Forecasting volcanic activity onset and eruption with the next generation of thermal infrared data

Thermal infrared (TIR) imaging of volcanic activity has become common over the past quarter century with the advent of smaller, inexpensive, ground-based cameras and greatly expanded orbital coverage. Because of these advances, TIR data are also now integrated into the standard set of monitoring tools at many volcano observatories. These data are acquired using permanent ground-stations, less frequent campaign mode deployments from the ground and air, as well as orbital remote sensing. However, the ability to forecast a new eruption using orbital TIR data remains unrealized despite decades of data acquisition, modeling, and analysis. Fundamentally, these data are limited due to the design metrics of the sensors such as spatial and/or temporal resolution. One endmember group of these instruments is defined by lower spatial, higher temporal resolution whose data can detect large-scale thermal change such as new lava on the surface. Sensors in this class are used to rapidly identify a new eruption and monitor its evolution, for example. The other endmember has sensors with higher spatial, lower temporal resolution data with sensitivity to detect subtle temperature changes (1-2 degrees) over small spatial scales. Our work examines decades of TIR data from this second endmember class to identify precursory thermal eruption signals. By including all data (day and night) screened for clouds, we produce a larger statistical dataset from which to extract thermal signal deviations from a standard baseline. This long time series orbital TIR data enable a unique opportunity to quantify low-level anomalies and small eruption plumes over long periods. Most significant is the finding that the smaller, subtle detections served as precursory signals in ~81% of eruptions for our five test locations, which we have now expanded to a wider range of volcanoes and activity styles. The results also serve as training for machine learning based modeling that is applied to different targets for this study. This model learns to identify discriminant thermal trends associated to unrest conditions preceding eruptions.  Over the next decade, several high spatial (~ 60 m) resolution orbital sensors are planned will  provide near-daily TIR data at every volcano, vastly improving thermal baselines and detection of new activity. One of these, the Surface Biology and Geology (SBG) TIR mission, contains an infrared instrument and a planned higher-level data product called the Volcano Activity (VA), which will be crucial for accurate daily monitoring of volcanic temperatures and degassing rates. However, despite the promise of SBG data, the next fundamental step-change in orbital volcanology will not come until high-speed, spaceborne data are possible. A proposed “hypertemporal” TIR mission would acquire these data at sub-minute scales to determine mass and thermal flux rates of gas emissions, eruptive ash plumes, and lava flows. With such a mission, data now acquired by current ground-based cameras will become possible from orbit for the first time.