Drought-induced Forest Dieback Increases Taxonomic and Functional Diversity But Not Phylogenetic Diversity of Saproxylic Beetles at Both Local and Landscape Scales

Context: Forest ecosystems worldwide are facing increasing drought-induced dieback, causing mortality patches across the landscape at multiple scales. This increases the supply of biological legacies and differentially affects forest insect communities. Objectives: We analysed the relative effects of local- and landscape-level dieback on local saproxylic beetle assemblages. We assessed how classic concepts in spatial ecology (e.g. habitat-amount and habitat-patch hypotheses) are involved in relationships between multi-scale spatial patterns of available resources and local communities. Methods: We sampled saproxylic beetle assemblages in commercial r forests in the French highlands. Through automatic aerial mapping, we used dead tree crowns to assess dieback levels at several nested spatial scales. We analysed beetle taxonomic, phylogenetic and functional diversity related to differing levels of multi-scale dieback. Results: In line with the habitat-amount hypothesis, taxonomic and functional diversity, but not phylogenetic diversity, of beetle assemblages signicantly benetted from forest dieback, at both local and landscape scales. Very few single or interaction effects were detected in the multiplicative models combining local and landscape variables, though a signicant positive effect of landscape-scale dieback on the abundance of cavity- and fungus-dwelling species was consistent with a spill-over effect. Increased landscape-scale dieback also caused a functional specialisation of beetle assemblages, favouring those related to large-diameter, well-decayed deadwood. Conclusions: Increasing tree mortality under benign neglect provides conservation benets by heterogenising the forest landscape and enhancing deadwood habitats. Legacy retention practices could take advantage of unharvested, declining forest stands to promote species richness and functional diversity within conventionally managed forest landscapes.

Kuuluvainen 2016). Among them, saproxylic beetles are a highly diverse group of insects that depend on dead or decaying wood for at least part of their life cycle and play important ecological roles by participating in carbon and nutrient cycles or by complexifying trophic chains (Stokland et al. 2012). However, current silvicultural practices tend to reduce deadwood-related resources and microhabitats (Siitonen 2001;Grove 2002). As a consequence, saproxylic beetles are at considerable risk in intensively managed forests (Grove 2002), and 17.9% of saproxylic beetle species are now considered threatened in Europe (Calix et al. 2018).
Climate change alters natural disturbance regimes: warmer and drier conditions facilitate drought, wild re and insect outbreaks (Seidl et al. 2017). Lately, several large drought-related dieback events have been reported (Sangüesa- Barreda et al. 2015), and a further increase in drought-induced dieback and decline in terms of frequency, intensity and spatial extent is expected to arise (Allen et al. 2010; Samaniego et al. 2018). As a consequence, deadwood supply and the number of tree-related microhabitats are likely to increase and this may favour saproxylic communities ; Thorn et al. 2017;Sallé et al. , 2021Cours et al. 2021). Several local-scale (i.e. less than 0.5ha) studies have highlighted different positive relationships between saproxylic beetle populations and forest dieback: (i) pest-related dieback from large outbreaks of spruce bark beetles (Ips typographus, Linnaeus, 1758) resulted in an increase in saproxylic beetle species richness, including many red-listed species (Beudert et al. 2015;Cours et al. 2021) and (ii) drought-related dieback of Quercus spp. also increased saproxylic beetle species richness ). It has therefore been hypothesized that declining stands may improve habitat conditions for threatened forest communities such as saproxylic beetles (e.g. Müller  Phylogenetic diversity re ects the evolutionary history of a community through lineage relatedness while functional diversity re ects the diversity of the phenotypical traits selected by a particular environment, i.e. biotic and abiotic lters (Devictor et al. 2010; Kozák et al. 2020).
