Open Collections

An exploration of computational methods for classifying sediment patches within archived aerial photographs of gravel-bed rivers

University of British Columbia

Bed material within gravel-bed rivers, which consists of gravel and other sediment coarser than 2mm, determines channel morphology and provides important benthic habitat. The impacts of flooding and anthropogenic activity on bed material are often examined to determine their effect on the morphology and ecology of gravel-bed rivers. To truly characterize this relationship, bed material must be discriminated from sand and sediment finer than 2mm at reach-scale over time. The current remote sensing routines that are used to synoptically characterize fluvial sediment did not emerge until the early 2000s, and as a result, reach-scale assessments of sand and gravel, extending beyond the 2000s, are absent for most gravel-bed rivers. Fortunately, archived aerial photographs, can be used to analyze past landscapes. Traditionally, these analyses are carried out, manually, by photo interpreters, but multiple studies have shown that image processing techniques can be used to extract meaningful information from scanned aerial photographs. This study provides an exploration of semi-automated and automated image classification routines and their ability to replace manual interpretation for delineating patches of sand and gravel within scanned archived aerial photographs. Results indicate that patches of sand and gravel within contemporary digital aerial imagery that has been degraded to mimic the characteristics of analog aerial photography can be consistently classified with overall accuracy above ~93% using automated object- and pixel-based classification routines. However, these same classifications only agree with manual interpretations of archived aerial photographs between ~45-70%. In contrast, the semi-automated routine provides measures of agreement that range almost entirely between ~75-85% when compared to the manual interpretations as well as the automated routines. Together, this demonstrates that a semi-automated routine should be used to classify scanned archived aerial photographs into sand and gravel.

Text

2019-08-21

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/71386

Geography

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/