Control room applications for multi-areas systems, e.g. oscillations detection, are dependent on models and measurements of neighbouring areas. Given inherent limitations in state-of-the-art data-sharing architectures, such dependencies can suffer from model imperfections and low observability. This paper presents a study on the use of the Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) for centralised estimation of rotor angles, in support of applications such as oscillation detection. The study focused on the Kalman filter variants ability to overcome low observability and model imperfections, implemented on the Kundur two-area four-machine network and the Nordic-44 bus network. To study the effect of low observability and model imperfection limitations, the study included three different data sharing architectures as proposed by ENTSO-E. The simulated test cases demonstrated that the EKF is unable to capture the dynamics under low observability conditions during model imperfections, while UKF captures the dynamics for all presented cases.