Optimal Multi-Satellite Planning for Collecting and Downlinking Data with Limited Storage, for Wildfire Danger Monitoring

We present a novel Mixed Integer Linear Program formulation which produces optimal plans for a challenging climate science problem: wildfire danger monitoring. We are motivated by the need to solve a real-world mystery: What is the optimal plan for collecting wildfire-related sensor data using NASA’s currently active CYGNSS satellite constellation?

Our planner generates coordinated plans for data collection and downlinking, for multiple satellites with limited storage capacity and energy constraints. Each potential observation target is associated with a science reward based on a wildfire probability index produced by the U.S. Geological Society. The planner maximizes the aggregate rewards collected for all observed targets on all satellites.

We present a novel interval-based abstraction which eliminates time-indexed variables and enables the first optimal solutions to this problem to be found. We compare optimal MILP results vs. suboptimal baselines produced using Monte Carlo Tree Search. We present experiments producing optimal 24-hr plans for a constellation of 8 satellites, which capture 99% of the∼23,000 available targets