The Next Decade of Physics-Based Eruption Forecasting

Abstract:

Over the last decade, interest has grown in the development of eruption forecasts that are based directly on our understanding of the governing physics (National Academies of Sciences, 2017). In a physics-based eruption forecast, the future state of a volcanic system is predicted using a deterministic, physicochemical (physics-based) model. Although analogous approaches have proven highly successful in fields such as weather forecasting, almost all short- to intermediate-term eruption forecasts today are still based on identifying patterns in monitoring data. Current limitations include: imperfect knowledge of the conditions (stress, temperature, composition) within volcanic systems; the heavy computational burden of physically realistic models that may need to explain critical processes operating over a huge range of spatial and temporal scales; the wide variety of eruptive styles (even at a single volcano) which may be governed by very different dominant physical processes; and the inherent nonlinearity of volcanic systems which can result in very different future outcomes due to even small variations in initial conditions.

We expect that important strides will be made over the next decade in several key areas. Increasingly sophisticated physics-based models that are constrained by more diverse observations will improve our ability to resolve conditions within volcanic systems. Mixed deterministic-stochastic models will be constructed to overcome limitations in physical understanding. Emulators (fast surrogates) will be used in place of computationally expensive models and incorporated into probabilistic data assimilation algorithms such as those based on the Kalman filter. The outputs of these algorithms will in turn be incorporated into holistic forecasting frameworks that merge physics-based forecasts with those based on empirical machine learning and other approaches. We anticipate that by 2030 physics-based forecasting techniques will increasingly be honed and validated using data from previous eruptions (hindcasting), and also implemented alongside more traditional forecasting techniques as a test of their reliability and efficacy. This progress will lay the groundwork for the wider use of physics-based eruption forecasts in the decades that follow.