Improving Empirical Pressure Estimation Over Mainland China and Surrounding Regions Using Machine Learning
Abstract
Accurate pressure is required in high-precision geodetic applications, including global navigation satellite systems (GNSS) and very long baseline interferometry (VLBI). Widely used empirical models provide pressure estimates from location and time alone, but their predefined formulations leave residual spatiotemporal variations unresolved, limiting accuracy. In this work, a machine-learning approach (random forest) is used to learn these residual spatiotemporal variations and to construct an enhanced empirical pressure model, EGPT3, for Mainland China and surrounding regions. Against an independent radiosonde dataset, EGPT3 achieves a bias of—0.2 hPa and a root mean square error (RMSE) of 3.2 hPa. RMSE is reduced by 87.0% relative to the benchmark GPT3 model and by 67.7% relative to T0, which requires in situ pressure and temperature measurements. Performance is found to be more reliable across spatial, temporal, and vertical domains than for GPT3 and T0. In particular, EGPT3 delivers substantial gains under large height differences and outperforms T0 in these regimes, and in some cases even achieves higher accuracy at large height differences than at low height differences. Furthermore, the EGPT3-derived ZHD exhibits superior accuracy and stability compared with the T0- and GPT3-derived ZHD. Overall, EGPT3 provides stable and accurate pressure estimates from the surface to the tropopause, thereby supplying reliable pressure inputs for geodetic applications, including GNSS and VLBI, over the complex terrain of mainland China and its surrounding regions.