A regional/single station ZTD combined forecasting model based on machine learning algorithm

Aiming at the temporal and spatial characteristics of zenith tropospheric total delay (ZTD), a combined regional/single station ZTD prediction model based on BP neural network and long-term memory network (LSTM) algorithm is proposed. Taking the observation data of 18 stations in Hong Kong continuously operating reference stations (CORS) network for 14 consecutive days as an example, the regional, single station and combined ZTD prediction models are studied by using BP neural network, LSTM and the algorithm proposed in this paper. The prediction results of HKWS station show that the root mean square error (RMSE) of regional, single station and combined ZTD prediction models are 10.2 mm, 10.4 mm and 8.5 mm respectively, and the prediction accuracy of the combined model is improved by 17.2% and 18.4% compared with the regional model and the single station model, respectively.