Abstract: Accurate modeling of the Zenith Tropospheric Delay (ZTD) is crucial for enhancing the accuracy of Global Navigation Satellite System (GNSS) positioning and for meteorological applications. Due to the extensive geographical range from north to south along the east coast of China, the spatio-temporal variation of ZTD is intricate. However, the existing GPT3 model struggles to accurately describe the complex nonlinear changes of ZTD on different space-time scales. In this study, two advanced deep learning algorithms, namely, back-propagation neural network (BPNN) and radial-basis-function neural network (RBF), were employed to establish a highly accurate tropospheric delay model for the eastern coast of China using their excellent nonlinear fitting capabilities.The model made full use of the ZTD calculated from regional GNSS data and the corresponding ZTD estimated from the Global Pressure and Temperature 3 (GPT3) model, and adopted the strategy of seasonal and daily modeling. We used data from 92 stations (80%) for the eastern coast of China provided by the Crustal Movement Observation Network of China from 2016 to 2019 for model construction, and the remaining 24 stations (20%) for model testing. Results showed that (1) the proposed model had a good internal coincidence accuracy. The root mean square error (RMSE) of the BPNN_GPT3 and RBF_GPT3 models were 2.70 cm and 1.94 cm, which had an improvement of 46.9% and 61.8% over the GPT3 model. (2) The external validations in non-modeling stations demonstrated the superiority of the proposed approach. The RMSE values were 2.78 cm and 2.28 cm, which had an improvement of 46.5% and 56.2% over the GPT3 model.