You are on page 1of 12

Environmental Modelling & Software 66 (2015) 57e68

Contents lists available at ScienceDirect

Environmental Modelling & Software
journal homepage:

Assessment of the internal dynamics of the Australian Water Balance
Model under different calibration regimes
Li Li a, 1, Martin F. Lambert a, *, Holger R. Maier a, 2, Daniel Partington b, 3,
Craig T. Simmons b, 4
School of Civil, Environmental and Mining Engineering, The University of Adelaide, North Terrace, Adelaide, South Australia 5005, Australia
National Centre for Groundwater Research and Training and School of the Environment, Flinders University, GPO Box 2100, Adelaide,
South Australia 5001, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Conceptual rainfall runoff models are used extensively in practice, as they provide a good balance be-
Received 1 August 2014 tween transparency and computational and data requirements. However, the degree to which they are
Received in revised form able to represent underlying physical processes is poorly understood. This is because the performance of
14 December 2014
such models is generally assessed based on their ability to match total streamflow, rather than
Accepted 15 December 2014
Available online 13 January 2015
component processes. In this paper, the ability of the Australian Water Balance Model (AWBM) to
represent baseflow and quickflow is assessed for 66 synthetic catchments with different physical char-
acteristics and hydrological inputs under seven calibration regimes utilising a shuffled complex evolution
Australian Water Balance Model
(SCE) algorithm. The “observed” total-, base- and quick-flow hydrographs for these catchments are
Calibration generated using HydroGeoSphere. The results indicate that while AWBM is generally able to match total
Baseflow streamflow well, the same does not apply to baseflow and quickflow, suggesting that these processes are
Regression models not represented well by AWBM.
HydroGeoSphere © 2014 Elsevier Ltd. All rights reserved.
Virtual laboratory
Integrated surface water/groundwater

1. Introduction Black box modelling approaches, such as artificial neural net-
works (Abrahart et al., 2012; Maier et al., 2010; Wu et al., 2014), are
Modelling approaches for estimating runoff from rainfall and at one end of the spectrum, while physically based approaches are
evapotranspiration (ET) can be traditionally classified into three at the other end. Black box models produce streamflow outputs
main groups: black box models, physical process based models and solely as a function of their inputs and transfer characteristics,
conceptual rainfall runoff (CRR) models (Beven, 2005). While all of without any knowledge or understanding of the underlying phys-
these approaches have been shown to be able to predict total ical processes. However, they are generally computationally effi-
streamflow successfully, the degree to which they are able to cient and can be developed using limited data. Physically based
represent underlying streamflow generating mechanisms is highly approaches attempt to simulate the detailed mechanisms of the
variable (Chen and Adams, 2006; Ferket et al., 2010; Refsgaard and component physical processes within the hydrologic cycle using
Knudsen, 1996). well-established physical laws, with numerical solutions of the
mathematical representation of these processes (Jayatilaka et al.,
1998). Such approaches include fully integrated surface water/
groundwater (SW/GW) models, such as InHM (VanderKwaak and
* Corresponding author. Tel.: þ61 (0)8 8313 4320; fax: þ61 (0)8 83034359.
E-mail addresses: (L. Li), martin.lambert@adelaide. Loague, 2001), MODHMS (HydroGeoLogic, 2000), HydroGeo- (M.F. Lambert), (H.R. Maier), daniel. Sphere (HGS) (Therrien et al., 2009) and SHE (Abbott et al., 1986). (D. Partington), However, there are problems with the application of these models
(C.T. Simmons). in practice due to the difficulties and expense associated with
Tel.: þ61 0406745484; fax: þ61 (0)8 83034359.
2 obtaining the data required (e.g. due to limitations of existing
Tel.: þ61 (0)8 8313 4139; fax: þ61 (0)8 83034359.
Tel.: þ61 (0)8 8201 2648; fax: þ61 8 8201 7906. instrumentation and intrinsic uncertainty in measurements), as
Tel.: þ61 (0)8 8201 5509; fax: þ61 8 8201 7906. well as their high computational demands.
1364-8152/© 2014 Elsevier Ltd. All rights reserved.

Finally. While the methodology is Among the different rainfallerunoff modelling approaches illustrated for a particular case study. As shown. 2012. It mance of CRR models using total streamflow time series (Ferket should be noted that while the AWBM is used as the CRR model in this study. baseflow quickflow. such as baseflow and means of quantifying the absolute volume of the flow components (e. A methodology is given in Section 2. They conclude that no clear picture emerges of which processes explicitly. As part of the the lack of transparency of black-box models. AWBMs are developed for the same catchments by using the same cesses are modelled accurately. The parameters and production of generally acceptable results. Details of each step in the methodology are given in subsequent sections. very few attempts have been made to use baseflow or quickflow estimates for CRR model internal dynamic performance 2. baseflow and quickflow. due to the difficulty of accurately measuring baseflow The 66 synthetic catchments with different physical characteristics and hydro- or quickflow in the field (Dukic. they provide the best number of additional states and fluxes. Next.58 L. 1999). These catchments have drainage Recently. A unique feature is that multi-objective optimisation is used (e. such as hydraulic benchmark against which the internal dynamics of CRR models can be assessed (e. as they represent the assumed underlying that the MWARPE calibration algorithm and the HBV model lead to physical processes in a conceptual manner. into. CRR models are the most widely utilised in could easily be adapted and applied to other CRR models around practice.. Li et al.. for example. by comparing the rigorous representation of the underlying physical processes of hydrologic systems (Brookfield et al. 2007. whether flow components that make up total streamflow are predicted accu- While there have been many studies comparing the perfor. quick-flow hydrograph prediction. Sulis et al. similar total stream.and series to be useful in water resources assessments. such as assessing (i) how well the Australian Water Balance Model. Therefore. 1991. 1982. SIMHYD (Chiew et al. which are measurable in theory. without consideration of the appropri- providing a means of assessing the performance of the internal dynamics of AWBM ateness of their quantities and dynamics. This is a reasonable assumption. Ranatunga et al. two optimisation algorithms (SCE-UA and MWARPE) are computationally efficient and less data intensive than process used to calibrate the models by matching total streamflow to ob- based models... While it is acknowledged that fully integrated SW/GW models are in themselves total streamflow time series. at least when compared with the results obtained using the mathematical equations to conceptualise the movement of water Furey and Gupta filter (Furey and Gupta.. 2013. 2014). or an surface flow with the Richards' equations.. modelled total streamflow time series with the corresponding Therrien and Sudicky. 2008. 2006. due to their relatively simple structure. 2009. in a directed towards obtaining a well-defined optimal parameter set. as fully integrated SW/GW models provide a rameters are generally estimated by calibration. 1985). 2010. Ahmadi et al. its generic nature means it mentioned above. they can be used to obtain estimates of different flow components for catchments with different characteristics (see Partington et al. Troutman. These models have been able to different calibration regimes that take internal model dynamics capture enough of the dynamics of rainfall runoff simulation time into account in different ways on the accuracy of total-. The remainder of this paper is organised as follows. obtain total streamflow. discussion in Section 3 and summary and conclusions in Section 4. 2005). / Environmental Modelling & Software 66 (2015) 57e68 CRR models represent a compromise between the high data and the internal dynamics of two CRR models (HBV and PDM) for a computational requirements of physical process based models and subcatchment of the Dender catchment in Belgium. has been obtained.1. base... McCallum et al. Venant equations. the performance of flow time series can be obtained with very different combinations the AWBMs calibrated using the different methods is compared in terms of the ability to predict total-. difficult to measure in practice. they are more transparent than model produces the best results of simulating total streamflow. there may be many combinations of parameter values and quickflow) currently available (see Partington et al. and is caused by problems such are able to provide the first step towards being able to assess the internal dynamics as over-parameterisation of models. even though the structure of CRR The steps in the methodology adopted for assessing the internal dynamic per- models is based on a conceptual representation of underlying formance of AWBM are given in Fig. 2014. as a good match to total properties and hydrological inputs in order to ensure the results are as generic as streamflow does not necessarily mean that the component pro. 1. conductivity and porosity. physically realistic fashion (Therrien et al. 2012. generally in the form of the best baseflow estimates.g. are usually baseflow (see Partington et al. quickflow and of the catchment (e.. evaporation. rately). lakes and streams). although some The underlying premise of the proposed methodology for assessing the internal of the CRR model parameters. including overland flow. the Soil Moisture and Accounting Model (SMAR) (Tuteja flow for 66 synthetic catchments with different catchment char- and Cunnane. All of the governing flow equations including local-type direct search optimisation methods and implemented by fully integrated SW/GW models are solved simultaneously to globally based optimisation methods (Duan et al. 1993). giving the best internal model dy- a number of interconnected storages that are linked with empirical namics. infiltration and recharge. that give similar objective function values. 2001). base.. as they do not attempt to represent all physical servations. same approach should also be used to test the internal dynamic performance of other CRR models. followed by the results and common feature of CRR models is that some of their model pa. base. 2012. there are some physical parameters. 1993. They are more study.and quick-flow hydrographs accurately. They typically represent 3D variably saturated sub- observed data until an acceptable fit to the objective function. 2010).g. 1992).. hydrological inputs. between the storages of and out of a catchment. 2014). 2009). while internally they calculate a an approximation of the actual processes in real catchments.. a commonly used CRR. streamflow. However. making them ideal candidates for because CRR models are usually calibrated using only observed assessing the internal dynamics of CRR models. Methodology physical processes are lumped together. but whether flow components that make up total streamflow are predicted accurately). as shown in .g.and quick- 2004). 1996). Significant research effort has been as subsurface discharge to surface water features (e. initially synthetic total streamflow physical processes.. the V-catchment test case. small number of the world. is able to represent total-.. catchment area. Furman. such CRR model pa. but different calibration methods.. Li et al. which is the Australian Water Balance Model (AWBM) (Boughton. For example. (e. how well these processes are represented by (qobs obs obs T ). In addition. Therefore they called ‘equifinality’ (Beven.. Gibbs such models can simulate the partitioning of rainfall into different components. the et al. As a result.g. 2001) to validate surfaceesubsurface hydrology problem. baseflow (qB ) and quickflow (qQ ) hydrographs are generated using a fully integrated SW/GW model for a number of catchments with different physical calibrated models is generally unknown. as well et al. Many This study builds on the research by Ferket et al. possible. data limitations and structural of CRR models under a range of physical conditions in a controlled manner. Ferket et al. due to the fact that many complex catchment 2. representing the physical properties dynamics of CRR models is that fully integrated SW/GW models can be used to obtain reasonably accurate estimates of actual total streamflow. Post et al.g. Catchment characteristics and hydrological inputs assessment. Knapp et al. faults in the model. 2008). (2014) are used. 2002) and GR4J acteristics and hydrological inputs and (ii) the impact of seven (Oudin et al. 2013). surface slope). rameters have limited physical interpretation (Delleur. and 1D and 2D surface flow with the acceptable trade-off between objective functions in cases where diffusion wave approximation to the St.g. thereby of baseflow and quickflow.. 2013. This phenomenon is In addition. logical inputs developed by Li et al. However.. (2010) by different CRR models have been proposed in the literature. 2012). Li et al. Partington et al. thereby providing a measurable. (2010) use baseflow estimated from a areas ranging from 6 to 192 km2 and are loosely based on a benchmarked integrated physically-based digital filter (Furey and Gupta. 2010.. but black-box models.

