Hydrological process knowledge in catchment modelling –

lessons and perspectives from 60 years development

Jens Christian Refsgaard1, Simon Stisen1, Julian Koch1
1
Department of Hydrology, Geological Survey of Denmark and Greenland, Copenhagen, Denmark
Accepted Article

Abstract
Hydrological process knowledge has advanced significantly during the past six decades. During the same
period catchment models have undergone major developments including simple black box models, lumped
conceptual models, hydrological response unit models, spatially distributed process-based models and,
recently, the emergence of machine learning hybrid models. This development has been enabled by
improved understanding of hydrological processes together with ever increasing computer power and
improved availability and accessibility of data. During the first couple of decades, a key assumption
motivating the development towards increasing complexity of model codes was that more detailed process
description would lead to more accurate model simulations and enable prediction of impacts from human
activities that previous models were not able to provide. Subsequently, scientific tests showed that this is
very often not the case, leading towards a recognition of the importance of careful model evaluation
accounting for key uncertainties in data, model parameters and model conceptual understanding. We have
reviewed 54 model studies from the past 60 years and characterized them with respect to model type,
spatial discretization and model evaluation techniques. This showed clear development trends and
different strategies for enhancing hydrological process knowledge in models. In addition, we present a case
study, where we use two models for the same catchment. The models are identical except for the spatial
discretisation of 100m and 500m, respectively. The two models have an apparent equal performance
measured against standard calibration metrics, but nevertheless show large differences when considering
detailed process information such as partitioning of streamflow components and water table depth
patterns, that was not considered during the model calibration process. The paper discusses perspectives
for enhancing hydrological process knowledge in future catchment modelling concluding that the
emergence of big data is likely to become a major game changer.

Key words
catchment model, hydrological process understanding, spatial discretization, model evaluation, big data,
machine learning

Correspondence
Jens Christian Refsgaard, Department of Hydrology, Geological Survey of Denmark and Greenland, Øster
Voldgade 10, 1350 Copenhagen, Denmark.

Email: [email protected]

This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process which may lead to
differences between this version and the Version of Record. Please cite this article as doi:
10.1002/hyp.14463

This article is protected by copyright. All rights reserved.

 Introduction
During the first hundred years of hydrological modelling, models relied on hand calculations (Mulvany,
1850; Sherman, 1932; Nash, 1959). These models were empirical and developed from analyses of
concurrent input and output data without consideration of hydrological processes. With the introduction of
computer technology, many so-called conceptual model codes were developed. This development was
pioneered by the Stanford Watershed IV (Crawford & Linsley, 1966) and followed by a large number of
model codes that are commonly used even today, e.g. Sacramento (Burnash, Ferral & McGuire, 1973), HBV
Accepted Article

(Bergström & Forsman, 1973) and NAM (Nielsen & Hansen, 1973). The conceptual models were based on
sound hydrological process knowledge, although point scale process equations and data could not directly
be exploited because the computational unit of the models was the entire catchment. The next step
towards development of model types that could encapsulate more hydrological field data and process
knowledge were triggered by a blueprint from Freeze and Harlan (1979) and the development of the first
spatially distributed process-based (or physics-based) model codes, e.g. SHE (Abbott, Bathurst, Cunge,
O’Connell & Rasmussen, 1986a,b), IHDM (Beven, Calver & Morris, 1987) and THALES (Grayson, Moore &
McHahon, 1992).

During the first couple of decades of development towards ever increasing model code complexity, a key
hypothesis has been that more detailed process description with improved hydrological knowledge would
lead to more accurate model simulations and enable prediction of impacts from human activities that
previous models were not able to provide (Abbott, et al, 1986a). In the absence of hard facts from suitable
tests, the scientific discussion was to a large extent based on expectations and qualitative arguments and
the hypothesis was questioned (Beven, 1989). Since then, several studies have documented that this
hypothesis cannot be confirmed for cases, where the simulation target is confined to river discharge
(Perrin, Michel & Andreassian, 2001; Reed et al., 2004; Refsgaard & Knudsen, 1996). Instead of complex
models with sophisticated process descriptions, large data requirements and many parameters several
authors argued that models should be parsimonious, i.e., they should be as simple as possible and include a
minimum of parameters requiring calibration (Beven, 1989; Bergström, 1991; Jakeman & Hornberger,
1993; Perrin, Michel & Andreassian, 2003).

