In this study, the impact of dropwindsonde observations on track forecast of Hurricane Bonnie (1998) is investigated in terms of the number and distribution of dropwindsondes around the storm center using the MM5 model and the observing-systems simulation experiments (OSSEs) approach. The control and sensitivity simulations are initialized with the NCEP and TOGA analyses, respectively, which give rise to two quite different tracks at the end of 5-day integrations. To avoid the "identitcal twin" problem, we treat the analysis with the best simulation as the "ground truth," and add different data from this "truth" to another analysis to see the impact of different variables and configurations of "observations" on the simulated tracks. Our results show that increasing the number of dropwindsonde observations around the storm center improves track forecast; however, excessively dense data may not give any further improvement. In addition, enhanced observations in any semicircle in the vicinity of storm center results in less improvement than those distributed with uniform distance from the storm center.
In the context of data assimilation, our results imply that proper observational network (i.e., distribution) and density (i.e., number) is important in improving the track forecast of tropical cyclones. Therefore, the flight routes for deploying dropwindsondes should be carefully planned. The results also address the importance of conducting adaptive observations in finding more effectively where and how one should deploy dropwindsondes and enhance other observing systems.
Supplementary URL: