Global variation in unique and redundant mammal functional diversity across the daily cycle

Organisms primarily influence ecosystems through their functional traits when they are physically active. Following the nocturnal bottleneck, the expansion of mammals into the daytime expanded mammalian functional diversity (FD), however there is also notable overlap in trait space across diel niches leading to redundant FD. We explore how the unique and redundant contribution of each diel niche varies spatially and in relation to natural variation in light and temperature.


| INTRODUC TI ON
Following the K-Pg event (66 Ma) that saw the extinction of all nonavian dinosaurs, most surviving mammals remained nocturnal, like their ancestors, though many others evolved to be active during the daytime (Gerkema et al., 2013;Grossnickle et al., 2019;Maor et al., 2017). Mammals active during the day faced different abiotic (e.g. light levels, temperature) and biotic (e.g. competition, predation) regimes to those solely active at night (Hut et al., 2012), and during this geological period they also experienced rapid diversification in functional traits such as body mass, modes of locomotion and diet (Grossnickle et al., 2019). Indeed, to maximise their fitness, mammals active at different times of day evolved somewhat different combinations of functional traits to those that remained active in the night, although there is also notable overlap in global trait space across diel niches (Cox et al., 2021).
A species' functional traits determine its ecological role and contribution to ecosystem function (Violle et al., 2017); mammals support a variety of ecosystem processes such as nutrient cycling, predation, herbivory, seed dispersal and pollination (Luck et al., 2012;Lundberg & Moberg, 2003). The time of day during which a species is physically active determines when it primarily exerts its influence on ecosystem functions and processes [e.g. predator-prey relationships  herbivory (McMunn, 2017), pollination (Tremlett et al., 2020)]. What is presently unknown is the extent to which diversification into different diel niches may have served to increase the overall functional diversity (FD) of local or regional mammal assemblages (increasing unique components of FD at different times of day), potentially increasing their overall impact on ecosystems.
Also unknown is the extent to which this diversification increased levels of redundancy in the FD of mammal assemblages, with species active at different times of day making similar contributions, and thereby potentially increasing ecosystem stability (Houadria et al., 2016;Mougi, 2021). The possibility of either outcome is likely to be shaped spatially by the opportunities and constraints on being active at different times of day posed by climatic conditions (particularly cold nights and hot daytimes; Bennie et al., 2014), and thus there could be biogeographic structure to the contributions of different diel niches to unique and redundant components of FD.
Here we address two primary research questions: (1) What are the biogeographic contributions of diel niches towards diel unique and diel redundant components of mammalian FD? (2) How do these components covary with prevailing light and temperature regimes?
We show that although the diversification of mammals into the daytime resulted in the creation of new FD, within assemblages much of this expansion is restricted to higher latitudes (cathemeral) and where there are more hours of biologically useful daylight (diurnal). Instead, a greater proportion of non-nocturnal FD replicated nocturnal FD (i.e. is diel redundant), which suggests that these species have not in this sense diversified far from their nocturnal ancestors. Biogeographic patterns of unique and redundant FD of each diel niche covary with natural cycles of light and temperature.
Almost half of mammalian FD is uniquely nocturnal associated with more hours of biologically useful moonlight, whereas unique diurnal FD is positively correlated with biologically useful daylight. At higher latitudes, regions that receive fewer hours of biologically useful moonlight and daylight, cathemeral FD dominates.

| Summary
For 5033 extant terrestrial mammal species we focused on five major functional traits (body mass, litter size, habitat breadth, foraging strata, diet breadth) to calculate FD as the total dendrogram branch lengths (Petchey & Gaston, 2002, 2006 for each diel niche (nocturnal, crepuscular, cathemeral and diurnal). Together, these five traits reflect the spatiotemporal distribution of resource capture and release and dictate a species' influence on ecological and biogeochemical processes (Cooke et al., 2019;Cox et al., 2021). We calculated: (i) the biogeographical distribution of the proportion of total mammalian FD that was unique to that time of day (i.e. thereby reflecting emphasis on different ecological strategies across the daily cycle and potentially increasing the overall impact of mammals on ecosystems), and the proportion that was redundant across other diel niches (i.e. similar functional roles are carried out by species at different times of day, thereby potentially increasing ecosystem resilience). (ii) We then modelled associations between spatial variation in each dimension of FD against five potential predictors, variation in "biologically useful" natural light (daylight, twilight, moonlight) constrained by temperature (and, for moonlight, cloud cover), elevation above sea level, and the presence of habitat with closed (with a tree canopy e.g. forest, woodland) or open (other landcover types without trees) cover due to its influence on natural light intensity (Endler, 1993). All data processing and analyses were performed in R software for statistical computing v4.0.5 (R Core Team, 2021).

| Functional traits
For 5033 extant terrestrial mammals we obtained data on diel niche from our recently compiled database for mammals (Cox et al., 2021; that in turn sourced data from Cooke et al.,  and diet breadth (a continuous synthetic trait generated from 10 diet categories; see Appendix S1; Data S1). For taxonomy, we followed PHYLACINE 1.2.1, a taxonomically integrated platform containing phylogenies, range maps, trait data, and threat status for all 5831 known mammal species that have lived since the last interglacial. We excluded extinct, fossorial and marine mammals, the latter two likely being reliant on different light cues than above-surface terrestrial species. More species have been 'discovered' since PHYLACINE 1.2.1 was released, however many are the result of taxonomic splitting of existing species, thus share similar functional traits and so the impact on the biogeography of mammalian FD will be low.
Data were not available for all species and traits but, overall, less than 5% of trait values were missing (diel niche, 3%; body mass, 1%; litter size, 24%; foraging strata, 1%; habitat breath, 0.1%). Excluding species with incomplete data reduces sample sizes (and consequently the statistical power of the analysis) and may introduce bias (Penone et al., 2014). To achieve complete species-trait coverage we imputed missing data using the Multivariate Imputation with Chained Equations (MICE) package (Table S1) based on the ecological (the transformed traits) and phylogenetic (the first 10 phylogenetic eigenvectors extracted from trees obtained from PHYLACINE 1.2.1 database; Faurby et al., 2018;Penone et al., 2014) relationships between species. The MICE procedure produced 25 imputed datasets (Cox et al., 2021). Owing to the high computational costs of running null models for all associated datasets, for all analyses we used a single randomly selected dataset (dataset 24; Data S1). The reliability of the imputed data compared to a data-deletion approach (exclusion of species with missing data) has previously been validated, and results were found to be qualitatively similar (Cox et al., 2021).

| FD
We calculated functional diversity as the total branch lengths in a functional dendrogram (termed FD; Petchey & Gaston, 2002). FD provides a single index that encompasses how dispersed a set of species are in trait space, describes the functional relationships shared by species (Petchey & Gaston, 2002), accounts for clustering of species (functional redundancy) and is widely used with presence/ absence data such as analysed here. Once a master dendrogram has been generated for an assemblage, species branches can be removed ('pruned') without changing the relative positions of remaining species. Thus, FD provides a highly flexible approach for calculating the contribution of subsets of species (i.e. diel niches) to the FD of the entire assemblage, while also allowing comparable measures of the unique and redundant FD in each subset to be calculated across diel niches (e.g. Figure S1).
We developed a master functional dendrogram for all 5033 species following well-established approaches (Barbet-Massin & Jetz, 2015;Jarzyna et al., 2021;Jarzyna & Jetz, 2017;Pavoine et al., 2009;Podani, 1999). We calculated multivariate trait dissimilarity using Gower's distance for each pairwise combination of all 5033 species in the dataset using the gowdis() function in the 'FD' package (Table S1). Equal weight was given to each of the traits.
The functional dendrogram was built using the hclust() function with UPGMA clustering, which ensures the most faithful preservation of the original distances in the dissimilarity matrix (Mérigot et al., 2010). Using a dendrogram approach might in some cases artificially increase the functional distances between species that have similar trait values (Magneville et al., 2022;Maire et al., 2015;Villéger et al., 2017). Compared to other clustering methods (Ward, Single, Complete, WPGMA, WPGMC and UPGMC), UPGMA had the highest cophenetic correlation coefficient (c = 0.79) and the lowest mean absolute deviation (0.047) and lowest root mean squared deviation (0.067; Table S2) indicating a minimal loss of information.

