All possible 20-year long subsets of our full dataset were taken (black lines), as well as all possible 20-year long subsets of the data for each individual trap (grey lines). We calculated change in biomass per year in two ways (regression over all years versus difference between first-last years), as for Extended Data Fig 7. Panels show a regressed percentage change per year, and b percentage change per year between first and last years, against onset of the data subset. Twenty-year time sequences that originate before the late 1970s (that is, well before the 1982 peak) consistently suggest increases in biomass over time, whereas sequences originating later than this typically indicate declines in biomass. The two-sample estimates (black zig-zag line in b) are much noisier than those based on continuous sampling (black curve-like line in a).