Ecological and life-history traits predict temporal trends in biomass of boreal moths

1. Dramatic insect declines, and their consequences for ecosystems globally, have received considerable attention recently. Yet, it is still poorly known if ecological and life-history traits can explain declines and whether insect decline occurs also at high latitudes. Insects ’ diversity and abundance are dramatically lower at high latitudes compared to the tropics, and insects might benefit from climate warming in high-latitude environments. 2. We adopted a trait-and biomass-based approach to estimate temporal change between 1993 and 2019 in Finnish macro-moth communities by using data from 85 long-running light traps. We analysed spatio-temporal variation in biomass of moth functional groups with Joint Dynamic Species Distribution Models while accounting for environmental variables. 3. We did not detect any declining trends in total moth biomass of moth functional groups, and most groups were stable over time. Moreover, biomass increased for species using coniferous trees, lichens, or mushrooms as hosts, multivoltine species, as well as monophagous and oligophagous species feeding on trees. We found that length and temperature of the growing season, winter climatic conditions, and habitat structure all partially explained variation in moth biomass. 4. Although boreal moth communities are rapidly changing due to species turnover, in terms of total biomass they seem to contradict the trend of dramatic insect declines observed globally. This may lessen the immediate possibility of negative bottom-up trophic cascades in boreal food webs.


INTRODUCTION
Declining insect abundance following anthropogenic change, has been reported repeatedly in recent years (Habel, Trusch, et al., 2019;Hallmann et al., 2017;Janzen & Hallwachs, 2019;Lister & Garcia, 2018;Seibold et al., 2019;Wepprich et al., 2019). Such loss of insects has potentially severe consequences for ecosystem functioning (e.g., predator-prey relationships) and ecosystem services (e.g., pollination), since insects play ecologically important roles in almost all terrestrial ecosystems (Dirzo et al., 2014;Wagner, 2020;Walther, 2010). Although insect decline has received much attention, the responses of insects to environmental change are complex and heterogeneous, and insects are not declining everywhere (e.g., Crossley et al., 2020;Macgregor et al., 2019;Montgomery et al., 2020;van Klink et al., 2022). Furthermore, temporal abundance trends differ between insect taxa (Crossley et al., 2020;Engelhardt et al., 2022;van Klink et al., 2022), and biomass trends appear to depend on the spatiotemporal resolution of the studies. For example, Macgregor et al. (2019) concluded that estimates of long-term biomass change may be sensitive to the length of the time series and the number of sampling sites.
Global change affects distribution, occurrence, diversity and phenology of insects (Forister et al., 2010;Fox et al., 2014;Habel et al., 2021;Pilotto et al., 2020;Raven & Wagner, 2021;Wagner et al., 2021;Wilson & Fox, 2021). Although there are many insect abundance datasets from temperate environments (e.g., Hallmann et al., 2017;Wagner et al., 2021), northern high-latitude environments are underrepresented. Nonetheless, climate warming is more pronounced in these regions (IPCC, 2014;Rantanen et al., 2022) and species turnover is hence likely to be stronger at high latitudes (Antão et al., 2022;Pilotto et al., 2020). There are two main processes concerning insect abundance change in high-latitude environments. On the one hand, insects are generally a low-latitude taxon with the highest diversity and abundance in the tropics, and relatively low diversity and abundance at high latitudes. It is thus expected that climate warming increases insect abundance at high latitudes as it supports influx of new species. On the other hand, land use change, eutrophication, pesticide use, light pollution and climate change (negative effects on northern species) are expected to lead to insect decline. Both abundance declines Hällfors et al., 2021) and no change in abundance and biomass (Kohonen, 2020) have been reported in Finland. Here, we aim to investigate temporal biomass trends of moths from the boreal region in Finland and to see if any of these two processes could explain the changes we observe in the biomass of moth functional groups.
In addition to uncertainties in the abundance trends of insects in general, in specific areas, and/or specific taxa, there is a lack of knowledge on the association between ecological traits and temporal abundance trends. Consequently, a trait-based approach could potentially be informative of the ecological correlates of changes in abundance.
