Uncertainty-aware ground-based telescope observation scheduling at ALMA

Abstract

Automated scheduling tools for ground-based observatories must balance immediate observing conditions with long-term scientific priorities and global scheduling constraints. Existing dynamic schedulers often fail to enforce such global constraints, and typically do not account for uncertainty in forecasts of environmental or operational availability, thereby exhibiting suboptimal performance. We address these limitations with PULSAR: Predictive Uncertainty-aware Lookahead Scheduling with Adaptive Rebalancing, an automated scheduling algorithm for dynamic telescope operations.

PULSAR dynamically schedules observations, accounting for observatory priorities, global scheduling constraints, and uncertainty in forecasts. PULSAR introduces two key improvements:

• First, a dynamic weighting scheme that progressively increases penalties for global-constraint violations;
• Second, an anticipatory lookahead algorithm that plans over horizons of hours to days accounting for uncertainty in forecasts of weather and equipment availability. To our knowledge, PULSAR is the first dynamic telescope scheduling approach that includes these two factors.

We evaluate the effectiveness of PULSAR using simulations of the 2023-2024 observing cycle at the Atacama Large Millimeter/submillimeter Array (ALMA). Our evaluations reveal that while a simulation of ALMA’s current dynamic scheduling algorithm violates constraints by 2.2-7.1%, PULSAR satisfies global constraints within a tolerance of just 0.1%. PULSAR also improves satisfaction of scientific goals by completing approximately 100 additional high-priority observations and 5 additional high-priority projects relative to ALMA’s current scheduler. PULSAR is feasible for deployment, computing decisions in approximately 30 seconds on average.

All code for PULSAR is available at https://github.com/justinpayan/PULSAR.