For this project, we developed a deep learning (DL) approach to producing probabilistic forecasts, based on a novel distribution-learning DL architecture. Probabilistic – as opposed to point-estimate – forecast systems are essential for risk mitigation under uncertainty. Probabilistic forecasting systems can provide critical inputs to predict hurricane tracks, forecast electrical demand, track epidemiology, evaluate species endangerment, assess seismic activity, and many other similar use cases.  The project is addressing a research gap in that the typical DL approaches to forecasting yield only a point estimate. Generating probabilistic forecasting using DL models has not yet been explored, and probabilistic DL has not been utilized for forecasting. As a part of this research, we built a joint probability distribution of current state and delayed state from training data to develop a probabilistic forecast distribution. We used a DL-based Jacobian Tracker Neural Net (JTNN) to make the probabilistic forecast. The method was applied to predict wind velocity and speed using simulations produced by the Weather Research & Forecasting (WRF) code.