The development of a reliable structural damage prognostics framework, which can accurately predict the fatigue life of critical metallic components subjected to a variety of in-service loading conditions, is important for many engineering applications. In this paper a hybrid damage localization method is developed for prediction of cracks in aluminum components. The proposed fully probabilistic methodology combines a physics based prognosis model with a data driven localization approach to estimate the crack growth. Particle filtering is used to iteratively combine the predicted crack location from prognostic model with the estimated crack location from localization algorithm to probabilistically estimate the crack location at each time instant, while accounting for the uncertainties. At each time step, the crack location predicted by the prognosis model is used as a priori knowledge (dynamic prior) and combined with the likelihood function of the localization algorithm for accurate crack location estimation. For improving the robustness of the localization framework, temperature compensation is carried out. The model is validated using experimental data obtained from fatigue tests preformed on an Al2024-T351 lug joint at different temperatures. The results indicate that the proposed method is capable of estimating the crack length with an error of less than 1mm for the majority of the presented cases.