Atmospheric River Forecast Model Bias Correction

We propose to use machine learning algorithms to help improve weather forecasts of Atmospheric Rivers (ARs). ARs are filamentous bands of anomalous water vapor transport, which deliver 90% of the total water vapor from the tropics to the extratropics and midlatitudes. Landfalling ARs are associated with extreme precipitation, and strongly influence both flooding and water availability. ARs are particularly important for the North American West Coast, whose precipitation signal is dominated by the frequency of ARs, with the majority of the annual precipitation and extreme precipitation events being contributed by a few events each year. These storms are immensely important yet forecasting their landfall location remains a challenge. A 2013 study by Wick et al examined the ability of five operational ensemble forecast systems to accurately predict AR landfall location as a function of lead time out to 10 days and found landfall error ranges of up to 300 km with a forecast lead time of just 1 day. This project proposes to use deep learning techniques to help correct forecast model bias and improve the forecasting skill associate with landfalling ARs. The implications of an improved forecast are far reaching, touching the west coast domestic water supply, understanding of a changing climate and food security.