In spite of these apparent shortcomings and arguments for a different strategy, the developments towards
more detailed spatial representation and more sophisticated process descriptions continued, e.g., MIKE
SHE (Graham & Butts, 1995), SWAT (Arnold, Srinivasan, Muttiah & Williams, 1998), ParFlow-CLM (Kollet &
Maxwell, 2008), HydroGeoSphere (Therrien, McLaren, Sudicky & Park, 2010), mHM (Samaniego, Kumar &
Attinger, 2010). This resulted in model applications that are able to exploit almost all existing data on
climate, soil, vegetation, geology, topography and streams and hereby contribute to improved credibility
among stakeholders (De Lange et al., 2014; Højberg, Troldborg, Stisen, Christensen & Henriksen, 2013) as
well as models capable of assessing impacts of human activities such as hydropower development
(Refsgaard et al., 1998), groundwater abstraction (Sophocleous et al., 1999) and changes in climate and
land use (Mango, Melesse, McClain,Gann & Setegn, 2011).

Recent developments in data acquisition such as new geophysical techniques (Auken, Boesen &
Christiansen, 2017), new earth observation techniques (McCabe et al., 2017) and citizen data collection (Le
Coz et al., 2016) open a whole new chapter in hydrological modelling with easy access to an until now
unprecedented amount and type of data. Together with ever increasing computer power, this provides an

This article is protected by copyright. All rights reserved.

 opportunity to exploit more hydrological data and process understanding in hydrological modelling (Kollet
et al., 2010). How far this potential can be exploited remains to be seen.

Altogether, the ever-increasing computer power, the improved understanding of hydrological processes
and the improved availability and accessibility of spatial data have enabled and motivated the development
of increasingly complex hydrological models with finer and finer spatial resolutions (Liu & Gupta, 2007). At
the same time, it has been widely recognized that use of sophisticated, high-resolution models in itself is no
guarantee for more accurate simulations. As an example, improved availability of high-resolution spatial
data from satellites has made it possible to test how well spatially distributed models are able to reproduce
Accepted Article

observed spatial patterns. The tests revealed that models performing well against observations of
streamflow and groundwater heads often were not able to simulate spatial patterns in land surface
temperature and evapotranspiration particularly well (Stisen et al., 2018; Demirel et al., 2018), implying
that the internal process representation or parameterization is deficient and that it therefore could be
argued that the streamflow simulations to some extent provided the right answer for the wrong reasons
(Kirchner, 2006).

This illustrates that theoretical process descriptions are not a sufficient measure of hydrological process
knowledge. A second, and equally important component, is model evaluation to account for model
parameter and conceptualization uncertainty as well as catchment specific information. In this respect,
factors contributing to constraining the reliability of complex models are mismatch of spatial scales (Beven,
1995; Blöschl & Sivapalan, 1995), overparameterization leading to equifinality (Beven, 2006), lack of good
quality data and inadequate model calibration. Consequently, a proper evaluation of model simulations and
the associated uncertainties are crucial for the credibility of hydrological model simulations (Klemes 1986;
Beven, 1989; Refsgaard 1996; Beven 2002; Jakeman, Letcher & Norton, 2006; Refsgaard, van der Sluijs,
Højberg & Vanrolleghem, 2007) and hence for enhancing our hydrological process knowledge.

Our three basic hypotheses are that i) a lot of hydrological process understanding has not been utilised in
state-of-the-art catchment modelling until now, ii) currently, hydrological modelling is at a development
stage where a relatively larger share of the existing hydrological process knowledge can be exploited, and
iii) such a development will call for further enhancing hydrological process understanding and model
evaluation strategies.

The objectives of the present study are therefore: i) to learn about the major drivers and obstacles to
utilising hydrological process representation in catchment modelling; and ii) to analyse and discuss
perspectives for enhancing hydrological process knowledge in future catchment modelling.

Methodology
Literature review
The literature review focuses on development of hydrological models during the past 60 years. The review
does not aim at providing a comprehensive review of hydrological modelling literature. Instead, we use
only a few selected references attempting to describe key model developments, seen from a hydrological
process knowledge perspective.

