Calibration of hydrological models for ecologically relevant streamflow predictions: a trade-off between fitting well to data and estimating consistent parameter sets?

Abstract

Abstract. The ecological integrity of freshwater ecosystems is intimately linked to natural fluctuations in the river flow regime. In catchments with little human-induced alterations of the flow regime (e.g. abstractions and regulations), existing hydrological models can be used to predict changes in the local flow regime to assess any changes in its rivers' living environment for endemic species. However, hydrological models are traditionally calibrated to give a good general fit to observed hydrographs, e.g. using criteria such as the Nash–Sutcliffe efficiency (NSE) or the Kling–Gupta efficiency (KGE). Much ecological research has shown that aquatic species respond to a range of specific characteristics of the hydrograph, including magnitude, frequency, duration, timing, and the rate of change of flow events. This study investigates the performance of specially developed and tailored criteria formed from combinations of those specific streamflow characteristics (SFCs) found to be ecologically relevant in previous ecohydrological studies. These are compared with the more traditional Kling–Gupta criterion for 33 Irish catchments. A split-sample test with a rolling window is applied to reduce the influence on the conclusions of differences between the calibration and evaluation periods. These tailored criteria are shown to be marginally better suited to predicting the targeted streamflow characteristics; however, traditional criteria are more robust and produce more consistent behavioural parameter sets, suggesting a trade-off between model performance and model parameter consistency when predicting specific streamflow characteristics. Analysis of the fitting to each of 165 streamflow characteristics revealed a general lack of versatility for criteria with a strong focus on low-flow conditions, especially in predicting high-flow conditions. On the other hand, the Kling–Gupta efficiency applied to the square root of flow values performs as well as two sets of tailored criteria across the 165 streamflow characteristics. These findings suggest that traditional composite criteria such as the Kling–Gupta efficiency may still be preferable over tailored criteria for the prediction of streamflow characteristics, when robustness and consistency are important.

Predictability of flow metrics calculated using a distributed hydrologic model across ecoregions and stream classes: Implications for developing flow–ecology relationships

“…Magnitude HMs for average flow conditions were highly correlated between observed and simulated values and had IQRs that consistently fell within the acceptance criteria of ±30%. This is similar to the findings of previous studies indicating that magnitude is generally more reliable than the frequency, duration and rate of change of hydrologic regime components (Eng et al, 2017; Hallouin et al, 2020; Vigiak et al, 2018; Westerberg et al, 2016; Westerberg & McMillan, 2015). This result is expected given that these metrics represent the volume of streamflow, which is the target of most model simulation processes and calibration objectives, and the conditions expected within streams most of the time.…”

“…This result indicates this flow regime component was by far the most accurately simulated HM in terms of bias, where our TimeBias calculation considered the relationship between the start and end of the year and maximum value in Julian days. This result differs from other studies that had larger errors associated with timing of flow events (Eng et al, 2017; Hallouin et al, 2020; Vigiak et al, 2018). Given that the basins for which these HMs were assessed ranged in drainage areas from less than 20 km 2 to more than 3200 km 2 , the performance improvements seen over other studies are not solely due to routing of events in one scale of basin (i.e., small basins with quick response).…”

“…Following Arnold et al (2012), we verified that the split between the validation and calibration time windows was acceptable by (1) confirming that climate conditions in both windows contained similar wet and dry years as represented by the monthly Palmer Drought Severity Index available by state climate region through the National Center for Environmental Information and (2) ensuring that observed flow conditions were similar in summary statistics between the two windows (Figure 3). Hallouin et al (2020) followed a similar method of using a 30‐year period of full record and setting a minimum of 14 years of complete hydrologic data, which allowed for a minimum of 7 years for calibration and 7 years for validation within each of their assessed drainage basins. In their application of a distributed model examining HMs at two sites, Shrestha et al (2014) also chose a 30‐year period and used a subset of this period for calibration and validation.…”

“…Our selection of calibration statistics is supported by a number of recent studies focused on providing the most optimum representation of HMs from hydrologic modelling. For example, Hallouin et al (2020) concluded that traditional objective functions are still preferred over functions targeted to or using individual HMs. Similarly, Pool et al (2017) identified trade‐offs in HM accuracy when models are fit to a single HM.…”

“…Our choice in metrics used to evaluate the daily performance of the streamflow simulation was based on a compilation of recent studies focused on determining the optimum calibration statistics for hydrologic models in determining HMs (Hallouin et al, 2020;Hernandez-Suarez et al, 2018;Pool et al, 2017) and typical performance metrics for watershed-based models (Moriasi et al, 2007;Moriasi et al, 2015). With our focus on assessing a pool of HMs rather than targeting one or a select few HMs, we choose to first calibrate the model to daily observations and then determine how a well-calibrated daily simulation relates to the suite of predicted HMs.…”