[1P5-OS-16] OS-16

[1P5-OS-16-02] Accuracy Analysis of the Model-Bridge Data Assimilation

〇Keisuke Yamazaki¹ (1. AIST)

Keywords:data assimilation, simulation, surrogate model

Though data assimilation is an essential preprocessing for simulations in cyberspace, the high computational cost is required due to the repeated simulation runs. To tackle this issue, we have proposed the model bridge method, which learns the relation between the surrogate model and the simulation based on the past assimilation results and realizes to reduce the cost. However, its accuracy is still unknown, and there is no quantitative knowledge regarding factors for improving the method. This study provides the asymptotic analysis of the accuracy and clarifies the effect of the number of past datasets and the properties of the surrogate model.