Machine learning approach for predicting precise ZTD produced by GNSS broadcast ephemeris

Precise modelling of tropospheric delay plays an important role of in the precise Global Navigation Satellite System (GNSS) positioning application as well as meteorological studies and weather forecasting. In general, to calculate precise Zenith Total Delay (ZTD) utilizing GNSS system, extra supplemental data such as precise ephemeris and clock compensation are required in addition to GNSS observation data. For the real-time GNSS observation and positioning, continuous data communication for acquiring precise ephemeris leads additional power consumption. This paper proposes an estimation method based on machine learning approach for retrieving precise ZTD time series without using precise ephemeris and clock compensation data. Proposal shows that precise ZTD time series can be obtained from intermittent ZTD and surface meteorological data (pressure, temperature and relative humidity) with its root mean square error of less than 0.025 [m].