Abstract <p>Detection of subtle precursors to forecast eruptive events is one of the main issues in volcanic risk mitigation. Seismic interferometry of ambient noise allows to detect small changes in the medium, but is affected by numerous processes such as strain and meteorological variations that need to be identified to correctly interpret the observation for operational monitoring. To evaluate strain and meteorological effects on seismic velocity changes, we computed 10 years of daily velocity changes in the 0.5–4 Hz frequency range and analysed them with strain and meteorological changes at 21 Japanese volcanoes. We jointly inverted the respective contribution of strain, rainfall, snow load, temperature, atmospheric pressure, wind speed and sea level variations to the observed velocity change using a least-squares method combined to a Monte Carlo approach. Over the 21 volcanoes, we reduce the root mean square of the velocity changes by 15–19% on average. We determine that areal strain, rainfall, temperature and sea level are the parameters affecting the velocity the most, representing on <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\approx\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>≈</mo> </math></EquationSource> </InlineEquation>80% of the modelled variations. However, their effect is very variable from volcano to volcano, demonstrating the need for case by case analysis. Thanks to this analysis we unravel potential precursory velocity changes days to months before six eruptions that occurred at Asamayama, Kuchinoerabujima, Ontakesan and Satsuma–Iojima volcanoes. For four of these eruptions, velocity changes are the only precursor reported, highlighting the promising potential of the method for volcano monitoring. However, more physics-based model and collocated seismic and meteorological sensors are needed for an higher correction accuracy.</p> Graphical Abstract <p></p>