However, the trait-based approach has not been used widely in studies of biomass and abundance changes in Lepidoptera Coulthard et al., 2019;Roth et al., 2021). In temperate moth communities, the traits most strongly correlated with abundance decline are diet breadth, body size, voltinism, length of the flight period, dispersal ability and overwintering stage (Coulthard et al., 2019;Wagner et al., 2021). For instance, Coulthard et al. (2019) reported that larger-bodied moths are declining more often in the UK, and according to Roth et al. (2021), moth declines were more pronounced in host plant specialists and dark-coloured species. Here, we extend the trait-based approach to boreal moth communities.
To comprehensively understand temporal insect abundance change, there is a need to identify key environmental variables affecting short-term abundance fluctuations and long-term abundance trends. Environmental drivers that are commonly associated with global insect collapse include habitat destruction, climate change, agricultural intensification (including pesticide use), and invasive species (Fox, 2013;Thomas, 2016;Wagner, 2020;Wepprich et al., 2019).
Short-term abundance fluctuations are typically influenced by weather conditions (see e.g., Pöyry et al., 2011;Roy et al., 2001). In boreal environments, the growing season is short, winter is long and severe, and uniform snow cover may last up to 7 months. Duration and temperature of the growing season determine the physiological time available for insect growth and development and the number of generations appearing per year (i.e., voltinism) (Kivelä et al., 2013(Kivelä et al., , 2016Pöyry et al., 2011), which affects abundance. In addition to conditions during the growing season, winter conditions likely affect insect abundance in boreal environments too, both through cold mortality (Leather et al., 1993) and increased energy expenditure during diapause due to climate warming (Bosch et al., 2010;Nielsen et al., 2022). All resident species living in the boreal region are adapted to winter conditions, but overwinter survival may be compromised under exceptional winters (Abarca et al., 2019). On the other hand, duration and depth of insulating snow cover affect survival of species overwintering primarily under snow. Because climate change is rapid at high latitudes (e.g., polar amplification: stronger warming at high latitudes than the global average; IPCC, 2014;Rantanen et al., 2022), climate change is expected to adversely affect coldadapted insects, and this may translate into long-term abundance declines. Furthermore, anthropogenic changes in land cover and consequent effects on habitat structure may affect the long-term abundance trends of insects .
To address the poorly known temporal change in insect abundance and its trait-dependency in boreal environments, we take advantage of the spatially replicated and long-running (27 years) Finnish Moth Monitoring Scheme and of the detailed knowledge of traits of Finnish moth species. We aim to show how ecological and life-history traits as well as environmental factors affect abundances of boreal macro-moths over time. We classify species into functional groups based on species trait information for (1) larval host type, (2) diet breadth within host type, (3) overwintering stage, (4) voltinism and (5) body size. Then, to extract the spatio-temporal abundance trends of functional groups, we use Joint Dynamic Species Distribution Models (JDSDMs). Recent advancement in computational power has made this approach possible and has boosted community ecology studies (Ovaskainen et al., 2017;Thorson et al., 2016;Thorson & Barnett, 2017). Here, we take advantage of the VAST R package, which has been developed to model spatio-temporally dynamic communities (Thorson, 2019;Thorson & Barnett, 2017).
In the analysis, we also take growing season and winter climatic conditions into account, as well as habitat structure effects on moth abundance. Our aims in this study are (1) to investigate the temporal biomass trends in moth functional groups, (2) to explore the spatial variation in temporal trends of functional groups, and (3) to investigate the effect of climatic and land-cover covariates on the biomass variation of functional groups.

