Japanese animal first-appearance dates were 5.8 days later per unit share of nearby no-record years in four regions and two periods, 1953-2020; plant dates showed 1.3 days

Question: do first-appearance dates of animals in the Japan Meteorological Agency (JMA) phenology network contain a detection bias? A first date depends on the number of individuals and on search effort, not only on the timing of the population. Approach: we used 22 animal items (birds, insects, frogs, a lizard) and 31 plant items at 101 JMA stations from 1953 to 2020, with local temperature from GHCN-M v4. In a year when the station was active, a missing record inside the span of a series counts as non-detection. For each dated record we calculated the share of non-detected neighbour years (y-2 to y+2). Fixed-effects models held local 60-day temperature, the station-item series and each item-year constant; standard errors were clustered by station. Plants were the control. Results: animal first dates were 5.8 days later (95% CI 4.7 to 7.0) per unit share of non-detected neighbour years. The slope was positive in four regions (3.8 to 9.9 days) and in 1953-1986 (4.8) and 1987-2020 (5.1), and positive for all 22 animal items (13 with CI above zero). The plant slope was 1.3 days (0.2 to 2.3); the animal-minus-plant difference was 4.6 days (3.0 to 6.1). A within-series shuffle gave a null slope of 0.0 (range -0.7 to 0.7). Window length, adjusted temperature and on-site-only records did not change the animal slope; series x decade fixed effects reduced it to 2.4 days (1.3 to 3.6). After temperature control, animal first dates moved later by 1.6 days per decade and plant dates by 0.2 days per decade. Animal series with a rise in non-detection of more than 0.1 had a delay of 3.2 days per decade; series with no rise still had 1.2 days per decade. Limitations: a missing record can also come from programme or observer changes; we have no abundance or effort data, so we cannot separate fewer animals from less search effort. Non-detection explains only a part of the long-term animal delay. Climate studies that use these first dates must control for detection.

phenology · first appearance date · detection bias · observation bias · Japan Meteorological Agency · climate change · birds · insects · cicadas · frogs · population decline · GHCN-M · fixed effects · urbanization