Avalanche is a mass of snow slipping down from a mountain slope. Avalanches are frequent in snow-bound areas and cause severe damage to objects in their path. Avalanche forecasting aims to provide critical information for risk mitigation that includes locations of instabilities and their probability of release under different conditions. Machine learning models are frequently used to generate these forecasts. The outputs from these models help to quantify uncertainty and make inferences from limited data. Unavailability of long historical snow and avalanche records of a region can make their use infeasible for forecasting in that region. In this paper, we develop transfer learning models to address this problem. The models can use data from a support region to improve the forecasting performance of a region where the recorded data is insufficient to use conventional statistical models. We achieved operational performance using training data of one winter season from target region. This is valuable improvement over other models reported in literature which required at least 7-year training data of target region.