CW3E is an acronym for the Center for Western Weather and Water Extremes, an arm of Scripps Institution of Oceanography in San Diego, CA. It is a body with a particular focus on the analysis and prediction of weather patterns, especially precipitation, in the western United States and California in particular. Because California's mediterranean climate features very distinct wet and dry seasons, and because most of the state's precipitation comes from a handful of atmospheric river (AR) events in the winter, the state is highly susceptible to droughts in bad years or floods in periods of high AR activity. Therefore, accurate forecasting is tremendously important.
One major project CW3E is currently working on is EnsDL, a Deep Learning-based prediction algorithm that builds on a pre-existing major model ensemble. While ensemble forecasts are generally accurate, they are not tuned for the particular climatic and topographic conditions in the western U.S; thus, the EnsDL algorithm takes ensemble outputs and feeds them into another neural network, which makes an adjustment to improve the accuracy of the forecast. The output is a gamma distribution of forecasted potential precipitation, which can then be parsed by a CDF function into a mean forecast and various percentile forecasts.
I was initially accepted as a Summer Intern for CW3E, but my position has been extended to be a part-time one through the school year. Since joining CW3E in June, I've made a number of key contributions to the EnsDL codebase, including: