Evaluation of baseflow separation methods with real and synthetic streamflow data from a watershed

Highlights

• Performance of graphical & digital filter baseflow separation methods are compared.

• Methods are compared using real data with a subjective concept of hydrologic plausibility.

• Synthetic data from an integrated hydrologic model used for more rigorous assessments.

• Estimated baseflow hydrographs reveal seasonal variability.

• Filtered UKIH approach works best for most study locations but depends spatially.

Abstract

Baseflow originating primarily from groundwater is a critical streamflow component, although its accurate estimation is fraught with significant difficulties. This study estimates baseflow through existing graphical and digital filter methods, using actual streamflow data from a gauging station at the Alder Creek Watershed (ACW) and synthetic streamflow data at ten study locations within the same watershed simulated with HydroGeoSphere (HGS) (Aquanty Inc., 2018). There are four widely used graphical (Institute for Hydrology, 1980; Sloto and Crouse, 1996; Aksoy et al., 2008) and six digital filtering (Lyne and Hollick, 1979; Chapman and Maxwell, 1996; Furey and Gupta, 2001; Eckhardt, 2005; Tularam and Ilahee, 2008; Aksoy et al., 2009) baseflow separation approaches compared in this study. To determine the most optimal approach, baseflow estimates from real data are assessed based on the subjective concept of hydrologic plausibility, while baseflow estimates obtained from a HGS streamflow record with graphical and digital filtering methods are compared to those computed directly by HGS. Overall, results from this study indicate that baseflow hydrographs reveal a seasonal pattern at the ACW. During wintertime, streamflow is composed almost entirely of baseflow, whereas during summertime, baseflow only consists approximately 20% to 60% of streamflow. After comparing baseflow estimates with those computed by HGS, the most optimal approaches at the ten study locations are assessed. Results show that the best approach at six study locations is the FUKIH (Aksoy et al., 2009) approach, while at three locations, the Chapman and Maxwell (1996) approach and for one location, the Eckhardt (2005) approach performed the best. In conclusion, it is inferred that the most optimal approach within the ACW varies spatially.

Introduction

Baseflow is an essential component of streamflow, which originates primarily from groundwater discharging into streams (e.g., Hall, 1968; Freeze, 1972; Hayashi and Rosenberry, 2002, Eckhardt, 2005, Brodie et al., 2007). Under certain situations, baseflow can also result from delayed flow from surface water features such as wetlands and lakes, as well as through flow regulation and wastewater discharge (Piggott et al., 2005). As an important component of streamflow, baseflow plays an essential role in sustaining riparian and aquatic ecosystems as well as influencing stream water chemistry. In many streams, streamflow is primarily composed of baseflow during dry periods, and during wet seasons, the ratio of baseflow to streamflow could decrease significantly depending on site conditions. For example, in a dry climate, baseflow has been estimated to constitute approximately 80 percent of river flow within the Upper Colorado River Basin (Miller et al., 2016). In the baseflow and water use assessment report by Toronto and Region Conservation (2008), baseflow has been estimated to be approximately 40 percent from the Don River Watershed in a humid climate during summertime. Due to the spatiotemporal variability of baseflow from one watershed to another, predicting the contribution of baseflow relative to streamflow is essential for effective watershed management.

Research on baseflow characteristics is vital for comprehending runoff generation processes and understanding interactions among streamflow, groundwater, and other components of the water cycle. Moreover, investigations on spatial and temporal variability of baseflow could lead to better quantification of the significance of groundwater in streamflow processes (Tong et al., 2021). Baseflow recession has also been studied by researchers to estimate aquifer parameters from streamflow data (e.g., Brutsaert and Nieber, 1977, Troch et al., 2013, Liang et al., 2017).

Due to the importance of accurately estimating baseflow, several studies based on various approaches have been conducted (e.g., Winter et al., 1998, Cey et al., 1998, Kalbus et al., 2006, Rosenberry and LaBaugh, 2008). However, it is difficult to obtain accurate baseflow hydrographs directly and continuously in the field. Therefore, many different approaches have been developed to estimate baseflow based on disparate data available within a watershed.

Baseflow estimation methods can be roughly divided into three groups: direct measurements, tracer-based separation, and non-tracer-based separation methods. Direct measurements rely on different instruments such as bag-type or automated seepage meters, mini-piezometer, heat pulse meter, and ultrasonic meter to measure baseflow values at discrete points (e.g., Kalbus et al., 2006, Rosenberry and LaBaugh, 2008). These instruments are installed to measure water fluxes across the groundwater – surface water interface. While direct measurements are very desirable, they typically only provide point estimates. In addition, direct measurements are always time-consuming and are not viable in making measurements at multiple locations within a watershed and over long-time scales. Such point measurements are also logistically difficult to conduct in large rivers, where access to measurement sites can be limited and the safety of workers can be a concern.

