Snow water equivalent (SWE) is a crucial component of mountain hydrology but still faces large uncertainties in its quantification due to its high temporal and spatial variability. While machine learning (ML) has been applied to similar domains with notable results, its use for SWE prediction has been hindered by the quality, quantity and extent of the measured data. Hybrid models that integrate simulated data from physics-based models with ML have shown promising results in data-scarce scenarios. A comparison of hybrid models was performed, targeting temporal and spatial extrapolation of SWE. Crocus snow model simulations were used together with data from ten meteorological and snow observation stations throughout the northern hemisphere containing 7-20 years of data. Two main setups were tested; a common post-processor approach, where the outputs and state variables from Crocus were fed as additional predictors to the ML model at each time step, and a data augmentation approach, where Crocus simulations for stations at which no measurements are available were used as additional training points. The results show that the post-processor approach is best suited for predicting SWE in years excluded during training. However, when predicting in left-out stations, the data augmentation setup achieved the largest increase in performance, reducing the root mean squared error by 22% compared to Crocus and by 42% compared to a measurement-based ML model. A feature importance analysis reveals that the hybrid model predictions are most influenced by physically-sound variables, such as incoming radiation, current SWE status, air temperature and snowfall. After proving the ability of hybrid models for predicting SWE under data-scarce conditions, ongoing work will further assess their robustness and applicability on an extensive dataset covering the northern hemisphere.