The Sakurajima Volcano has experienced repeated Vulcanian eruptions since 1955 for a period of 69 years. We propose a forecasting system for volcanic ashfall prior to the onset of eruptions based on monitoring ground deformation. Inflationary strain changes are detected by strainmeters in an underground tunnel prior to eruptions. Statistical features of the ground deformation are represented by a log-logistic frequency distribution of the duration time from the onset of inflation to the onset of eruption, and the deflation volume ratio to precursory inflation. This indicates that the onset time and scale of eruption can be forecasted probabilistically. In the forecasting system, FALL3D is operated as the simulation engine, with a time interval of six minutes for 20 scenarios (four cases for onset time and five cases for scale) with probabilities given by the log-logistic distribution.