. fully integrated SW/GW models are used for assessing the in- physical catchment characteristics are represented using seven variables: catchment ternal dynamic performance of AWBM under a range of calibration approaches. Fully integrated SW/GW model be found in Li et al. SW/GW disconnection problems (Banks et al. Partington et al.2.. 2. which are assumed to represent characteristics considered are given in Table 1 and the characteristics of the hy- the “true” catchment physical processes. catchment hill slope (S1. Details of the different values of the physical catchment catchment characteristics and hydrological inputs. which includes Ks and van Genuchten parameters a and b. 2012). 2009). Schematic representation of tilted V-catchment flow problem (adopted from Panday and Huyakorn. Brunner et al. This area (A). 2011. 2004).. catchment is because they provide the best possible approximation to the physical processes of channel slope (S2. 1. and water flow within catchments and can therefore be used as an approximation to soil type. such as the characteristics and hydrological inputs. 2 (Panday and Huyakorn. Fig. 2004). from these catchment characteristics and hydrological inputs.. which is also used in this study by sampling comparison of baseflow estimation methods (Li et al. This is because HGS can be used to simulate drological inputs used are given in Table 2. Different As stated above. (2014) used Latin Hypercube hydrological processes within catchments in a physically based manner (Therrien Sampling (LHS) to generate the 66 catchments with different physical catchment et al. which are obtained from the Australian Bureau of Meteorology the 66 synthetic catchments' response to rainfall and ET inputs under different National Climate Centre. L. In this inputs are represented using the ratio of daily rainfall to ET from five Australian cities study. / Environmental Modelling & Software 66 (2015) 57e68 59 Fig. 2011. Li et al. thus only a brief overview is provided here. (2014). which is perpendicular to the channel). which is parallel to the channel). A detailed description of these catchments can 2. . Fig.. Li et al. Hydrogeosphere (HGS) is used as the fully integrated SW/GW model to model (Li et al. 2013. Schematic representation of steps in the proposed methodology. HGS has been applied successfully to various studies.. catchment aspect ratio (AR). The hydrological such processes subject to a variety of physical characteristics and forcings. 2009). 2014).

