Use of a WRF, MM5 and COAMPS multi-model ensemble to improve coastal weather predictions

A numerical experiment was conducted to develop a methodology to improve weather forecasts in coastal regions where the complexity of the interface induces difficulties in accurate predictions. A 15-day period in December 2008 was simulated using a multi-model ensemble approach using the WRF, MM5, and COAMPS models. All models were set up on the same grid with a 36 km resolution covering the western U.S. To examine the accuracy in using one high-resolution run vs. a coarser resolution ensemble mean, single runs for each model were completed with 12 and 4 km resolutions on a sub-domain. A total of 150 simulations were used to investigate possible advantages of using the multi-model ensemble as compared to the individual models. The ensemble members were selected by combining various setup options, planetary boundary layer parameterizations, microphysical options, radiation schemes, and cumulus parameterizations. In the initial simulation period with a strong frontal passage, the dynamics dominated and the ensemble members did not provide a sufficient spread. However, at later times the spread developed and the ensemble mean showed skill in the predictions. The WRF and MM5 ensemble members compared well with radiosonde measurements in the first period, but were not able to well represent the second frontal passage at the end of the period. The COAMPS members generally showed a warm bias, but at the later stage showed greater total spread growth. It is noticeable that the time growth of the members' spread did not show monotonic characteristics as well as the modeled ensemble errors. A novel approach of incorporating the climate radiosonde data was also introduced to enhance multi-model probability density functions and improve the accuracy of the ensemble prediction system.