Machine Learning-Based Enhancement of the NESDIS Snowfall Rate Product for Orographic Snow (Invited)

The NOAA-NESDIS operational land snowfall rate (SFR) product is derived from a constellation of polar-orbiting satellites operated by NOAA, NASA, and EUMETSAT: S-NPP, NOAA-20, NOAA-21, NOAA-19, Metop-B and Metop-C. The product is derived from measurements taken by passive microwave radiometers including ATMS, AMSU-A and MHS. The SFR algorithm consists of a snowfall detection model and a snowfall rate estimation procedure. The former is an eXtreme Gradient Boosting (XGB) machine learning model trained using the NOAA Integrated Surface Database (ISD) global ground observations. The snowfall rate algorithm is based on the one-dimensional variational retrieval (1DVAR) approach, and enhanced with a neural network (NN) bias correction model.

The NEDIS SFR algorithm has benefited significantly from continuous development in recent years. One of the newest developments is in the area of orographic snowfall. Snowfall induced by orographic effect is among the most difficult types of precipitation to retrieve. The SFR product has reduced performance in orographic snowfall regions, such as in the Appalachians where snowstorms are often lake effect snow because of the moisture carried to the region from the Great Lakes. We examined and selected a method for identifying orographic snowfall. The SFR product was then evaluated for its effectiveness in identifying snowstorms as a result of orographic enhancement. The goal of this study is to develop a machine learning model that incorporates orographic snow-relevant features, such as elevation gradient and wind, into the SFR algorithm to improve its performance for the challenging snowfall.