although a daily time step is used in this study. The method uses the fluid mass balance from ability to account for baseflow when predicting streamflow by using a baseflow the fully integrated SW/GW model of each node at each model time step to calculate index (BFI). Catchment characteristic Unit Explanation Values considered Ks m/s Saturated hydraulic conductivity 2.1. respectively. Li et al. it becomes runoff and is divided between surface runoff fluxes are correspondingly half of those expected when accounting for both sides of (quickflow) and baseflow. The AWBM is a typical lumped CRR model.. 2. configuration of three different surface storages.078 Darwin 014015 1707.008.4.99  1005. it is a reasonable assumption that it should be able to reproduce As mentioned previously.. affect the separation of rainfall excess into quickflow/baseflow storage and the baseflow component of total streamflow. the fraction is determined using the modified mixing rule (Campana routed surface runoff recession constant (KSurf) are used to describe the daily and Simpson. 2. which is the ratio of the amount of baseflow to the total amount of the fraction of water in the cell that comes from different streamflow generation streamflow and determines the proportion of excess moisture at each time step that mechanisms (e. streamflow (qobs obs obs T ). model dynamics into account in different ways is investigated in this study. surface attenuation store. When combining the HMC capacity variation over the catchment is described by the combination of the surface method with the fully integrated SW/GW model. fjðkÞ 2. As a result.161.2 catchments (see Partington et al.029. 8. Moisture graphs (i. As the aim of the AWBM is to represent the underlying the HGS parameters can be found in Li et al.11  1004. 2. If there is example of the catchment configurations considered in Fig. rainfall is all simulations are conducted for only half of the catchment. 192 S1 e Hill slope (perpendicular to the channel) 0. Australian Water Balance Model (AWBM) which is referred to as quickflow in this study. 0. baseflow and quickflow hydrographs). C2 tained using the integrated SW/GW model.793.. City Gauge Average annual rainfall Rainfall data period for Average annual potential ET R/ET (average annual rainfall/average annual no.75 b e van Genuchten parameter 1.270 A km2 Catchment area 6. 1. 0.004. for modelling rainfallerunoff relation- permeability systems (Schwartz et al. At each time step. 2010).25. a calibration approach that does not take internal ical processes used to describe the movement of water into and out of storages. 2011.01 AR e Ratio of catchment width to length (x/y) 0. 48. 120. 2013). A number of software frameworks have been developed for hydrological modelling and are in active use.12  1004. The quickflow (qobs Q ) is taken as the difference between the total streamflow (qobs obs T ) and baseflow (qB ).556. A2 and A3.16 1999e2009 2056. Further description of the code and its ships. and quick-flow hydrographs. and the reported any excess from any store. Details of the optimisation draulic information and is able to capture the storage effects and time lags within method and error measure used in the calibration methods are given Sections 2. C2 and C3.0025. (2011) as: AWBM parameters and their ranges used is given in Table 3. 0. In this study.97 0.3..e. as well Values of all of the eight parameters listed in Table 3 are obtained as part of the as the corresponding simulated total streamflow and baseflow hydrographs ob- different calibration methods investigated. 0.g.. 2013). with each surface storage.4. 2014).005. 4.02 S2 e Channel slope (along the channel) 0. ij denoting volume into cell j from cell i over As stated previously. model dynamics into account in different ways on the ability to predict total-. 2004. The baseflow hydrograph (qobs B ) is extracted from the model using the uses rainfall and actual evapotranspiration as inputs and principally consists of a Hydraulic Mixing-Cell (HMC) method (Partington et al. baseflow from fully integrated SW/GW models.008. 0 1  Pn N such as the Rainfall Runoff Library (RRL) and hydromad (Hydrological Model BV N1 Pm V N C  N1 j¼1 Vji N1 fjðkÞ Assessment and Development) package (Andrews et al.16 1984e1994 1470. V denotes the volume with the superscript denoting time state and subscript i denoting the cell. each node in the surface domain of storages and the related fractional areas. (2014).0. 1. steps. 1.44  1005.4. 2.3.70  1004 a e van Genuchten parameter 2. and KSurf affects the routed surface runoff. It also adopts the structure of transferring a fraction of the generated runoff numerical formulation can be found in Therrien et al. The depths of these storages The HMC method developed by Partington et al.811 .22 1997e2007 2077 1. A summary of the tion mechanism k at time N in cell i is given by Partington et al. 2014). correspond to the parameters C1.5 bank storage dynamic processes analysis (Doble et al.90 2012) and the study of dual and spatial variability of the abstractions. 0. calibration methods are summarised in Section 2. groundwater discharge to the stream). AWBM N B i j¼1 ij N1 C N1 fiðkÞ ¼B N  Cf þ @ V N Vi A iðkÞ Vi N is implemented using the RRL developed by the Cooperative Research Centre on i Catchment Hydrology (www. As can be seen.4.33 0.4. representing the impacts of antecedent wetness (Method 1). C1.576 Sydney 070062 1095.1.190 Brisbane 031011 2238. 2011). 3. Calibration methods The dataset of the 66 catchment characteristics and hydrological inputs. 3. physical processes. the HGS models are used to obtain the ‘observed’ the dynamics of the rainfallerunoff relationships generated synthetically using HGS.33 1968e1978 911. base- with interconnected storages and algorithms that mimic the underlying hydrolog. baseflow (qB ) and quickflow (qQ ) hydrographs for the 66 The detailed structure of the AWBM is given in Fig. the catchment used in this study is symmetrical. 1984). 5.. Calibration of Australian Water Balance Model (AWBM) in the neighbouring cell j.572.. The equation for each fraction f for each streamflow genera. model dynamics into account explicitly is used to provide a basis of comparison AWBM uses three surface stores. / Environmental Modelling & Software 66 (2015) 57e68 Table 1 Catchment characteristics considered (adopted from Li et al..toolkit. 3. and total streamflow is the sum of surface runoff and the stream. 2.347 Melbourne 086071 525.32. which provides a means of estimating and 2. A fraction of the total area is associated poral information of flow generation mechanisms to obtain the component hydro.60 L. 9. 2004) and is now one of the most widely used CRR models in Australia (Marshall In order to be able to assess the impact of calibration methods that take internal et al. 2004). the AWBM catchments.016. 1. Details of the calibration methods are The AWBM is a saturation overland flow model developed by Boughton (1993.5. 1. given in Table 4 and Fig. 11.370. N1 denotes fraction k at time N1 Where there are n sources and m sinks for cell i.012. 0. AWBM can be operated at either daily or hourly time As shown in Fig. discharge from baseflow and surface runoff (quickflow) storage. as shown for one added to each of the surface stores and evapotranspiration is subtracted. Detailed information of the catchment model with the selected values of baseflow (Boughton. as represented by the parameters A1. 0. can be downloaded as supplementary and C3) are related to the moisture capacity of the catchment. (2011) utilises the spatiotem.. 6. The daily baseflow recession constant (KBase) and daily streamflow.955.006. (mm/a) sampling (mm/a) potential ET) Adelaide 023011 510. the effectiveness of calibration methods that take internal the time step from N1 to N and ji denoting volume from neighbour j into i. (2009) and Brunner and direct to the baseflow store at the same time as the residual is transferred to the Simmons (2012). 0. An important feature of AWBM is the the model is treated as a mixing-cell. 1. 80. Ranatunga et al. 0. Five of these parameters (A1. For each component of is returned to the baseflow. The The HMC method extracts streamflow generation mechanisms using only hy. A2. 2008).37 0. two (BFI and KBase) material. Table 2 Hydrological inputs considered (adopted from Li et al.85 1982e1992 920.75.

in which BFI and KBase are calibrated to baseflow extracted from . shown in Fig. a two-step process is adopted as follows: method. In the second step. Fig. L. 2014). Method 1: of total streamflow. / Environmental Modelling & Software 66 (2015) 57e68 61 Fig. 3. Li et al. 5. an attempt is made to obtain the and that obtained using HGS (qobs T ) using the selected optimisation method. Consequently. 4. BFI and KBase. as best possible match to the baseflow hydrograph. In each (qsim obs T ) and that obtained using HGS (qT ) using the selected optimisation of these methods. In the first step. are The methods used for obtaining values of BFI and KBase for Methods 2e7 from estimated. 1. Methods 2e4 belong to of rainfall excess into quickflow/baseflow storage and the baseflow component the first category. BFI and KBase are selected for this step as they control the separation the above step 1 can be divided into three broad categories. all parameters are calibrated simultaneously so as to minimise the a direct effect on the baseflow hydrograph produced by AWBM. values of two of the model parameters. The aim of this calibration method is to provide the best possible fit 2. BFI and KBase are fixed at the values obtained in the first step to total streamflow. and the remaining parameters are calibrated simultaneously so as to minimise As can be seen from Fig. without any consideration of internal model dynamics. six different methods (Methods 2e7) are used. selected error measure between the total streamflow obtained using AWBM (qsim T ) by estimating values of these parameters first. 5. and are therefore the two parameters that have In Method 1. Structure of AWBM.. respectively. in order to take internal model dynamics into account the selected error measure between the total streamflow obtained using AWBM explicitly during calibration. An example of the 3D catchment model for the case study (adopted from Li et al.

