Recent computational advances have enabled substantial progress in large-data analysis and deep learning. These tools can be leveraged to discover patterns within the Earth system and to broaden our understanding of organized convection. Applications of deep learning for organized convection include storm detection within high-resolution numerical model simulations and the classification of possible storm hazards (e.g., hail and tornadoes) that ongoing convection could produce. However, as the climate continues to change, deep learning models trained using the historical climate may not perform with skill when faced with future extreme events. This talk will highlight a U-Net that can detect mesoscale convective systems within climate simulations, and a benchmark study that evaluates the ability of a convolutional neural network to skillfully classify severe convective storms of a future climate. Explainable artificial intelligence techniques were also applied to the trained neural network, which revealed that deep learning methods can learn physical information not prescribed during the training process, helping the network remain robust when encountering future outlier events. Given the significant role that organized convection plays within the climate system, particularly in the hydrological cycle and production of severe hazards with societal impacts, advancing our physical understanding of such phenomena using state-of-the-art tools remains of critical importance.