Abstract
A consequence of illegal immigration and border law enforcement practices has been the development of a complex foot trail network in remote areas of San Diego County near the US–Mexico border. Comprehensive monitoring of changes in trail networks requires multitemporal remote sensing analyses. Three semi-automated, commercially available object extraction routines were evaluated for delineating new trails. Three types of multi-temporal image products were generated from two dates of scanned colour infrared (CIR) images of two US–Mexico border study sites, at four spatial resolutions (15, 30, 60 and 120 cm). Accuracy was assessed using reference data created by manual image interpretation. The semi-automated routines captured most of the new trail objects, but also contained gaps and high commission error. Products derived using a spatial contextual-based neural network classifier with CIR image difference or red band layer stack and 15 or 30 cm spatial resolution image inputs yielded the best extractions of new trails.
Acknowledgements
Funding was provided by the National Aeronautics and Space Administration, Affiliated Research Center (NCC13-00002) and Research, Education, and Application Solutions Network (NCC13-03007) programs. Officers of the San Diego Sector of the US Border Patrol assisted in obtaining GPS and digital map data, and provided field assistance. Special thanks go to Zlatina Anguelova for digitizing reference trail maps, and to John Kaiser, who contributed a wealth of knowledge of US–Mexico border issues and remote sensing and GIS technology.