The subsurface ocean dynamics of Indian Ocean Dipole (IOD) is identified using regular and extended empirical orthogonal function (EOF and EEOF) analyses of the TOPEX/Poseidon altimeter sea-level anomaly and expendable bathy-thermograph (XBT) temperature data. An IOD event is initiated by an anomalous upwelling along the SumatraJava coast at the start of the normal upwelling season in May-June. This enhances cooling of sea-surface temperature (SST) in the eastern Indian Ocean, which couples with a westward wind anomaly along the equator and drives rapid growth of IOD. The wind anomaly and associated Ekman pumping generate off-equatorial Rossby waves that travel westward, deepen the thermocline, and warm SST in the western Indian Ocean, causing the peak of an IOD event a few months after it begins. The decay of an IOD event is characterized by a slow eastward propagation of warm anomaly along the equator, with warm SST leading deepened thermocline depth. The deepened thermocline arrives at the eastern boundary and reduces the rate of cooling during the next upwelling season. This causes a positive SST anomaly in the eastern Indian Ocean in the following year of an IOD event. The evolution of the event during two upwelling seasons involves the SST, wind, and subsurface temperature. The vertical and horizontal subsurface thermal structure of IOD is described in detail. There are upward propagations of temperature anomaly from subsurface to surface layer in the XBT EEOF analysis, which implies the potential role of ocean dynamics on the SST. From correlation analyses, the subsurface evolution in the tropical Indian Ocean (indexed by the zonal equatorial wind) is strongly affected by Indian Ocean SST zonal gradient, and to a lesser extent by El Nin˜o Southern Oscillation (ENSO). The observed behavior of the tropical Indian Ocean and the role of internal ocean dynamics suggest a coupled ocean/atmosphere instability, which may be initiated by ENSO or other anomalies during the early Sumatra-Java upwelling season; however, proof of its existence will require further research, including modelling and model-validation with these observations. r