit according to the same revisit schedule. Whilst it is not optimal to have spatial separation in the revisit locations, sites that are closer together in space will have higher covariance (through spatial autocorrelation) than sites that are more distal. Thus, the temporal signal is still maximised by minimising the spatial variance. An important, and possibly reasonably frequently occurring, exception to this general advice are for those measurements that have low spatial autocorrelation. In such situations, the temporal signal will require more sampling effort to determine and the assumption of repeat transects (which are not in fact repeats) will confound temporal trend with spatial variability (Perkins et al., 2017).
Legacy sites, which come from a previous randomised survey, can, and should, be incorporated into a temporal monitoring program. Our advice is to try and make sure that some legacy sites get sampled during each revisit for new sites. This has the effect of ensuring ‘a link back to the legacy site time-series’ for each revisit. If the biota change randomly/unpredictably and rapidly, even at the same spatial location, then there is little point revisiting sites. This is especially so for monitoring programs with substantial time between revisits. In summary, think carefully about the relative importance of the temporal signal versus the generality. This will reflect the number of revisits to perform.