CEE Ph.D. Defense Announcement: Disentangling Soil Moisture–Precipitation Coupling with High Dimensional Model Representation (HDMR) across CONUS

Abstract: Land–atmosphere interactions shape regional hydroclimate, with soil moisture–precipitation coupling (SMPC) being among the most influential yet least well quantified. Commonly used statistical approaches struggle to capture the complex, high-dimensional and nonlinear interactions among soil, vegetation and atmospheric variables that govern SMPC. This dissertation applies High Dimensional Model Representation (HDMR), a variance/covariance-based sensitivity framework, to disentangle the direct (structural) and indirect (correlative and cooperative) effects of soil moisture and related land–atmosphere variables on precipitation. Simple toy problems and benchmark experiments with hydrologic models demonstrate HDMR's ability to guide experimental design and distinguish between structural, correlative and cooperative contributions of input variables to model output. Applied to collocated SMAP soil moisture and GPM cloud profile observations over the central United States, HDMR shows that seven-hour antecedent soil wetting combined with warm surfaces boosts 1–3 km radar reflectivity by up to 4 dBZ, with pronounced spatial variability in both sign and magnitude. Continental analysis with the CONUS404 reanalysis reveals that morning soil moisture accounts for 20–40% of afternoon precipitation variance across the Great Plains, roughly double previous estimates, and identifies critical thresholds where interactions with temperature and humidity significantly influence precipitation amounts. Overall, the HDMR framework offers a novel and rigorous diagnostic tool for evaluating land–atmosphere coupling in weather and climate models, with significant implications for improving seasonal forecasts amid climate and land use changes.