Forest hoverfly community collapse: Abundance and species richness drop over four decades

To study insect decline, an important threat to biodiversity, long‐term datasets are needed. Here we present a study of hoverfly (Diptera: Syrphidae) abundance and diversity in a Dutch forest, surrounded by other forests, and analyse the variation in insect numbers over four decades. Between 1982 and 2021, abundance decreased by 80%. Until 1990, abundance showed a strong decrease of 10.9% per year, mainly in nationally rare species with carnivorous larvae exposed to air. From 1990, abundance stabilised, whereas from 2000, a second period of strong decline of 9.0% per year occurred, mainly in very common species. Species richness also declined strongly between 1979 and 2021: the total number of species observed in five monitoring days dropped by 44% over those 43 years. The characteristic set of dry‐forest hoverfly species disappeared over four decades. The number of nationally rare species observed at the study site declined from 19 to 9 early on, in a period (1979–1984) that coincided with intense nitrogen input and acidification caused by agriculture in the same region. The more recent decline is likely also caused by factors from outside the forest, as forest management and conditions remained constant. Continued influx of nutrients and pesticides at a regional level, as well as climate change are possible causes of the decline. Research is needed to quantify their relative effects.


INTRODUCTION
Animal diversity is decreasing in the world (Dirzo et al., 2014;Goulson, 2019) and Red Lists are a formalised tool to illustrate declines (IUCN, 2019). Knowledge on former and present populations is needed for Red Lists, and with well-studied species such as mammals and birds this tool has been successful. However, datasets to evaluate declines of insects are less available, due to the enormous diversity in these smaller species, the specialism required to identify them, and as a consequence fewer people to monitor species groups.
Since 2000 long-term research in insects illustrated declines in distribution in well-studied insects such as butterflies, bumblebees, moths, bees, and flies (Biesmeijer et al., 2006;Bourn & Thomas, 2002;Carvell et al., 2006;Conrad et al., 2006;Shortall et al., 2009). In more recent years, many case studies and reviews prove that the distribution and abundance of many other insect groups have declined by 30%-75% within a few decades Dirzo et al., 2014;Fox et al., 2014;Grabener et al., 2020;Hallmann et al., 2017Hallmann et al., , 2020Homburg et al., 2019;van Strien et al., 2019). Society realises that these insects are an essential link in the food web, and that they facilitate essential processes such as pollination and nutrient recycling (Habel & Schmitt, 2018;Raven & Wagner, 2021).
Because so many insect species exist, with a wide range of life history strategies, a decline in numbers can be explained with many hypotheses , for instance, due to the disappearance of a specific habitat or landscape structure (Keil et al., 2011;Moquet et al., 2018;Seibold et al., 2019). Many hypotheses focus on general changes in the environment by human activities like nutrient inputs, spread of pesticides used in agriculture, drought and climate change (Christopher et al., 2021;Wagner et al., 2021), pointing at regional rather than local drivers (Habel et al., 2019;Seibold et al., 2019). Mostly these hypotheses cannot be tested due to the absence of datasets on the spatial and temporal variation in insect abundance and/or equally detailed information on relevant environmental factors (but see Hallmann & Jongejans, 2021 for a spatiotemporal analysis of effects of weather, land use and micropollutants on aquatic insects). However, as an alternative, traits of species with either stable or declining populations can be analysed to see whether these are consistent with hypothesised mechanisms due to certain environmental changes. For instance, Bowler et al. (2021) found that German dragonfly species that are generally known to be cold-adapted or preferring standing waters had decreasing distributional ranges, whereas those of warm-adapted and stream-preferring species increased.
Within Diptera, the hoverflies (Syrphidae) are among the best investigated families with many records. They are mostly not difficult to identify, and the species are observed by relatively many people, resulting in knowledge of their distribution and links with ecosystems (e.g. Reemer et al., 2009;Speight et al., 2020). Biesmeijer et al. (2006) compared pre-and post-1980 species richness of Syrphidae (and bees) per 10 Â 10 km grid cells with rarefaction. They did find comparable numbers of grid cells with decreasing and increasing species richness of Syrphidae in the Netherlands. However, they found a significant decrease in rare species and an increase in common species. More recently, Powney et al. (2019)  In this article, we present a long-term (1979-2021) study on hoverfly abundance and species richness in a Dutch forest ('Boeschoten') with little human disturbance for many decades, meaning that local drivers for changes are absent. Most previous insect studies concentrate on agricultural landscapes and miss combined data on abundance and species richness over several decades. There are relatively few studies on insect trends in forests (e.g. Habel et al., 2019;Roth et al., 2021), while forest-dwelling hoverflies remain particularly understudied. Preliminary results from Boeschoten are discussed by Barendregt (2001). Here, the data are extended 21 years and used to answer the following research questions: Are there trends in forest hoverfly abundance and species richness, and do trends vary over the years? Do trends depend on species traits? Can environmental drivers of change be indicated based on the traits of declining hoverfly species?

