| Through advancements of digital photogrammetry the construction of terrain models has become a streamlined and efficient process. The development of automated terrain extraction methods has opened the door to the inexperienced user, where before a great deal of experience and technique has been required to create terrain models using digital photogrammetry. However the question remains: How accurate are automatically extracted terrain models when compared to the actual surface of the earth.;Using the Leica Photogrammetry Suite, marketed by ERDAS, this study was undertaken to: (1) determine how aerial photography and default software characteristics affect terrain model quality and accuracy and (2) evaluate and recommend techniques and methods to maximize the accuracy of terrain models based on default software settings.;Two test areas were selected to analyze the terrain model accuracy. A landslide which occurred in 1998 in the Cook Lake area was used to determine the ability of the software to portray slope movement, while static landslide areas in the Union Pass area were used to determine how well the software could accurately represent landforms using multiple photographic series. Accuracy was determined by comparing extracted terrain models to both the United States Geological Survey (USGS) 10 meter Digital Elevation Models (DEMs) and sub-meter Global Positioning System (GPS) data collected in the field.;Results show that the extracted terrains are within the USGS mapping standards for DEMs, and in most cases are as accurate as the USGS 10m DEM when both the extracted terrain model and the USGS 10m DEMs are compared to the sub-meter GPS data. Although accuracy is generally acceptable for all series of photographs tested, errors in terrain model elevations vary somewhat randomly but accuracy is considerably better for those series that have higher quality data on flight and equipment parameters. This variability in accuracy illustrates that mapping large-scale slope changes is possible, but small-scale events (less than 20m) are lost within the uncertainty inherent in the method used to automatically extract terrain models. |