and quick-flow hydrographs obs LHð0:925Þ produced by HGS. inputs than the corresponding filter parameter in the Boughton filter (Li et al. / Environmental Modelling & Software 66 (2015) 57e68 Table 3 Method 5: AWBM parameter description and ranges.. obs LHðoptÞ the “observed” baseflow is given by (qB ). (2014) used in Methods 6 and 7. a modified version of the procedure for improving baseflow esti- mation using RDFs suggested by Li et al. 4). except that Method 4 (Eckhart filter) is used RDFs. or by regression. 5 Estimate BFI and KBase based on regression relationships with catchment characteristics and hydrological inputs and then calibrate remaining parameters to total streamflow. BFI and KSurf are calibrated simultaneously so as to minimise the selected error Method 3: measure between the quickflow obtained using AWBM (qsim Q ) and that obtained Method 3 is identical to Method 2. KSurf.62 L. values of BFI and KBase are obtained directly (i. 2005) Model calibration is conducted using the shuffled complex evolution (SCE-UA) obs EckðoptÞ is used to obtain the hydrograph of “observed” baseflow (qB ). When people have actually tried to measure baseflow (although in (Benchmark 2). A1a Partial area 0e1 similar to the regression relationships for predicting filter performance and optimal A2a Partial area 0e1 filter parameters developed by Li et al.2. First. which determines the magnitude of the baseflow component within total an estimate of the optimal filter parameter as a function of catchment characteristics streamflow (Fig. use Method 3.e. the remaining parameters are set to the optimal values obtained Benchmark 2 (Quickflow): using Method 1. If the performance of the LH filter is comprised solely of baseflow. in which BFI predicted to be acceptable.0. as in Methods 2e4. including genetic algorithms (GAs) (Goldberg. 1999.0. accuracy of the RDF. 7 Identical to Method 6. 4 First calibrate BFI and KBase to baseflow obtained using Eckhart filter with optimal filter parameters and then calibrate remaining parameters to total streamflow. the regression total HGS streamflow using recursive digital filters (RDFs). Table 4 Summary of calibration methods.. However. In addition. Calibration Details method 1 Calibrate all parameters simultaneously to total streamflow. Method 3 is used to obtain estimates of BFI and KBase. Method 5 belongs to the second category. if this is not the case.9e0. Murphy et al. (2014) is used to determine whether the LH filter which determines the magnitude of the quickflow component within total can provide satisfactory estimates of baseflow for a particular catchment based on streamflow (Fig. 2004. in Method 7: which BFI and KBase are either estimated by calibration to baseflow extracted using Method 7 is identical to Method 6. Li et al.925. istics and hydrological inputs. the parameters that affect baseflow B ) and that obtained using the Lyne and Hollick (LH) RDF (Nathan and McMahon. (2014). 2014). without cali- bration) by developing regression models predicting KBase and BFI as a function of Parameter Description Parameter range the catchment characteristics (Table 1) and hydrologic inputs (Table 2) investigated. 0.98 are found to be reasonable for real catchments. as detailed below: only a few catchments). (2014) to determine whether the LH filter can provide satisfactory estimates of baseflow. 2 First calibrate BFI and KBase to baseflow obtained using LH filter with default filter parameter (0. based on the predicted instead of Method 3 (LH Filter).4. mated using the regression relationship developed by Li et al. except that the Eckhart filter (Eckhardt. 3 First calibrate BFI and KBase to baseflow obtained using LH filter with optimal filter parameters and then calibrate remaining parameters to total streamflow. estimated in step 1 of the procedure for obtaining the quickflow benchmark searchers since. The SCE-UA algorithm is based on the strengths of that can be estimated more easily from catchment characteristics and hydrological several existing search procedures. 2009). Details of the different methods for estimating BFI and KBase in In order to be able to assess the absolute performance of the different calibration the first step of the two-step process described above are given below. only difference is in the way the parameters affecting baseflow (i. so that the total streamflow produced by AWBM is and hydrological inputs. comprised solely of quickflow. . relationship developed by Li et al. using the selected optimisation method. Methods 6 and 7 belong to the third category. The resulting C3 Surface storage capacity 0e500 regression equations and scatter plots are shown in Fig. is set as 1. In order to do this. 4). Values of BFI BFI Baseflow index 0e1 are calculated as the ratio of the volume of baseflow to the volume of total C1 Surface storage capacity 0e50 streamflow obtained from HGS and KBase is obtained by performing recession C2 Surface storage capacity 0e200 analysis on the total streamflow hydrographs obtained from HGS. In Method 5. using the selected optimisation method.and quick-flow processes.925 is adopted are estimated in step 1 of the procedure for obtaining the baseflow benchmark by Nathan and McMahon (1990) based on comparison of the results of this filter (Benchmark 1) and the way the parameters affecting quickflow (BFI and KSurf) are with manual methods.e. and KBase are estimated using regression relationships with catchment character. Consequently. Franchini et al. If not. except that Method 4 is used instead of Method 3. because it has been proven to be both accurate and efficient in previous the LH filter.and quick-flow hydrographs.925 (qB ).. which provides KBase. BFI can be KBase Daily baseflow recession constant 0e1 predicted very well and KBase can be predicted reasonably well. 2014. using the selected optimisation method. (2014) is used in order to improve the baseflow estimates obtained using the LH filter in Method 5. Method 4: 2. rather than using the default value of 0. a value of the LH filter parameter of 0.98 is found to produce Benchmark 1 (Baseflow): a better fit (CSIRO and SKM. Optimisation method Method 4 is identical to Method 3. 6 Use the regression relationship developed by Li et al. use calibration Method 5. so that the total streamflow produced by AWBM is catchment properties and hydrological inputs.and quick-flow hydrographs Method 2: accurately. using Method 2. which has been recommended for the purposes of being one of the best search procedures for use in CRR modelling applications (Ajami estimating BFI for AWBM by Boughton (2004). as in Method 5. As can be seen..925 is the most measure between the baseflow obtained using AWBM (qsim B ) and that obtained with commonly accepted value in most published studies. 1998). and to be able to assess the degree to which the structure of AWBM is In Method 2. The LH filter is used to obtain the baseflow hydrograph as it is calibration methods. two benchmarks are developed. rather than algorithm. 1990). a A1 þ A2 must be less than or equal to 1. 6.. with and quickflow are calibrated to the “observed” base. 2001). Li et al.925) and then calibrate remaining parameters to total streamflow.0. In order to do this. except that the LH filter parameter is esti. using HGS (qqobs HGS ). which are also used to assess the performance of the different the filter parameter set to its default value of 0. Method 6: In Method 6. SCE-UA is widely recognised as Boughton filter (Boughton. BFI and KBase) It should be noted that the LH filter with a filter parameter value of 0. but has a filter parameter (BFImax) et al. 1989) and the Nelder & Mead Simplex downhill search scheme (Nelder and Mead. A two-step calibration process similar to that used for Methods one of the most widely used RDFs and has been found to give good results in a 2e4 is used in order to obtain the benchmark base. During the calibration process HGS (qBobs HGS ). BFI and KBase are calibrated simultaneously so as to minimise the able to represent base. If so. Although LH filter parameter values in the range BFI and KBase are calibrated simultaneously so as to minimise the selected error of 0. methods in terms of their ability to predict base. and this value has been widely adopted by a number of re. is set as 1. As selected error measure between the baseflow obtained using AWBM (qsim part of the benchmark development process. Method 5 is used. 2010). 1993). These results also KSurf Daily surface flow recession constant 0e1 indicate that BFI and KBase are closely related to catchment characteristics and hydrological inputs. The number of case studies (Arnold and Allen.. The Eckhart filter is used as it is mathematically identical to the studies (Hapuarachchi et al.

au/rrl). equifinality is a potential problem and to ensure near globally optimal solutions are obtained. Initial testing dem- gorithm. and therefore. Summary of AWBM calibration methods. daily streamflow.. A The length of streamflow data available for AWBM calibration is 10 years (Li detailed description of this method can be found in Duan et al. the models to account for residual water in the catchment by partially filling their tion parameters to reduce the chance of premature termination of the search In this study. Nonlinear regression models for the prediction of BFI and KBase. The ranges of all of the et Ef values are calculated by comparing the difference Fig. in which the The choice of an appropriate error measure is very important (e. Chrono- recommended values in Duan et al. the warm up period directly precedes the calibration period. but also introduces the concept of complex shuffling (Duan et al. each calibration run is repeated ten times. 2014). In order to check whether parameter logically. (1992).4.toolkit. storages. 5. . 6. 1965). which enables toolkit. is calibrated against In this study. The calibration period requires a warm up period. 2013).3. the NasheSutcliffe coefficient (Ef) (Nash and Sutcliffe. see Bennett algorithm searches for the optimal parameter combination. (1994).. which primes the models for the calibration period. / Environmental Modelling & Software 66 (2015) 57e68 63 Fig.. AWBM runs on a daily time step. The number of complexes is set equal to the number of calibra. Li et al.g. SCE-UA is implemented using the Rainfall Runoff Library (www. measures in hydrology. as it is one of the most highly used performance Library (www. All of the other parameters are set to the onstrates that a 1 year warm up period is sufficient for model calibration. as suggested by Kuczera (1997). Another important aspect in the 2. L. Error measure application of SCE-UA is that a parameter space needs to be defined. 1992). eight AWBM parameters used are based on the suggestions in the Rainfall Runoff 1970) is chosen for this purpose. et and are listed in Table 3.

