Abstract

Based on the near real-time Global Positioning System (GPS) precise pointing positioning (PPP)-inferred water vapor system recently developed at the University of Calgary, Calgary, Alberta, an Ordinary Kriging procedure has been developed to predict the local and regional precipitable water vapor (PWV) and describe its distribution over Canada using limited available data. The Ordinary Kriging procedure includes five steps: (1) quantifying PWV spatial structure by calculating an experimental semivariogram; (2) fitting semivariogram models (spherical, exponential, and Gaussian) with nonlinear, weighted least squares; (3) determining the best-fitted model with cross-validation analysis; (4) estimating whole PWV maps by Ordinary Kriging interpolation; and (5) outputting kriging standard error maps. The 24 h variogram analysis shows that the correctly calculated experimental semivariogram is essential to the accuracy of the kriged maps, which depends on the configuration of the sites, lag step, and lag tolerance. The optimal lag step and lag tolerance for the current Canadian GPS network configuration are 5° and 2.5°, respectively. Among the three semivariogram models, the spherical model fails in its performance most of the time, and the best hourly fitted semivariogram model is either the exponential (90%) or the Gaussian (10%) model. The Gaussian-model-based Ordinary Kriging process produces more detailed maps. The surface maps of the kriging standard errors indicate that the area between longitudes –125° and –60° and latitudes 44° and 54° has higher accuracy due to higher availability of data.

You do not currently have access to this article.