A long-lived tropical cyclone that occurred on North West Pacific called In-fa has been studied. Two kinds of data, one is ERA-5, another is FNL are used to produce the initial values for model integration. The ERA-5 data is taken as reference, the difference between FNL and ERA-5 is taken as perturbation. By decomposing the perturbation into different kind of scales, we reveal the impacts of different-scale initial perturbations on the forecasts of tropical cyclone. The results indicate that, whether it is intensity forecasting or track forecasting, the impact of mesoscale errors is most significant in the first ten days of forecasting. After adding mesoscale errors, the typhoon track tends to add up all the errors, and the intensity also tends to add up all the errors, indicating that mesoscale errors played a decisive role in the forecast for the first ten days. In the next four days of forecasting, the contributions of large-scale and small-scale errors gradually become prominent, and the effects of the three scale errors are equivalent. In addition, mesoscale errors in the early stage caused the typhoon track to be eastward and stronger, while large and small scale errors caused the typhoon track to be westward and weaker. The analysis of the reasons shows that the mesoscale error field has a higher similarity with the entire error field, that is, the information of the entire error field is mainly reflected in the mesoscale error, so the role of mesoscale error in the early stage is significant. In addition, the superimposed mesoscale error field has a stronger initial strength for the background wind, and the eastward track results in a later landing time. Therefore, mesoscale error leads to a stronger typhoon intensity. However, due to the large and small scale errors causing the typhoon track to be westward and make landfall earlier, the overall strength of the typhoon is weaker.