| Biogeography of diel FD
We used estimates of species ranges downloaded from IUCN (N = 5021; https://www.iucnr edlist.org/resou rces/spati al-datadownload), and where range maps were unavailable, PHYLACINE 1.2.1 (N = 12; Faurby et al., 2018). Range maps were rasterised to a resolution of 96.5 km by 96.5 km and projected to Behrmann cylindrical equal area (Appendix S1). Based on the range maps, we created a presence-absence matrix of all land pixels, where columns were species (N = 5033) and rows were pixels (N = 14,438).
For each pixel, we first calculated the FD of all species within the pixel (FD pixel ) by pruning from the master dendrogram branches of species that were not present in the pixel using the prune() function in the 'dendextend' package, before summing the lengths of remaining branches in the resulting dendrogram (Petchey & Gaston, 2002, 2006 using the treeheight() function in the 'vegan' package (Table S1). For each pixel and diel niche we pruned the functional dendrogram of branches of all species in the pixel except (1) those of the diel niche of interest, before summing the remaining branch lengths to give the FD of the diel niche (FD diel ), (2) those within the pixel that were not of the diel niche of interest, before summing the remaining branch lengths to give the FD of all species excluding the diel niche of interest (FD excl-diel ). In each pixel the unique contribution of the diel niche to mammalian FD (FD unique ) was then calculated as the total FD minus the FD of species in other diel niches (i.e. FD pixel − Fd excl-diel ). Finally, the FD that is redundant across diel niches (FD redund ) was calculated as the FD of the diel niche minus the unique FD of that diel niche (i.e. FD diel − FD unique ). To give the proportion of total mammal FD in each pixel that was unique or redundant to each diel niche, we then simply divided the resulting FD unique or FD redund by FD pixel .
For example, if 70% of mammalian FD in the pixel is nocturnal, with 40% of FD being unique to the nighttime, then 30% of nocturnal FD will be diel redundant (see Figure S1 for a further example). When species richness is low the proportional FD may be overly weighted by the contribution of single species. Therefore, to remove this potential source of bias, we excluded pixels that contained 10 species or fewer (N = 2908 pixels; 5.9% of all land pixels; primarily occurring at the upper latitudes, central Sahara and New Zealand). Inevitably there are pixels where mammalian species richness is >10 but there is only one species in the diel niche of interest. Because no multivariate dispersion can be calculated for one species, these pixels were given an FD diel of 0.

| Biologically useful light
We generated global maps representing the annual total of biologically useful hours of daylight, twilight and moonlight within a temperature range of 0 and 35°C. Following Bennie et al. (2014) it was assumed that outside of these limits activity is either greatly reduced or imposes high physiological or adaptive costs. We downloaded 4× daily (6-hourly) global mean air temperature and cloud cover data for 2017, also on a T62 Gaussian grid, from the NOAA website (Kanamitsu et al., 2002) and applied the native spline() function in base R to derive hourly values. For each pixel location on a global 94 (latitude) × 192 (longitude) T62 Gaussian grid we calculated the hourly sun position above the horizon (getSunlightPosition()), the hourly moon position above the horizon (getMoonPosition()) and the hourly illuminated fraction of the moon (getMoonIllumination()) using functions from the 'suncalc' package (Table S1). We then calculated the hours of daylight (solar altitude >0°), twilight (solar altitude ≥−12° and ≤0°, therefore incorporating both civil and nautical twilight) and full moonlight (solar altitude <−12°, moon altitude >0° and moon illumination >75%). Moonlight was adjusted for cloud cover obscuring the face of the moon, so that adjusted moonlight duration was calculated as hourly moonlight duration × 1-hourly cloud cover ( Figure S2). We excluded hours of daylight, twilight and moonlight where the corresponding hourly temperature value for a grid pixel was <0°C or >35°C. Finally, we summed the hours of daylight, twilight, and moonlight that remained following exclusions for temperature and cloud cover.
We created a binary layer of pixels where the habitat was domi-  (Table S3). A global Digital Elevation Model (DEM) layer was downloaded at a spatial resolution of 0.50° from the Registry of Open Data on AWS using the get_dem() function in the 'microclima' package (Table S1).
Ocean pixels and small islands were masked on all layers. All layers were then projected to Behrmann cylindrical equal area, resampled to 96.5 × 96.5 km resolution (see Appendix S1) and converted to a data frame for spatial analysis.

