Long runs
A dataset that starts this year is already more valuable than one that starts next year. After fifty years, there is no substitute.
Why time is the variable you cannot fake
Ecology runs on repetition. A single count tells you what was there on that day; a decade of counts begins to show a trend; half a century begins to separate signal from noise — the underlying shift from the annual fluctuation that would otherwise mask it. This is why long-running datasets occupy a different category from everything else in the discipline.
The Common Bird Census, begun by the British Trust for Ornithology in 1962, is one of the clearest examples. By the time analysts could see farmland bird populations collapsing in the 1970s and 1980s, the data behind that finding already spanned two decades. Without the earlier baseline, the collapse would have been invisible — or at least undatable, and therefore unprovable. The transects and point-count methods used in its successor schemes only mean something because the same methods have been applied, season after season, on the same ground.
The same principle holds in plant ecology. The Rothamsted long-term experiments, some running since 1843, demonstrate what fixed-plot recording accumulates that a short study simply cannot: slow processes, regime shifts, recovery arcs measured in generations rather than grant cycles.
The practical problem is institutional fragility. Long runs require continuity of method, continuity of observer (or rigorous handover protocols), and funding that survives reorganisations. Method changes are the quiet killer — a recorder who shifts from one identification standard to another mid-series, or a survey that moves its visit date by a fortnight, can introduce a step change that is indistinguishable from a biological event. Retrospective correction is possible but always approximate.
This is why ecologists treat a fifty-year dataset with something close to reverence, and why the loss of one — through funding withdrawal, site development or simply no one picking up the clipboard — is genuinely irreversible. You can restart a survey. You cannot recover the years that passed unrecorded. The data that does not exist yet, for the trend that will matter in 2070, is being either gathered or lost right now.