Global retrieved snowfall properties using a neural network and GPM-DPR

Abstract:

With the launch of the Global Precipitation Measurement mission’s Dual-frequency Precipitation Radar (GPM-DPR) in 2014, there is an opportunity to investigate the global distribution of snowfall and its various properties. A novel neural network retrieval that is formulated using state-of-the-art ice particle scattering models and in-situ observations is used to retrieve the characteristic size and mass flux of near-surface snowfall. Six years of dual-frequency measurements in falling snow (over 8 million profiles) show that the neural network retrieval predicts 87% higher climatological mass accumulation than the operational retrieval algorithm, agreeing in sign with the published underestimation when compared to CloudSat. Within this new snowfall climatology, a thermodynamic analysis of MERRA-2 reanalysis data show that 71% of continental snowfall profiles are produced from stable thermodynamic environments while 59% oceanic snowfall profiles are produced by conditionally unstable environments. Furthermore, conditionally unstable environments have significantly (p < 0.05) larger characteristic size and mass flux compared to stable environments. Spatial distributions of the near surface snowfall characteristic size show that regions containing orography (e.g., Tibetan Plateau; Rocky Mountains) have anomalously small characteristic sizes compared to the global average, while the North Atlantic has anomalously large characteristic sizes. Additional investigations of the climatology using k-means clustering will elucidate the dominate snowfall modes which will then be characterized by the vertical profiles of radar observables (e.g., Ku-band reflectivity), MERRA-2 reanalysis and retrieved snowfall properties.