Investigating a New Stochastic Physics Suite for Use in the NCEP Global Ensemble

Numerical weather ensembles based on initial condition variation alone are inherently under-dispersive; in order to increase ensemble spread towards the desired value ensemble systems must include additional components that represent the uncertainty in the model processes. One group of methods for including this uncertainty is stochastic methods, which perturb ensemble members during the model integration via a random process. The current generation of the NCEP global ensemble forecast system (GEFS) uses a process known as stochastic total tendency perturbation, where each member's tendency is perturbed by a random combination of the other members' tendencies. While this has proven moderately successful at increasing spread, there are both scientific and technical reasons a replacement is desired. NCEP is currently evaluating a new stochastic physics suite that combines stochastically perturbed physics tendencies (SPPT), stochastic humidity perturbation (SHUM), and stochastic kinetic energy backscatter (SKEB) to replace the current STTP. Here we present preliminary results from these tests for a number of meteorological variables including 500-hPa height, temperature, wind, and precipitation.

Preliminary results show that the new stochastic suite considerably improves spread-skill ratio in the tropics and summer hemisphere, and for some fields improved the skill as well. CONUS precipitation forecasts are also substantially improved in the summer. However, results for the winter are more mixed. There is also still underspread for week two forecasts of some fields, particularly 500-hPa height. This week two underspread will be discussed in a separate presentation.