Decision support systems are of paramount importance to propose traffic control strategies and reliable information to road users. They usually take benefits from data-driven methods which represent a reliable way to gather real time traffic information and make predictions for recurring situations. Data Assimilation (DA) consists in considering both observed data and a traffic flow model to monitor and forecast traffic state. Traditionnaly, Kalman filtering methods with macroscopic traffic models have been widely used. More recently, mesoscopic approaches proved to be relevant for large scale network applications. However, errors on traffic state is a key information for DA methods. The paper proposes a DA framework that accounts for two types of error: (i) errors from the data collection system and (ii) errors that are propagated and amplified in time and space by the dynamic traffic flow model. When applied on a basic network, the proposed framework demonstrates its ability to track errors and the traffic state information is enriched with a level of uncertainty. It paves the way for new traffic indicators and new solutions for traffic forecast applications.