Development of a Deep Full-Scale Connected U-Net for Reflectivity Inpainting in Spaceborne Radar Blind Zones

CloudSat's Cloud Profiling Radar (CPR) is a valuable tool for remotely monitoring high-latitude snowfall, but its ability to observe hydrometeor activity near the Earth's surface is limited by a radar blind zone caused by ground-clutter contamination. This study presents the development of a deep learning 3Net+-style convolutional neural network (i.e. blindpaint) to predict reflectivity profiles within the blind zone using seven years of CloudSat-calibrated Ka-band ARM Zenith radar (KaZR) data from two Arctic locations. The blindpaint network learns to predict the presence and intensity of near-surface hydrometeors by coupling latent features encoded in blind zone-aloft clouds with additional context from collocated atmospheric climate variables (i.e. temperature and specific humidity). The results show that blindpaint predictions outperform traditional interpolation methods, with a 33% higher Sørensen–Dice coefficient (significant at alpha < 0.05) and power spectral density curves two orders of magnitude closer to observations. Moreover, blindpaint demonstrates high wintertime pixel-level accuracy, with mean absolute error reductions of 7% when trained using the reanalysis-derived atmospheric covariates in addition to reflectivity, compared to when the model is trained using reflectivity as the sole predictor. Training the network on a combination of CPR-calibrated KaZR surface radar datasets at multiple locations facilitates the development of a generalized model with applications to spaceborne observations from CloudSat in Arctic regions. The proposed machine learning-based inpainting technique has the potential to enhance current and future spaceborne remote sensing snowfall missions by providing a better understanding of the nonlinear relationship between blind zone reflectivities and the surrounding atmospheric state.