Impact of Parameter Variations on Tropical Forecasts in a Global Atmospheric Ensemble

There are several complementary methods to account for model error and uncertainty in ensemble forecasts. The one examined here is varying parameters within the parameterizations of sub-grid-scale processes. We examine the impact of these parameter variations on the performance of the Navy Operational Global Atmospheric Prediction System ensemble forecasts using several metrics. The first set of variations occurs within the parameterization of deep convection only. The second set has variations in both convection and boundary layer parameterizations. It is shown that there is significant spatial variability in the impact of the parameter variations, with little impact in the extra-tropics, but significant improvements in performance in the tropics. In particular, parameter variations significantly increase ensemble spread and reduce extraneous rank histogram outliers in tropical wind and temperature fields. These improvements are obtainable even when time-mean forecast errors are removed, indicating that the improvements are not simply due to introducing a wider range of biases amongst ensemble members with different parameter values. Brier scores for low-level wind speed and temperature are also improved. While improvements to the Brier scores are expected when increasing the spread of an under-dispersive ensemble through improvements to the reliability, in this case it is found that the improvements to the Brier scores are obtained primarily through improvements to the resolution. This indicates that the parameter variations are producing improvements in the flow-dependent estimates of forecast uncertainty and error. There are also small but significant improvements in the ensemble mean tropical cyclone track forecasts. In general, it is found that the second set of variations, with changes in both the deep convection and boundary layer parameterizations, outperforms the first set of variations in the parameterization of deep convection only. The impact of parameter variations on the forecasts of the Madden Julian Oscillation will also be examined and presented.