| Statistical analysis
For each diel niche, we modelled the proportion of FD that was unique (FD unique ) or redundant (FD redund ), termed FD_prop (niche) in each pixel i as a function of mammalian FD (FD pixel ), and five measures of natural variation in light and temperature [hours of biologically useful daylight, twilight and moonlight, habitat cover (open, closed), elevation] using a beta regression approach, that is, assuming that the response variable could be modelled using a beta probability distribution (Douma & Weedon, 2019). Beta regression requires that the response variable is on the open interval (0,1) and small violations of this assumption can be dealt with through data transformations (Douma & Weedon, 2019). However, on inspection it was evident that these data were consistently zero-inflated due to grid pixels with fewer than 11 species, and those with only one species in the diel niche, being assigned a value of zero FD (see above).
Thus, to accommodate this feature of the data, we modelled the relationship between FD_prop(niche) and the covariates using zeroinflated beta regression models. The mean-variance relationship varied with FD pixel and, therefore, it was necessary to model the phi precision parameter as a function of this variable. The zero-inflation parameter zi was also modelled as a function of FD pixel because the probability of FD_prop(niche) being equal to zero in a grid pixel was correlated with FD pixel in that pixel.
Residual spatial autocorrelation was high in models when there was no attempt to correct for spatial autocorrelation. To reduce its effect on parameter estimates we chose to implement the residual autocovariate (RAC) approach, which is computationally tractable and has been shown to result in good inferential performance (Crase et al., 2012). The RAC approach is an extension of the commonly used autocovariate method, but aims to account for residual autocorrelation using a covariate based on the model residuals (rather than the values of the response variables) after the environmental covariates have had the opportunity to account for spatial autocorrelation in the response variable (Crase et al., 2012). This approach was implemented for each diel niche model by fitting a model including only the FD pixel and environmental predictors, and then extracting and spatially smoothing (mean across a 9 × 9 pixel moving focal window) these residuals to create a RAC variable that was then used as a covariate in a second model. To reduce further the effects of spatial autocorrelation and to reduce the computational demands of model fitting, models were fit in turn to four spatially balanced and disaggregated blocks. Blocks were spatially balanced by first dividing the model space into a checkerboard (2 × 2 pixels) and assigning one pixel from each checker at random to one of four blocks ( Figure S3). Iteratively, pixels with adjacent neighbours were reassigned at random to a new block until each block has approximately the same number of pixels (range 3440-3484) and a minimal number of adjacent pixels. The same blocks were used to fit all models. FD_prop(niche) i = a 0 + a 1 FD pixel i + a 2 landcover i + a 3 elevation i + a 4 moonlight i + a 5 twilight i + a 6 daylight i + a 7 RAC i + i , a categorical variable indicating whether the dominant vegetation in a pixel is either open or closed and elevation in metres, annual hours of biologically useful moonlight, twilight and daylight are continuous variables and i is the error. All predictors were centred and scaled prior to model fitting. Models were fit in a Bayesian framework using the R package 'brms' (Table S1) and for each model two chains were run for 8000 iterations per chain, with a burn in of 4000. Horseshoe priors were applied to all the main effects (except the intercept) to shrink unimportant parameter estimates towards zero (Carvalho et al., 2010).
For each diel niche, rather than reporting results for four separate models built on different data blocks, we report results after combining the posterior distributions of the four models using the com-bine_models() function from 'brms' (Table S1). We report the Bayesian R 2 (Gelman et al., 2019) using the bayes_R2() function as the mean across the four models. In all models, the Moran's I test statistic for spatial autocorrelation significantly deviated from expectation in the model without the RAC covariate, but this deviation was considerably reduced, although not eliminated, with the addition of the RAC term.
We present parameter estimates from both the models with and without the RAC term so that the effect of the RAC term on parameter estimates can be compared and we provide the full model results in the supplementary material (Table S4).
Conversely, across all diel niches 33.3% of FD is redundant, with only 2.5% being redundant across non-nocturnal species (Table S5).
Nocturnal FD is highest across the tropics, and lowest at higher latitudes on the Tibetan plateau and in hot dry regions (Figure 1a).
Although strictly crepuscular FD is relatively rare, aggregations occur in East Africa, and along the West Coast of North America ( Figure 2a). Cathemeral FD is greatest in Siberia, South-East Asia, and East Africa (Figure 3a), whereas although diurnal FD is highest in tropical forests it is also common in northern North America ( Figure 4a).

