Scientists at the five U.S. Volcano Observatories conduct geologic investigations and real-time monitoring of volcanoes to address these questions. Longer-term forecasts, or hazard assessments, are commonly based upon an individual volcano's geologic past. Scientists assess the frequency, magnitude, and style of eruptive events by cataloguing the sizes and ages of well characterized past eruptions. On the shorter-term, real-time monitoring can provide scientists with data necessary to forecast possible style and timing of an eruption within a period of volcanic unrest.
Key to success in short-term volcanic eruption forecasting is being able to recognize when a volcano is moving away from its background level of activity. To do so, volcanologists must collect volcano-monitoring data during times of quiescence. With reference to these background measurements, scientists are better able to interpret changes that are caused by magma movement or pressurization, including shifts in seismicity, appearance of ground deformation, and change in character or rate of gas emissions. Through global volcanic database efforts, comparisons are also made between patterns of change in monitoring data with those during unrest periods from other volcanoes around the world.
In some cases, physical and numerical modeling of volcanic eruption processes helps scientists to forecast eruption processes and impacts. Such models have included distribution and thickness of ash clouds, pryoclastic flows, lahars, and lava flows.
Eruption forecasting is not limited to estimating the timing of eruption onset—once an eruption begins, activity can last a few days or years and the style of volcanic eruption can change over this interval. Furthermore, because every volcano erupts in multiple different styles, any forecast necessarily includes a range of possible outcomes, some more likely than others. To address this range of possible behaviors, volcanologists employ probabilistic methods of event forecasting, which incorporate the uncertainty in forecast estimates.
An approach to displaying probable unrest scenarios is called volcanic event trees, where each downstream branch in the tree describes a different unrest scenario, including difference in event size and impact area. The scenarios are continually updated as unrest unfolds or after an eruption occurs.