Moth data
We used macro-moth abundance data from Finland. Almost 90% of these data come from the National Moth Monitoring Scheme (Nocturna), which is coordinated by the Finnish Environment Institute (SYKE). This monitoring scheme has extensive spatial coverage throughout Finland and has been running annually since 1993 in 248 locations (Leinonen et al., 2017). Sampling covers the entire adult moth activity period from early spring to late autumn, and light traps are usually emptied weekly. We supplemented these Nocturna data with data from the Värriö nocturnal moth monitoring scheme (10% of data), collected daily at the Värriö research station in Värriötunturi, north-eastern Finland (Hunter et al., 2014). This dataset has been collected since 1978 starting with 11 light traps, two of which are still running. When combining these two monitoring schemes, we only considered data from those 85 traps (out of total of 262 traps) located in a forest environment with ≥5 years of sampling between 1993 and 2019, yielding 1452 trap Â year combinations (Figure 1e). These selection criteria were chosen to minimise habitat variation and to avoid any biases that might be caused by short-running traps (many short-running traps were running only in the very beginning or the very end of the time series). The final data for the abundance trend analysis included observations of more than 4.3 million individuals and 736 species (Trait data file in Supplementary material).
Biomass represents the overall functionality of a species, or a group of species, within the community better than abundance, as biomass is a key variable in terms of energy flow, productivity, and food-web dynamics (Brown et al., 2004;Saint-Germain et al., 2007).
Hence, we converted annual abundance into annual fresh biomass (mg) by multiplying the annual totals of species-and trap-specific numbers of individuals by species-specific body mass estimates. We used a linear regression model to estimate body mass (mg) as a function of species wingspan (mm) and body plan (i.e., stout or slender) using empirical data on fresh masses of 1542 specimens across 164 genera (9 families; 244 species; R-squared = 0.93, Table S1). We F I G U R E 1 (a) Spatial variation in mean growing season length, (b) spatial variation in temporal trends (colour gradient) in growing season temperature sum, (c) snow cover duration and (d) number of cold days (mean daily temperature < À5 C) in Finland. Panel a shows a purely spatial variable as is, while panels b-d show rates of change over time in variables with yearly values with more reddish colours showing stronger warming of summer and shortening of winter. Panel e shows the locations of light traps (circles) and mean proportions of forest and other habitat types within a 2.5 km radius from the trap sites (i.e., pie charts within the circles). The landcover has yearly values, but temporal variation in it is relatively small. then used this model to predict the fresh body masses of all species based on their wingspans and body plan (Kinsella et al., 2020;Kohonen, 2020), as given in literature (e.g., Silvonen et al., 2014). We calculated annual total biomass per species per trap location and used the pooled biomass of species grouped together based on their traits (see below) as the response variables in JDSDMs.

Moth functional groups
We aggregated moth species into functional groups in five separate ways; we selected five ecological and life-history traits and classified all macro-moth species according to the values of these functional traits Finnish

Environmental data
We used spatially interpolated weather data on daily mean temperatures and snow depths in a 10 km grid (Finnish meteorological institute; https://paituli.csc.fi/, 2021). We defined the length of the growing season as the number of days between the dates when the daily mean temperature rises above 5 C for seven consecutive days for the first time in a year and falls below 5 C for seven consecutive days in the latter half of the year. We used the mean growing season length over the period 1993-2019 to describe the spatial variation in climate across the latitudinal gradient ( Figure 1a). To describe temporal changes in summer thermal conditions, we defined growing season temperature sum anomaly as the difference between sum of day degrees above a 5 C threshold accumulated during the growing season as defined above, and the mean growing season thermal sum over the period 1984-2020 (note that a longer time period was used to derive a solid estimate of the baseline climate than the duration of the moth time series; cf. Figure

Joint Dynamic Species Distribution Model
We analysed spatial and spatio-temporal variation in macro-moth biomass with JDSDMs, fitted as Vector-Autoregressive Spatio-Temporal models by using the VAST package (release number 8.2.0; Thorson, 2019;Thorson & Barnett, 2017) in R version 4.1.2. (R Core Team, 2021). JDSDMs, as implemented in VAST, simultaneously analyse correlated spatial and spatio-temporal variation in population densities of multiple species while accounting for effects of environmental variables on occurrence and abundance of species (Thorson et al., 2016;Thorson & Barnett, 2017). VAST is a powerful tool that can be used to model spatio-temporally dynamic moth biomass data.
