This study describes a machine learning workflow that brings together a variety of surface mass balance (SMB) measurements and a SMB model to improve model skill and to evaluate the performance of the trained and untrained model. The research makes a novel contribution to glacier mass balance modelling, with the potential for transferability across diverse climatic and glaciological settings. The workflow also provides a valuable framework that could inform the wider development and application of hybrid modelling approaches to glacier evolution that remain physically grounded.