In this paper we use the term hydrological process knowledge as comprising two elements: i) theoretical
process understanding in terms of equations and literature knowledge on parameter values, and ii)
catchment specific information in terms of data, observations, and conceptual understanding. By combining

This article is protected by copyright. All rights reserved.

 these two elements a catchment specific model may be able to help inferring new process knowledge. The
implication of this terminology is that we consider hydrological process knowledge to be domain and
catchment specific, albeit with many generic aspects. To evaluate how hydrological models contribute to
enhancing hydrological process knowledge one therefore must consider both elements. While theoretical
process understanding typically is built into model codes in terms of process equations and literature-based
default parameter values, the catchment specific information is typically incorporated in connection with
model calibration and uncertainty assessment procedures.

Our analysis focus has been on how the level of process knowledge has evolved during the past 60 years.
Accepted Article

Catchment models can also be classified into continuous simulation models and event models (Singh,
1995). Because event models are mainly relevant for flood simulation and often dominated by uncertainty
on initial conditions, we have chosen to focus on continuous simulation models and, with a few exceptions,
only include model studies with simulation periods longer than one year. Similarly, we have confined our
analysis to catchment models and excluded point scale models and global hydrological models.

To support the review, we define four proxies (code class, N, Δx, evaluation level - see below) for the
degree of hydrological knowledge embedded in a hydrological model.

Model code classification

Fundamentally, we distinguish between a model code as a generic software package and a model as a site
and purpose specific setup of a model code for a particular catchment. This implies that we for instance talk
about the SWAT model code, but about a specific SWAT model, when the code is populated with
catchment specific data and parameter values for a catchment application. This distinction is important so
that we can talk about the reliability of a particular site-specific model without providing universal claims
about the model code (Refsgaard & Henriksen, 2004).

Numerous hydrological model codes have been developed and a large number of classification schemes
have been proposed such as Woolhiser (1973), Singh (1995), Refsgaard (1996), Todini (2007), Pechlivanidis,
Jackson, McIntyre & Wheater (2011), Paniconi & Putti (2015), Hrachowitz & Clark (2017). In this paper we
use a classification scheme illustrated in Figure 1 that reflects the degree of hydrological process
understanding that can be built into the model code.

 Black box models (BB) are empirical with mathematical equations and coefficients that are
developed and assessed purely from analyses of concurrent input and output time series without
considering hydrological processes understanding.
 Lumped models (LU) include some concepts reflecting hydrological process understanding such as
root zone capacity, overland flow and baseflow, that are upscaled to represent hydrological
response at catchment scale. These models have often been denoted conceptual models, but we
do not prefer this term here, because the term conceptual is generally used to describe the
understanding of the system (Refsgaard & Henriksen, 2004; Hrachowitz & Clark, 2017) and other
hydrological model types also include conceptual understanding.
 Process-based models (PB) apply point scale (or very small scale) equations of hydrological
processes and a spatially distributed representation of the catchment. These models have often
been denoted physically-based.
 Hydrological Response Unit (HRU) models group areas with the same combination of some key
characteristics such as soil type, vegetation, topography and climate forcing within a catchment

This article is protected by copyright. All rights reserved.

 into one calculational unit (HRU), so that a catchment on one hand includes several HRUs with
detailed process descriptions and on the other hand has a lumped-conceptual representation of
other processes.
 Coupled model codes integrating surface water models with full 3D groundwater models (HRU-GW,
PB-GW), water quality components (HRU-WQ, PB-WQ) or atmospheric model components (PB-
ATM).
 Hybrid models (ML-HM) are combinations of machine learning techniques (sophisticated BB
models) and other model types. This includes models where machine learning models use
simulation results from other model types or where LU/HRU/PB models or hydrological process
Accepted Article

understanding are built into in the machine learning architecture.

The upper row of three boxes in Figure 1 (BB, LU, PB) are basic categories. The names of these classes
should be considered relative rather than absolute. For example, there are elements of lumping in all
process-based models, because the spatial representation will always be larger than point scale (Beven,
1989). In addition, many model codes do not fit nicely into one of these three boxes, because they include
components from more than one class. The commonly used HRU model codes constitute hybrids of LU and
PB with some process descriptions fitting into the LU box and others into the PB box. Finally, to provide
more clarity in our description and discussion of model developments in the remaining part of this paper
we have chosen to make separate classes for some of the most commonly used coupled model codes
including groundwater, water quality and atmospheric process descriptions.

