Enhancing Snowfall Accumulation Prediction in the Northeastern United States: Integrating NWP, Probabilistic Forecast and Machine Learning

The Northeastern United States is frequently confronted with the consequences of winter storms, characterized by varying snowfall density and trees surrounding critical infrastructure. Accurate snowfall predictions remain elusive, compounded by the potential for high winds to damage trees, power lines, residences, and essential systems. In this context, we present a comprehensive study where the Weather Research and Forecasting (WRF) model is employed to simulate historical winter storms. Despite its capabilities, the WRF model encounters challenges inherent to deterministic Numerical Weather Prediction (NWP) models, particularly in accurately capturing snowfall. To enhance predictions, we examine the Probabilistic Winter Precipitation Forecast (PWPF) products generated by the Weather Prediction Center Winter Weather Desk (WWD). These PWPF forecasts, collaboratively refined with National Weather Service Weather Forecast Offices (WFO), extend 72 hours into the future, aiding in accumulation predictions. Similar to the NWS WFO collaboration, our aim is to assess the probable implementation of the PWPF forecasts to enhance our own snowfall accumulation predictions.

We embark on a two-fold approach: first, assess the PWPF products and our in-house WRF model output against the National Snowfall Analysis (NSA). Subsequently, we propose the integration of various atmospheric variables from WRF, PWPF products, and NSA datasets using machine learning algorithms to further refine snowfall accumulation predictions. This research holds the potential to significantly bolster snowfall forecasting accuracy, thus contributing to better-informed decision-making and improved preparedness for winter weather events in the Northeastern U.S.