Scalable and robust wide-field facet calibration with LOFAR's longest baselines

Abstract

Recent work has successfully achieved sub-arcsecond wide-field imaging with high-band observations from the Low Frequency Array (LOFAR). However, the scalability of this work remains limited due to the need for manual intervention, poor calibration solutions for the Dutch LOFAR stations, and high computational costs. We address these issues by: (1) improving automated self-calibration using a signal-to-noise metric and a neural network for image artefact detection; (2) implementing a refined calibration strategy for the Dutch LOFAR stations; and (3) cutting computational costs by optimising the data processing strategy. We demonstrate the effectiveness of our automated processing strategy by reprocessing one previously reduced dataset and a new dataset from the ELAIS-N1 deep field, which features more severe ionospheric conditions. We find calibration artefacts across facet boundaries to be reduced with our improved automated calibration strategy and achieve a computational cost reduction of about a factor of 4 to 6 compared to previous work, where the exact factor depends on whether a single observation is processed or multiple observations of the same sky area are combined. Further optimisation and improved handling of data with baseline-dependent averaging could reduce this in the near future by another factor of two, bringing the total cost for an 8-hour observation below 30,000 CPU core hours. This work enables ultra-deep imaging at sensitivities on the order of a few $\mu$Jy/beam. Furthermore, it also lays the foundation for a fully automated survey pipeline for sub-arcsecond wide-field imaging of the northern sky with LOFAR.