Varied diets, including broadleaved forage, are important for a large herbivore species inhabiting highly modified landscapes.

Diet quality is an important determinant of animal survival and reproduction, and can be described as the combination of different food items ingested, and their nutritional composition. For large herbivores, human landscape modifications to vegetation can limit such diet-mixing opportunities. Here we use southern Sweden's modified landscapes to assess winter diet mixtures (as an indicator of quality) and food availability as drivers of body mass (BM) variation in wild moose (Alces alces). We identify plant species found in the rumen of 323 moose harvested in Oct-Feb, and link variation in average calf BM among populations to diets and food availability. Our results show that variation in calf BM correlates with variation in diet composition, diversity, and food availability. A varied diet relatively rich in broadleaves was associated with higher calf BM than a less variable diet dominated by conifers. A diet high in shrubs and sugar/starch rich agricultural crops was associated with intermediate BM. The proportion of young production forest (0-15 yrs) in the landscape, an indicator of food availability, significantly accounted for variation in calf BM. Our findings emphasize the importance of not only diet composition and forage quantity, but also variability in the diets of large free-ranging herbivores.

2mm and 4mm fractions). To assess how consistently this method produced results from different sub-portions of an original rumen sample, the same procedure was repeated (by the same person) on an additional sub-portion from 16 of the individuals already analysed. The analyst did not know the original sample ID. We used a two-sample t-test to assess whether the relative proportions of plant categories (% dm) differed between the two subportions.
To establish whether calf BM reflects population reproductive status, we tested the correlation between population mean calf BM (N=7) and two estimates of reproduction recorded routinely by Swedish hunters (Morellet et al. 2011): mean observed number of calves per Felton, A. M., Holmström, E., Malmsten, J., Felton, A., Cromsigt, J.P.G.M., Edenius. L., Ericsson, G., Widemo, F., Wam, H.K. Varied diets, including broadleaved forage, are important for a large herbivore species inhabiting highly modified landscapes. Supplementary Appendix female, and proportion of females observed with at least one calf . We extracted these data from the systematic hunter-reported statistics in the Swedish national online databases www.viltdata.se (for the observations) and www.algdata.se (for the carcass weights; Table   4SI) for five hunting seasons (2012/13 -2016/17).
We inferred food availability at the subpopulation level from landscape scale habitat variables available in publically available data source from the Swedish Environmental protection agency. Data was calculated by GIS for a circular area, with a 10 km radius (314 km 2 ) from the centre point of each subpopulation, using Esri software ArcGis Desktop software 10.4.1 Area of land use (forest, agriculture, urban, water, mire) and forest type (within the land use "forest", see below), was retrieved from a raster layer with pixel resolution of 25×25 m (CadasterENVSweden 2015). Forest types were divided in coniferous or broadleaved forest, and further divided in Scots pine or Norway spruce forest, and slow-(mainly oak or beech) or fast-growing broadleaves (mainly birch and aspen).

Statistical Analyses
Statistical analyses were conducted in R ×64, 3.4.3 (RCoreTeam 2017). All variables were visually explored using Cleveland dotplots to detect outliers in variables, as well as boxplots and interaction plots for factors (Zuur, Ieno & Elphick 2010). We omitted one outlier for calf BM (96 kg carcass weight, which is physically not possible, assuming it was a yearling wrongly classified by hunters into calf). A first check of data (regression analysis) was made in order to detect bias on BM due to harvest date, which was not significant, and therefore ignored in further analyses.
We tested the potential correlation in the data set between the mean % dm of "unidentified wood" per moose population (n = 7) and mean calf body mass (BM) of the population using Pearson correlation. Potentially, a large share of wood fragments that could not be assigned to a genus or species in rumen samples could be related to a high browsing pressure in the area, as heavily browsed trees may offer moose twigs devoid of buds to a greater extent.
We wanted to see whether or not it is legitimate to include rumen results from all age-sex classes in a subpopulation average measure of diet composition. If age-sex classes differ, they should not be merged. We therefore conducted a principal component analysis (PCA), including individuals with known sex and age (n = 302) and a subset of the 44 plant categories identified. This subset included all plant categories that were represented with at Felton, A. M., Holmström, E., Malmsten, J., Felton, A., Cromsigt, J.P.G.M., Edenius. L., Ericsson, G., Widemo, F., Wam, H.K. Varied diets, including broadleaved forage, are important for a large herbivore species inhabiting highly modified landscapes. Supplementary Appendix least 1% of dm across all subpopulations (14 categories). In addition, because of our a priori interest in supplementary fed root vegetables (due to their suspected disproportionate effects on moose digestion and forage selection 31 ), we also included within this subset "all root vegetables", which surpassed the 1% threshold when B. vulgaris, D. carota and S. tuberosum were combined (0-14% of dm among subpopulations, mean 1% ± 0.6% SE). Together these fifteen plant categories represented 96% of dm. While any inclusion threshold (i.e 1% of dm) is to some extent subjective, in this case it enabled us to remove the noise of rare food items, increase the explanatory power of the model, while simultaneously capture the dominating food items, including supplementary foods. For each moose individual, the % dm per plant category was included, and its respective demographic class (calf = 1, yearling = 2, adult female = 3, adult male = 4).
The first removal and sorting of covariates in the landscape scale analyses was done using multipanel scatterplots when there was collinearity between covariates. For all correlations > |0.6| (Zuur et al. 2009), we kept the variable with the lowest sum of correlations across covariates. We did not include, for example, area of conifers because it was strongly correlated with total forest area. All subpopulations were included in the landscape scale analyses, irrespective of sample sizes of moose data.

