This work focuses on dynamic sustainable system optimum (SSO). When carrying out dynamic SSO assignment, instead of minimizing their own travel cost of each user, we aim to define an SSO status to minimize the on-road traffic emissions of the whole system. The method is based on an system-optimization program. This work proposes a simple, intuitive, simulation-based methodological framework for solving the SSO problem. The main contribution of this paper is to propose a statistically reliable learning process of link marginals that permits to derive link performance function from any DTA simulator. Then the obtained functions can be used to compute network equilibrium including SSO. A test case is carried out on a toy network. Numerical results show that with the help of the proposed method, we can reduce 5.79% of carbon monoxide (CO) emission on the whole system under SSO equilibrium, when compared with UE.