A comparison of robust bayes and classical estimators for regional lake models of fish response to acidification

Abstract

Empirical models of fish response to lake acidification were recently fit to a large historical data set from the Adirondack region of the United States using classical and Bayesian methods. The models may be used to predict species presence/absence for brook trout and lake trout as a function of acid-precipitation-related water chemistry, using a logistic function. To evaluate the effectiveness of the models in the prediction of presence/absence due to regional lake acidification, new data sets were used for cross validation of the candidate models. Based on this evaluation, the robust Bayes models, which are based on a compromise estimator between Bayes and empirical Bayes, were found to be the best predictors of species presence/absence in lakes.