Study site
Boeschoten is an (extensively used) agricultural enclave within large mixed deciduous-coniferous forests in the centre of The Netherlands, 40-50 m a.m.s., west of the village of Garderen (52 13 0 24 00 N, in the ground layer the vegetation changed in some species after the period of intense acid rain ('Waldsterben') around 1985, when Galium saxatile and Deschampsia flexuosa decreased and Rubus increased in abundance. This forest was selected because it was representative of forests in the Veluwe region, because it was (and still is) a rather homogeneous old forest (i.e. no large roads, no disturbance due to recreation, no nearby intensive agriculture), and because of its proximity to the home of the first author. The hoverfly community in this forest did not appear to be richer or poorer than in other parts of the Veluwe region in 1975Veluwe region in -1985. This Veluwe region (50 Â 25 km) is the largest forest-heathland area in the Netherlands. In the surroundings of Boeschoten, there are some smaller arable fields; 5 km to the south and west there is intensive livestock farming.

Data collection
Within the 'Boeschoten' forest, the same permanent route of approximately 3 km has been inspected for the presence of hoverflies (Diptera: Syrphidae) in the second half of the morning (10:00-13:00), for a duration of approximately 2 h. The forest was included up to 30 m from the route to obtain a complete inventory of the ecosystem; a complete list of all present hoverflies was aimed for each time. Monitoring was done only on sunny days, independent from temperature.
All observed specimens have been counted and collected with an insect net, species and sex identified in the field, or preserved for identification later on (with e.g. Barendregt, 1978;van der Goot, 1981;Bot & van de Meutter, 2019;Speight et al., 2020).
Vouchers are deposited in the first author's collection. During the survey, Syrphus nitidifrons was found for the second time ever, only after the original description by Becker in 1921 (Barendregt, 1983). While monitoring started in 1974, it was only from 1979 onwards that complete species lists were kept, and from 1982 onwards that the number of observed individuals per species was recorded ( Figure 1). The sampling days were irregularly distributed over the years (Appendix A of Data S1) and throughout the whole flight season (April-October, with September slightly under-sampled).

Abundance trend analysis
Daily totals of the number of observed hoverflies were analysed to test whether there was a trend in total hoverfly abundance over the years. As total hoverfly counts showed a clear bimodal pattern throughout the season (with peaks in spring and summer), we first fitted a generalised additive model (GAM) to all daily counts as a function of the day of the year (irrespective of year and weather). This GAM was then used to predict day-of-the-year-specific relative hoverfly abundance ('season score'), from 0 in winter to 1 at the time of the highest peak (summer). This season score, and a continuous year variable, were included in a generalised linear mixed-effect model of the daily counts ( Figure 2). Year was also a random (categorical) factor to account for the nestedness of multiple visits within a year. We assumed a negative binomial distribution, as the variance of the daily counts was much larger than the mean.
Because flight activity of Syrphidae is known to vary with weather (Gilbert, 1985), we also explored models that included sets of weather variables, to study whether potential weather effects on daily counts could have affected trend estimates. These weather variables and analyses are explained in detail in Appendix B of Data S1.
All visits since 1982 were included in the hoverfly abundance trend analysis ( Figure 3). A large number of these visits date from the first 9 years (139 = 55% in 1982-1990). No visits were done in the periods 2001-2009 and 2012-2019, resulting in two temporal gaps in the dataset. In addition to the main analysis of all data in the 1982-2021 period, we also fitted the same type of model to the 1982-1990, 1982-2000 and 1982-2011 subsets (to study whether the trend estimate changed with the increasing length of the time series ;Didham et al., 2020Didham et al., ) and 1990Didham et al., -2000Didham et al., and 2000Didham et al., -2021 (to see whether trends in total counts differed among time periods). We also performed trend analyses of total hoverfly abundance separately for data from the spring peak period (19 April to 3 June) and summer peak period (13 June to 8 September). Hoverflies can be univoltine or bi-or polyvoltine, and we studied the combined abundance trend of all univoltine species, as well as that of the group of bi-and polyvoltine species.
The same model structure was used to quantify trends for subsets of the species. For instance, all species were categorised (according to the non-zero fuzzy scores of Speight et al., 2020) into one of five larval feeding strategies: carnivorous (mainly on aphids; living in either trees and shrubs or the herb layer), phytophagous, saproxylic, aquatic, and 'other' (see Appendix C of Data S1). The 'other' group included species with larvae living in rotting plants, manure, or insect nests. Species were also grouped depending on whether their larvae are mainly (i) exposed to water, (ii) exposed to air, or (iii) hidden within plants or insect nests. Furthermore, we fitted separate models for rare (including very rare and fairly rare), common (including fairly common), and very common species in the Netherlands. Rarity was based on classification by Reemer et al. (2009). And finally, abundance trends were analysed separately for the 10 hoverfly species with the highest number of individuals observed over the 1982-2021 period (Table 1).

