In an effort to tackle with the climate crisis, since 2014, CO2 and H2S emissions from the Hellisheiði Geothermal Power Plant at Iceland are injected and stored underground as part of the CarbFix project. The CarbFix project has proven to be efficient in reducing emissions of CO2 and H2S by turning these gases into stable minerals through reactions with basaltic subsurface rocks. Previous studies related to the CarbFix project were mainly focused on analyses of the chemincal reaction and flow paths of the injected fluid. In this study, we monitor the subsurface structure down to the geothermal reservoir utilizing the ~2-year-long continuous recordings, starting from Oct 2018, of six seismic stations deployed near the geothermal power plant. To monitor the subsurface seismic wave velocity changes, we first compute autocorrelation function (ACF) for every 10 minutes of recording. Clear reflected waves are observed in ACFs of all stations with a two-way travel time of ~2 s and ~3 s in the period band of 0.4-0.8 s and 1-2 s, respectively. We then estimate the temporal pattern of relative velocity changes (dv/v) using the reflected waved identified in the ACFs for each station and frequency band. The resolved dv/v curve yields a complex pattern with magnitudes of both short-term (a few days) and long-term (monthly to annual) changes up to ~5%, which also varies with frequency band and station location. Analysis of dv/v using coda waves in ACFs and ambient noise cross-correlation functions, and the dominating mechanism for the resulting temporal patterns of dv/v will be reported in our presentation.