A case-study of in-stream juvenile salmon habitat classification using decision-based fusion of multispectral aerial images

Abstract

Cataloging and monitoring the quality and quantity of freshwater rearing habitat for Pacific salmon Oncorhynchus spp. is essential for maintaining self-sustaining stocks of these species. Field-based habitat monitoring in remote areas can be expensive, impractical, and subjective. We demonstrate the usefulness of high spatial resolution multispectral images acquired in the visible, near infrared and thermal infrared regions in effectively delineating habitat classes important to juvenile salmon rearing habitat. Our study showed that different spectral bands have complementary strengths, with visible bands being best for delineating main-channel habitats, visible and thermal infrared for off-channel habitats, and near infrared for landcover classes. Although no individual classification result captured all the habitat elements, a decision-based fusion that uses selected classes from all classified images provides a meaningful product that shows the spatial distribution and quantity of habitat classes. An accuracy analysis using ground truth reference data demonstrated that habitat classes were identified with an overall accuracy of 82.5%, with in-stream classes having an overall accuracy of 76.0%. This study has broad implications as it demonstrates that low-cost airborne images can be used to effectively determine the quality and quantity of juvenile Pacific salmon rearing habitat and decision-support in fisheries management.