In this study, a series of sensitivity experiments were performed for two tropical cyclones (TCs), TC Longwang (2005) and TC Sinlaku (2008), to explore the roles of locations and patterns of initial errors in uncertainties of TC forecasts. Specifically, three types of initial errors were generated and three types of sensitive areas were determined using conditional nonlinear optimal perturbation (CNOP), first singular vector (FSV), and composite singular vector (CSV) methods. Additionally, random initial errors in randomly selected areas were considered. Based on these four types of initial errors and areas, we designed and performed 16 experiments to investigate the impacts of locations and patterns of initial errors on the nonlinear developments of the errors, and to determine which type of initial errors and areas has the greatest impact on TC forecasts.

Overall, results from the experiments indicate the following: (1) The impact of random errors introduced into the sensitive areas was greater than that of errors themselves fixed in the randomly selected areas. From the perspective of statistical analysis, and by comparison, the impact of random errors introduced into the CNOP target area was greatest. (2) The initial errors with CNOP, CSV, or FSV patterns were likely to grow faster than random errors. (3) The initial errors with CNOP patterns in the CNOP target areas had the greatest impacts on the final verification forecasts.