Can Signatures Predict Hydrologic Model Performance in Validation Mode?

Abstract:

Hydrologic signatures have been receiving increasing attention in the modeling literature as a potential tool for improving the transfer of model parameters. Most commonly such applications focus on parameter transfers under a predictions in ungauged basins type scenario, where parameters are transferred spatially. In this study, we focus on utility of hydrologic signatures as predictors of model performance in a calibration-validation testing scenario (i.e., the temporal transfer of model parameters). The Probability Distributed Model, a simple conceptual hydrologic model, was used to test the approach across a number of catchments included in the MOPEX data set. We explored the change in model performance across calibration and validation time periods and contrasted it to the corresponding change in several hydrologic signatures. Results were explored in finer detail by utilizing a moving window approach to calibration and validation time periods, while also considering direct, signature-based parameter estimation. This investigation allowed us to develop a more in-depth understanding of the relationship between hydrologic model predictions and hydrologic signatures and provided insights into the potential added information that signatures can provide.