We model the shear wave velocity structure of the Hales-discontinuity beneath the Eastern Dharwar Craton and Southern Granulite Terrain in Southern India using P-wave receiver function (P-RF) analysis and joint inversion with Rayleigh wave phase velocity dispersion. For this study we use data from seismological stations HYB, GBA and KOD. We isolated P-RFs where the Hales phase is distinct and model them using joint data analysis. We also use common conversion point stack profiles constructed by depth migrating P-RFs through the velocity model and show that the Hales discontinuity is undulatory in nature. We perform petrological modeling of the Hales discontinuity and fianlly propose a geodynamic model for its evolution. 

Seismological analysis we use broadband teleseismic data from the Geoscope station Hyderabad (HYB) and the Indian Institute of Astrophysics & University of Cambridge temporary network stations Gauribidanur (GBA) and Kodaikanal (KOD). Stations HYB had been operated for a period of 18 years (1989-2007). Broadband data for HYB was downloaded from the Data Management Center of the Incorporated Research Institutions in Seismology (IRIS-DMC). GBA was operated from 1998 to 2001, and KOD was operated from 2002 to 2006. Both stations were equipped with Guralp CMG-3T seismometer (120~s--50~Hz flat velocity response), with Reftek data logger at GBA and Guralp Data and Communication Module (CMG-DCM) at KOD. The station location and time stamp to the recorded data were provided by Global Positioning System (GPS) receiver connected to the seismometer.

In this study we use teleseismic earthquakes with magnitude greater than 5.5, in the distance range of 30 to 90 degree to model the sub-crustal lithospheric mantle structure. A total of 604, 52 and 44 seismograms from HYB, GBA and KOD, respectively, have been used for P-wave receiver function (P-RF) analysis.  

The three component broadband waveforms are pre-processed by rotating the horizontal components (North-South and East-West) into radial and tangential directions using 
the back-azimuth angle. The vertical component waveform is deconvolved from the radial and tangential components to remove source and propagation effects, except for those immediately beneath the station. This deconvolved radial waveform contains the effects of the structure beneath the receiver site and is termed the P-wave receiver function. We use the iterative time domain deconvolution technique where the P-RF is created using a spike train, constructed by cross-correlating the radial with the vertical component waveform. This spike train is convolved with the observed vertical component waveform to form a synthetic radial component waveform. The synthetic and 
observed radial component waveforms are compared in the least-squares sense and a misfit value calculated. Based on this misfit, the spike train is iteratively updated. Once the misfit value fall below an acceptable limit (chosen as 0.001 in this study) or the maximum number of iterations (set at 200) are completed, the process stops. This iteratively updated best fitting spike train is the computed P-wave radial receiver function. Observed waveforms have high frequency noise, which are eliminated using a low-pass filter. This also assists in stabilizing the time domain deconvolution process. A Gaussian filter of width 2.5 (corner frequency $\sim$0.46~Hz) is used as this low pass filter \citep{Ammon91}. The P-RF obtained through the iterative time domain deconvolution is assessed for quality by the percentage fit to the observed radial waveform. In this study we use P-RFs with fit of 70\% and above. 

This dataset is the best P-RFs obtained from the analysis described above. 

Software needed: Seismic Analysis Codes: http://ds.iris.edu/files/sac-manual/