Plant-Soil interactions under Global Warming: Learning Mechanisms from Multiyear Field Experiments and Natural Gradients.

Soil carbon is the largest terrestrial carbon (C) store but remains a large source of uncertainty in climate models due to lack of process understanding. Using a natural temperature and precipitation gradient in a sage-steppe ecosystem, we characterize the change in soil carbon dioxide (CO₂) flux and production to explain environmental controls of ecosystem C storage and loss. We quantified soil CO₂ efflux, soil water, and temperature and how they co-vary with 1) depth within the soil profile, 2) elevation within the watershed, and 3) seasonality. Our three field sites were located at long-term experimental sites in the Reynolds Creek Experimental Watershed and Critical Zone Observatory at elevations ranging from 1425 to 2111 m, with corresponding mean annual temperature and precipitation ranges of 9.4 to 5.6 degrees Celsius and 292 to 800 mm, respectively. Snow cover ranged from ephemeral to seasonal with important impacts on soil freeze-thaw dynamics. Sagebrush species dominated at each site (Wyoming big sagebrush, Low sagebrush, and Mountain big sagebrush), with parent material, aspect, and slope all similar. Co-located eddy covariance indicate a large productivity gradient with elevation. At each site, soil profile CO₂, water content and temperature sensors were installed at 4-6 depths to a depth of about 1 m and recorded every 30 minutes between October 2016 and 2017. Rates of CO₂ efflux and production used modeled soil porosity and measured hydraulic characteristics to account for changes in diffusivity with variable soil moisture. Cumulative soil C flux differed between sites, varying more with a change in soil water than temperature. Changes in CO₂ production with soil depth varied between sites, related to both soil water and temperature thresholds. Our results highlight the use of existing sensor capabilities in understanding dynamic near-surface processes with high temporal variability to aid in dynamic C models.