Supplementary Results
There was no relationship between the mean % of unidentified wood in rumen samples and mean calf BM of the moose population (Pearson correlation = 0,464, P = 0,294, n = 7). There was a large overlap in plant identifications between the first and the second round of analysing 16 rumen samples with macroscopy. In 91% of cases, if a plant item was present or absent in one round of analysis, it was likewise present or absent in the second round. Eight novel plant items for the whole study were identified when the first and second round were compared. With one exception, these eight novel identifications represented items that were present in very small proportions of the sample's total dm, ranging from 0.07% to 0.96%. The exception was "unidentified bark" which was found in 2 of the original samples (both ca 2% dm) but not at all in the second round. There was no significant difference in the relative proportions of plant categories present in the two rounds (t-value = -0.19, DF = 65, p = 0.846). Regarding the ten most dominant plant categories, the same general diet pattern was obtained from both assessment rounds (Fig. 4SI). Only % dm of M. domestica (apple) Felton, A. M., Holmström, E., Malmsten, J., Felton, A., Cromsigt, J.P.G.M., Edenius. L., Ericsson, G., Widemo, F., Wam, H.K. Varied diets, including broadleaved forage, are important for a large herbivore species inhabiting highly modified landscapes. Supplementary Appendix differed considerably. We believe this had little bearing on our findings, because apples were found in only 5% of the rumen samples.
Principle component analysis of macroscopy results of 302 individual rumen samples indicated that individuals are representative of their area's overall diet regardless of their age or sex. First, the four different age-sex classes did not form distinct clusters in terms of % dm of different plant categories. The great majority of individuals fall within the main cluster of data points (Fig. 3SI). Of the 11 data points that stick out from the main cloud of data, 54% were calves, which is similar to their proportion in all samples (53%), which indicates that age class did not influence the pattern in diet composition. Only 22% of the total variation was explained by the first two components in this PCA. Second, the variable "age-sex class" (1-4) did not obtain high values in any of the principal components (Table 6SI). In the following analyses, all age-sex categories were therefore merged.  Felton, A. M., Holmström, E., Malmsten, J., Felton, A., Cromsigt, J.P.G.M., Edenius. L., Ericsson, G., Widemo, F., Wam, H.K. Varied diets, including broadleaved forage, are important for a large herbivore species inhabiting highly modified landscapes. Supplementary Appendix Figure 2SI. Linear fit (effect plots with 95% confidence bands) of the most parsimonious model of association between likelihood of moose subpopulations having calves with low body mass and a) proportion of forest in the landscape and b) the proportion of forest being young. Low BM was defined as < 0.5 SD from the mean BM across subpopulations (i.e. <51.9 kg dressed carcass). BM data from N= 222 calves (6-9 months old) in Southern Sweden, Oct 2014-Feb 2015. Forest data was collected from GIS data of a circular area with r=10 km from the centre point of the habitat range of each subpopulation.
Felton, A. M., Holmström, E., Malmsten, J., Felton, A., Cromsigt, J.P.G.M., Edenius. L., Ericsson, G., Widemo, F., Wam, H.K. Varied diets, including broadleaved forage, are important for a large herbivore species inhabiting highly modified landscapes. Supplementary Appendix Figure 3SI. A) Loading plot from Principal Component Analysis (PCA) of the diet compositions of 302 moose individuals. For each individual rumen sample the percentage of dry matter per plant category (15 categories) identified by macroscopy is included in the model, as is the code of their respective age-sex class. B) Score plot from the same PCA showing that the age-sex categories of the individuals (calf = 1, yearling = 2, adult female = 3, adult male = 4) were evenly distributed along the loadings. Of the 11 data points that stick out from the main cloud of data, 54% were calves, which is similar to their proportion in all samples (53%).
Felton, A. M., Holmström, E., Malmsten, J., Felton, A., Cromsigt, J.P.G.M., Edenius. L., Ericsson, G., Widemo, F., Wam, H.K. Varied diets, including broadleaved forage, are important for a large herbivore species inhabiting highly modified landscapes. Supplementary Appendix Figure 4SI. Comparison of the mean composition of plant categories, in terms of % dry matter (dm), identified in a first subportion (A) and a second subportion (B) of the same 16 original rumen samples. Plant categories that represent at least 2% of total dm are described by name.  Tables   Table 1. Samples from moose in southern Sweden were collected in seven (A-G, see Fig. 1 Main text) moose management areas (MMA, each representing a moose population), divided into multiple moose management units (MMU; subpopulation). The number of MMU from which we obtained samples is indicated ( a ), as well as their summed area ( b ) in terms of tha (1000 ha), mean area ( c ) and the proportion of the MMA which was included in the study ( d ). Samples were collected from moose harvested 23-Oct-2014 and 22-Feb-2015, with the majority during Oct -Nov (some samples lacked date information ( e )). Of the total number of obtained samples ( f ), the proportion representing calves is indicated ( g ). The number of samples ( h ) from these seven moose populations, used for macroscopy analysis of rumen contents, was 302 (21 of the 323 individuals for which we obtained macroscopy results came from areas outside of the seven main study populations).* In the multivariate analysis of diet composition ( Fig. 4 Main text) results from 252 rumen samples were used (only including MMUs for which we had at least 5 rumen samples with macroscopy results).