| Biogeography of diel unique and diel redundant FD
Proportionally, across much of the land surface approximately half of mammalian FD is unique to the nighttime (median across pixels (median pixel ) 45.9%, Lower Quartile (Q 1 ) 37.4%, Upper quartile (Q 3 ) 57.7%; Figure 1b). The proportion of unique nocturnal FD increases in hot dry regions but decreases at higher latitudes and on the Tibetan plateau (Figure 1b). Redundant nocturnal FD is lower than unique nocturnal FD (median pixel 23.5%, Q 1 20.2%, Q 3 28.2%).

| Covariation of FD with natural cycles of light and temperature
The contributions of diel unique (FD unique ) and diel redundant (FD redund ) FD to total mammalian FD (FD pixel ) strongly covary with natural cycles of light and temperature ( Figure S2a In regions where habitat is defined by closed cover (Figure S2d), the proportion of redundant nocturnal and redundant diurnal FD increases (Figures 1c and 4c). The proportional contribution of redundant nocturnal FD and unique and redundant cathemeral FD to total mammalian FD increases at higher elevations, while there is a decreased contribution of unique and redundant diurnal FD (Figures 1,   3 and 4).

| DISCUSS ION
Our findings demonstrate how FD in a major taxon is structured biogeographically across the diel cycle, and covaries with natural cycles of light and temperature. We show that, although globally the diversification of mammals into the day resulted in the creation of new FD, much of this expansion occurred at higher latitudes where uniquely cathemeral FD dominates, and where there are more hours of biologically useful daylight unique diurnal FD is more common.
Over much of the land surface redundant FD is more common than unique FD in non-nocturnal species, likely increasing ecosystem stability and making the system more resilient to species loss because similar functions and processes are carried out at different times of day (Mori et al., 2013). It also suggests that many functional trait combinations assessed here are equally successful regardless of when species are active.
During the Mesozoic, mammals were largely restricted to nighttime activity, which constrained taxonomic diversification (Wilson et al., 2012) but allowed time for the evolution of a broad diversity of ecological strategies (Cox et al., 2021). Consequently, over most of the land surface mammals exert their major influence on ecosystems at night, with approximately half of FD being uniquely nocturnal. Evidence suggests that mammals began to diversify first into cathemeral and then into diurnal niches during the decline and then extinction of the dinosaurs (Maor et al., 2017), F I G U R E 1 Biogeography mammalian nocturnal FD. (a) is nocturnal FD, (b) is the proportion of mammalian FD that is unique to the nighttime, and (c) is nocturnal FD that is redundant across the daily cycle presented as a proportion of mammalian FD. The histograms give the frequency distribution across pixels. Pixels with 10 or fewer species and those with only one nocturnal species have been removed from (b) and (c). The pie chart in the top right shows the non-spatial proportional contribution of nocturnal FD to total mammalian FD, where the darker colour is unique FD and the lighter colour is redundant FD. The grey demonstrates the proportion of mammalian FD that is not nocturnal. Model outputs show the standardised regression coefficients and 95% credible intervals of covariation between nocturnal FD with five predictors of natural variation in light and temperature for the spatial (black) and non-spatial model (grey/diamond). The map projections are Behrmann's equal area.
with the transition into the day starting in earnest following the Palaeocene-Ecocene Thermal Maximum event 55 million years ago (Bininda-Emonds et al., 2007;Maor et al., 2017). Competition with, and predation by, other taxa such as birds may have slowed the movement into the daytime and restricted the number of species able to do so (e.g. mammals with small body masses disproportionately remained nocturnal). Where hours of biologically useful daylight are reduced (in very hot or very cold places), the proportional contribution of uniquely nocturnal FD is governed by the availability of biologically useful moonlight. In hot dry regions high thermoregulatory and water budgets make activity during the day costly (Campbell & Norman, 1998) and restrict the diversity of possible ecological strategies, so that mammalian activity occurs almost exclusively in the cooler hours of darkness. In contrast, in the high northern latitudes, low nighttime temperatures result in prohibitively high energetic costs for strict nocturnality and thus nocturnal FD is reduced, and uniquely nocturnal FD is rare.
Globally, almost all crepuscular functional trait space overlaps with nocturnal, and to a slightly lesser extent diurnal trait space (Cox et al., 2021), and we find that spatially this holds true for FD.
Regions where there are more hours annually of biologically useful twilight have fostered the evolution of a greater diversity of uniquely crepuscular ecological strategies. However, most crepuscular FD is replicated at other times of day, and a high proportion of predominantly crepuscular species are flexible in when they are active (Cox et al., 2021). This further supports the assertion that crepuscular activity is largely a behavioural adaptation to specialised foraging at dusk and dawn, rather than strongly constrained by physiology, and is likely driven by a combination of predator-prey relationships and competition (Bonnot et al., 2020;Pei et al., 2018).
