The climatic conditions, particularly the sources of precipitation that enabled extensive glacial growth during the Last Glacial Maximum (LGM) in the European Alps, remain poorly constrained. Here, we apply an inversion method to reconstruct equilibrium line altitude (ELA) fields using glacier footprints, such as the moraines deposited by Alpine glaciers during the LGM. By employing a machine-learning emulator trained on outputs from a glacier-evolution model, we predict glacier maximal thickness. The emulator is integrated into a gradient-based inversion scheme to derive ELA fields consistent with LGM footprints. The results show that the reconstructed ELA fields align with those from previous studies, validating the robustness of our approach. Unlike existing inversion methods, our approach is more general and avoids restrictive assumptions. Notably, by incorporating the transient response of glaciers to climate variability (we do not assume steady state), we show that the cold spell period is crucial for interpreting the reconstructed climate patterns—an aspect previously overlooked. Our findings provide new insights into climatic variability during the LGM, particularly concerning the interaction between precipitation patterns and the cold spell period. Furthermore, the computational efficiency of our method makes it applicable to large-scale paleoclimate reconstructions based on glacier footprints.