A number of high-priority Earth science investigations benefit from rapid, reactive observations in multiple spectral bands and across multiple spatial scales. Examples include thermal detection of volcanic activity and wildfires followed by visible-light follow-up observations of the plume; joint tracking of flooding events with SAR and visible-light imaging; and detection of atmospheric events with full-disc, low-resolution imaging triggering narrow, high-resolution follow-ups.
In recent years, the LEO economy has seen a proliferation of commercial providers with dense constellations offering data-as-a-service, with high spatial and temporal resolution in visible light, hyperspectral bands, and SAR. However, no single provider offers all of these observing modes in a single constellation.
The convergence of these two trends offers a unique opportunity to perform multi-spectral, reactive science campaigns using "virtual constellations" overlaying existing commercial offerings. To realize this vision, autonomous brokers are needed to translate high-level science workflows into observation requests to individual providers. These brokers should adapt to the commercial providers' real-time availability (without assuming knowledge of the providers' full schedule); maintain custody of the workflows, replanning in response to new observations as well as changes in the commercial operators' availability; and do so significantly faster than current ground-based, human-in-the-loop approaches, interfacing directly with commercial providers' scheduling systems on the ground and on individual satellites.
In this talk, we will present the initial implementation of one such broker, developed as part of JPL's Federated Autonomous MEasurements (FAME) project. Simulation results suggest that the broker is able to effectively perform reactive observations of fast-changing phenomena such as volcanic eruptions and flooding across multiple constellations and spectral bands, effectively executing a high-level scientific workflow across a virtual constellation, and vastly outperforming human-in-the-loop scheduling. Initial deployment on satellites belonging to FAME's federated entities is planned for 2026.