Evaluation of Models Predicting Mixing Depth

Abstract

Since mixing depth affects many aspects of lake productivity, including nutrient recycling, I evaluated the predictive power of 17 empirical models that relate mixing depth to morphometric variables to identify the best predictor. These models were tested empirically by compiling data from 123 temperate lakes of differing morphometry, geometry, and trophy. Four statistical indices of precision and bias, indicate that the model published by Shuter et al. (1983) using maximum effective length of the lake was the best published model for predicting mixing depth, although it is slightly biased. I then examined the effect of alternate predictors, reflecting lake configuration, basin shape, and geographical indices, to formulate an improved model. The best single predictor of therrnodine depth (THER) was maximum effective length (MEL): Log THER = 0.336 Log MEL − 0.245. No improvement in predictive power was obtained by combining other variables. This model is statistically superior to that of Shuter et al. (1983) beause it is not biased, it represents a greater number of lakes, and it covers a broader range of lake sizes and shapes over a more extensive geographical region. Two other models, using lake area and length of shoreline are proposed as alternate predictive tools, if MEL is not readily available.