EVALUATING STATISTICAL APPROACHES TO QUANTIFYING JUVENILE CHINOOK SALMON HABITAT IN A REGULATED CALIFORNIA RIVER. (20th December 2012)

• Record Type: Journal Article
• Authors: Beakes, M. P.; Moore, J. W.; Retford, N.; Brown, R.; Merz, J. E.; Sogard, S. M.
• Is Part Of: River research and applications. Volume 30:Number 2(2014:Feb.)
• Page Start: 180
• Page End: 191
• Publication Date: 2012-12-20
• Subjects: Rivers -- Regulation -- Periodicals; Rivers -- Periodicals; 551.483
• DOI: 10.1002/rra.2632
• Languages: English
• ISSNs: 1535-1459

Abstract

<abstract abstract-type="main"> <title>ABSTRACT</title> <p>Decisions on managed flow releases in regulated rivers should be informed by the best available science. To do this, resource managers require adequate information regarding the tradeoffs between alternative methodologies. In this study, we quantitatively compare two competing multivariate habitat models for juvenile Chinook salmon (<italic>Oncorhynchus tschawytscha</italic>), a highly valued fish species under serious decline in a large extent of its range. We conducted large‐scale snorkel surveys in the American River, California, to obtain a common dataset for model parameterization. We built one habitat model using Akaike Information Criterion analysis and model averaging, 'model G', and a second model by using a standard method of aggregating univariate habitat models, 'model A'. We calculated Cohen's kappa, percent correctly classified, sensitivity, specificity and the area under a receiver operator characteristic to compare the ability of each model to predict juvenile salmon presence and absence. We compared the predicted useable habitat of each model at nine simulated river discharges where usable habitat is equal to the product of a spatial area and the probability of habitat occupancy at that location. Generally, model G maintained greater predictive accuracy with a difference within 10% across the diagnostic statistics. Two key distinctions between models were that model G predicted 17.2% less useable habitat across simulated flows and had 5% fewer false positive classifications than model A. In contrast, model A had a tendency to over predict habitat occupancy and under predict model uncertainty. The largest discrepancy between model predictions occurred at the lowest flows simulated and in the habitats most likely to be occupied by juvenile salmon. This study supports the utility and quantitative framework of Akaike Information Criterion analysis and model averaging in developing habitat models. Copyright © 2012 John Wiley & Sons, Ltd.</p> </abstract>