Furthermore, many previous studies have highlighted the fact that the diversity and structure of local saproxylic beetle communities rely on local habitat conditions, though they may also depend on large scale, i.e. landscape, conditions (e.g. Økland et al. 1996; Gibb et al. 2006;Franc et al. 2007;Haeler et al. 2021). Sampling area size in uences the detection of biodiversity responses to environmental conditions, a phenomenon known as "the scale of effect" ( relationship between spatial patterns of available resources and the number of associated species has been explored through several concepts based on ecological mechanisms. Within a given sampling area, the "habitat-amount hypothesis" predicts that the cumulative amount of habitat patches at the landscape scale better explains species richness than does local patch size (Fahrig 2013;Seibold et al. 2017). In contrast, the "habitat-patch hypothesis", based on island-biogeography theory, assumes that local species richness is mainly restricted by local patch size and isolation (MacArthur and Wilson 1967; Fahrig 2013; Seibold et al. 2017). The "resource concentration hypothesis" predicts that the occurrence of a particular resource patch in the landscape induces a concentration of the species specialising on that resource, while at the same time, over-availability of that particular resource, exceeding the reproductive and colonizing capacity of the associated species, could lead to a "dilution effect" (i.e. a large amount of substrate could lead to a reduction in the species load colonising the substrate) (Otway et al. 2005).
Hence, assessing the relative contribution of both local and landscape conditions on local biodiversity may well be critical; unfortunately, it is often challenging (Ammer et al. 2018). In recent decades, remote sensing and aerial photography have been widely used to monitor forest conditions and forest disturbances such as res, defoliation or deforestation at large and nested spatial scales. However, few studies have monitored insect responses to forest disturbances at these various scales, and even fewer combine taxonomic, functional and phylogenetic responses (Kozák et al. 2020).
In our study, we assessed how multi-scale forest dieback shaped local saproxylic beetle assemblages. We analysed aerial photographs with machine-learning algorithms to map dead tree crowns and monitor dieback in silver r forests in the French Pyrenees. After assessing dieback level at several nested spatial scales, we focused on the taxonomic (α-diversity), phylogenetic and functional responses of local saproxylic beetle assemblages to the multi-scale spatial structure of the forest dieback. As a consequence of the increase in deadwood amount and light availability, we expected positive responses from the saproxylic beetle assemblages along the gradient of forest dieback at several spatial scales Bouget et al. 2014;Seibold et al. 2016). We expected an increase in species richness resulting from the "Species-Area Relationship" (MacArthur and Wilson 1967) and an increase in abundance from the "More-Individuals Hypothesis" (Srivastava and Lawton 1998;Müller et al. 2018). We also expected contrasted responses according to functional guilds (e.g. trophic; Percel et al. 2019). Therefore, in this study, we addressed three major questions: i. Are the effects of forest dieback globally positive on the community metrics for saproxylic beetle assemblages? ii