We are aware that, even with these extensions to the three basic classes, some model codes cannot really
be fitted into one of our proposed classes. All classification schemes have some shortcomings, and so does
our proposed scheme, which has been chosen to support our objectives of discussing how the use of
hydrological process understanding in hydrological models has evolved during the past decades.

Number of spatial calculational units in the model (N)

While this is simple to assess in surface water models, it is trickier to describe by one number in
integrated/coupled models. We have included all spatial grids/elements in the surface water module and
added computational units in modules which are computationally heavy. Here we have included
calculational units in i) 3D groundwater models, ii) unsaturated zone in cases of the full Richards’ equation,
and iii) 3D atmospheric modelling. Thus, we have not added water storages such as interception and a
couple of root zone boxes, if calculations here are based on relatively simple, explicit water accounting.
Furthermore, we have not included calculational units in the river network. Although river routing can be
computationally very heavy, we have ignored this, because the focus in the present paper is on catchment
processes rather than river hydraulics.

Length of spatial calculational unit in the model (Δx)

In some models that couple surface water and groundwater there are differences in the calculational unit
size between compartments, e.g. a Richards’ equation discretisation of the unsaturated zone is typically in
the order of a few cm vertically, while the underlying saturated zone discretisation may be much larger,
both vertically and horizontally. In some model codes, e.g. in case of hydrological response units or
irregular grids, the size of the units may vary considerably. Here we have estimated the length scale as the

This article is protected by copyright. All rights reserved.

 square root of the average horizontal calculational unit, which in some cases is calculated by dividing the
entire model area with the number of calculational units.

Model evaluation
This proxy is a measure of the extent to which catchment data and observations have been incorporated in
the modelling process to evaluate the reliability of the model. We define four levels with increasing
amounts of model evaluations (i.e. runs) to characterise this:
Accepted Article

 Single (Level 1). In this case the hydrological model has only a single deterministic configuration for
a single catchment. The parameter values have been assessed from literature (no calibration) or
from relatively simple manual adjustments based on a single objective function (typically with
discharge as calibration target).
 Multiple (Levels 2-4). In this case multiple evaluations have been performed with respect to three
aspects i) multiple data types used in objective functions for calibration; ii) multiple parameter sets;
and iii) multiple model structures. The level reflects how many of the three aspects have been
considered as multiple. As an example, a model study using one parameter set (no parameter
uncertainty), calibration only against discharge and three model codes will have level 2, while a
model using several parameter sets, that have been estimated through calibrated against both
discharge and soil moisture data and evaluated for three alternative model codes will have level 4.

Case study
The case study has been designed to investigate the importance of spatial discretization as one of the
potential obstacles in utilizing hydrological process representation in catchment modelling. We use two
models for the same catchment. The models are identical except for the spatial discretisation, which differs
by a factor of five in each horizontal direction. We then analyse how the hydrological process knowledge of
the two models differ due to the differences in spatial representation. First, we evaluate the performance
of the two models using standard metrics focusing on the accuracy of simulated streamflow and
groundwater heads when compared to field data used for calibration and validation tests. Next, we
evaluate how the two models simulate shallow groundwater depth as well as partitioning of streamflow
into baseflow, drainflow and overland flow, which are variables that have not been subject to calibration.
Finally, we analyse how the differences in spatial discretisation leads to differences in smaller scale
hydrological processes and discuss to which extent this may limit the model applicability at scales smaller
than those targeted by the calibration data.

Literature review
History of model code development
The description is divided into five periods, where we focus on what we perceive as dominating trends
within each of these periods. We acknowledge that many trends that dominate in one period typically start
with experimental developments in an earlier period and continue with consolidating developments in later
periods. Thus, many developments are overlapping a couple of periods. Hence, the timing of what is
described as developments in these periods should be seen as indicative only.

This article is protected by copyright. All rights reserved.

 The early days – BB models (before 1970)
The history of hydrological modelling dates back to the pre-computer era. Some of the early key
achievements were the Rational Method (Mulvany, 1850) and the Unit Hydrograph (Sherman, 1932), which
were applied in flood modelling. These simple models were suitable for hand calculations. The model
parameter values were estimated based on measured input (precipitation) and output (runoff) data with
attempts to estimate parameters from catchment properties to allow flood prediction in ungauged
catchments. Thus, the models were data-driven without making inferences from hydrological process
Accepted Article

knowledge and therefore belong to the Black box (BB) class (Figure 1).

