Effects of sensor resolution on mapping in-stream habitats

Legleiter, C.J., Marcus, W.A. and Lawrence, R.L.

Photogrammetric engineering & remote sensing, Vol. 68 Issue 8 pp. 801-807

2002

Effects of spatial, spectral, and radiometric resolution on remote mapping of fourth-order in-stream habitats were evaluated by comparing hyperspectral imagery to simulated multispectral data. Spectral resolution was more important than spatial or radiometric resolution in improving classification accuracies, although overall accuracies never exceeded 62 percent. Overall accuracies were significantly greater for (1) hyperspectral data (7.2 percent) compared to simulated multispectral imagery, (2) 1 -m pixels (4.7 percent) compared to 2.5-m pixels, and (3) 11-bit data (0.8 percent) compared to &bit data. Higher spatial resolution also enabled removal of transitional areas between units by using interior buffers, improving accuracy by up to 15.6 percent. We believe low overall accuracies were primarily due to the subjective and oversimplified nature of the polygon-based field maps used as ground reference data, and high-resolution imagery might provide a more detailed representation of in-stream habitats. Improved methods of collecting ground reference data, utilizing a point-based approach, should be developed for assessing the accuracy of classifications derived from fine spatial resolution (less than 5-m) imagery.