Machine learning process
We manually constructed vector training data through on-screen interpretation (Fig. 2c), resulting in 4,256 polygons for four land cover classes: 1,743 polygons of dead crowns (41%), 1,606 of living trees (37.7%), 212 of shady areas (5%) and 695 polygons of bare soil (e.g. meadows; 16.3%). We implemented a machine learning algorithm with the Orfeo Toolbox (OTB) software and we applied both a Pixel-based image analysis (PBIA) and an Object-based image analysis (OBIA) (Grizonnet et al. 2017).
In the PBIA, we ran a Random Forest (RF) classi cation model (Breiman 2001

Measuring forest dieback at the landscape scale
We used our RF classi cation model with the PBIA approach to identify the dead crown pixels over large areas around our study plots. Our approach did not allow us to assess dead-tree density so we estimated the cumulative surface area of the dead and dying tree parts, i.e. the dead crowns (Larrieu et al. 2018). We then mapped and summed the dead crown pixels to assess a level of forest dieback over several spatial scales. We designated several nested buffer zones around our study plots in order to describe forest dieback from the local to the landscape scale. The zones had radii of 25, 200, 500, 800, 1100 and 1500 m; we added the dead crown pixels in each of these buffer zones (Tab. 1).

Statistical analyses
Data analysis was conducted with R software 4.0.0 (R Core Team 2021). Firstly, we calculated abundance and species richness for the substrate and trophic beetle guilds. We also calculated abundance and species richness for both common and rare saproxylic beetle species as well as the total species richness.
Secondly, to assess functional diversity indices for the community, we extracted quantitative values for preferred deadwood diameter and decay level for larval development, canopy openness preference and mean body size for each of the captured saproxylic beetle species, as in Gossner et al. functional traits (i.e. preferences in deadwood diameter and decay and canopy openness, and mean body size) (Tab. 1): i) functional richness (FRic), i.e. "the range of functional space lled by the community"; ii) functional divergence (FDiv), which "relates to how abundance is distributed within the volume of functional trait space occupied by the community"; and iii) functional evenness (FEve) or "the evenness of abundance distribution in a functional trait space". These three indices should be able to quantify the functional changes occurring in a community after a disturbance (Mouillot et al. 2013).
Thirdly, we calculated Faith's standardized phylogenetic diversity index (SES Faith's PD) to obviate the relationship between Faith's PD and species richness (Pearson's r = 0.98, P < 0.001). We also calculated two phylogenetic species-diversity metrics: phylogenetic species variability (PSV) and evenness (PSE). PSV "is one when all species are unrelated and approaches zero as species become more related"; PSE "is one when species abundances are equal and species phylogeny is a star" (Tab. 1; Kembel et al. 2020). DNA barcode consensuses were mined from the BOLD data system (Ratnasingham and Hebert 2007) whenever accessible for each saproxylic beetle species morphologically identi ed. When multiple records and BINs (Ratnasingham and Hebert 2013) were available for a given species, a choice was made according rst to geographic area of sampling, and second to sequence length and quality. Close geographical areas were favoured as were high-quality 658bp-long sequences (N < 1%) whenever possible. The dataset of the records we used for phylogenetic diversity is publicly available at the between linear and logarithmic regressions). Since six different spatial scales were compared (R = 25 m, glmmTMB model and performed AICc comparisons between the models to select the best landscape scale for each tested response variable. We checked for non-collinearity between local and landscape scales with the "check_collinearity" function from the performance R-package (Lüdecke et al. 2020) and always observed a variance in ation factor (VIF) below three. We then performed multiple regressions with generalized linear mixed models with "site" (Aure Valley or Sault Plateau) as a random variable and including as xed variables, the measure of local dieback and the best landscape scale. As three different terms in the multiplicative models were involved, we performed Post-Hoc Holm adjustments on each of their p.values. Our purpose was to evaluate the potential interactive effect between the local (25 m) and the most appropriate landscape scale (i.e. the landscape scale with the lowest AICc value) with a multiplicative interaction model (Equation 1). Since the 200-m scale was highly correlated with the 25 mscale (Pearson's r = 0.82, P < 0.001), we excluded this metric from our analysis (Fig. S8).
which is the multiplicative equation assessing whether there is an interactive relationship between X 1 and X 2 .
In multiplicative interaction models, hereafter referred to as "multiplicative models", β 1 and β 2 are signi cantly different from zero when X 1 and X 2 are respectively equal to zero (H 1 : β 1 ≠ 0 when X 2 = 0 and vice-versa) (Braumoeller 2004). We associated the results of the multiplicative models with the ecological mechanisms affecting the spatial pattern of the saproxylic beetles. When H 1 : β 1 ≠ 0 could not be rejected, we hypothesised an effect of local resource concentration since we were assessing the effect of local forest dieback on saproxylic beetles in the case of limited dieback at the landscape scale (Fig.  3a). When H 2 : β 2 ≠ 0 could not be rejected, we hypothesised a spill-over effect since we were assessing the effect of forest dieback at the landscape scale in the case of limited local dieback (Fig. 3b). We hypothesised that a signi cantly positive interaction term (β 12 ) re ected a synergistic/amplifying effect ( Fig. 3c & 4) and that a signi cantly negative interaction term would support a dilutive/saturated effect and the habitat-patch hypothesis (

Results
Our nal dataset comprised 50,067 specimens of 393 saproxylic beetle species belonging to 50 families. To assess the relevance of the sum of dead crown pixels as an indicator of forest dieback, we evaluated the accuracy of the relationship between the local sum of dead crown pixels (R=25 m) and eld measurements of forest dieback carried out according to the ARCHI protocol (Drénou et al. 2013). The ARCHI protocol classi es trees in a gradient from healthy to dead, based on their architecture (Drénou et al. 2013). In our study, we applied the protocol to the 20 r trees closest to each plot centre. We observed a signi cant relationship between the two variables (β-Estimate = 0.7; P < 0.01; Fig. S6), which indicates that our classi cation model of dead crown pixels and the sum of these pixels provided a consistently accurate description of the local forest dieback assessed on site. In addition, we validated our landscapescale estimates of forest dieback by cross-checking the relationship between the sum of dead crown pixels at the landscape scales (i.e. R=200 to 1500 m) with the European disturbance map edited by Senf and Seidl (2021), which is based on a time series analysis of the spectral band values of Landsat satellite photographs (P < 0.001; Fig. S7).

