Compound heat and drought events have severe socio-economic impacts on human health, agriculture and electricity supply. While these compound extremes are projected to intensify under climate change, our understanding of their subseasonal predictability remains limited compared to that of individual heat or drought events. In this study, we evaluate the predictability of compound heat and drought events over Europe using the subseasonal prediction system of the European Centre for Medium-Range Weather Forecasts (ECMWF). We find that the physical coupling between heat and drought contributes up to 10 % towards an increase in forecast skill when heat and drought co-occur, relative to a baseline that assumes independence between extremes. However, in regions where the physical coupling between heat and drought via land-surface interaction is misrepresented, compound skill can be lower than when drought are predicted in isolation. These findings highlight the critical role of accurately simulating land-surface feedbacks to improve the reliability of the subseasonal prediction for compound extremes.