Species richness trend analysis
For each of the 105 hoverfly species recorded since 1979, we noted whether it was observed in each of six time periods : 1979-1982, 1983, 1984-1987, 1988-1994, 1995-2011, and 2020-2021. These combinations of years were chosen to minimise the variation among the periods with respect to the number of monitoring days: 45, 45, 45, 45, 36, and 63 days, respectively. This allowed us to visualise the turnover of species.
Since monitoring days were differently distributed over the seasons in different years, while also the number of monitoring days differed between study years, standard methods for estimating and comparing species richness among years were not applicable (i.e. the assumptions of methods like Chao2 (Béguinot, 2014;Chao et al., 2017) were not met). Instead, we visually inspected differences in species accumulation curves between years (starting in 1979), and performed simple statistical trend analyses, as explained below ( Figure 4). First, we created year-specific species accumulation curves by randomly reordering the monitoring days of a particular year and  T A B L E 1 Summary of abundance trend analyses for different subsets of the hoverfly species and time periods.

Annual abundance change
Hoverfly group 1982-1990 1982-2000 1982-2011 1982- To statistically test for potential trends in species richness, we used the number of unique species observed during five randomly chosen monitoring days as a year-specific statistic of species richness.
Generalised linear regression models (with Poisson distribution) were fitted using the mean estimates of year-specific species-richness in 5 days, based on a 1000 random draws of 5 days per year ( Figure 5).
The choice for a threshold of 5 days was the result of balancing the need for multiple monitoring days to arrive at a robust estimate of species richness, and the aim to include as many years as possible in the analyses (i.e. only excluding years with less than five monitoring days). We also plotted species accumulation curves based on the spring peak days only, and on the Summer peak days only. For each of the peak periods the same minimum of five monitoring days was required for a year to be included in a trend analysis of species richness.

RESULTS
In 254

Abundance trends
The GAM fitted to the daily total number of hoverfly individuals (as a function of day of the year), indeed showed a bimodal pattern with spring and summer peak periods (Figure 2). The GAM was then applied to predict relative hoverfly abundance ('season score') at the day-of- abundance trend starting in 1982. The strongest decline was seen over the first 9 years (1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)): a statistically very significant À10.9% per year (with a À14.3% to À7.4% AE1 standard error range; Figure 3; significance levels given in Table 1). Daily numbers, although highly variable, appeared to stabilise and even slightly improve in the 1990s, leading to non-significant decline over the 1982-2000 period: À2.0% per year (À3.8% to À0.1%). Adding the low daily numbers of 2010 and 2011 reduced the trend to a nearly significant À2.5% per year (À3.7% to À1.2%). Over the entire study period (so including the very low numbers in 2020 and 2021), there was a highly significant decline in individuals: À4.1% per year (À4.8% to À3.3%), resulting in an overall loss in individuals of 80% over 40 years (À85% to À74%).
We observed that this decline is varying over time. As indicated earlier, the strongest loss was found over the first 9 years, resulting in a loss of 60% over 9 years (À71% to À46%). Looking at the 1990-2000 period, we see that the total number of hoverfly individuals was stable (i.e. slightly positive trend, but not significantly: +0.9% per year [À3.1% to +5.0%]; Table 2). On the other hand, the 2000-2021 period showed a highly significant negative decline of À9.0% per year (À10.6% to À7.3%). The differences in trend between observations in spring and summer were small: both followed the time-period-specific trend based on all monitoring days (Table 2), while in the 2000-2021 period, the negative trend was stronger in summer (À11.9% per year [À15.2% to À8.5%]) than in spring (À7.1% [À9.4% to À4.8%]). In the 1980s, the abundance of univoltine species declined more strongly than that of species with more than one generation per year (Table 1).
However, with new years of data added, the differences in the negative trends of these two groups of species disappeared.
Next, we fitted separate models for nationally rare (including very rare and fairly rare), common (including fairly common), and very common species. Rare can be used as an indication of special conditions and very common of the conditions that are generally present. The rare species especially declined significantly in the period 1982-1990 with À26.8% (À31.8% to À21.4%), and to 2021 with À5.1% per year (À6.8% to À3.5%; Table 1). The common species declined in all periods, and to 2021 by À4.6% per year (À5.5% to À3.7%), whereas the very common species only declined significantly when the whole study period was considered: À3.6% per year (À4.5% to À2.8%). Over any period of time studied starting 1982, the (relative) decline in numbers of rare species is the strongest, those of very common species the weakest. To illustrate that species respond individually in time, we added in Table 1

