An Adaptive Four-Dimensional Incremental Analysis Update Method for the GSI Based Hybrid 4DEnVar Data Assimilation System to Improve HWRF Hurricane Prediction

Spin-down has been a known issue for HWRF intensity forecast when assimilating the inner-core observations. The spin-down is a significant Vmax drop (greater than 5m/s per 6 hours), which jeopardizes the benefit of the more realistic analyses produced through inner-core DA. With the implementation of the advanced, continuously cycled hybrid DA system for HWRF, the spin down issue is alleviated, but not totally cured (Lu et al. 2017). Our previous work (Lu and Wang 2017) suggested that the spin-down issue with HWRF inner-core DA is likely to be attributed to the mismatch between the HWRF model and the analyses produced from the inner-core DA. Incremental analysis update (IAU) method is therefore implemented and explored in this study.

In the traditional 4DIAU method, increments are pre-determined by subtracting the background forecasts from the 4DEnVar analyses. Then the pre-determined increments are gradually added to the background forecast at each time step during the model integration. These pre-determined increments require linear or near linear evolutions of the weather systems during the 6-hour time window. However, the linear evolution assumptions are usually not valid for the rapid-changing situations like RI or eyewall replacement in hurricanes. Therefore, an adaptive 4DIAU method is proposed in this study to adapt to the rapid, non-linear evolution of hurricane over the DA window. In this adaptive 4DIAU method, the increment at each analysis interval (e.g. 1 hour) is re-calculated by subtracting the 1-hour adaptive background forecast from 4DEnVar analyses. The re-calculated increment will then be added to the adaptive background at each time step during the model integration until the end of this hour. A 1-hour free forecast will be launched to provide adaptive background forecast for the next hour.

Experiments with hurricane Patricia (2015) showed that the traditional 4DIAU method is able to significantly alleviate the spin-down issue in the intensity forecast with reduced model imbalances. However, large position errors are also found in the traditional 4DIAU analyses due to the use of pre-determined increments. The newly proposed adaptive 4DIAU method is found to further improve track forecast while maintaining intensity forecast improvement from traditional 4DIAU. These findings together with further diagnostics will be presented at the conference.