The optimisation of short-term scheduling of science observations at Paranal observatory (VLT and ELT)

TL;DR

This paper discusses the optimization of short-term scheduling for scientific observations at Paranal observatory, introducing a simulator and exploring machine learning predictions to enhance scheduling efficiency.

Contribution

It presents a new STS simulator and evaluates the impact of machine learning predictions on scheduling effectiveness at Paranal and future ELT.

Findings

Machine learning predictions of atmospheric seeing improve scheduling efficiency.

Dynamic scheduling models can better incorporate medium-term constraints.

The simulator enables testing of various model assumptions for optimization.

Abstract

The efficiency of science observation Short-Term Scheduling (STS) can be defined as being a function of how many highly ranked observations are completed per unit time. Current STS at ESO's Paranal observatory is achieved through filtering and ranking observations via well-defined algorithms, leading to a proposed observation at time t. This Paranal STS model has been successfully employed for more than a decade. Here, we summarise the current VLT(I) STS model, and outline ongoing efforts of optimising the scientific return of both the VLT(I) and future ELT. We describe the STS simulator we have built that enables us to evaluate how changes in model assumptions affect STS effectiveness. Such changes include: using short-term predictions of atmospheric parameters instead of assuming their constant time evolution; assessing how the ranking weights on different observation parameters can be changed to optimise the scheduling; changing STS to be more `dynamic' to consider medium-term scheduling constraints. We present specific results comparing how machine learning predictions of the seeing can improve STS efficiency when compared to the current model of using the last 10\,min median of the measured seeing for observation selection.