Instead of using species-specific biomasses as response variables, we defined the response variables of the multivariate JDSDMs to be the functional-group-specific biomasses. With this approach, we reduced the species pool from 736 species to three to eight functional groups, depending on the trait categories in the five alternative groupings, which in turn facilitated JDSDM modelling, and enabled us to draw inferences on how ecosystem-level services and functions may have changed in the past three decades. We repeated the analysis for each of the five different functional groupings of species, based on the above-mentioned ecological and life-history traits (see Table 1).
On the grounds of observed biomass of functional group c at site s and year t, VAST estimates the biomass density (biomass per unit area), d(s, c, t) for each functional group.
Four climatic variables (i.e., anomaly of growing season temperature sum in the previous year, growing season length, anomaly of snow duration in the previous winter and anomaly of cold days in the previous winter) and one land cover variable (i.e., percentage of forested land within a 2.5 km radius around trap sites) were set as covariates in the analyses. All the covariates were standardised prior to analyses (i.e., subtracted by mean and divided by standard deviation).
We specified the VAST model as a Poisson-link delta model that approximates a Tweedie distribution (Thorson, 2018). In delta models, the probability distribution for data b s, c, t ð Þis parsed into two components representing (i) occurrence probability r 1 s, c, t ð Þ for location s, functional group c and year t, and (ii) biomass of functional group c, r 2 s, c, t ð Þ, given that functional group c is predicted to occur at location s and year t.
Lognormal and Gamma distributions for biomass were evaluated.
Gamma distribution models fitted the data better than models using lognormal distribution (ΔAIC > 2; Table S2). Therefore, the probability distribution of biomass data was specified as: where a i is the sampling area (km 2 ) that is used to estimate the functional groups' densities (biomass/km 2 ). We assumed that a light trap samples moths within a circular area of 0.031 km 2 , that is, within 100 m radius from the trap. We acknowledge that the effective sampling area likely varies among species (see e.g., Merckx & Slade, 2014), habitats, dates (due to seasonal variation in night duration and darkness) and sites (due to spatial variation in night duration, darkness and habitat characteristics). However, in this analysis, the sampling area only affects the scale of the density estimates (i.e., biomass/km 2 ), and not the relative densities among sampling locations and times, the relative differences being important for our inferences. Hence, the specific choice of the sampling area does not affect our inferences.
In this specification of the delta model, density of individuals while the second linear predictor models average individual body We used the 'grid' method for spatial modelling and set 100 spatial knots (except for 95 in voltinism group due to model convergence issues). In this method, VAST defines the locations of knots as the centroids of grid cells that have at least one sampling site, which produces a systematically distributed network of knots to be assigned to each sample. Spatial ω i (s, f ) and spatio-temporal ε i (s, t, f ) random effects were modelled at the knot locations as Gaussian random fields following multivariate normal probability distributions. VAST uses stochastic partial differentiation equations (Lindgren et al., 2011) to approximate Gaussian random fields using package R-INLA (Lindgren & Rue, 2015). The Gaussian random fields are used for estimating all spatial processes (see Thorson et al., 2015 for details).
We defined Gaussian random fields to specify the spatial variation as constant over time (Equation 6), and spatio-temporal variation was set without temporal structure (i.e., independent among years; Equation 7) for both linear predictors: where R i is a spatial correlation matrix approximating spatial covariation in density of individuals (i = 1) and average individual body mass (i = 2), among locations s (s = 1, …, k). Spatial and spatio-temporal Gaussian random fields (σ 2 ω and σ 2 ε , respectively) had the variance of 1 to ensure that the covariance among functional groups is defined by the loadings matrix for that term. Spatial correlation between neighbouring locations generally declines with an increased distance between the two locations (Toblers' law of geography; Tobler, 1970).
