Abstract: Limited by the inherent density and variability of radar-extended targets, traditional methods often suffer from reduced accuracy in observation partitioning and matching under dense clutter and environmental noise. Additionally, the spatiotemporal correlations among uncertainty sources such as observation, clutter, and interference are difficult to quantify effectively. To address these challenges, this paper proposes a nonlinear state observation data uncertainty prediction modeling method based on Dynamic Bayesian Network (DBN), which effectively quantifies the uncertainty in nonlinear state observation data of extended targets while fully uncovering the underlying patterns in the observation data. Comparative experiments demonstrate the effectiveness of the proposed method. Further analysis shows that our approach enables flexible applicability in multi-task uncertainty processing for radar-extended target observation, supporting dynamic problem analysis such as target state estimation, track maintenance, and radar resource scheduling.