Continuum: A New Observing Strategies-Based Framework for Hyper-Local Situational Awareness

Abstract:

Cyclone Amphan recently caused significant destruction in Bangladesh, which was already reeling from the COVID-19 pandemic. Fortunately for Bangladesh, these compounding conditions were not further aggravated by an attack of locust swarms, which cannot be said of many East African nations. Extreme events such as hurricanes, coastal storms, and intense inland precipitation events and accompanying floods are on the rise, and increasingly overlapping with existing stressors. The need for global situational awareness to monitor and track these events is well established. Satellite constellations are providing terabytes of Earth observations data, enabling a lot of those functionalities. But rather than inundating disaster responders with data, there is value to providing them with actionable information instead. The response is often managed by the local community, and thus, that information needs to be locally relevant and applicable. Continuum is a framework that features a swarming-based new observing strategy for satellites, deep learning-based damage assessment, and cloud characterization, developed to provide hyper-local situational awareness about disaster events -- down to the zip code-level.

Hurricane Matthew (Haiti; 2016) is presented as a use case to demonstrate Continuum’s applicability. A JHU/APL-developed deep learning algorithm is deployed on a representative space flight architecture to classify building damages using post-hurricane imagery. JHU/APL’s Compact Midwave Imaging System (CMIS) is an instrument being developed to measure cloud brightness temperatures, cloud top heights, and cloud motion vectors. In a software-in-the-loop (SWIL) environment, CMIS-equipped satellites work in conjunction with damage assessment-capable satellites to form a heterogeneous network. Techniques from the multirobot swarming community are used to tip and cue satellites for dynamic tasking and re-tasking. The presented use case further demonstrates how inputs, such as historical cloud products (VIIRS), winds (UCAR), precipitation (GPM), population density and other socio-economic factors (SEDAC), and geospatial disease risk estimations (University of Rhode Island), are ingested into the framework and intelligently selected at appropriate times for effective decision support.