To account for this correlation, we specified a Matèrn correlation function with smoothness v = 1, and an estimated rate parameter κ i (i = {1,2}) specifying the distance at which locations are essentially uncorrelated, as well as an estimated matrix H specifying geometric anisotropy (asymmetry in decorrelation distance to different directions; see Thorson et al., 2016;Thorson & Barnett, 2017 for details).
We set functional group-specific intercepts β i c ð Þ to be constant among years and treated them as fixed effects. Variance in spatial (σ 2 ω ) and spatio-temporal (σ 2 ε ) variation, factor loadings matrices (L ωi , L εi ), effect of environmental covariates γ j and δ j and the parameters governing the geometric anisotropy (H) and decorrelation distance (к) in the Matèrn correlation function were also treated as fixed effects.
Spatial ω i s, f ð Þ and spatio-temporal ε i s, f,t ð Þ variations were treated as random effects.
All parameters were estimated with maximum marginal likelihood using Template Model Builder (R package TMB; Kristensen et al., 2016). TMB uses Laplace approximation (Skaug & Fournier, 2006) to estimate fixed effects by maximising the log-marginal likelihood of fixed effects in the R statistical environment (R Core Team, 2019) after integrating over the random effects. The likelihood is then optimised, and standard errors are obtained using generalisation of the delta method. Random effects are then predicted based on the joint likelihood of random effects and data, given the maximum likelihood estimates of the fixed effects (more details in Thorson et al., 2015). We considered all parameters whose 95% confidence interval did not encompass zero as statistically significant.
The Autumnal moth (Epirrita autumnata) shows outbreaks with intervals of 8-11 years in northern Fennoscandia (Jepsen et al., 2008;Tenow, 1972;Tenow et al., 2007). Models including this specieseven when excluding the highest observations exceeding 10,000 individuals per trap per year-did not converge. Hence, we completely excluded E. autumnata from our analyses, which is a practice that was previously done in other Finnish moth studies (Dallas et al., 2020;Kohonen, 2020). We repeated the analysis for each of the five functional groupings based on the five ecological and life-history traits (Table 1) and used all five environmental covariates in each case.
However, the model for voltinism-based grouping did not converge when including all five environmental covariates. Therefore, we fitted models to the voltinism-based grouping data so that one of the environmental covariates was excluded at a time, and then chose the model with the lowest AIC value for inferences. None of the models for body size grouping converged, most probably due to high within group variations, thus we do not report any results from this grouping.
We investigated goodness-of-fit of the VAST models on the grounds of the diagnostic residual plots produced with the tools available in the R package 'DHARMa' (Hartig, 2020). Lastly, we used a 7-fold cross-validation procedure to evaluate the model performance, and calculated proportion of deviance explained by using tools in the VAST package to measure model explanatory power (Table S3).

Analysing temporal biomass trends
The year-specific, Finland-wide total moth biomass estimates, produced by the VAST models, were used as response variables in generalised least squares linear models to analyse the temporal biomass trends of these functional groups. The function 'gls' from R package 'nlme' normalise residuals. Temporal autocorrelation in functional-groupspecific total biomass estimates was modelled with the 'corAR1' autocorrelation function (Pinheiro and Bates, 2022).
To depict spatial variation in temporal biomass trends of 22 functional groups (Figure 3), we used the predicted yearly moth biomasses at the VAST models' spatial knots (systematically arranged 95 locations in voltinism group and 100 locations in three other groups). We then used linear regression to ln-transformed biomass time series at each knot. Finally, we took the estimate (slope for year effect) at each knot and interpolated the values to cover the whole area of Finland (Voronoi diagram of the knots) using nearest neighbour (five nearest) interpolation.

Temporal biomass trends in moth functional groups in Finland
The biomass of 15 functional groups was stable, while seven functional groups increased in biomass over the study period  in Finland. Adult biomass of species using coniferous trees and lichens as larval hosts increased, but all other functional groups in host type trait were stable over time (Table S4, Figure 2a). Half of the diet breadth within host type functional groups (i.e., lichen, monophagous-trees, mushroom and oligophagous-trees; note that lichen-feeding moths are present in two of the traits groupings; in host type and in diet breadth within host type) showed positive biomass trends, while the four other groups remained stable (Table S4, Figure 2b). Biomass of all functional groups based on the overwintering developmental stage remained stable too (Table S4, Figure 2c). The biomass of multivoltine species increased, while univoltine and semivoltine species were stable over time (Table S4, Figure 2d).
