Tropospheric delay is a GNSS error source caused by signal propagation through the troposphere, the lowest layer of Earth’s atmosphere extending from the surface to approximately 12 kilometers altitude where weather phenomena occur. Unlike ionospheric delay, tropospheric delay affects all GNSS frequencies equally (non-dispersive), preventing removal through multi-frequency combinations and requiring dedicated modeling approaches for precision applications.
Tropospheric delay consists of two components with different physical origins and modeling characteristics. The hydrostatic (or dry) component, caused by the dry gas mixture of the atmosphere, accounts for approximately 90% of total delay and is relatively stable and predictable from surface pressure measurements. The wet component, caused by atmospheric water vapor, accounts for approximately 10% of delay but varies rapidly and unpredictably with weather conditions, making it the primary challenge for tropospheric modeling.
At zenith (directly overhead), total tropospheric delay is approximately 2.3 meters. As satellite elevation decreases, the path length through the troposphere increases, and delay grows according to mapping functions that relate zenith delay to delay at arbitrary elevations. At 10° elevation, delay may exceed 10 meters. This elevation dependence motivates the common practice of excluding low-elevation satellites from position solutions, though this reduces satellite availability.
Tropospheric modeling in GNSS applications uses various approaches depending on accuracy requirements. Standard models like Saastamoinen or Hopfield estimate delay from surface meteorological data or standard atmosphere assumptions, achieving accuracy of several centimeters. Network RTK and PPP-RTK services estimate residual tropospheric delays at reference stations and provide corrections appropriate to user locations. High-precision processing software may estimate zenith tropospheric delay as a parameter in the positioning solution, improving accuracy but requiring sufficient observation geometry and time to constrain the additional unknown.
