Smart low flow signature metrics for an improved overall performance evaluation of hydrological models

Abstract

Hydrological models have to be calibrated accurately to provide reasonable model results. For a concise model evaluation, the different phases of the hydrograph have to be considered in multi-metric frameworks with appropriate performance metrics. Low and high flows need to be reproduced simultaneously without neglecting the other phases of the hydrograph. In this paper, we highlight the relevance of very low and low flows with separate performance metrics. We present a multi-metric evaluation framework to identify calibration runs, which represent the different phases of the hydrograph precisely. A stepwise evaluation was done with commonly used statistical performance metrics (Nash–Sutcliffe, percent bias) and signature metrics, which are based on the flow duration curve (FDC). In order to consider a fairly balanced evaluation between high and low flow phases, we divided the flow duration curve into segments of high, medium and low flow phases, and additionally into very high and very low flow phases. The model performance in these segments was evaluated separately with the root mean square error (RMSE). Our results show that this evaluation method leads to an improved selection of good calibration runs to enhance the overall model performance by the refined segmentation of FDC. By combining performance metrics for high flow conditions with low flow conditions, this study demonstrates the challenge of calibrating a model with a satisfactory performance in high and low phases simultaneously. Consequently, we conclude that an additional performance metric for very low flows should be included in model analyzes to improve the overall performance in all phases of the hydrograph.