The energy of the seismic coda generated by multiple scattering at small scale heterogeneities in the Earth is generally uniformly distributed in space at some late lapse time. 3-component seismogram envelopes of volcano tectonic B-type earthquakes recorded at Merapi volcano, on the contrary, show a systematic decrease of the coda amplitude with distance from the summit of the volcano. We present two different models to explain this observation. In the first model seismic energy transport is described as a diffusion process and the observed coda localization is explained by an inhomogeneous distribution of the diffusion coefficient (D = D(r)). As a simple analytical model we use a strongly scattering cylinder (representing the volcano) embedded in a homogeneous half-space (representing the surrounding crust). In the second model we assume the disorder within the strato volcano to be so large that diffusion theory breaks down and a transition to the so called 'Anderson localization regime' occurs. This model predicts a concentration of seismic energy near the source region and can thus also explain the observed coda localization.