We processed 79 days of GPS data collected at four stations on Miyakejima volcano (Japan) before and during its eruption in 2000. This period includes a strong seismic swarm (27 June to 8 July) followed by five eruptions (8 and 14--15 July, 10, 18, and 29 August). Surface motion is strong during the seismic swarm, then remains almost steady during eruptions. However, during each eruption, we found anomalous values of the LC (ionosphere free) double differences, especially on 18 August, with a peak of 0.6 cycles (i.e., an apparent 290 mm increase of the satellite-station range). We cannot explain these anomalies by temporary ground displacements during the events. Instead, we propose that they are due to path delay effects in the hot volcanic plume. By adapting a seismic tomography algorithm, the refractivity anomaly and the inferred temperature in the plume is mapped in time and space.