Involving human forecasters in numerical prediction systems

Abstract

Human forecasters routinely improve upon the output from numerical weather prediction models and often have keen insight to model biases and shortcomings. This wealth of knowledge about model performance is largely untapped, however, as it is used only at the end point in the forecast process to interpret the model-predicted fields. Yet there is no reason why human forecasters cannot intervene at other earlier times in the numerical weather prediction process, especially when an ensemble forecasting system is in use. Human intervention in ensemble creation may be particularly helpful for rare events, such as severe weather events, that are not predicted well by numerical models. The USA/NOAA SPC/NSSL Spring Program 2003 tested an ensemble generation method in which human forecasters were involved in the ensemble creation process. The forecaster highlighted structures of interest and, using an adjoint model, a set of perturbations were obtained and used to generate a 32-member ensemble. The results show that this experimental ensemble improves upon the operational numerical forecasts of severe weather. The human-generated ensemble is able to provide improved guidance on high-impact weather events, but lacks global dispersion and produces unreliable forecasts for non-hazardous weather events. Further results from an ensemble constructed by combining the operational ensemble perturbations with the human-generated perturbations shows promising skill for the forecast of severe weather while avoiding the problem of limited global dispersion. The value of human beings in the creation of ensembles designed to target specific high- impact weather events is potentially large. Further investigation of the value of forecasters being part of the ensemble creation process is strongly recommended. There remains a lot to learn about how to create ensembles for short-range forecasts of severe weather, and we need to make better use of the skill and experience of human forecasters in this learning process.