The ability to be active across the very different environmental and sensory conditions of day and night has arisen over much of the land surface where high temperatures do not preclude daytime activity. Cathemerality is thought to have evolved either as a flexible response to fluctuating environmental conditions (Curtis & Rasmussen, 2006), or due to high energetic demands in very small (Halle, 2006) or very large species (Owen-Smith, 1998;Ramesh et al., 2015). We find that support for both hypotheses F I G U R E 2 Biogeography of mammalian crepuscular FD. (a) is crepuscular FD, (b) is the proportion of mammalian FD that is unique to periods of dawn and dusk, and (c) is crepuscular FD that is redundant across the daily cycle presented as a proportion of mammalian FD. The histograms give the distribution across pixels. Pixels with 10 or fewer species and those with only one crepuscular species have been removed from (b) and (c). The pie chart in the top right shows the non-spatial proportional contribution of crepuscular FD to total mammalian FD, where the darker colour is unique FD and the lighter colour is redundant FD. The grey demonstrates the proportion of mammalian FD that is not crepuscular. Model outputs show the standardised regression coefficients and 95% credible intervals of covariation between crepuscular FD with five predictors of natural variation in light and temperature for the spatial (black) and non-spatial model (grey/diamond). The map projections are Behrmann's equal area.
varies geographically, with unique cathemeral FD being highly correlated with temperature seasonality (Pearson's correlation = 0.62; Figure S6). At the higher latitudes, very low winter temperatures combined with the seasonal absence of day or night make exclusively diurnal or nocturnal activity energetically costly. In these regions, cathemeral FD (unique and redundant) contributes most to mammalian FD, with unique cathemeral FD dominating. In response, species have evolved a diverse array of specialised adaptations, from continuous foraging (Halle, 2006), or seasonally switching between nocturnal and diurnal activity (Blix, 2016;Guiden & Orrock, 2020) to more thermally efficient large body sizes combined with thick winter coats and fat storage (Blix, 2016;Clarke & Rothery, 2008).
In contrast, cathemerality as a flexible response may have evolved largely in the tropics, where warmer temperatures allow behavioural flexibility to adapt activity to varying food availability (Ramesh et al., 2012). Indeed, activity during both the daytime and nighttime is prevalent in large carnivores (69% of 29 species >15 kg) and megaherbivores (>45 kg; Martin, 1989; Data S1), and we find high functional redundancy with nocturnal and diurnal species.
The commonness of diurnal activity across much of the tropics and mid-latitudes is associated with more hours of biologically useful daylight, supporting the suggestion that diurnality in mammals evolved independently on multiple occasions (Anderson & Wiens, 2017;Smale et al., 2003), driven in part by species escaping intense ecological pressures at night (Wu et al., 2018). This may particularly be the case under forest and woodland canopies where there is reduced light intensity and spectral composition, while cloud cover generated by forests decreases the availability of lunar light (the strongest natural influence on light intensity at night; Veilleux & Cummings, 2012). In these regions a greater proportion of both diurnal and nocturnal FD is redundant, doubtless as similar species avoid competition through temporal niche partitioning (e.g. Ferreguetti et al., 2015;Frey et al., 2017;Nakabayashi et al., 2021) or predatorprey relationships (Botts et al., 2020). Despite diurnality being more common than cathemerality (N = 896 vs. N = 526 species, respectively), outside of the upper latitudes both diel niches contributed similar levels to mammalian FD with a greater proportion of FD being uniquely cathemeral than uniquely diurnal. This suggests that being F I G U R E 3 Biogeography of mammalian cathemeral FD. (a) is cathemeral FD, (b) is the proportion of mammalian FD that is uniquely cathemeral, and (c) is cathemeral FD that is redundant across the daily cycle presented as a proportion of mammalian FD. The histograms give the distribution across pixels. Pixels with 10 or fewer species and those with only one cathemeral species have been removed from (b) and (c). The pie chart in the top right shows the non-spatial proportional contribution of cathemeral FD to total mammalian FD, where the darker colour is unique FD and the lighter colour is redundant FD. The grey demonstrates the proportion of mammalian FD that is not cathemeral. Model outputs show the standardised regression coefficients and 95% credible intervals of covariation between cathemeral FD with five predictors of natural variation in light and temperature for the spatial (black) and non-spatial model (grey/diamond). The map projections are Behrmann's equal area.
active during the night and day plays a greater role in trait diversification than fully switching to the daytime.
Complex niche construction within communities combined with sensory and environmental differences across the daily cycle suggest that the functioning of ecological systems is strongly partitioned between day and night (Gaston, 2019). However, somewhat surprisingly, very little attention has been paid to understanding large scale patterns of diel variation in FD (Gaston, 2019). Many ecosystem functions have evolved to target species active during specific periods of the day [e.g. attraction of pollinators (Fleming et al., 2009) or the evolution of strong odours or contrasting colours in fruits (Corlett, 2011)]. Although, functional redundancy may also occur across taxa, how processes and functions are expressed vary across taxa, for example nocturnal mammals often disperse seeds over a greater distance than diurnal birds (Medellin & Gaona, 1999;Tsunamoto et al., 2020), birds and bats forage for different invertebrate herbivores (Kalka et al., 2008;Kolkert et al., 2021) and efficiency of pollination varies between nocturnal and diurnal communities (Jaca et al., 2020;Tremlett et al., 2020). The consequences for ecosystem functioning in the context of the daily cycle has rarely been explored, however as diel unique and diel redundant species are lost the ecosystem impacts may be profound.