as discussed previously.g. / Environmental Modelling & Software 66 (2015) 57e68 between the ‘observed’ (e. However.64 L. whereas the results with parameters that result in poorer internal model dynamics (e. As can be seen in Fig. for baseflow.5). the pattern of baseflow obtained with param- eter set 1 is very similar to the pattern of quickflow obtained with parameter set 2). in the practical application of AWBM.e. the parameters of Set 2 can be avoided. 4. as indicated by significantly better matches to the base. Ef values between 0. In Fig.0 indicating a model that can make predication of the same Overall. This is because there are two distinct sets of model parameters that result in similar model performance in terms of total streamflow during the 10 calibration trials conducted for some of the 66 catchments. Performance of AWBMs for the different calibration methods investigated.8 result in poor model performance (Fig. . 3. Ef values between 0. similar total streamflow can be ob- tained by exchanging parameter values for KBase and KSurf. Gupta and Kling (2011) indicate that high values of Ef can give poor model performance. The per- formance of models developed using the different calibration methods is assessed using Ef and by visual inspection. the internal dynamics are much better when parameter set 1 is used. there is no control on internal model dy- namics and the only objective is to find a set of model parameters that provides the best match to the “observed” total streamflow hydrograph. The ability of AWBM to represent quickflow inspected following optimisation. An exact criterion of ranges of Ef values using all calibration methods investigated. the performance of the model can be judged as satisfactory when Ef values catchments (54%e68%) “good” performance is obtained.4. although more subjective than the use of statistical catchments and “good” performance for fewer than 20% of the measures of goodness-of-fit. (qobs B and qobs Q )) (Fig. baseflow and quickflow from AWBM. with “poor” model indicting model performance does not exist in the literature. This is because experienced modellers would usually subjectively assess the values Fig. Consequently. For most of the indicate ‘acceptable’ model performance. cedure (i.5 show ‘acceptable’ model per.g.g. the performance of AWBM models calibrated using the different methods outlined in Section 2. the AWBM total streamflow is the sum of routed surface runoff and baseflow.7e0. as observed in the calibration re- sults for Method 1. as shown in Fig. 7(b)). 8(b)) are represented by Set 2.0 and 0. This is because when Method 1 is used. It can also be seen that two sets of results are presented for Method 1. 1). each with a single parameter (KBase for baseflow and KSurf for routed surface runoff). as in Fig. However. the results show that total streamflow is predicted well quality as the mean of the observations. where “poor” performance is obtained for more than formance and ‘poor’ model performance is represented by negative values of Ef. Results and discussion The performance of the AWBMs calibrated with the seven different methods investigated.and quick- flow hydrographs. good overall model performance is obtained for both parameter sets in terms of matching total streamflow. 8.e. all of the results with parameters that result in better internal model dynamics (e. half (58%e74% (ignoring Method 1 with parameter set 2)) of the However. outputs from HGS simulations) and simulated time of the KBase and KSurf parameters based on the calibration pro- series for each time step for total streamflow. Chiew and Method 5 (Regression) is used. how well the AWBM generated baseflow (qsim sim B ) and quickflow (qQ ) hydrographs match the corre- sponding hydrographs obtained using HGS) (i. are greater than 0. Method 1) and only accept values that seem reasonable. except when empirical guidelines are given when implementing Ef as the metric. 7. with 0. although various performance for fewer than 10% of the catchments. Evaluation of model performance As mentioned previously.g.0 (perfect model performance). with good performance for 32%e41% of the catchments when the calibration methods that consider internal 2. while Moriasi et al. “acceptable” (0  Ef < 0. (2007) suggest.e. The Ef value ranges from ∞ (very bad model performance) to 1. as well as that of the two bench- marks. Consequently.5. An example of this is given in Fig. 8(a) and (b). In this study. Li et al. 7. However.1 is compared in terms of overall model predictive performance (i. resulting in model equifinality. 8(a)) are represented by Set 1. in general. the results are presented in terms of the percentage of models developed for the 66 catchments resulting in “good” (Ef  0. for each of the total-. is slightly better.and quick-flow hydrographs. It is clearly evident that the baseflow and quickflow patterns are reversed for the models with the different parameter sets (e. 7(a)). However. is summarised in Fig.5) and “poor” (Ef < 0) performance. both of which are modelled in an identical fashion. base. how well the total streamflow generated using AWBM (qsim obs T ) matches the corresponding streamflow generated using HGS (qT )) and the accuracy of the resulting internal model dynamics (i. plots of simulated and observed hydrographs are also catchments (Fig. in which case 15% of catchments McMahon (1993) and Ladson (2008) suggest that Ef values above about 0. especially model performance. 7.5 and 1.e.0 correspond to ‘good’ the internal model dynamics are not represented as well. As can be seen.5. as in Fig.

the catchments. As shown in Fig. Methods 2e7) are used and poor “poor” model performance. although Method 1 results in a signif- performance for fewer than 30% of the catchments (12%e29%) icantly smaller number of catchments with “good” performance (Fig. base.and quick-flow hydrographs obtained for two distinct parameter sets obtained using Method 1 for one of the 66 catchments investigated. / Environmental Modelling & Software 66 (2015) 57e68 65 Fig. the performance is only obtained for 12% of catchments.e. base. (a) gives the total-. Set 2) by accepting the parameter set the method calibrates all of the model parameters simultaneously that seems reasonable. model dynamics explicitly (i. As discussed above. This indicates that the RDFs are able to produce reason- performance of Method 1 is comparable with that of the other ably accurate estimates of baseflow.g. experienced modellers could avoid an inap- of matching total streamflow (Fig. there is only a small percentage of catchments with baseflow produced by HGS. Methods 2-4) result in the best match to model performance for quickflow estimation. overall performance of the method is best in terms of total able” performance is achieved for the vast majority of the streamflow prediction and reasonable compared with that of the remaining catchments (71%). with poor performance for only 17% of other methods in terms of baseflow and quickflow prediction. This is because very in order to obtain the best match to total streamflow. “poor” per- is at the expense of internal model dynamics. the baseflow. the method results in the second highest percent. Li et al. Nevertheless. Example of total-. irrespective of the method used. However. as propriate parameter set (i. used. However. 7(c)). “good” compared with the other calibration methods.and quick-flow hydrographs simulated using AWBM with the “good” set of parameters from calibration Method 1.e. the equifinality problem requires careful attention.and quick-flow hydrographs simulated using AWBM with the “bad” set of parameters from calibration Method 1. where BFI and KBase are calibrated to the estimation.e. 7. this poor internal model dynamics can be obtained (e. but “accept. 8. which is evidenced by the fact methods in terms of baseflow estimation. base. It should be noted that when Method 1 is used. which is the same baseflow used for . age of catchments with “poor” model performance for baseflow The methods that calibrate the BFI and KBase to the baseflow estimation and the lowest percentage of catchments with “good” extracted using RDFs (i. This is not surprising. L. (b) shows the total-. formance for 100% of the catchments in terms of baseflow predic- there is a potential problem with equifinality as a result of the tion and for 50% of the catchments in terms of quickflow structure of AWBM. even if the results for Set 1 prediction) if the inappropriate parameter set (Method 1 (Set 2)) is are considered. as baseflow is estimated that the performance of Methods 2e4 is only slightly worse than poorly. Method 1 is the best-performing calibration approach in terms although in practice. In relation to quickflow that of Benchmark 1 (i. 7(a)). Consequently.e. as discussed above.

