Abstract: The root water uptake process is a key link in the hydrological process within a continuous groundwater-soil-vegetation-atmosphere system. However, the limitations of traditional methods such as experiments and numerical simulations still lead to great uncertainties in the study of the root water uptake process. Deep learning models have demonstrated the potential to surpass traditional methods in the field of hydrological process simulation, but relatively few studies have investigated the root water uptake process. A deep learning model (CNN-LSTM) combining a convolutional neural network (CNN) with a long short-term memory network (LSTM) was proposed in this study. The spatiotemporal distribution of root water uptake of a typical xerophytic plant, Salix, during three growth cycles was simulated with easily accessible hydrometeorological data. The CNN-LSTM model accurately characterized the spatiotemporal variation dynamics of root water uptake. The determination coefficient was not less than 0.86, and E_RMS was less than 0.01, significantly enhancing the prediction performance of root water uptake. The CNN-LSTM model can capture the spatiotemporal dynamic root water uptake processes with lower computational cost and higher accuracy. This model has a stronger generalizability and can adapt to conditions lacking basic data, such as soil hydraulic parameters and vegetation parameters. This study explored the application of deep learning methods in simulating complex root water uptake and achieved good results. This study provides reference data for theoretical research, method innovation for vegetation water consumption and ecological practices, and provides reference data for deep learning research on other hydrological problems.