| CON CLUS IONS
Here, we demonstrate that although across much of the global land surface approximately three-quarters of FD is unique to specific times of day, almost half of this occurs at night, underscoring the functional importance of nocturnal mammals for ecosystems. In the face of the current intense anthropogenic pressures on mammals (e.g. Ceballos et al., 2017), maintenance of much of the mammalian influence on ecosystems will require a focus on protecting nocturnal species. However, nocturnal mammals are under increasing threat. Over much of the world, the nighttime is warming more quickly than the daytime (Cox et al., 2020), reducing thermal constraints and thus increasing habitable space for novel nocturnal species to invade and change existing competition dynamics and predator-prey relationships (Bonebrake et al., 2020). This effect will be compounded by an increasing number of day-active species escaping into the night in response to human F I G U R E 4 Biogeography of mammalian diurnal FD. (a) is diurnal FD, (b) is the proportion of mammalian FD that is unique to the daytime, and (c) is diurnal FD that is redundant across the daily cycle presented as a proportion of mammalian FD. The histograms give the distribution across pixels. Pixels with 10 or fewer species and those with only one diurnal species have been removed from (b) and (c). The pie chart in the top right shows the non-spatial proportional contribution of diurnal FD to total mammalian FD, where the darker colour is unique FD and the lighter colour is redundant FD. The grey demonstrates the proportion of mammalian FD that is not diurnal. Model outputs show the standardised regression coefficients and 95% credible intervals of covariation between diurnal FD with five predictors of natural variation in light and temperature for the spatial (black) and non-spatial model (grey/diamond). The map projections are Behrmann's equal area. disturbance and hunting (Gaynor et al., 2018), or higher thermoregulatory and water demands from increasing daytime temperatures (Bonebrake et al., 2020;Cox et al., 2020;Levy et al., 2019). Further, the prevalence of artificial light at night brings a broad range of negative impacts to species, populations and communities (Sanders et al., 2021), along with increasing unsustainable harvesting pressures (Bowler et al., 2020). Altogether, these factors may shift the distribution of mammalian FD across the daily cycle with currently unrealised but potentially profound consequences for ecosystems. No permits were required for this research.

CO N FLI C T O F I NTE R E S T
The authors declare no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
The trait data were principally extracted from our recently compiled database (Cox e al., 2021). The trait data were compiled from puted data with data sources and updated activity patterns in Data S1, which is also available at https://doi.org/10.5281/zenodo.7476391.