e.e. methods investigated. This study offers an analysis of seven calibration methods under hart filter (Method 7) overall. might Eckhart filter performs slightly better. prediction of quickflow is much formance and second highest percentage of catchments with better.66 L. the LH method with optimal filter parameter values produces the 4. then internal model dynamics are likely a non- worst. The stances. This improvement is compromise in terms of performing reasonably well on all three particularly significant in terms of producing “good” estimates of hydrographs (i. previously. the impact of seven different calibration methods method compared with Method 2 is that information on catchment on the ability of AWBM to predict total-. Which calibration that they are more complex to apply than the other methods. as discussed above. the improvements ob- method in each case. dicting total streamflow over short time periods (days to weeks) Of the three methods considered. the regression equations for predicting filter performance. / Environmental Modelling & Software 66 (2015) 57e68 performance assessment). a disadvantage of this In this paper. perform better. black-box models. total-. the suggestion by Boughton (2004) that the Boughton filter. (Method 6) provides the best overall trade-offs in terms of In relation to total streamflow. which uses the LH within the bounds (minimum and maximum flow) for which it has filter with its default filter parameter value of 0. The use of Methods 2e4 results in a significant increase in the per. their performance is more consistent than that of any methods. there is also a slight with optimal filter parameters and the regression approach increase in the percentage of catchments with “bad” performance. performs been calibrated. This supports This highly difficult challenge is strived for where the model use is. Method 2. indicate that total streamflow can generally be predicted to an The regression method (Method 5) results in the best perfor. as method should be used is based which component of the flow is of they require baseflow extraction using a RDF.997. base. total streamflow prediction is the most . However. slight reduction in performance compared with Method 1. If the model purpose requires the model to be “right centage of catchments with “good” performance and a slightly for the right reasons”. it is also the worst-performing 88%. AWBM represents internal physical processes could be improved by drological inputs is needed in order to apply the regression equa. Li et al. with the percentage of catchments for which good the best-performing methods for any of the three hydrograph performance is obtained increasing from ~10% to ~40% for most components. using additional information about the catchment. smaller percentage with “poor” model performance. The results making it more difficult to apply. “poor” performance for 57% of the catchments) tends to catchment characteristics and hydrological inputs using AWBM. It also indicates that there is some catchment characteristics and hydrological inputs in order to apply benefit in terms of improving internal model dynamics by con. in such in- Eckhart filters with optimal filter parameters) is very similar.and quick-flow properties and hydrological inputs is needed in order to apply the hydrographs is assessed for 66 synthetic catchments with regression equations used to obtain optimal filter parameter values.and quick-flow). Overall. with acceptable performance levels ranging from 71% to “good” performance. with a slightly larger per. due to problems imposed by the con. However. Median BFI ¼ 0. while the performance of the other two methods (i. Of the methods investigated. acceptable level for more than 90% of the 66 synthetic catchments. the changes in performance are due to the calibration method or Overall. as part of which While use of Methods 2e4 is able to produce the best results in BFI and KBase are calibrated to the baseflow obtained using the LH terms of baseflow estimation. calibration process. baseflow.and quick-flow among Methods 2e4. the method using the LH tained by using Methods 2-7 are only small (generally < 5%). Summary and conclusions best trade-offs in performance between matching total-. performing only slightly In contrast.925. The relatively poor performance for Bench. This raises questions about the way the processes asso- straints of the model structure in AWBM preventing simulation of ciated with these streamflow components are conceptualised in the whole range of baseflow dynamics resulting from the complex AWBM. Generally. A disadvantage of these methods is which model internal consistency is assessed. This disparity in performance between baseflow and quick- method in terms of matching total. is well suited to use with AWBM. different physical properties and hydrological inputs. filter (Method 6) performs better than the method using the Eck. as shown in tions used to obtain values of BFI and KBase as part of the calibration Methods 2e7. base. While they are not quickflow. such as artificial neural networks. base. Of the two methods.e. The method mance levels ranging from 26% to 41% for the different calibration results in the smallest percentage of catchments with “poor” per. as AWBM is unable to provide good predictions of the other methods and not far from that of the best-performing of baseflow in general. Even though performance in terms of baseflow hydrograph prediction. obtain- straining BFI and KBase to ensure that baseflow is matched as well ing the optimal filter parameters and obtaining direct estimates of as possible. However. However. However. An advantage flow prediction is despite the fact that the hydrographs for the 66 of the method is that it does not require estimates of baseflow synthetic catchments consist of a wide range of BFI values hydrographs (only total streamflow hydrographs are needed). the benchmarks are used to test how much poor. streamflow can be predicted very well over a wide range of mark 1 (i. but a (0. mance in terms of matching quickflow. If the model purpose is only pre- and depletion. as mentioned simulating flows outside of the calibration range.53). the results quantitatively demonstrate that while total the model structure. baseflow is predicted poorly. which for example: aiding catchment behaviour understanding and/or is mathematically identical to the Eckhart filter. LH and issue and it is just a curve fitting exercise.and quick-flow. provide the best the ability to match total streamflow and quickflow. then process representation is important. the suggest that the improvement that is possible by adopting this component hydrographs are not modelled very well. particularly approach is rather limited. The pertinence of such questions is dependent on the underlying physical processes associated with baseflow generation purpose of the model though. regression method (Method 5) provides the best match to the centage of catchments with “good” performance in terms of quickflow hydrographs and the hybrid method using the LH filter quickflow prediction compared with Method 1. BFI and KBase in the case where predicted filter performance is As stated previously. use of Methods 2e4 also results in a matching all three hydrographs. suggesting that the way disadvantage is that information on catchment properties and hy. Methods 3 and 4. as well as data on most interest. there are some trade-offs in terms of and Eckhart filters with optimal filter parameters. Use of the calibration methods that take internal model dy- The two methods combining aspects of the filter (Methods 2-4) namics into account explicitly (Methods 2-7) results in improved and regression (Method 5) methods (Methods 6-7) provide a good prediction of the component hydrographs. with acceptable perfor- worse than the quickflow benchmark (Benchmark 2).006 < BFI < 0.

