Abstract. Many ice shelves in Antarctica experience surface melt each summer, with potentially severe consequences for sea level rise. However, large interannual and regional variability in surface melt increases uncertainty in predictions of how ice shelves will react to climate change. Previous studies of surface melt have usually focused on either a process-level understanding of surface melt through energy balance investigations, or used regional melt metrics to quantify interannual variability in satellite observations of surface melt. Here, we use an approach that helps bridge the gap between work at these two scales. Using daily passive microwave observations from the AMSR-E and AMSR-2 sensors, and the machine learning approach of a self-organising map, we identify nine representative spatial distributions (“patterns”) of surface melt on the Shackleton Ice Shelf, East Antarctica, over the previous two decades (2002/03–2020/21). Our results point to a significant role for surface air temperatures in controlling the interannual variability of summer melt, and also reveal the influence of local controls on driving melt. In particular, prolonged melt in the south-east of the shelf and along the grounding line shows the importance of katabatic winds and surface albedo. Our approach highlights the necessity of understanding both local and large-scale controls on surface melt, and demonstrates that self-organising maps can be used to investigate the variability of surface melt on Antarctic ice shelves.