The models estimated moth biomass over the whole area of Finland biomass added by oligophagous tree-feeding species was 41.2 times larger than that of monophagous tree-feeders, 13.45 times higher than that of lichen-feeders, and 527 times higher than that of the mushroomfeeding moth. We highlight that these biomass predictions are countrywide and should not be misinterpreted as trap-level biomass predictions.

Spatial variation in temporal biomass trends of moth functional groups in Finland
There was spatial variation in the temporal biomass trends of all functional groups. The increase in conifer-feeders occurred country-wide, with the lowest rates in Central Finland, and the highest rates in the north and the south, while moths feeding on deciduous trees declined in much of the southern half of Finland (Figure 3a). The lichen-feeding species increased in biomass almost in all parts of the country, except for coastal areas in the south and the west (Figure 3a). F I G U R E 4 Effect of five environmental covariates on the density of individuals (solid line) and average body size (dashed line) of functional groups. (a) Effect of mean growing season length (i.e., the number of days between the dates when the daily mean temperature rises above 5 C for seven consecutive days for the first time in a year and falls below 5 C for seven consecutive days in the latter half of the year. (b) Effect of growing season temperature sum anomaly of previous year (i.e., the difference between sum of day degrees above a 5 C threshold accumulated during the growing season, and the mean growing season thermal sum over the period 1984-2020). Note that the effect of this variable on voltinism could not be assessed in our models due to convergence issues. (c) Effect of anomaly of snow duration (i.e., the number of days during the previous winter when snow depth exceeded 10 cm, subtracted by the location-specific mean snow duration for 1984-2020. (d) Effect of anomaly of cold days (i.e., number of days with mean daily temperature below À5 C during the previous winter, and the mean number of cold days over the period 1984-2020. (e) Effect of percentage of forested land. Points and squares stand for the estimated coefficient and whiskers the 95% confidence intervals. Red and blue colours indicate positive and negative effects, respectively. Grey colour stands for no effect.
functional groups in the diet breadth and host type grouping seem to follow the same spatial pattern but at different scales (Figure 3b), with the weakest biomass trends occurring along the coast, as well as in the northwest and southeast. The steepest biomass increases for lichen-eating moths in the diet breadth and host type grouping took place in the southern half of the country, except for coastal areas in the south and southwest; however, the increasing trend is countrywide ( Figure 3b). The increase in mushroom-eating, monophagous understory-feeding, and oligophagous understory-feeding moths seem to be aligned with the decreasing trend in the number of cold days and increased towards the east (compare Figure 3b with Figure 1d). In adult-overwinterers, the increasing temporal biomass trend was strongest in the north, and there was biomass loss in the central-western part of the country (Figure 3c). Egg-overwinterers increased steepest in the northern and north-eastern part of the country (Figure 3c). In multivoltine species, temporal biomass increase was lowest in the south-west, a pattern repeated slightly more strongly for univoltine moths (Figure 3d). Semivoltine species had more dramatic declines both in the southern and northern parts of the country.

Effect of covariates on biomass of functional groups
The effects of covariates on biomass of functional groups were separately estimated for the two linear predictors of the model (Table S5): density of individuals (i.e., density of individuals per unit area; hereafter density) and average body mass (i.e., average body mass of an individual belonging to the group; hereafter body size). In the functional group context, covariate effects on the body size predictor (second linear predictor) can be interpreted as environmental effects on species turnover: for example, longer growing season might favour increasing abundance of larger-bodied or smaller-bodied species in each functional group. Here, we explain the results for those functional groups that were associated significantly with the environmental variables considered. However, Figure 4a-e shows the results for all functional groups, and we ask readers to check the non-significant results from the figures (see also Table S5).
