The potential of digital surface models based on aerial images for automated vegetation mapping

Abstract

Segmentation of vegetation patches was tested using canopy height models (CHMs) representing the height difference between digital surface models (DSMs), generated by matching digital aerial images from the Z/I Digital Mapping Camera, and a digital elevation model (DEM) based on airborne laser scanner data. Three different combinations of aerial images were used in the production of the CHMs to test the effect of flight altitude and stereo overlap on segmentation accuracy. Segmentation results were evaluated using the standard deviation of photo-interpreted tree height within segments, as well as by visual comparison to existing maps. In addition, height percentiles extracted from the CHMs were used to estimate tree heights. Tree height estimation at the segment level yielded root mean square error (RMSE) values of 2.0 m, or 15.1%, and an adjusted coefficient of determination (adjusted R²) of 0.94 when using a CHM from images acquired at an altitude of 1200 m above ground level (agl) and with an along-track stereo overlap of 80%. When a CHM based on images acquired at 4800 m agl and an overlap of 60% was used, the corresponding results were an RMSE of 2.2 m, or 16.0%, and an adjusted R² of 0.92. Tree height estimation at the plot level was most accurate for densely forested plots dominated by coniferous tree species (RMSE of 2.1 m, or 9.8%, and adjusted R² of 0.88). It is shown that CHMs based on aerial images acquired at 4800 m agl and with 60% along-track stereo overlap are useful for the segmentation of vegetation and are at least as good as those based on aerial images collected at a lower flight altitude or with greater overlap.

Acknowledgements

The authors would like to thank Mikael Johansson, Anders Andersson, Berth Hedqvist, and Christer Strandberg at Lantmäteriet in Gävle, Sweden, for providing the Z/I DMC images and for the matching of DSMs. The authors are grateful to Associate Professor Sören Holm for discussions regarding statistical methods and Dr Heather Reese for revision of the language, both at the Swedish University of Agricultural Sciences, Umeå, Sweden. The authors would also like to thank three anonymous reviewers for their valuable comments.

Additional information

Funding

This study was part of the programme, Environmental Mapping and Monitoring with Airborne Laser and Digital Images (EMMA), supported by the Swedish Environmental Protection Agency [Dnr 08/234].