More sophisticated BB models were subsequently developed using statistical techniques. Examples of such
models include i) gauge to gauge correlation methods (Searcy, 1960); ii) Antecedent Precipitation Index
(WMO, 1994); and iii) Autoregressive Integrated Moving Average models (Box & Jenkins, 1970). Some of
these methods were originally developed for hand calculations while others were supported by mainframe
computer calculations.

These BB models have been widely used since then, both as stand-alone tools and as components
integrated in more sophisticated models. In addition, a large range of new BB-models have been developed
subsequently by use of first hydroinformatics and, in recent years, machine learning techniques (see
below).

Widespread use of LU models (1970-1985)

The development of the first lumped conceptual (LU) model codes were initiated in the late 1950s with
pioneering efforts made by the SSARR (Rockwood, 1964; Speers, 1995) and Stanford Watershed IV
(Crawford and Linsley, 1966) followed by widespread international developments (Becker, 1977; Porter &
McHahon, 1971; Sugawara, 1967). During the 1970s many of the today most commonly used LU codes
were developed, e.g. Sacramento (Burnash et al. 1973), HBV (Bergström, 1976), NAM (Nielsen & Hansen,
1972) and Xinanjiang (Zhao, 1980). The widespread use and promotion of these models in operational
practice is illustrated by a series of intercomparison studies conducted by the World Meteorological
Organisation (WMO, 1975, 1988, 1992).

PB models emerge and HRU models widely used (1985-2000)

The first attempt to outline the potentials and some of the key elements in process based (PB) catchment
scale models were made by Freeze and Harlan (1969). The first attempt to apply a PB model to make
transient simulations of saturated-unsaturated contributions to streamflow at a field site was performed by
Stephenson and Freeze (1974) using a model based on numerical solution to the 3D Richards’ equation
(Freeze, 1971). Initially, the PB models were constrained by lack of computer power. Therefore, the first
published results from PB models using SHE (Bathurst, 1986), IHDM (Beven et al., 1987) and THALES
(Grayson et al., 1992) were all confined to small catchments ranging from a few ha to a few km2 and
simulation of events with a couple of days duration. A few years later Jain, Storm, Bathurst, Refsgaard &
Singh (1992) reported a SHE application simulating a 6-year period for the 820 km2 Kolar catchment in India
using a coarse (2 km) model grid. Towards the end of the century Refsgaard et al. (1998) reported a
comprehensive study using a coupled PB-GW-WQ model for the 3,000 km2 Danubian Lowland in Slovakia.

This article is protected by copyright. All rights reserved.

 HRU models were introduced around 1980 (Lundquist, 1978; Refsgaard & Hansen, 1982) and gradually
became the preferred modelling tool for many practitioners for discharge simulations, because they were
much easier to operate and generally achieved just as good results as PB codes (Refsgaard & Knudsen,
1996). SWAT (Arnold et al., 1998) and HBV-96 (Lindström, Johansson, Persson, Gardelin & Bergström, 1997)
became two of the most commonly used HRU codes.

The emergence of new sophisticated PB codes, coupled models and new parametrization schemes
(2000-2015)
Accepted Article

Facilitated by increasing computer power and improved numerical techniques a number of advanced PB
models were developed during the first decade of the millennium, e.g. INHM (VanderKwark & Loague,
2001), CATHY (Bixio et al., 2002; Camporese, Paniconi, Putti & Orlandini, 2010), GEOtop (Rigon, Bertoldi &
Over, 2006), ParFlow (Kollet & Maxwell, 2006) and HydroGeoSphere (Therrien et al., 2010).

Compared to one of the most advanced PB codes commonly used also before 2000 (MIKE SHE) these new
codes had more advanced numerical solvers and in some respects more detailed process descriptions, e.g.
3D instead of 1D Richards’ equation for variably saturated-unsaturated zone flow. Furthermore, some of
the codes were developed with couplings to land surface schemes and numerical weather models, e.g.
ParFlow-Noah-WRF (Maxwell et al., 2011).

