Consensus Weather Forecasting: The Next Generation of Statistical Model Post-Processing (Invited Presentation)

In the past 15 years, operational weather forecasting advancements in the post-processing of numerical weather models have significantly improved the weather forecasts provided to the public. The development of consensus forecasting techniques has probably led to the largest reduction in forecast errors since the advent of Model Output Statistics (MOS) in the 1960s. The Dynamic Integrated ForeCast System (DICast), developed in the late 1990s at the National Center for Atmospheric Research (NCAR), is a completely automated consensus forecast system that was modeled on the human forecast process. It considers multiple forecast inputs and continually compares its forecasts to observations in a machine learning approach that generates objective forecasts that outperform its ingredient forecasts, including NWS and ECMWF products. DICast now drives the forecast engines of several of the largest weather forecast providers in the United States and, to an increasing degree, internationally.