Application of a Distributed Large Basin Runoff Model to Lake Erie: Model Calibration and Analysis of Parameter Spatial Variation

Abstract

The distributed large basin runoff model (DLBRM) was designed to simulate the hydrological processes of the Great Lakes watersheds. As part of its development, the DLBRM was recently applied to 18 watersheds in the Lake Erie basin, where it was first calibrated to reproduce the observed discharge in 1950–1964 and then applied to 1999–2006. Four different calibration objective functions: root mean squared error (RMSE) minimization, mean absolute error (MAE) minimization, correlation maximization, and Nash-Sutcliffe index maximization were tested, revealing RMSE minimization as the most successful method and able to achieve results very close to its global minimum. Further, the distribution of the main DLBRM parameters in the 18 watersheds was consistent with regional patterns, although each watershed was calibrated individually, thus adding credibility to the calibration process. Model performances, while generally good, varied across the basin according to a series of environmental factors, including climate, watershed shape, topography, and land cover and observation factors such as gauging station distribution. Gauging station coverage proved to be extremely important in the ability of the model to track flow variability. The DLBRM proved to be able to replicate well the 1999–2006 hydrologies of most watersheds without recalibration. However, its performance declined in heavily urbanized watersheds, where the landscape changed the most. The results described in this paper will lead to improved model performance and increased practical applications of the DLBRM, providing important information to researchers and decision makers for efficient water management programs in Great Lakes watersheds.