Fieldwork and sampling
Measurements were taken from a 20x20 m area at each site. After the tree leaves had fully expanded (late June), the B. pubescens leaf area index (m2 leaf m-2 ground, LAI) was estimated with the Hemisfer software, version 2.2 (Schleppi et al. 2007; Thimonier et al. 2010) based on ~10 hemispherical images per site (camera: Nikon Coolpix 4500; lens: Nikon Fisheye converter fc-e8 0.21x). In addition, we used ground slope correction (Schleppi et al. 2007), and the most recent algorithm for LAI estimation following manual thresholding (Gonsamo et al. 2018). We randomly selected ~30-50 fully expanded leaves from the lower canopy (~1.5-2.5 m above the ground) from a minimum of sevenB. pubescens trees per site to estimate specific leaf area (m2 g-1 dry mass, SLA) by scanning the fresh leaves to obtain the area, and then drying (40°C for 48h) and weighing. Annual B. pubescens leaf production (g dry mass m-2 yr-1, LP) per site was assumed to be equivalent to fully expanded canopy biomass in this deciduous species, which was estimated by dividing the LAI with the SLA (LP=LAI/SLA). After scanning and drying, the leaves were chemically analysed (see below). In the late growing season (late August), ~30-50 freshly fallen or senesced yellow leaves still on the branches were collected from a minimum of 7 trees per site in order to determine leaf chemistry at senescence (see below). All leaves were dried (40°C for 48h) and ground before chemical analyses. In the late growing season, 4 large organic horizon samples (soil from ~1 m2 each sieved through 6 mm mesh) from each site were collected for a range of chemical analyses as indicators of site fertility (see below). Subsoil samples were collected in sampling rings (3 x 100 cm3 per site) and composited before analyses.
The insect herbivory level was estimated using a modified version of the method described by Crutsinger et al. (2008). At each site, we randomly selected 7 trees in the late growing season (August/September), after the vast majority of insect herbivory had terminated. Trees were selected from a distance (~15 m) where insect herbivory could not be detected, to minimize sampling bias (Zvereva and Kozlov 2019). On each tree, 3 branches were randomly selected on which we visually estimated the percentage of leaf area lost to herbivory (0%, 1–5%, 5-10%, 10–20%, 20–30% … 90–100%) in the lower canopy (~1.5-2.5 m above the ground). Ten leaves were surveyed per branch, starting from the first full-sized leaf and down the branch surveying every other leaf. When calculating the average herbivory level for each site, each observation was assigned the median value of its interval, e.g. 10-20% = 15%.
For site characterisation, the ground vegetation cover was visually estimated in the early growing season (late June/early July) as the average coverage (%) within 3 randomly selected squares of 3x3 m per site. Dwarf shrubs were determined to the species level, while mosses, lichens, graminoids and forbs were identified to functional group level.
The Luftwebb record for 2000-2014 shows a range in mean annual air temperature across our transects between -1.1±0.2 and 0.3±0.2 °C (mean ± SE), and the mean annual precipitation ranged from 447±17 to 1366±34 mm year-1. The mean growing season temperature ranged between 9.9±0.3 and 11.3±0.3 °C. To capture the smaller scale variation in climate at the site level, the soil temperature was measured with iButtons (Maxim Integrated, San Jose, CA, USA) installed in the topsoil (5-10 cm depth, 1 at the centre of each site) from early September 2015 until late August/early September 2017. We used the mean growing season soil temperature as a predictor (averaged between 15/6-15/9) to get the most relevant temperature for aboveground processes, and to reduce the legacy effects of varying snow-cover between sites. The soil moisture was estimated as an average of ~15 measurements per site with a Campbell Hydrosense II moisture sensor (~10 cm depth) (Campbell Scientific Inc., Logan, UT, USA) at every site visit. For statistical analyses, we used the early growing season moisture content (late June/early July); as the topsoil was very dry at some sites in the late season, which might make these measurements less representative as a proxy for root zone water availability.