An Evaluation of Atmospheric Models for GPS data Retrieval by Output from a Numerical Weather Model

抄録

Atmospheric delays in GPS processing are estimated by fitting parameters of atmospheric models to the observed delays. GPS-related error sources include multipath effects and phase center variations, as well as uncertainties in atmospheric models. An atmospheric model must accurately capture the delay distribution if errors are to be minimized. Large GPS position errors occurred on 7 March 1997 on the Izu Peninsula and at Hatsu-shima, about 100 km southwest of Tokyo, concomitant with a mountain lee wave. The coinciding of the large position errors and the mountain lee wave suggests that small-scale fluctuations in water vapor and air density associated with lee waves could cause large position errors.
This study used a high-resolution non-hydrostatic model to simulate the mountain lee wave. Slant delays for the GPS satellites were calculated from the reproduced water vapor and air density fields using a ray-tracing method. The atmospheric delays are obtained by fitting the atmospheric models to the reproduced slant delays, instead of the actual delay data. The atmospheric models were evaluated by determining the difference in position error between the reproduced delays and the model-fitted delays. The mountain lee wave case was used to evaluate three different atmospheric models. The first, the “constant model”, has only zenith delay as an unknown parameter. Large position errors occurred when the constant model was used to fit the data in this case. The “linear gradient model”, adds two horizontal gradient parameters to the zenith parameter, and yielded significantly reduced horizontal position errors. Horizontal and vertical position errors were reduced further with a “second order model”, which adds second-order terms. Evaluation of the atmospheric models using the mountain lee wave case indicated that 1) the linear gradient model cannot express complicated atmospheric disturbances; and, 2) a second-order model reduces horizontal and vertical position errors.