Tracer-based separation methods mainly rely on various isotopic and chemical tracers to explore the generation processes of each water cycle component (e.g., Yu and Schwartz, 1999) by separating streamflow into surface runoff and baseflow. In previous studies, these approaches have been widely used to determine baseflow values, although they are always laborious, have high data and sampling requirements, and cannot be applied to past events due to the lack of required chemical data (Gonzales et al., 2009). Moreover, the assumptions embedded in this approach may not be satisfied. For example, tracer-based methods may contain relatively large uncertainties from chemical reactions during the mixing of components, tracer measurements, and elevation effects on the isotopic composition of precipitation (Gonzales et al., 2009). These uncertainties in chemical reactions could result in tracer concentration changes during water movement through the watershed, thus leading to less reliable baseflow estimation results. Therefore, alternative non-tracer-based methods are needed.

Non-tracer-based baseflow separation methods could be subdivided into several groups, including graphical (e.g., Institute of Hydrology, 1980, Sloto and Crouse, 1996) and digital filter methods (e.g., Lyne and Hollick, 1979, Chapman and Maxwell, 1996, Furey and Gupta, 2001, Aksoy et al., 2009, Eckhardt, 2005, Tularam and Ilahee, 2008). In graphical methods, different criteria are used to separate streamflow into baseflow and surface runoff through the analysis of a streamflow hydrograph. In digital filter methods, numerical approaches are utilized to filter streamflow into different portions of the hydrograph.

In previous research on baseflow estimation, various approaches have been utilized and then compared to evaluate results (e.g., Nathan and McMahon, 1990, Cey et al., 1998, Chapman, 1999, Arnold et al., 2000, Smakhtin, 2001, Conant, 2004, Schwartz, 2007, Gonzales et al., 2009, Indarto et al., 2016, Lott and Stewart, 2016, Xie et al., 2020; Kissel and Schmalz, 2020). For example, Cey et al. (1998) compared four field approaches to measure baseflow values at a small watershed in southern Ontario, Canada. Approaches compared include the use of the velocity-area technique, mini-piezometer measurements, seepage meter measurements, and analyses of electrical conductivity as well as isotope data. Among the first three techniques, the velocity-area technique resulted in best baseflow estimates. From the analyses of isotope and electrical conductivity data, it was shown that during storm events, pre-event water contributed approximately 64 % to 80 % of total stream discharge, and antecedent moisture conditions of the catchment were found to largely affect the percentage of event- and pre-event water in streamflow.

Gonzales et al. (2009) compared various baseflow estimation methods, including both tracer- and non-tracer-based methods in a lowland area of Netherlands. The tracer approach revealed that groundwater responded quickly to rainfall events in this area, and surface water contributed to most of measured discharge during flood events. Moreover, estimated results were compared with baseflow values determined through the tracer-based method. In their study, Gonzales et al. (2009) concluded that the rating curve method and the recursive filtering method proposed by Eckhardt (2005) resulted in reliable baseflow values.

Indarto et al. (2016) reviewed earlier work on baseflow estimation and used seven recursive digital and two graphical methods to streamflow records from a watershed in East Java, Indonesia to determine optimal parameter values, baseflow index, and the appropriate method for the investigated watershed. Results revealed that the exponentially weighted moving average (EWMA) approach (Tularam and Ilahee, 2008), the Lyne and Hollick method (1979), and the local-minimum method (Sloto and Crouse, 1996) performed better in this area.

Xie et al. (2020) estimated baseflow values with four graphical and five digital filter methods for 1,815 catchments in the United States. An evaluation criterion was established to determine the true baseflow, and this evaluation criterion was used together with performance metrics to analyze the accuracy of each separation method. In this evaluation criterion, they selected streamflow values during low flow conditions and treated these streamflow values strictly as baseflow values. Low flow conditions were defined as the condition when quick flow, which includes interflow and overland flow, has ceased in a catchment. Through this evaluation criterion, Xie et al. (2020) concluded that the Eckhardt (2005) method had the best performance across the contiguous United States based on the evaluation results for 1,815 catchments.

In these previous studies, several baseflow separation methods were utilized and evaluated. However, as mentioned previously, baseflow values are notoriously hard to quantify over long-time scales especially over a large study area. Actual baseflow values from a watershed are always absent to help determine the best separation technique. To circumvent this issue, most studies have determined the optimal baseflow separation technique based on the qualitative concept of hydrologic plausibility. In this study, hydrologic plausibility means that features of a given baseflow hydrograph separated from a streamflow record should be consistent with anticipated natural conditions, such as: 1) its less variable nature compared to streamflow; 2) its delayed response relative to interflow and overland flow; and 3) that it does not exceed streamflow or exhibits unusually high increasing/decreasing rates during rainfall events. Emphasis is made that the concept is subjective, as there are no quantitative metrics that one can rely on, thus could potentially lead to biased results.

