Field Sites and Experiments.

Field experiments were conducted concurrently in two grasslands chosen for their contrasting mixtures of C₃ and C₄ grasses. The mixed-grass prairie near Hays, KS, is a C₄-dominated native grassland (73.1% C₄ based on ANPP measurements from control plots during the pretreatment and four experimental years) located within the epicenter of the 1930s Dust Bowl drought and where drought-induced increases in C₃ grasses were documented historically (Fig. 1 and (11, 12)). The grassland near Cheyenne, WY, is a C₃-dominated grassland (69.8% C₃), located at the periphery of the most severe extent of the Dust Bowl drought. In addition to their geographic locations at the edge and center of Dust Bowl drought, these sites were chosen because P. smithii (the C₃ grass that increased extensively during the Dust Bowl (11, 12)) was present at both sites; as a dominant species at the C₃ site and as a widespread, moderately abundant member of the plant community in the C₄ grassland. Both grasslands were lightly grazed prior to a pretreatment sampling year (2013) and were protected from domestic grazers during the subsequent 4-y experimental period. Although most grasslands are naturally grazed by native or domesticated large herbivores, including their activities as part of this experiment was not feasible.

Drought was experimentally imposed at each site for 4 y using large rainfall exclusion shelters (Fig. 1A and 68). At each site, 30 plots, each 36 m², were established across a topographically uniform area. Each plot was hydrologically isolated from the surrounding soil using aluminum flashing to a depth of 20 cm and 6-mil plastic barriers installed to a depth of 50 cm. Drought was imposed in 20 plots per site by installing large shelters (10 × 10 m) that were used to block 66% of incoming rainfall during each growing season—this is roughly equivalent to a 50% reduction in annual precipitation given that 60–75% of MAP falls during the growing season in these ecosystems. The remaining 10 plots per site were trenched and hydrologically isolated as well, serving as controls that received ambient rainfall. Drought was imposed by reducing the same quantity of growing season rainfall inputs in two different ways. We removed a portion (66%) of each rain event (68) during the entire growing season (April–August; chronic drought treatment, n = 10 plots/site) or we fully excluded all rain events for a shorter portion of the growing season (May through mid-August, intense drought treatment, n = 10 plots/site). Soil moisture was measured in the center of each plot at a depth of 0–15 cm and 0–30 cm with 30-cm time-domain reflectometry probes (Model CS616, Campbell Scientific, Logan, UT) inserted at a 45° angle or perpendicular to the soil surface. Responses in soil moisture and ratios of C₃ vs. C₄ ANPP to each type of drought were consistent (SI Appendix, Figs. S5 and S6), thus results were combined into a single drought treatment. Treatment infrastructure was installed in the spring of 2012, but drought treatments did not begin until 2014 because 2012 was a natural extreme drought year and not appropriate as a pretreatment year. Rain gauges were established in a subset of control and treatment plots to confirm the efficacy of the infrastructure in removing rainfall inputs to experimental plots. Infrastructure effects on the microclimate were also monitored, particularly photosynthetically active radiation and soil temperature, and these effects were similar to those documented for many other studies of rainfall exclusion infrastructure effects (10–20% reductions in photosynthetically active radiation, depending on the time of day; minimal differences in air, soil, and leaf temperatures; no differences in leaf‐to‐air vapor pressure differences under the infrastructure compared with controls; 68–70).

At the end of each growing season, all aboveground biomass was harvested in three quadrats (0.1 m²) placed randomly in new locations each year in two 2 × 2 m subplots designated for destructive measurements. A different subplot was harvested each year. Biomass was sorted to remove the previous year’s growth, dried for 48 h at 60 °C, and weighed to estimate total ANPP, as well as ANPP of dominant plant growth forms (C₄ grasses, C₃ grasses, and C₃ forbs). Subshrubs and succulents, comprising < 5% of ANPP at both sites, were excluded from calculations due to measurement difficulties, but C₃ sedges, as graminoids, were included in the C₃ grass category.