A model based technique for echo shift-based ultrasound temperature estimation is presented. First, the dynamic model is derived from direct discretization of the Pennes' bio-heat equation. This model is then used in a Kalman filter setting for temperature tracking using the incremental temperature changes acquired at each discrete grid point as the measurements. In addition to tissue heterogeneity, natural motions and deformations during in vivo ultrasound thermography play a significant role in degrading the performance of the ultrasound thermography. Using the model based approach, we present the results of temperature estimation during sub-therapeutic high intensity focus ultrasound (HIFU) shots in the hind limb of Copenhagen rats in vivo. The results show continuous tracking of the temperature via pure prediction during significant error cycles or displacement tracking failure. These results represent an early validation of a fully adaptive spatial-temporal filtering of thermography data to compensate for tissue motions and deformations in vivo.