Accuracy Assessment of GEOID09 abd GEOID03 in Connecticut

Date of Completion

January 2010


Geodesy|Engineering, Geological




This study assessed the accuracy of the two U.S. national hybrid geoid models (GEOID09 and GEOID03) in Connecticut. I compared National Geodetic Survey (NGS) North American Vertical Datum of 1988 (NAVD 88) orthometric heights with orthometric heights determined by subtracting modeled (GEOID09 and GEOID03) geoid heights from NAVSTAR Global Position System (GPS)-derived ellipsoid heights. The GPS observation campaign (2007–2008) followed NGS guidelines for ellipsoid heights, and resulted in 60 days of GPS observations on 50 First-Order bench marks and 22 temporary bench marks with transferred elevations. I determined ellipsoid heights with least-squares adjustments of baseline vector networks using TOPCON Pinnacle GNSS observation processing software package, holding 12 Continuously Operating Reference Stations (CORS) as control. I used On-line Positioning User Service (OPUS)-ellipsoid heights as a check, and found no statistically significant differences (2- σ ) between adjusted ellipsoid heights and OPUS ellipsoid heights. I examined the effects of phase center variation (PCV) by computing ellipsoid-height differences between the network with PCV corrections and the same network without PCV corrections. The differences were statistically significant (2- σ ) and the uncorrected PCV network produces biased-solutions. I investigated the effect on ellipsoid heights resulting from allowing vectors not meeting NGS quality standards into the network. There were no statistically-significant differences (2- σ ) in a network of a single-day's data nor in a network of all data. Adjusted GPS-derived orthometric heights were subtracted from NGS published heights, and I performed residual analyses and two residuals at bench marks on the edge of the study area appeared as outliers. Analysis of residuals at bench marks belonging to NGS specified stability classes A, B and C revealed statistically significant differences between each class. The maximum residual value increases with decreasing stability class. I also analyzed the residuals by fitting a planar regression model, computing a datum transformation model, and by using kriging estimation. The results suggest that the residuals are random with no significant underlying spatial structure, which indicates that GEOID09 and GEOID03 perform within NGS-published specifications (Roman et al., 2009 and 2004) with no evidence of bias. ^