Calderas represent the sites of the world's most serious volcanic hazards. Although eruptions are not frequent at such structures on the scale of human lifetimes, there are nevertheless often physical changes at calderas that are measurable over periods of years or decades. Such calderas are said to be in a state of unrest, and it is by studying the nature of this unrest that we may begin to understand the dynamics of eruption precursors. Here we review combined gravity and elevation data from several restless calderas, and present new data on their characteristic signatures during periods of inflation and deflation. We find that unless the Bouguer gravity anomaly at a caldera is extremely small, the free-air gradient used to correct gravity data for observed elevation changes must be the measured or calculated gradient, and not the theoretical gradient, use of which may introduce significant errors. In general, there are two models that fit most of the available data. The first involves a Mogi-type point source, and the second is a Bouguer-type infinite horizontal plane source. The density of the deforming material (usually a magma chamber) is calculated from the gravity and ground deformation data, and the best fitting model is, to a first approximation, the one producing the most realistic density. No realistic density is obtained where there are real density changes, or where the data do not fit the point source or slab model. We find that a point source model fits most of the available data, and that most data are for periods of caldera inflation. The limited examples of deflation from large silicic calderas indicate that the amount of mass loss, or magma drainage, is usually much less than the mass gain during the preceding magma intrusion. In contrast, deflationary events at basaltic calderas formed in extensional tectonic environments are associated with more significant mass loss as magma is injected into the associated fissure swarms.
