Tests of various schemes for representing model uncertainty in the GFS

Jeff Whitaker


Three different methods for treating model uncertainty in ensemble prediction have been tested in the GFS. They are stochastically-perturbed physics tendencies (SPPT, the operational ECMWF scheme), vorticity confinement (VC, a technique using in computer animation, but also being tested at UKMET) and stochastically-perturbed boundary layer humidity (SHUM). 20 member ensembles were run for the month of July 2012, with each of these schemes individually and all of them together. The results are compared to a control ensemble with no treatment of model uncertainty, and to the operational GEFS scheme (stochastic total-tendency perturbations, STTP). It is found that VC and STTP mainly increase spread in middle-latitudes, while STTP and SHUM are effective at increasing spread in the tropics. SPPT and SHUM are complementary, since SPPT modulates the physics tendencies where the physics is already active, while SHUM can activate or deactivate convection by perturbing the trigger function in the convection scheme. However, SHUM introduces a warm (dry) bias in the tropical upper (lower) troposphere. SPPT and SHUM have a significant impact on tropical cyclone track spread, while VC and STTP do not.