Computer simulations of the paths of flowing lava are achieved using a program, FLOWFRONT, that describes the behavior of flow and digital models of the terrain. Two methods of application of simulations to the hazards posed by lava flows are described. The first, deterministic, method requires that program parameters such as vent position, minimum flow thickness, and thickness/slope relationship be based on the ambient eruptive conditions so that the future course of a specific lava flow can be simulated. This is illustrated using retrospective modeling of the first 21 days of the eruption of an andesitic lava flow at Lonquimay volcano, Chile, in 1988--1989. The usefulness of this method for real-time predictive modeling is likely to be limited by the lack of accurate field data on flow characteristics, the simple nature of the model, and the sensitivity to parameter choice of the final planimetric form of the model flow. The second application is probabilistic in nature and creates a map of the likelihood of inundation by lave flows that is useful for long-term land use planning. This method uses the historical record of past eruptions to constrain a series of Monte Carlo simulations and is illustrated using data from Etna volcano in Sicily.

A multivariate statistical analysis of nine parameters for the 1763--1989 eruption catalog using simulated annealing permitted a classification of Etna's flank eruptions into two types: A and B. Type A eruptions are short-lived and produce linear lava flows; type B eruptions are long-lived, and produce lava flows that are much broader in shape, and their vents are restricted to the eastern flank of the volcano. The simulation method consists of creating a probability surface of the location of future eruption vents and segmenting the region according to the most likely historical eruption on which to base the simulation. Analysis of the autocorrelation of the historical eruptions shows that type A eruptions are strongly autocorrelated but type B eruptions are only weakly autocorrelated. A library is created of simulation parameters that give reasonable retrospective fits to each of the historical lava flows. Monte Carlo sampling of the simulation space using this library and the above spatial constraints produces a map of probability of lava inundation. ¿ American Geophysical Union 1994