Authors: Savanna Wolvin – University of Utah
Title: Prediction of Derived and Crowdsourced Precipitation Phase Observations of the Western CONUS via Machine Learning
Presentation Type:
Abstract: Precipitation phase has strong implications for the estimation of snow water equivalent (SWE) and snow depth in the western Contiguous United States (CONUS). Historically, the western CONUS receives a majority of its cool season precipitation as snowfall in the mountains, which then melts in the warm season, replenishing freshwater reservoirs and lakes. Cumulative errors in precipitation phase determinations can cause errors in estimated SWE and snow depth in the western CONUS, subsequently producing uncertainties in peak SWE, snow cover duration, downstream runoff, and other hydrological processes. For instance, rain-on-snow events can cause snowmelt and additional runoff, leading to enhanced flood risk and natural hazard fatalities. Thus, accurate predictions of precipitation phase are vital. The National Oceanic and Atmospheric Administration’s National Water Model (NWM), designed to predict streamflow for 2.7 million river reaches, estimates the precipitation phase empirically via the surface air temperatures. However, numerous studies of precipitation phase compared to surface temperatures show significant spatial and temporal variability across the western CONUS; additionally, surface air temperature cannot account for physical processes above the surface. Prior studies have called for the integration of above-surface conditions and the use of novel data to improve precipitation phase prediction. Our study aims to develop a tree-based machine learning algorithm to predict derived and crowdsourced observational precipitation phase from above-ground and surface output of the High-Resolution Rapid Refresh (HRRR) forecast simulations to improve the NWM’s hydrological output. The derived observational precipitation phase comes from SNOw TELemetry (SNOTEL) daily observations, where increases in daily SWE indicate snow and decreases in daily SWE indicate rain, when precipitation accumulation is observed. Crowdsourced observational precipitation phase comes from the Mountain Rain or Snow (MRoS) and the Meteorological Phenomena Identification Near the Ground (mPING) projects, in which precipitation phase is recorded in real-time and categorized by citizen scientists. We present on the intercomparison of the precipitation phase datasets, development of a machine learning algorithm, and the effects of differing target and predictor data sets.