Challenges and Successes of Snowfall Forecasting Using Machine Learning and Numerical Prediction for the Northeast United States
Abstract
Winter storms accompanying snowfall of varying density have severe impacts in the Northeast United States, given the densely forested areas that surround power lines and essential infrastructure. However, one recurring challenge is the lack of accurate snowfall and snow density predictions that when coexisting with high winds, can be damaging to trees and, thus, power lines, houses, and infrastructure. The goal of this work is to explore a novel integrated machine learning (ML) approach to assess and evaluate the accuracy of snowfall prediction. We have developed the ML model by feeding different atmospheric variables at surface and different pressure levels from the Weather Research and Forecasting (WRF) model, trained and validated using snowfall observations. WRF provides the 3D atmospheric variables necessary for the prediction of snowfall including temperature, humidity, horizontal and vertical wind, and total accumulated precipitation. The Global Historical Climatology Network (GHCN) and National Snowfall Analysis (NSA) are two data sources that are used as observations. The first ML algorithm of choice is random forest (RF) which has shown very promising results for other dynamic variables such as wind gust. For this research, we have investigated thirty-two snowstorms (mostly Nor'easters) that have affected the NE U.S. in the past 10-15 years. We trained and implemented the ML model using the leave-one-storm-out cross validation approach. For some storms, the approach has shown improved snowfall prediction, and some have faced challenges compared to the standard utilization of microphysics schemes in the context of numerical weather prediction modeling. We will present the challenges and successes of the NWP-ML approach for snowfall as well as future steps to gain additional improvements.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H35M1288K