Abstract: Zenith tropospheric delay (ZTD) is a key factor affecting global positioning system (GPS) positioning accuracy. In order to improve the prediction accuracy of ZTD, a Gaussian process(GP) regression prediction model based on phase space reconstruction is proposed.In view of the chaotic characteristics of ZTD time series, using the ZTD data provided by the International Global Navigation Satellite System Service (IGS) stations.Firstly, the embedded dimension is determined using Cao method, phase space reconstruction of ZTD data is carried out, and the precision and accuracy of ZTD using GP model for 12 IGS ststions at different latitude levels in the southern and northern hemisphere are explored.Then, in order to verify the effectiveness of GP model, the prediction results are compared with the original data and prediction results of the back propagation (BP) neural network model, and the influence of different time on the prediction accuracy of ZTD is further explored. Finally, the influence of longitude and altitude on the prediction accuracy of ZTD is analyzed.The results show that the root mean square error (RMSE) of GP model prediction results reaches mm level, the correlation between GP model and theoretical value reaches 0.997, and the prediction accuracy index is obviously better than that of BP neural network model. The prediction accuracy of GP model in the southern hemisphere is higher than that in the northern hemisphere, and RMSE in the high latitude area is less than 3.6 mm, which is more suitable for the tropospheric delay prediction in the high latitude area. In the time domain of the study, the prediction accuracy of GP model at night is higher than that in the day at most sites, the longitude has no obvious influence on the prediction accuracy of ZTD, and the altitude is proportional to the prediction accuracy of ZTD. Therefore, GP model has better practicability and feasibility for the prediction of tropospheric delay.