Mean growing season length was negatively associated with the density of semivoltine, mushroom-feeding, oligophagous-understoryfeeding and polyphagous-understory-feeding species, and positively with adult-overwinterers (Figure 4a). It was negatively associated with the body size of adult-and egg-overwinterers and positively with the body size of conifer-feeders, species feeding on other-hosts, as well as monophagous-understory-feeding and oligophagous-understoryfeeding species (Figure 4a).
Anomaly of growing season temperature sum in the previous year was negatively associated with the body size of monophagous-treefeeders and positively associated with the body size of the mushroom-feeder (Figure 4b). This suggests that warmer growing seasons increase the abundance of smaller-bodied monophagous-treefeeders ( Figure 4b). The results on the body size of the mushroomeating moth should be interpreted with caution, as our data did not include body size variation for this single species (Parascotia fuliginaria). For egg-overwinterers, the anomaly of growing season temperature sum in the previous year was negatively associated with density ( Figure 4b) and positively with body size (Figure 4b). Hence, warmer growing seasons favour lower densities and large-bodied species of egg-overwinterers (Figure 4b).
Anomaly of snow duration in the preceding winter was positively associated with the density of species feeding on coniferous trees (Figure 4c), suggesting that a long-lasting snow cover benefits these species. Conversely, anomaly of snow cover was negatively associated with the densities of species feeding on lichens and other hosts (from the host type grouping), lichen-feeding species (from the diet breadth within host type grouping), and multivoltine species (Figure 4c), suggesting that these groups of species benefit from shorter-lasting snow cover. Anomaly of snow duration was negatively associated with the body size of species feeding on coniferous trees, while its association was positive with the body size of multivoltine species (Figure 4c), indicating that a long-lasting snow cover helps small conifer-feeders and large multivoltine species. Anomaly of cold days in the preceding winter was negatively associated with the body size of species using lichens as hosts, lichen-feeders and monophagous tree-feeders

DISCUSSION
We did not detect any country-wide declining trends in biomass of macro-moth functional groups when analysing the long-term systematic monitoring data in Finland from the past three decades. With our functional-group-and biomass-based approach, we found a countrywide increase in biomass of species feeding on conifers or lichens, multivoltine species, monophagous and oligophagous tree-feeders and the single mushroom-eating moth. Country-wide biomass remained stable in all the other functional groups considered in this study. These results are consistent with Hunter et al. (2014) who showed that abundances of 90% of 80 forest moths were either stable or increasing over 32 years in a subarctic location, and with Kozlov et al. (2010) who reported that 74% out of 42 abundant species were stable, 14% increased, and 14% decreased in another subarctic location over 26 years. However, a species-level analysis using the same monitoring data as the present study, but over a shorter period (1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012) found an abundance decline ; but see Kohonen, 2020).
However, Kohonen (2020) investigated four more years  and did not detect any long-term trends in either abundance or bio- The trait-based approach is informative of the ecological correlates of abundance changes, and the biomass approach measures abundance changes in units relevant to ecosystem functioning. Biomass is a key variable in terms of energy flow, productivity and foodweb dynamics (Brown et al., 2004). Hence, biomass better represents the overall functionality of a species or a group of species within the community, and biomass of a specific trophic level better represents resource flows within an ecosystem than abundance (Saint-Germain et al., 2007). Using moth biomass allows us to draw inferences about the consequences of moth abundance changes, or lack of changes, on ecosystem functioning. As moth biomass seems either stable or increasing across some functional groups, food availability for insectivores should be stable or increasing, thus supporting ecosystem functionality. However, despite the benefits of the trait-and biomassbased approach, we cannot exclude the possibility that some species are declining. In fact, we know that there are more declining than increasing moth species in the dataset in 1993-2012 , but some species have increased a lot, balancing the overall change in total abundance and biomass (Kohonen, 2020).  (Roy et al., 2001). The climatic gradient explained density and body size variation in some groups, but not for multivoltine species. The unexpected lack of effect for multivoltine species can be explained if summer warming occurs mainly in late summer when it is unlikely to affect the induction of diapause versus direct development. Note also that we could not include growing season temperature sum in the previous year as a predictor when modelling voltinism-based groups due to model convergence issues. The overall increasing trend in biomass of multivoltine species shown here was expected because increasing voltinism results in an increasing number of individuals flying during the growing season, and climate warming expands the area where multivoltinism is possible (Altermatt, 2010;Pöyry et al., 2011;Virtanen & Neuvonen, 1999). With the northward expansion of many lepidopteran species, it is highly likely that new multivoltine species are also arriving in Finland (Pöyry et al., 2009(Pöyry et al., , 2011.
