SIParCS 2019 - Behrooz Roozitalab
Analog Ensemble Probabilistic Forecasting Using Deep Generative Models
Reliable weather forecasts can provide insightful information for decision makers and save many lives and resources. Hence, Deterministic Numerical Weather Prediction (NWP) models are widely used for this purpose. However, each NWP model run can represent only one imperfect future state of the atmosphere due to errors in initial conditions and scientific gaps in the model. As a result, probabilistic forecasts have been explored to overcome this issue. Analog Ensemble (AnEn) is a method that uses a historical dataset of the predictand to correct the forecasted value. Although this method has removed the necessity of running one deterministic NWP for multiple times, it still requires to keep the whole dataset in the memory. Moreover, it has to search the dataset for every correction. In this work, we try to replace the whole historical dataset with a model that has the capability to learn the Probability Density Function (PDF) of that dataset. Specifically, we utilize a Conditional Variational Autoencoder (CVAE) deep generative machine learning model in order to correct the wind speed forecasts of North American Mesoscale (NAM) forecasting system. As a result, we feed the values forecasted by the NWP model as a condition to our CVAE and generate an ensemble used to correct the forecasted value in constant time and with small memory usage.
Mentors: Alessandro Fanfarillo, Davide Del Vento