Evaluation of multispectral, fine scale digital imagery as a tool for mapping stream morphology
Wright, A., Marcus, W.A. and Aspinall, R.J.
Geomorphology, Vol. 33 Issue 1-2 pp. 107-120
Multispectral digital imagery acquired from Soda Butte and Cache Creeks, Montana and Wyoming was used in conjunction with field data to classify and map hydrogeomorphic stream units on four stream reaches. The morphologic units that were field mapped were eddy drop zones, glides, low gradient riffles, high gradient riffles, lateral scour pools, attached bars, detached bars, and large woody debris. Unsupervised and supervised classifications of the imagery were used to develop a Maximum Joint Probability classification and an Alternative Joint Probability classification of the stream reaches. The Maximum Joint Probability classification allowed only one of the image classes to represent each hydrogeomorphic unit on the field map and resulted in relatively low overall accuracies for identification of these units of 10% to 50%. The Alternative Joint Probability classification allowed each image class to represent any geomorphic unit where the probability of a correct classification was greater than random. In this technique, two or three image classes were assigned to represent each hydrogeomorphic unit, resulting in higher overall accuracies of 28% to 80%. Accurate classification of hydrogeomorphic units was hampered by poor rectification of imagery with the field maps because of inadequate ground control points. In general, the largest hydrogeomorphic units were most accurately classified, whereas units that were small in area or spatially linear were least likely to be accurately classified. The results of this study demonstrated that multispectral digital imagery has the potential to be a useful tool for mapping hydrogeomorphic stream units at fine scales. Imagery to be an effective tool, however, careful measures such as accurate documentation of ground control points must be taken to ensure accurate rectification of the imagery with field maps.