We used anomalies of snow cover duration and number of cold days to describe overwintering conditions. The duration of snow cover is shortening, especially in southern Finland, and the number of cold days is decreasing, especially in eastern and northern Finland, which results from nitrogen deposition. Habitat structure effects were captured by the annual percentages of forest coverage around the sampling locations. As expected, forest-dwelling species benefited from increasing forest coverage. Moreover, density of semivoltine species also increased with increasing forest coverage, which is expected as most of the semivoltine species are forest dwellers.
We found no associations between overwintering developmental stage and biomass trends. Earlier studies hypothesized that the eggand adult-overwinterers benefit from warmer summer conditions, and species that overwinter as a pupa are more likely to produce more generations per year (Teder, 2020;Virtanen & Neuvonen, 1999 (Blackshaw & Esbjerg, 2018;Pöyry et al., 2011;Van Dyck et al., 2014). Species overwintering as a pupa are vulnerable to extension of the warm period in the end of summer and autumn (Nielsen et al., 2022), and this may hold for egg-overwinterers too. Our results suggest that egg-and adult-overwinterers are smaller in southern than northern Finland. Moreover, increasing biomass of adultoverwinterers in northern Finland suggests that they benefit from longer-lasting snow cover. Biomass of egg-overwinterers increased in northern and north-eastern Finland, which coincides with the pattern of a decrease in the number of cold days. The growing season is, on average, getting warmer all over the country, which favours higher relative abundance of larger egg-overwintering species, but in lower densities. Therefore, these two opposing factors-density versus body sizemay have levelled out biomass trends of the egg-and adultoverwintering groups. Furthermore, increase in energy expenditure during diapause may be another mechanism that prevents milder winters to be beneficial for adult-, egg-and pupal-overwinterers. Unlike species which overwinter as larvae, they are unable to replenish energy reserves during winter. Another issue many adult-, egg-and springflying pupal-overwinterers might face is matching larval hatching with tree budburst (Both et al., 2009;Visser & Holleman, 2001). Such phenological mismatching is suggested by the spatial pattern seen in adult overwintering and deciduous tree-feeding species-many of which are shared between the groups-where the southern half of the country displays negative trends.
A possible caveat in the Nocturnal moth monitoring dataset is that the sampling sites do not have either a random or systematic dis- In our study, we showed that boreal moth functional groups perform better than expected on the grounds of recent results from many other parts of the world. Some functional groups even increased in biomass. These results suggest that moth-mediated ecosystem functions and services are not under immediate threat in Finland. Yet, we need to be cautious because, with our approach, we cannot exclude the possibility that some species are declining, or that there are negative and time-lagged community-level effects of climate and habitat changes (see e.g., Hanski et al., 1996;Tilman et al., 1994). The increasing total moth biomass trends in functional groups reported in this study should not be misinterpreted as increase in abundance at the species level, because species-level patterns cannot be inferred from our community-level approach. Ecological and life-history traits of moths explain some temporal biomass trends, and thus call for similar approaches in other studies on insect abundance change. Traitbased approaches will help in understanding the causes and consequences of insect abundance change and insect response to (a)biotic environmental change (but see Tordoff et al., 2022). As such, we strongly encourage such approaches to be used in future studies.

SUPPORTING INFORMATION
Additional supporting information can be found online in the Supporting Information section at the end of this article.