Measuring precipitation, especially winter snow accumulation, remains highly uncertain in remote regions, yet is critical for advancing the understanding of climate- and land-use-driven changes in energy and hydrological cycles. In northern and boreal environments, uncertainties in precipitation directly affect snow storage, runoff generation, and land–atmosphere exchanges, thereby limiting the reliability of hydrological and land-surface modelling. Retrievals from remote sensing techniques therefore hold strong promise for providing observational constraints on surface precipitation in data-sparse regions. To exploit the synergistic strengths of multiple instruments for enhanced retrieval accuracy, we develop a missing-aware three instrument synergy model, PrecipFusion, based on convolutional neural networks (CNNs) that estimates 20-min average accumulated surface precipitation from an ensemble of ground-based instruments (Doppler radar, Doppler lidar, and microwave radiometer). Across a wide range of precipitation intensities, PrecipFusion consistently benefits from three-instrument synergy, achieving an average 88% improvement in the coefficient of determination (R²) relative to radar-only retrievals. Diagnostic analyses further reveal distinct and complementary contributions from each instrument, including lidar sensitivity to small hydrometeors and its characteristic missing structures under intense precipitation. This research is conducted at the University of Waterloo and funded by the Canadian Space Agency in support of initial HAWC science development activities. Future work will extend the PrecipFusion framework to incorporate coincident HAWC/AOS measurements, enabling improved precipitation estimation across Canada’s North.