Species richness trends
The results in species richness from Boeschoten are summarised in  (starting in 1982 and ending in 1990, 2000, 2011 and 2021). The fitted models account for seasonal patterns ( Figure 2) and a random effect of year. Annual change (%) and statistical significance can be found in Table 1. Upward-pointing triangles indicate days during the spring-peak period, downward-pointing triangles indicate days during the summer-peak period. Monitoring days outside those periods are indicated with circles. A log-scale version of this graph can be found in Appendix D of the Data S1 To test the species richness by year, species accumulation plots were calculated (Figure 4). T A B L E 2 Summary of abundance trend analyses of the total number of hoverflies per monitoring day for three time periods.
T A B L E 3 Summary of species numbers in six time periods with comparable numbers of sampling days Time period 1979Time period -1982Time period 1983Time period 1984Time period -1987Time period 1988Time period -1994Time period 1995Time period -2011Time period 2020Time period -2021 Number In this dry mixed forest, we observed two waves of decline in species richness. In the 1980s, the diversity of rare, characteristic species decreased by 27%. Most common species remained present at least until 2000, after which the number of observed species in that category also declined. In recent years, the total number of individuals has decreased dramatically, mainly due to decreased abundance of the previously common species (Table 1).
It appears that in this process of overall collapse of the hoverfly community, the specialised species are already lost in an early per year for the whole season, spring peak period and summer peak period, respectively, with p < 0.002 in all cases). The 1000 grey lines in each panel represent models fitted to each of the 1000 sets of obtained estimates by randomly selecting five sampling days each time. In all cases, the year effect was negative. In 95.5%, 99.8% and 100% of the cases, respectively, the p-value was below 0.05 insect group, it is known that specialist species decline first, whereas generalists stabilise or increase (Habel et al., 2016).
Beside the important ecological role of rare species in an ecosystem (Jain et al., 2014;Leitao et al., 2016;Mouillot et al., 2013), our data suggest that the loss of rare species may be an indicator of ecosystem development many years before the number of individuals in common species is reduced.
The decline in species richness at Boeschoten cannot be explained by local changes in land use and management: this forest and its surroundings have not changed for decades. A clue to the explanation could be that in the period 1979-1984 especially the rare hoverflies with carnivorous larvae living on aphids in open air declined, but not the species with hidden larvae that live in plants, dead trees or the ground.
It seems that the origin of the change in hoverfly fauna is due to airborne transport; this period is known for the extreme emissions of nitrogen (acid rain) from agriculture, which was reduced by national leg-  The second study with which we compare our results is that of Flanders (based on declining distributions). The majority of these 29 species belong to the group of species with carnivorous larvae that was decimated at Boeschoten.
In conclusion, we can add hoverflies to the groups of insects that have recently declined by 70%-80% in western Europe (e.g. Hallmann et al., 2020;Homburg et al., 2019;van Strien et al., 2019). The fact that we cannot pinpoint an obvious direct effect of local human interference in our forest study, can only lead to the conclusion that large-scale processes such as influx of nutrients and pesticides, acidification and/or climate change contribute to insect declines . To substantiate these strong suspicions with data, it is critical that long-term monitoring of insect populations and potential environmental drivers is conducted at a multitude of locations, for