Artificial Intelligence and Machine Learning (AI/ML) techniques are powerful tools, which can be applied to forecast spatial imagery in sequence. In this study, we applied a Convolutional Neural Network (CNN) model to predict the next few hours of rainfall observation imagery from NASA's Global Precipitation Measurement mission and its IMERG Early Run product in assist of hurricane predictions over the U.S.-Mexico Gulf coastal region. The IMERG Early Run is a satellite-based rainfall retrieval technique, providing rainfall measurement with a half-hourly temporal resolution, near-global scale coverage, and 4 h observation latency rainfall intensity imagery. The goal of this study is to extend the observation latency so that the IMERG Early Run product could potentially be used for real-time hurricane prediction and flash flood warning. In this study, we first build a sub-dataset from the IMERG Early Run product containing total 37 past hurricane events that hit the contiguous U.S. (CONUS) bordering the Gulf of Mexico from 2002 to 2019. Then, two CNN models with different model structures are built and tested to forecast the 37 hurricane events in a retrospective manner. Sensitivity experiments are conducted to select the optimal hyperparameters of the two CNN models, i.e., the number of convolution layers, filter size, minibatch size, number of filters, and pooling size. The prediction results show that the two employed CNN models generally provide satisfactory performance, showing averaged Accuracy in the categorical metrics is above 90% and averaged NSE in the continuous metrics is above 0.5. We found that the forecasting performance of the two CNN models is not significantly different; however, the CNN model without pooling layers always shows slightly better performance than the CNN model with pooling layers. We also found that both CNN models can predict the spatial range of hurricane rainfall well, but there are limitations to underestimate in larger rainfall rates. This study will serve as a novel investigation on how AI/ML models could help with real-time hurricane predictions, and provide a future reference for selecting the optimal CNN model parameters for processing spatial rainfall data.