Understanding the climate response of the Antarctic Peninsula ice sheet is vital for accurate predictions of sea-level rise. However, since climate models are typically too coarse to capture spatial variability in local scale meteorological processes, our ability to study specific sectors has been limited by the local fidelity of such models and the (often sparse) availability of observations. We show that a high-resolution (5.5 km × 5.5 km) version of a regional climate model (RACMO2.3) can reproduce observed interannual variability in the Larsen B embayment sufficiently to enable its use in investigating long-term changes in this sector. Using the model, together with automatic weather station data, we confirm previous findings that the year of the Larsen B ice shelf collapse (2001/02) was a strong melt year, but discover that total annual melt production was in fact ~30% lower than 2 years prior. While the year before collapse exhibited the lowest melting and highest snowfall during 1980–2014, the ice shelf was likely pre-conditioned for collapse by a series of strong melt years in the 1990s. Melt energy has since returned to pre-1990s levels, which likely explains the lack of further significant collapse in the region (e.g. of SCAR Inlet).