. pp. Simmons..K. S. providing insight as to whether they streamflow yield estimates. although structure. there is a trade-off between realism and the ability saturated processes between surface water and groundwater and their impact to assess model dynamics accurately.. 1984. particularly when they are sets are reasonable. 47 (11). D..J.H.A. N.F.. L. Marsili-Libelli. K. Model. 2005.. the on the state of connection. Model...F. E. Vegetation controls on variably alism. Adams.015. Wilby. Y. When all three hydrographs are important for thank the three anonymous reviewers for their detailed comments.J.. Newcastle.... Jakeman. Voinov..G. 1e20. . Zuidema. it would be modelling in a fully integrated surface/subsurface framework. it is worth anarchy? Emerging themes and outstanding challenges for neural network river forecasting. Model.J. S.J. N. 1993.. 318 (1e4). L. Hydrol. Ground Water 50 (2). Hydrol. Shin et al. F... Assessing the adequancy of catchment dynamics of CRR models. rates using a discrete-state compartment model and C-14 data. Arnold.. the fact that ground water recharge from streamflow records.D. 87 (1e2). 1999. However. J. P. distributed reasonable improvements in internal model dynamics in such modelling system. RDFs). P. Of these methods. use of Method 2 (LH filter with Abbott.. However. applied in settings outside those used for model calibration.J. Frei and Fleckenstein. K. Firstly. An open software environment for hydrological model assessment and development. transformations could be applied to the data (e. complexity (see e. 4 and 6). C. uncertainty could be considered (10). R. synthetic data are used in this study limits its complexity and re- Banks.and quick-flow hydrographs are generally unknown Jakeman.A..g. A.. pp. during the AWBM calibration process (e.1016/j. KBase directly (Methods 5 and 6) and estimating the level of per- formance of the LH filter with optimal filter parameters (Method 6).W. the problem of equifinality needs to be Water Resour. Anctil. method 1 (i. A hydrograph-based model for estimating the water yield of acteristics and hydrological inputs. A. 2006. uniform rainfall. Res. the method that uses the Eckhardt filter with the optimal filter parameters obtained Acknowledgements using the regression equations developed by Li et al.H.g. 2004. B.A.J. Andreassian.D.. Beven...T.. 2012. References which makes them difficult to apply if the required information is not readily available. V. Automated methods for estimating baseflow and Fonseca et al. Finally. while the computa.A. similar to that of black- all parameters to total streamflow) is likely to be the most appro. Seppelt. Abrahart. B. S. water management. J. A. Allen. Robson. Res. Brunner. 2014.T. internal dynamics of conceptual rainfallerunoff models.. Environ. B. Softw. Guariso. Hydrol. N. 2013). the hybrid method using the Lyne and Hollick which have greatly improved the quality of the paper. 45e59.. HydroGeoSphere: a fully integrated. NSW.K. these values are also ungauged catchments. Guillaume. C. 112e135. Wagener.T. Croke. P. Hydrology and Water Resources Symposium.S. P.G. pursuing other calibration methods that aim to take the internal Ahmadi. 943e956. Ascough Ii.... 41e51.g.. A. 26 (Romanowicz.-J. In addition. Environ.. J. Solomatine. Cook. Coulibaly.e. O'Connell. 16 (1). 2004. tional experiments are conducted for a range of catchment char. F. J. Simpson. actual base. As mentioned in 171e185. Croke. Hydrogeologic controls on disconnec- presented in this study with alternative approaches to assessing the tion between surface water and groundwater.g. For real catchments. Norton.. E. Fath. the base. tracers. Prog. flood prediction).. Water Resour. J. T. Ajami.B. Newham. Soil Res.. or whether they are able to provide a reasonable priate.g.g. Boughton.. (2014) (Methods 4) performs best.. 480e513. J. Thermal transport should be investigated in future studies. M.. Aust. such as low flow forecasting. Water Resour. However. Arabi.. 72. Softw. C.. of Engineers Australia. uniform slopes etc... An default filter parameter of 0. and have to be estimated using a variety of indirect methods (e. 232e249. 1986.doi. Toth. the methodology presented in this paper should be and recharge estimation.) and the effect of additional Boughton. A. a first step towards the more rigorous assessment of the internal Chiew... Park.Y. “SHE”. B. the methodology presented in this paper provides Chen.. It should be noted that most of the methods that perform best Appendix A. 40 (0). The Australian water balance model. Calibration of a semi- of the total streamflow hydrograph are captured. G. (Methods 2-4) should be used. Softw.. P. while the focus of this paper is on the assessment of the Campana. 2010)).T. physically based hydrological model. use a BoxeCox J. Cunge... / Environmental Modelling & Software 66 (2015) 57e68 67 important. 36 (4). T. Rasmussen. 2009. For example.. V. Phys. See. Perrin. Model... Gupta..E. box models.e. Encyclopedia of Hydrological Sciences. M..L. In addition. Institution represented in a simplified manner (e. 2012.A.. and therefore. in order to obtain reasonable internal model dy. the Introduction. L. predicting BFI and dx. geneous soils. 317e324. Brunner. 298. (LH) filter with optimal filter parameter and the regression method (Method 6) is likely to give the best performance. It is also important to highlight a number of limitations of this Shamseldin. W.. representation of the underlying processes. 2011). Groundwater residence times and recharge internal model dynamics of the AWBM.H. Conant Jr. Bathurst.. 19 (10). Sudicky. M. or different distributed hydrologic model for streamflow estimation along a river system. such as W01422. 2011. sensitivity analysis (e.C.envsoft.. Hydrol. Ahmadi et al. the approach adopted in this study enables Characterising performance of environmental models.. Process. Assoc. Simmons. Adv. R. cases. M. 2011. 1993. If the latter applies.. 665e680. McMahon. P. changes improved calibration of watershed models: multisite multiobjective measures could be made to the objective function to ensure known features of information. Hydrol. For situations where baseflow is of most in. Wiley and Sons. 2014. B..A. In addition. Hamilton. C.M. C. J. (Ed.925) can be used in order to achieve introduction to the European Hydrological System d Systeme Hydrologique Europeen. E.. The authors also gratefully should be used. S.). 45 (1).C.S. Beven. 2014. E. J. Water Resour. 2150e2164. there can be greater confidence in the experienced modellers usually have an idea of which parameter results obtained from CRR models. Rainfall-runoff modeling: introduction.P. Integration of artificial neural networks with conceptual models in rainfall-runoff modeling. simultaneously calibrating primarily provide a curve-fitting service. Softw. Prophecy.. Simmons.G. P.. W11517.E. reality and uncertainty in distributed hydrological enabling an accurate assessment of the internal dynamics of AWBM modelling.W. 1171e1185. R..W. as suggested by their namics. J. terest. 2014) on the results Brookfield.H. the generic approach pre. Environ. J.. 1e12. to be obtained in a simplified setting. 1993. Geogr. Li et al.12. Am. Lastly. the regression method (Method 5) (CSC) and the University of Adelaide. sented is equally applicable to other CRR models..C.. Engel.C. Pierce.. 59 (0). W. transformation to shift the focus from high flows to low flows Andrews. J.G. 411e424. 35 (2). interesting to compare the results obtained using the approach 23 (15). 31. Sorooshian. J.and quick-flow hydrographs to be known with certainty. J. Supplementary data require information on catchment characteristics and hydrological inputs in order to apply the regression equations for obtaining Supplementary data related to this article can be found at http:// optimal filter parameters (Methods 3.. Brunner. However. Bennett.. 135e145. 2009.E.A.2014. B. Mount. If quickflow prediction is the primary This research was supported by the Chinese Scholarship Council objective (i. Fontane.T. Dawson. 1: history and philosophy of a physically-based. F. In: Anderson. identification of source areas Consequently. Environ.P. B. Toward model dynamics into account explicitly. homo.. M. Two decades of study that provide avenues for future studies. D. 170e176.K. the methods that calibrate BFI and KBase applied to other CRR models in order to assess their internal to baseflow extracted using recursive digital filters (RDFs) dynamics.W. 2013.

