In North American BBS route counts (1966–2024), the variance of change grows to 55-year lags; observer changes cause most apparent trend reversal and the sub-linear growth exponent

Question: does the variance of change in local bird counts keep growing with the time lag, and how much of the lag pattern and of local trend reversal comes from observer changes? Data: North American Breeding Bird Survey 2025 release (doi:10.5066/P14SNUV4), standard runs, 983 routes surveyed from 1966–1975 to 2015–2024, 35,260 species-route series (presence at least 75%, mean count at least 3), 3,950 observers; replication with Swedish fixed routes 1996–2025 (161 routes). Method: semivariance of log(count + 1) by lag, split into same-observer and different-observer pairs; models fitted to 1966–1995 pairs and tested at lags 30–55; correlations of disjoint adjacent increments; correlations of 1966–1994 and 1996–2024 slopes, from OLS and with one intercept for each observer; route bootstrap; simulations on the real observer sequences. Results: the semivariance increased from 0.242 at lag 1 to 0.907 at lag 50 and did not stop increasing. A power law fitted to early pairs (exponent 0.52) predicted lags 30–55 (mean error −0.000, 95% interval −0.030 to 0.031). A saturating model under-predicted (−0.068), and a linear model over-predicted (+0.116). A change of observer added 0.184 to the semivariance. Inside each pair class, the growth exponent was 1.26 (same observer) and 0.96 (different observer). Thus the low all-pairs exponent comes mainly from the increasing share of observer-change pairs. With species means removed, the slope correlation between halves was −0.134 for OLS slopes and −0.018 for within-observer slopes. Increments that contain an observer change had correlations of −0.09 to −0.17. Increments from one observer had correlations near zero. Swedish routes gave the same direction (−0.166 against −0.071). Simulations of a random walk with observer offsets and drift reproduced these patterns. A stationary AR(1) model did not. Limitation: within-observer comparisons do not remove slow drift in one observer's counts. Thus the results combine population change with detection drift, and weak population regulation is not excluded. The results apply to common species on long-surveyed routes and describe means over series.

breeding bird survey · observer effects · population variability · variogram · time series length · spectral reddening · density dependence · trend persistence · random walk · Swedish Bird Survey · measurement error