Relationship among taxonomic, phylogenetic and functional diversity
We did not observe a relationship between total saproxylic-beetle species richness and the phylogenetic diversity metrics (Fig. S1). Nevertheless, we found a positive relationship between total species richness and FRic (Fig. S1), but not with FDiv (Fig. S1) and FEve (Fig. S1). We did not observe any relationship between the phylogenetic diversity metrics and FRic, though there was a signi cantly negative relationship between PSE and FDiv (Fig. S1) and a positive relationship between PSE and FEve (Fig. S1).

General metrics
In the univariate models, both local and landscape dieback metrics had positive effects on total species richness, abundance and richness of common species, and on abundance of rare species (Tab. 2; Fig.  S2). In contrast, rare-species richness did not respond to forest dieback at any scale (Tab. 2; Fig. S2). Almost no effect was detected in the multiplicative models. Nonetheless, the abundance of common species still responded positively to forest dieback at the local spatial scale, thus supporting a concentration effect. (Tab. 2; Fig. 3).

Feeding and substrate guilds
In the univariate models, the abundance of wood-eating species was positively affected by local forest dieback only (Tab. 2; Fig. S2). In contrast, the richness of wood-eating species positively responded to both local-and landscape-scale dieback (Tab. 2; Fig. S2). In the multiplicative models, neither the abundance nor the species richness of wood-eating beetles signi cantly responded to forest dieback; this supports the habitat-amount hypothesis (Tab. 2; Fig. 3). Cavicolous and fungicolous species responded positively to forest dieback at the local and landscape scales, both in terms of abundance and richness, in the univariate models (Tab. 2; Fig. S2). In the multiplicative models, we detected a signi cant positive effect of dieback level at the landscape scale on fungicolous and cavicolous abundance in agreement with the spill-over effect (Tab. 2; Fig. 3).

Response of phylogenetic diversity to dieback
In both the univariate and multiplicative models, none of the phylogenetic diversity metrics responded to forest dieback at either the local or landscape scale. (Tab. 3; Fig. S2).

Response of functional diversity to dieback
In the univariate models, we observed positive effects for both local and landscape forest dieback on FRic  We did not observe any signi cant interaction between local and landscape forest dieback (Tab. 2). According to Seibold et al. (2017), the lack of interaction between these two spatial scales should support the habitat-amount hypothesis since the amount of habitat at both scales is merely additive (Fig. 4). This is further supported by the fact that local and landscape effects alone cancelled each other out in our multiplicative models, while most of the univariate-model effects for taxonomic diversity were signi cant (Tab. 2; Fig. S2). The habitat-amount hypothesis predicts that "species richness in a sample site is independent of the area of the particular patch in which the sample site is located (its local patch)" (Fahrig, 2013). Therefore, in our study, local scale alone (i.e. without any dieback areas in landscape) should not have been su cient to detect dieback effects on saproxylic beetle biodiversity, even if it appears as the potentially scale of effect. Nevertheless, the opposite is also true:

Contrasting responses of different biodiversity dimensions to forest dieback
We did not detect any response to forest dieback for phylogenetic diversity. However, PSE was negatively correlated with FDiv and positively correlated with FEve, both of which were in uenced by landscape forest dieback (Tab. 3; Fig. S1-2). Therefore, forest dieback did not seem to induce any loss or gain in the range of evolutionary history occupied by saproxylic beetle assemblages, or if so, only indirectly by in uencing functional diversity. Nonetheless, the use of DNA barcodes alone may be insu cient to estimate real phylogenetic diversity and the inclusion of multigene phylogenies may better estimate phylogenetic diversity and its response to ecological processes (Liu et al. 2019). In contrast, taxonomic and functional diversities were in uenced by forest dieback (Tab. 2 & 3). Consequently, the diversity and quantity of habitats and resources released by forest dieback increased species richness and more heterogeneous functional assemblages, as suggested by the more-individuals and the habitatheterogeneity hypotheses (Seibold et al. 2016), without signi cantly increasing phylogenetic diversity.
Furthermore, phylogenetic response to disturbance may be such a long-term process that the effects of forest dieback on this component could not be observed in our study (Purschke et al. 2013).
However, Kozák et al. (2020) showed that phylogenetic diversity of saproxylic beetles was positively affected by canopy openness, which in turn was positively in uenced by recent disturbances. Our results also suggest that forest dieback clustered the assemblages even more when it was severe at both the local and landscape scales, even if local dieback had no effect in the univariate and multiplicative models (positive synergistic effect in the multiplicative model; Fig. 4, Tab. 3). Therefore, forest dieback, especially at the landscape scale, seemed to promote and enhance local functional heterogeneity and thus diversify the functional niches of saproxylic beetles, at our study sites in the Pyrenean mountains. Intensive management generally leads to functional homogenisation, which is often driven by the decline of specialist species in favour of generalists (Clavel et al. 2011). Our plots were managed, and we found that forest dieback induced a functional heterogenisation accompanied by a specialisation of the studied assemblages at the boundaries of the functional space. We hypothesize that the functional heterogenisation was driven by the high resource availability and habitat diversi cation subsequent to the forest dieback. At the landscape scale, the dieback logically resulted in a matrix of remaining live trees, acting as disturbance refugia (Krawchuk et al. 2020), and discrete patches of open woodlands with standing dead trees and snags, logs, large deadwood, tree-related microhabitats, etc. Ultimately, this promoted the coexistence of a wide variety of ecological niches (Swanson et al. 2011), allowing the cooccurrence of functionally diverse saproxylic beetle assemblages (Thorn et al. 2018;Kozák et al. 2020).

Functional responses of assemblages to forest dieback: heterogenisation and specialisation
In addition, we observed functional specialisation in species preference for deadwood diameter: when forest dieback increased at the landscape scale, local assemblages preferred larger deadwood and functional dispersion was lower (CWM and FDis Diameter; Tab. 3). A previous study showed that the functional specialisation of saproxylic beetles towards large-diameter and well-decayed deadwood occurs when the overall amount of deadwood increases (Gossner et al. 2013). This functional specialisation might not account for the needs of species that prefer small-diameter deadwood. Nevertheless, these species still bene t from a relatively high amount of deadwood and are also less sensitive to intense forest management in the surrounding area (Gossner et al. 2013

Application and conclusions
Our study revealed that the taxonomic and functional diversity of saproxylic beetle assemblages in Pyrenean mountain r forests signi cantly bene tted from forest dieback, at both local and landscape scales, mainly thanks to landscape heterogenisation, to a large build-up of deadwood and to more canopy openings (Bouget et   "DCP" = Dead crown pixels; "CWM" = Community-Weighted Means; "FDis" = Functional Dispersion; "DW" = eadwood; rare species = patrimonial value ≥ 3 in France. Table 2. Results from the univariate (on left, column 3) and multiplicative interaction models (right-hand column) of the effects of local (R=25 m) and surrounding-landscape (highest scale of effect) dieback on the taxonomic diversity of saproxylic beetles.* Table 3. Results from the univariate (on left, column 3) and multiplicative interaction models (right-hand column) of the effects of local (R=25 m) and surrounding-landscape (highest scale of effect) dieback on saproxylic beetles (phylogenetic and functional diversity * Effects were tested with generalized linear mixed models (with "site" as a random variable). Values shown are z.values from glmmTMB models. Colours represent the direction of the effect: green for a positive effect of dieback on the considered variable, red for a negative effect and grey when no significant effects were detected. DW = "dead wood"; ns = P > 0.05; * = P < 0.05; ** = P < 0.01; *** = P < 0.001    Hypothesised saproxylic beetle responses to the terms of the multiplicative interaction models (inner circle = local conditions (associated estimates = β1); outer circle = landscape conditions (associated estimates = β2); light grey = healthy forest area; dark grey = disturbed forests; arrows represent species uxes). a) β1 > 0 = concentration effect; b) β2 > 0 = spill-over effect; c) β12 > 0 = synergistic effect; d) β12 < 0 = dilution or habitat-patch effect; e) β12 = 0 = habitat-amount hypothesis. Moreover, if there is a signi cant local effect in the univariate model but not in the multiplicative model, or a signi cant landscape effect in the univariate model but not in the multiplicative model, it supports the habitatamount hypothesis.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. 4Supplementarymaterials.docx