Some researchers have employed fully integrated three-dimensional surface water/groundwater physical models under varying hydrological conditions to simulate baseflow, and the synthetic baseflow were assumed to be the true baseflow to test various baseflow separation techniques (e.g., Partington et al., 2012, Li et al., 2014, Su et al., 2016). For example, Partington et al. (2012) used four approaches to estimate baseflow including the HYSEP approach (Sloto and Crouse, 1996), the PART program (Rutledge, 1998) constructed based on a graphical approach, the BFLOW program (Arnold and Allen, 1999) constructed based on the Lyne and Hollick (1979) approach, and the Eckhardt approach (2005). To test the performance of each approach, HydroGeoSphere (HGS) (Aquanty Inc. 2018) in conjunction with a hydraulic mixing-cell (HMC) approach were used to obtain the synthetic, true baseflow values for a simple, monotonically sloping V-shaped catchment. Partington et al. (2012) found that the performance of different baseflow estimation approaches varied under eight different scenarios with different hydrological conditions, but overall, the HYSEP sliding-interval approach showed the best results in most scenarios for this study.

Similar to the work of Partington et al., 2012, Li et al., 2014 used synthetic results from HGS for a simple V-shaped catchment to test the accuracies of several recursive digital filters (Lyne and Hollick, 1979, Chapman and Maxwell, 1996, Boughton, 1993, Chapman, 1999, Eckhardt, 2005). Results showed that baseflow estimates obtained through the Lyne and Hollick filter could better match the HGS synthetic baseflow under a wider range of catchment hydrological characteristics, and optimal parameters varied based on hydrological conditions.

Su et al. (2016) investigated the utility of hydrological signatures to calibrate the Eckhardt filter method (Eckhardt, 2005) and tested seven possible hydrological signatures of baseflow, comparing against the synthetic baseflow values simulated with HGS by Li et al. (2014) again for a tilted V-shaped catchment. Results showed that the Eckhardt filter had better performance after a hydrological signature-based calibration.

In these previous studies (Partington et al., 2012, Li et al., 2014; and Su et al., 2016), HGS was used to simulate baseflow with the HMC approach to help evaluate the performances of baseflow separation approaches. Although the use of synthetic baseflow from a model solved the problem of obtaining true baseflow estimates from an actual site, synthetic baseflow used in these studies was not simulated for a real watershed. Instead, the analysis was conducted based on a monotonically sloping V-shaped catchment, that simplified the intricate environmental conditions in actual watersheds subjected to seasonal hydrologic variations.

The primary purpose of this study is to assess baseflow estimates obtained from streamflow data at the Alder Creek Watershed (ACW) in southern Ontario, Canada, using various baseflow separation techniques. Baseflow separation is conducted through ten different approaches including four graphical and six digital filter approaches. Graphical methods include: the 1) United Kingdom Institute of Hydrology (UKIH) method (Institute of Hydrology, 1980, Aksoy et al., 2008); 2) three hydrograph-separation (HYSEP) methods, which are fixed-interval (HYSEP1), sliding-interval (HYSEP2), and local-minimum (HYSEP3) methods (Sloto and Crouse, 1996). Digital filter methods include the: 1) Lyne and Hollick (1979); 2) Filtered United Kingdom Institute of Hydrology (FUKIH) (Aksoy et al., 2009); 3) Chapman and Maxwell (1996); 4) Eckhardt (2005); 5) Furey and Gupta (2001); and the 6) EWMA (Tularam and Ilahee, 2008) approaches. Details to each of these approaches are provided in the Supplementary Information (SI) section.

Baseflow estimates obtained through ten approaches using actual streamflow data from a real streamflow gauge installed within the ACW are first compared and assessed utilizing the qualitative concept of hydrologic plausibility. For a more quantitative comparison, actual baseflow estimates are necessary. However, as actual baseflow estimates from the ACW are not available to properly evaluate the performance of these ten baseflow separation techniques, synthetic streamflow and baseflow data obtained from HGS are assumed to be true data and utilized for a more rigorous comparison to determine the most optimal approach for the ACW. Unlike the simple V-shape catchment models used by Partington et al. (2012) and Li et al. (2014), actual hydrological and geological conditions of the ACW are simulated with HGS that rigorously considers the coupling of surface water, groundwater flow, and other hydrological conditions for the ACW (Tong et al. 2021). Although numerical models may not be able to provide actual baseflow values for a given site, a fully 3D integrated hydrological model should still generate good independent conceptualization of watershed flow dynamics under different conditions until better baseflow estimation tools or observation techniques are developed (Partington et al., 2012).