This paper examines the current challenges of using Lamb wave interrogation methods to localize fatigue crack damage in a complex metallic structural component in the presence of temperature variations. The goal of this research is to improve damage localization results for a structural component interrogated at an unknown temperature by developing a probabilistic and reference-free framework for estimating Lamb wave velocities. The proposed approach for temperature-independent damage localization involves a model that can describe the change in Lamb wave velocities with temperature, the use of advanced time-frequency based signal processing for damage feature extraction, estimation of the actual Lamb wave velocities from transducer signals, and a Bayesian damage localization framework with data association and sensor fusion. The technique does not require any additional transducers on a component and allows the estimation of the velocities for the actual Lamb waves present in a component. Experiments to validate the proposed method were conducted using an aluminum lug joint interrogated with piezoelectric transducers for a range of temperatures and fatigue crack lengths. Experimental results show the advantages of using a velocity estimation algorithm to improve damage localization for a component interrogated at both known and unknown temperatures.