2.2. Experimental design

At each site, 20 circular plots (7.1 m²) were randomly assigned to one of four climate treatments with five replicates each: control (ambient temperature and precipitation), warming (canopy temperature raised by 2.5°C), warming with additional precipitation (warming + ppt; plots irrigated to fully offset a warming‐induced drying effect), and drought (annual precipitation reduced by 40%). The southern and central sites were part of a previous experiment from 2010–2012 with a different set of treatments consisting of control, warming by 2.5°C, increased precipitation intensity by 20%, and warming by 2.5°C + increased precipitation intensity by 20% (Pfeifer‐Meister et al., 2013, 2016; Reynolds, Johnson, Pfeifer‐Meister, & Bridgham, 2015). However, the precipitation intensity treatments had almost no effect on either plant or ecosystem responses since most of the additional water was applied during the wet season (Pfeifer‐Meister et al., 2013, 2016; Reynolds et al., 2015). Thus, the current experiment has the same control and warming treatments at the two southernmost sites, but the enhanced precipitation intensity plots became the drought plots, and the warming plus enhanced precipitation intensity plots became the warming + ppt plots of the current experiment. The northern site was newly established for this experiment.

Warming treatments were achieved using six 2000‐W infrared heaters per plot, as described in Pfeifer‐Meister et al. (2013). The warming + ppt plots used an automated sprinkler system (with rainwater collected on site) designed to irrigate these plots for 30 min each night that the volumetric water content was below 95% of the control plot average. The drought treatment used a common fixed rain‐out shelter design, with clear acrylic shingles (MultiCraft Plastics, Eugene, OR) covering 40% of the plot area to prevent 40% of annual rainfall from reaching the plot. The acrylic material has high light transmittance, reducing microclimatic impacts such as shading concerns or temperature buffering (Gherardi & Sala, 2013; Yahdjian & Sala, 2002). The 40% reduction in annual precipitation represents an “extreme” drought, consistent with a one‐in‐100‐year event for the three sites, determined using the Precipitation Trends and Manipulation tools from Drought‐Net (Lemoine, Sheffield, Dukes, Knapp, & Smith, 2016). Drought treatments were installed in February 2016, all warming treatments initiated by summer 2016, and irrigation initiated during summer 2016. Heaters were turned off in August and September 2017 at all three sites due to fire hazard. We used dataloggers to record continuous canopy temperature, soil temperature (at 10 cm depth), and volumetric water content (to 30 cm depth) within each plot. To compare soil moisture across sites with considerably different soil characteristics, we calculated soil matric potentials as described in Saxton and Rawls (2006). See Supporting Information Figure S2 for data on soil temperature and matric potential in plots during the study. Due to heater malfunctions in one of the central‐site warming plots for a period of the 2017 growing season, we excluded data from this plot for phenological analyses occurring during that time.

Between October 2014 and January 2015, all plots at the southern and central sites were mowed and raked while the new northern plots were treated with Glyphosate 2% (a total of three times) to remove standing biomass. By February 2015, all plots were seeded with a common mix of 29 native grass and forb species found in PNW prairies (Pfeifer‐Meister et al., 2013). Additionally, in fall of both 2015 and 2016, we seeded between 80–200 seeds per species of 14 range‐restricted species within each plot for the purposes of a separate demography experiment. These species were selected for having medium to high fidelities to upland prairies with geographic range distributions within the PNW (~41–50° latitude). Due to low establishment of six of these 14 species at all sites, only eight were used as focal species in this study (Table ​(Table1).1). For each species and site, we used seeds from the nearest available source population. Four species (Collinsia grandiflora, Festuca roemeri, Microseris laciniata, and Plectritis congesta) had unique sources for each site; the remaining four species (Achyracheana mollis, Plagiobothrys nothofulvus, Ranunculus austro‐oreganus, and Sidalcea malviflora) had single sources for all sites.

Characteristics of the eight focal species analyzed for flowering phenology observations (Jaster, Meyers, & Sundberg, 2017)