Population trends
We calculated the population trends from their collated annual abundance
indices. This was done separately for butterflies and moths.
For the butterflies, at each site, annual indices were computed from the
weekly counts, following the method described in Dennis et al.(2013, 2016) and implemented in the rbms R package (Schmucki et
al. 2020). Missing week counts were derived from a Poisson generalized
linear model (GLM) that included the regional flight curve as an offset
(Schmucki et al. 2016). Collated annual abundance indices were
then estimated with a weighted Poisson regression, accounting for site,
transect length and using the proportion of the flight curve monitored
as weight. Thereby, sites with many missing counts during the flight
period had lower weight than well monitored sites. For each species, we
calculated the long-term trends with a linear model that we fitted on
the log10 transformed collated annual indices, starting
with the year the species was first recorded until the last within the
1999-2017 period.
For the moth species, population trends were estimated using the TRIM
software (Pannekoek & Van Strien 2005), as implemented in the rtrim R
package (Boogart et al. 2020). TRIM uses Poisson regression to
estimate annual abundance indices, while accounting for missing
observations, site differences, overdispersion and temporal
autocorrelation. As a long-term trend estimate, TRIM calculates a
regression through the annual indices, and this linear trend slope (on
the log scale; the “additive” slope in TRIM) was used as a measure of
population trend for the moth species over 1993-2016. Four species
appeared in the dataset after 1993, with the first year of occurrence
marking the start of the timeframe for trend calculation.