Abstract —Tropsheric delay is one of important error sources in GPS positioning and InSAR applications. Accurate measurements of ZTD (Zenith Tropospheric Delay) can be used for correcting the atmospheric delay on GPS and InSAR as well as atmospheric science research and applications (e.g. numeric weather prediction). Traditionally, the GPS ZTD estimations were obtained based on the least squares (LS) principle, where the functional and stochastic models of GPS measurements need to be defined precisely. The functional models for GPS measurements have been investigated in considerable detail in the past two decades. However, most scientific GPS processing software packages, e.g. GAMIT, BERNESE or GIPSY, the stochastic models of GPS observation data are simplified, assuming that all the GPS measurements have the same variance, and that they are statistically independent in time and space. Such assumptions are unrealistic and will result in unreliable ZTD estimations as GPS observations from different satellites cannot have the same accuracy due to varying noise levels. In addition, it is impossible to model all systematic errors in the functional model, and therefore, modeling some systematic errors into the stochastic model is a current challenging topic to further realize the full potential of increasingly more accurate GPS positioning applications. This paper aims to improve the GPS ZTD estimations by modeling GPS systematic residuals into the stochastic model