brating the Xinanjiang watershed model. 1997.. K.. 1015e1031.F.. CSIRO: Water Baseflow for Catchment Simulation (Phase 1. C. 2006.W.. Softw. Q. H. McCallum. Resour. U.. Y. J... Werner. Hydrol. Res.L.A. Yang.S.. E. C. J. 1998. 54 (0). Dandy. R. B. G.H. 13 (12/13). J.. Baseflow Assessment for the Murray-Darling Basin. Code Documentation and User's Guide. and Surface Flow and Solute Transport. Effective and efficient global optimization Nelder.-J. 1994. Hydrol. J. Lambert. 25 (8). 28e39. 885e900.. for small catchments.M. to baseflow separation. A physically based filter for separating base flow from methods against simulated baseflow from a physically based. Hydrol. Brunner. 2001. A. Gupta. Boaventura.. Arnold..G.. P. Environ. Water Resour.. J. Am. H. 7 (2). J.. groundwater flow model. Simmons. 2010. C. R. Water.. T.. / Environmental Modelling & Software 66 (2015) 57e68 Chiew. Hydrol. 78 pp.V. A quasi physical snowmelt run-off modelling system ments with different physical characteristics and hydrological inputs. R. 1970. Australia. 2014. Gupta. Environmental Earth Science.Y. P. Solute dynamics during Temperate Headwater Catchment. alyses in R. IHACRES. S. Waterloo. Sudheer. 1961e1975. C. Li.. F. 2001. Publications. Regionalising the Hydrologic Response of Vadose Zone J. Anctil. 21 (8). 158e173. 2011. Meyerhoff. F. J. Peel. 39e52. J. Optimal use of the SCE-UA global 2005. M. Brunner. Maier. C. application to the assessment of the quality of the ANN model development Marshall. Sorooshian.. Chiew. Panday. C. Fleckenstein.. 2009.. Wu. 2012.. Model. Hydrol...B. 43 (3). 5501e5519. SKM..... Water Resour. Australian Rainfall and Runoff Technical Committee.W..P.S. Boston. Loague. Ames. Water Resour. 2012. D. 50 (3). Brunner. Inc. Water Resour. 135e152. and Machine Ranatunga. and Sacramento Rainfall- Gibbs. and recession analysis. Fleckenstein. 2014. K.E. Murphy. J. Hydrol. Contam. Nation. Barratt... M. R. 2009. S. 33 (1). Res. C. rainfall-runoff model SIMHYD. Cunnane. Comput. Hydrologic-response simulations for the R-5 Maier. 28 (4). 52 (0).J. 2002. Sharma.B.W. Hervieu. Evaluation of automated techniques for base flow and unsaturated flow effects on bank storage. Genetic Algorithms in Search.W. MODSIM 2007 International Congress on Modelling and Simu- regionalization in ungauged catchments.. Simmons.C. 2013. Process. T. J. Botelho.. B.R. J. A discussion of principles. 2005. R. 1996... S. B.. C. Knapp.. Process. Hill. Hydrol. Environ. Oudin.. Ferket. Mudgway. Model. Sci.. M. D. Partington. Kling. surface water- streamflow time series. 46 (7). Maxwell. Errors and parameter estimation in precipitation-runoff Li.R. The downhill simplex method. Neal. catchment models.. L.H.. 96. R. Environ. R. 265e284. Three-dimensional Numerical Model Describing Fully-integrated Subsurface Australia & New Zealand.. E..V. Model.. Water Resour.R. An analysis of river bank slope Nathan. Hydrol. Wang. Dandy. In: Singh.. Maier. USA. Representing effects of micro-topography on runoff water-groundwater flow models. Res. 2001. J. Oxford University Press.. Graszkiewicz. Res. P. Softw. Pauwels.. 456e467. A. J. Putti. H..F. Paniconi. R. D.. P. Geologic Survey. 5e18. F.. 886e898. Model..W. Sudicky.. Christchurch.... Duan. Sci... V. D. H. p. H.. 1e14. D.A. Andreassian. F. Knudsen. Maier.C. B. Environ.A. Gupta. V. S. Mclaren.J. A. L. K... J.. 2010... T. Berra.D. Project 7: CSIRO.. Addison-Wesley Longman Publishing co. H.. 952e964.. A. A. 2007.. Huber. for Stormwater Management. AGU. 999e1013. M. ments.. Kollet. K.. Simmons. V. HydroGeoLogic Inc.. Frei.. 458e459.. Dandy. Duan. 2013. component of streamflow within spatially distributed fully integrated surface Frei.. part 1. Hydrol. Bingner. 2008.. River flow forecasting through conceptual models: Urban Stormwater Hydrology. Sudicky. Environ. Samain.A. 2000. W07541.R. J. C. 2014.R. Lambert. flow estimation with application to the Lyne and Hollick filter. 177e185.C. A. Res. Littleton. Hydrol. 2189e2202.R. Harmel.. Environ. B.M. Frevert. 2008. Ground Water 50 (1). Post. Delleur. J. Sutcliffe.J. An application of a log-transformed low-flow (LTLF) model Hapuarachchi. Jain.. D. model part 2dtowards a simple and efficient potential evapotranspiration Dukic. Softw. 2007. Softw. (Eds. Softw.N. P. McLaren. DC. 208. Herndon.. R. 12 (4). Mathematical Model evaluation guidelines for systematic quantification of accuracy in Models of Small Watershed Hydrology and Applications. Three-dimensional analysis of variably-saturated assessing and improving the performance of recursive digital filters for base. 2534e2540. V... 2004. 37 opment of neural networks for the prediction of water resource variables in (4). Adv. Water Resour. 77e86. 108e130... sensitivity. M. model for rainfallerunoff modelling. D. J.. R. Res. Loumagne. Galeati. 163e175. J.R. uted model for evaluating surface/subsurface flow. Vilar. C. pp.. P. M. Water 48 (3). S. G. Perrin. A Review of Rainfall-runoff Modeling Sulis. Biol. W10601. VA. Maier.. 2011. 2004. Mead. Shin. Model. J. eration mechanisms from integrated surface-subsurface flow models of a ri- shed model parameter estimation and uncertainty in data-limited environ. 1e12.H. 27e28 (0). Li. Panday. 2010. Review of soil water models and Learning... Framework for Therrien.. 1985. Which potential evapotranspiration input for a lumped rainfallerunoff optimization method for calibrating watershed models... L. Michel. Nathan. river system: current status and future directions..W. L. A fully coupled physically-based spatially-distrib- Eckhardt. 26 (7). Efficient subspace probabilistic parameter optimization for water-groundwater interactions. D. Sorooshian. 2014. Protocol for developing ANN models and its 891e909...K. M. Illinois District. 7. J. 2010..G. Water Resour. Colorado. P. J.68 L.C. 47 different types of hydrological models. S. 443e458..M. H.. 23 (1e2). . 335e367.R. Vaze. Z. Dandy. A. 158. Therrien. G... S. Res. Approach). Liew.... 2010. Environ.-J. R. Modelling coupled surface-subsurface flow processes: a review. Hydrology: an Australian Introduction. Brunner.. Water necticut State University.L. 1965.K. parian wetland and catchment. Global optimization techniques for the Maier.. Jakeman. Water Resour. Modeling of base flow of the Basin of Kloubara River in Serbia. C.R. S. Berhane..C. Franchini.C. 1e34. Water Resour.. G. Application and testing of the simple Moriasi.G. Water Resour. M. Ground Modelling System. A comparison of two physics-based numerical models for simulating surface Kuczera. J. (2). W02501... Therrien. Carlo methods for conceptual rainfall-runoff modeling. M. E. Hydrol. 1992. Softw.. A hydraulic mixing-cell method to quantify the groundwater calibration of conceptual rainfall-runoff models. McMahon. D.E. J.. Hydrol.A... Schwartz.. P... 741e756. Veith.. Water Resources watershed simulations. 26 (7). N. Nash... MA. Res.S. Apted. 2008. Zuidema. H. 55 (6). their applications in Australia. Performance modeling: 1. Dandy. Interpreting streamflow gen- Fonseca.G.C. M.G.T. 84e93. G. Ungauged Catchments Using the SIMHYD.R. Theory. Tuteja. Guillaume. 41 (0).T. 10 (3). Evaluation of outputs from automated baseflow separation Furey. Goldberg. flow and solute transport in discretely-fractured porous media. Partington.. 40 108e127.A. and normalization of Mean Refsgaard. J. B. p. 19 (2). H.. E. F. Model. Doble. (10). Res.. V.. V. 2010. Werner. Gupta.. Adv. Hydrol.R.L. Eastern Connecticut State University: Eisenhower. Environ.. S.T. 54. 2013. Durgunoglu. Werner. pp. H. Eng..V. 308e313.. for conceptual rainfall-runoff models.). assessment and improvement of recursive digital baseflow filters for catch.. 327 (1e2). lation. Eastern Con- bank storage flows and implications for chemical base flow separation. HydroGeoLogic. HydroGeoSphere: a Ladson. Sudicky. VanderKwaak... rainfallerunoff models using baseflow separation... 32 (7).. T. Softw... 1990. 1999.T. C..C. R. Ortel.Selection of Baseflow Separation for a Healthy Country National Research Flagship. Troutman.. D.. 23 (9).J. R. Res.. 2012. 1998. 1996... P.D.P. Res.. 366e379.P. V. R. Lake Sci. 24.. Softw. 1989. R.. storage height variations. J. Western. D.. Simulation of water flow on irrigation questions about conceptual rainfallerunoff models with global sensitivity an- bay scale with MIKE-SHE. Application of SCE-UA method for cali. G.. G.. N. Optimization. R. Croke... Park... Marier.T. 1982.. 2010. 381. Agric. Therrien. A. Cook.. 1182e1206. A. H. Water Resour.N. Ladson.G.. V.S. G. 1195e1213. Addressing ten Jayatilaka. A generic framework for regression runoff Models. Storm. 304e314. Identifying Groundwater Contributions to Baseflow in a McCallum. 290e306. S. generation and sub-surface flow patterns by using superficial rill/depression Partington. Model.K. How to construct recursive digital filters for baseflow separation. 27 (4).D. H. pp. D... 33 (4). D. Water Resour. L. M. S. Maier.H..F. Romanowicz. 303 (1e4).V. Methods used for the devel. Modelling and Simulation Society of Australian and New Zealand. P.. 51 (0). R. 503 (0). A.. Nott. J. 37 (11). On typical range. Li et al. Simmons. Soc. MODHMS: a Comprehensive MODFLOW-based Hydrologic Ambiguous hydaulic head and 14C activities in transient regional flow. Q. Viney. 49 (9). Simmons. 2709e2722. 361e382. 2011.. 1e44. Internal validation of conceptual Partington. Softw.... Furman. 1991. 282e290. A.V..J. Huyakorn. Cook. Introduction to urban hydrology and stormwater management. Environ. A comparative study of Markov chain Monte process in drinking water quality modelling. 507e515.. Brunner. W. Washington.D. New Zealand. Christchurch. Operational validation and intercomparison of Squared Error and Nash-Sutcliffe Efficiency type metrics. catchment with a comprehensive physics-based model. Z. P. Model. 1465e1473. Model.. Partington..E... Therrien. Dandy.