Urban block-scale wind field prediction is important for urban disaster prevention and for improving the safety of drone transportation. However, predictions based on high-resolution simulations involve high computational costs. To address this issue, methods using neural networks have been proposed. Onishi et al. (2019) [1] proposed a super-resolution simulation method that combines super-resolution techniques with numerical simulations. Yasuda et al. (2023) [2] extended this approach to three dimensions, enabling real-time prediction of three-dimensional flow fields at the urban block scale. On the other hand, these methods rely on building-resolving simulations or high-resolution input data. As a result, a method that can accurately and generically predict three-dimensional urban block-scale wind fields using only mesoscale meteorological fields as input has not yet been established. We propose a prediction method for three-dimensional temperature and wind velocity fields around buildings in urban areas using three-dimensional super-resolution (3DSR) based on convolutional neural networks (CNNs). The proposed method eliminates the need for building-resolving large-eddy simulations (LES) during urban block-scale flow prediction. We evaluate the accuracy and generalization performance of the proposed method using training data based on building-resolving LES. Furthermore, we examine the effectiveness of fine-tuning with a limited amount of data to mitigate performance degradation when applying the model to different regions. Our proposed method has achieved an 11-fold speedup compared to an existing method [2] and a 680-fold speedup compared to high-resolution building-resolving LES while predicting urban block-scale flow fields. The mean absolute error of temperature near the structure surface is approximately 0.3 K, and that of wind speed is about 1.0 m/s across all heights. In addition, we have demonstrated that fine-tuning enables accurate predictions even in different regions. These results suggest that CNN-based super-resolution enables accurate real-time prediction of urban block-scale flow fields from mesoscale meteorological fields.