We will analyze a seven year archive of MODIS data in order to develop, test, and implement a new method for the early on-orbit detection of thermal unrest at all of Earth's ~1500 active and potentially active volcanoes. The results will be used to i) improve inventories of eruptive and non-eruptive volcanic heat fluxes (allowing us to better constrain volcanic magma budgets and models of volcanic "plumbing"systems), ii) to identify low temperature thermal precursors to large basaltic lava flow-forming eruptions, and iii) to establish the significance of short term (i.e. day to month) variations in background levels of volcanic power output as precursors to significant escalations of eruption intensity (i.e. transition from "normal"eruptive conditions to a more explosive eruption or a voluminous lava flow-forming eruption). The results will also of great significance to hazard managers and civil defense authorities as the increased sensitivity of the proposed algorithm to low temperature volcanic phenomena (including fumarolic activity and other hydrothermal processes) will allow the onset of volcanic activity at previously dormant volcanoes to be detected at an earlier stage.
The method we propose will use time-series analysis of co-registered MODIS images (Pergola et al., 2004) to detect the thermal "anomalies" associated with volcanic activity. This method uses such a time-series to establish the "normal" thermal "behavior" (and the natural range of variability) of all pixels contained within each image. In other words, we will characterize, in a statistically robust manner, the thermal characteristics of the ground contained within each 1 km MODIS pixel. Once this baseline has been established, the pixels within each subsequently acquired image are analyzed to determine whether they are thermally "similar" or "different" to how they have appeared in the past. Pixels that appear significantly more radiant than normal are flagged as hot-spots. Because the method uses time-series analysis it minimizes sources of natural variability that constitute noise in the detection process, thus lowering its detection threshold internally to match the natural level of scene variability. It will also run as a single-band point operation. Thus, the algorithm is computationally very simple. We anticipate that the algorithm will supersede the highly successful MODVOLC algorithm (Wright et al., 2002) that currently runs as part of EOSDIS Core System PGE-03 at NASA's Goddard Space Flight Center.
Previous accomplishments by the PI in this area include the development and successful implementation of the MODVOLC algorithm (Wright et al., 2002). The PI is also a member of the JPL's Volcano Sensor-Web team, which uses the MODVOLC system as a driver for rapid on-orbit re-tasking of the EO-1 Hyperion sensor in response to volcanic eruptions (Davies et al., 2006).
