A Results species models

The following sections display plots with model predictions for each species. The response value in these plots is the probability of presence for a species (expressed as a percentage). These plots can be used to judge for individual species how the chance of occurrence in 1 km x 1 km squares (where the species has been seen at least once) evolved over time. The map of the spatial smooth can be used to discriminate between areas where the species is on average less likely to occur (blue-ish) and more likely to occur (red-ish), or does not occur (empty areas).

Each plot for a smoother is conditional on the effects of the other smoothers. To aid at what values of the covariates (year, list length, x and y) the smoothed effect is zero, the reader should look at the dashed lines in the figures. For the year-smoothers, the horizontal dashed line corresponds with the intercept value plus the effect when list-length is set to 130 in the response scale (the spatial smooth is set to zero). The vertical dashed line(s) are drawn where the smoothed fitted curve crosses the horizontal dashed line and allow to read of the year(s) when this occurs. The confidence bands surrounding the year smoothers represent 30% (darkest), 60% and 90% (lightest) confidence levels. In the case of a two dimensional smoother, the dashed line indicates the intercept (grand mean) plus the conditional effect for the final year of observation plus the conditional effect for list-length equal to 130 in the response scale (the probability value is indicated on the dashed contour line). Colours are diverging from this dashed contour line and delineate areas of higher probability of occurrence (red) from lower probability of occurrence (blue).

The plots visualizing spatial smooths should not be mistaken for a distribution map giving real chances of occurrence in a km-square in a certain area. This is because the models use only km-squares where the species has been observed at least once during the entire period. Including also km-squares where the species has never been observed within a species’ extent of occurrence would allow us to make distribution maps that better show actual chances of occurrence in a km-square in a certain area, but this was not the goal. Models that focus instead on km-squares where the species has been observed at least once during the entire period - as in this report - better allow to show temporal change.

Because of the nature of these data (occurrence data), these models are expected to be more sensitive to an increase in the area of occupancy of a species than to a decrease in the area of occupancy. The reason is that expanding or colonizing species need only be observed in low abundance in km-squares where it used to be absent. On the contrary, when a species is in decline, it will first show signs in declining abundance and it is only after the last (detectable) individual of the species is lost from a km-square that it will be noted as absent from that square. Moreover, many plant species are known to withstand adverse conditions for a long time, which leads to a so-called (local) extinction debt.

The models presented here are not adjusted for detectability of plant species. Inconspicuous species may be easier missed during a survey of a km-square. However, because we only use km-square - year combinations which have been thoroughly surveyed (defined as a list with more than 100 species observed, i.e. list-length > 100), we can put more confidence in claiming that a plant is absent from some km-square in a certain year. Still, larger list-lengths will be the result of two processes which we cannot discern with our data: (1) a km-square may be more speciose per se (for instance due to many different habitats, soil types, …), increasing the chance that a plant species is present, and (2) the km-square has been more thoroughly surveyed (e.g. more visits, more surveyers, longer visit duration, observer experience, …) or is better surveyable (e.g. good accessibility to all habitats), increasing the chance that a plant species is detected. List-length was included in these models and will thus capture a combination of both processes.

Finally, in the early decades (’50 - ’80) for some species, there can be a noticeable lower number of well-surveyed km-squares in these years (where the species has been observed at least once) compared to years from later decades. Therefore, for these species, apparent short-term trends in the smoothed trend pattern in these earlier decades will be less reliable compared to later decades and should be treated with caution. This is, for instance, the case for Adoxa mosschatellina L..