Accurate short‐term weather prediction, essential for many aspects of life, relies mainly on forecasts from numerical weather models. Here, we report results supporting strongly deep learning as a viable, alternative approach. A 3D convolutional neural network, which uses a single frame of meteorology fields as input to predict the precipitation spatial distribution, is developed based on 39‐years (1980–2018) data of meteorology and daily precipitation over the contiguous United States. Results show that the trained network outperforms the state‐of‐the‐art weather models in predicting daily total precipitation, and the superiority of the network extends to forecast leads up to 5 days. Combining the network predictions with the weather‐model forecasts significantly improves the accuracy of model forecasts, especially for heavy‐precipitation events. Furthermore, the millisecond‐scale inference time of the network facilitates large ensemble predictions for extra accuracy improvement. These results demonstrate the promising prospects of deep learning in short‐term weather predictions.

Cloud diurnal variation (CDV) affects cloud radiative effects significantly as clouds reflect shortwave radiation only during the daytime but trap outgoing longwave radiation in both daytime and nighttime. Meanwhile, CDV also rectifies atmospheric variations of longer time scales via interactions with other physical and dynamic processes. These make CDV a valuable aspect for diagnosing climate model performance. Here, we evaluate the accuracy of simulated CDV in state‐of‐the‐art global climate models (GCMs) by comparing CDV in the historical simulation of 32 GCMs from 20 institutes participating the Coupled Model Intercomparison Project Phase 6 (CMIP6) with observations from the International Satellite and Cloud Climatology Project‐H product. While good agreement is found over the oceans, significant biases exist over land (notably deserts and plateaus), where the models simulate excessive nighttime clouds and insufficient daytime clouds and miss the observed peak of cloud fraction in the early afternoon. These biases persist throughout the year. It is illustrated that correcting the CDV biases tends to reduce the known model biases of smaller shortwave cloud radiative effect over the midlatitude Africa‐Europe‐Asia continent, South America, and vast ocean areas. Inter‐model comparisons show that the CDV biases vary significantly among models from different institutes and present similar characteristics among models from the same institutes, and suggests that the biases are more likely to be attributed to deficiencies in cloud‐related physical parameterizations rather than the model treatment of resolution, ocean, and chemistry. The improvement of CMIP6 models against their CMIP5 counterparts in simulating CDV is also discussed.