Accurate prediction of the freezing level (FZL) is essential for hydrometeorological forecasting systems and precipitation phase estimation, and it influences runoff generation and reservoir management decisions. In this study, we develop a deep learning based postprocessing framework using the Unet convolutional neural network (CNN) architecture to refine the FZL forecasts from the West Weather Research and Forecasting (West-WRF) model. The proposed framework leverages reforecast data from West WRF and FZL estimates from the California Nevada River Forecast Center (CNRFC) to train a deterministic Unet model over the Yuba-Feather watershed, a hydrologically critical basin in northern California. We introduce two variants of our model, Unet-Log and Unet-GMM, which utilize the logarithm of the hyperbolic cosine of Error and Gaussian Mixture Model loss functions, respectively, to enhance FZL forecast accuracy beyond an RMSE based benchmark. Results indicate that the Unet based postprocessing framework significantly improves FZL forecast skill across diverse atmospheric conditions and complex topography. Compared to the raw West-WRF output, our model achieves reductions in RMSE of up to 25% and increases the forecast observation correlation by about 10% over the Yuba-Feather watershed. Furthermore, it effectively captures the spatiotemporal variability of the FZL across different elevations, mitigating systematic biases inherent in the West-WRF model. This novel deep learning based postprocessing approach demonstrates a promising pathway for integrating machine learning into hydrometeorological forecasting and decision support within the Forecast Informed Reservoir Operations (FIRO) framework.