Another development during this period was the development and widespread use of water quality
modules coupled to HRU codes. Examples of such HRU-WQ codes include SWAT (Arnold et al., 1998) and
HYPE (Lindström et al., 2010).

A larger focus on scalability and spatial pattern consistency paved the way for fundamentally different
parametrization schemes in PB models, such as the multiscale parameter regionalization scheme (MPR)
(Samaniego et al., 2010). By linking parameter values to attainable bio-geo-physical properties in adequate
spatial resolutions, this approach enabled seamless parametrization and better scaling properties of fluxes
(Kumar, Samaniego & Attinger, 2013). These developments have since been taken up by a range of model
codes and facilitated continental and global scale modelling (Mizukami et al., 2017).

Recent developments (after 2015)

A significant development during the last few years is the fast-increasing use of machine learning
techniques (Nearing et al., 2021; Shen, 2018; Tyralis, Papacharalampous & Langousis, 2019) both as stand-
alone tools and coupled to other model types. Machine learning techniques are data driven and until now
the most advanced version of BB models. This development has been possible due to the increasing access
to large sample and spatially consistent data. Further, the full potential of hydrological big data, which are
multivariate and spread across spatial and temporal scales, cannot be exploited by the stringent model
structures of traditional hydrological models in terms of input data and parametrization (Nearing, Mocko,
Peters-Lidard, Kumar & Xia, 2016). This created the need for more flexible models that can exploit the
abundance of the available data sources most efficiently.

Recent efforts have demonstrated that hydrological process understanding can be consolidated with
machine learning algorithms. This builds a novel class of process- or knowledge-guided machine learning
models (ML-HM) that, for instance, use simulation results from hydrological models as input to the
machine learning model (Koch, Berger, Henriksen & Sonnenborg, 2019; Konapala, Kao, Painter & Lu, 2020),

This article is protected by copyright. All rights reserved.

 build first-order conservation laws into the machine learning model (Hoedt et al., 2021; Read et al., 2019)
or directly integrate hydrological models into machine learning architectures (Kraft, Jung, Körner &
Reichstein, 2020). Further, current research focuses on explainable machine learning to overcome the black
box notion to ultimately apply machine learning to generate new hydrological process knowledge (Kratzert
et al., 2019; Nearing et al., 2021).

Developments of models and process knowledge

To illustrate some of the enormous developments since the 1960s a number of selected hydrological model
Accepted Article

studies, which advanced the scientific frontier of hydrological modelling at the time of publishing, are
shown in Table 1 together with the four proxies. The table has been confined to include models of the LU,
PB and HRU classes, while the older BB and the very recent ML-HM models are not included. Based on the
models listed in this table Figures 2 and 3 show the developments in number of spatial calculation elements
(N) and their size (Δx), while Figure 4 shows the development in the evaluation level.

To prepare Table 1 we have made literature searches and attempted to select an ensemble of models that
can characterise the development of hydrological models. Any such selection comprises elements of
subjectivity and, given the self-citations in Table 1, we cannot claim to have avoided the same bias as when
hydrologists select a model code, namely that prior knowledge and legacy may play a role (Addor &
Melsen, 2018). The model studies listed in Table 1 should therefore not be considered one by one as the 54
most important model studies during the past 60 years. However, we would argue that the ensemble of the
54 studies reflects a reasonably representative sample of key model studies during this period.

The relationship between hydrological models and hydrological process knowledge is dual. On the one
hand, more and more advanced process descriptions have been built into model codes and more and
better local data have become available for modelling. On the other hand, more sophisticated
methodologies for evaluating parameter values and reliability of model simulations have contributed to
enhancing hydrological process knowledge for numerous specific catchments and as such also generically
for hydrology.

The two most important external factors that have enabled creation of enhanced hydrological process
knowledge during the past decades are increased computer power and better availability of data. In the
hydrological scientific community these new opportunities have been exploited by use of different
strategies. Looking at the proxies listed in Table 1 and shown in Figures 2-4, some key development tracks
appear:

 Process descriptions in model codes. The level of theoretical process understanding imbedded in
the model codes increases progressively from BB to LU and PB model classes. From Figures 2-3 it is
seen that the first LU models were developed in the 1960s, while the HRU models appeared in the
1980s. The first PB models with continuous simulations of full catchments appeared in the 1990s,
while the first PB models appearing in the 1970s and 1980s due to computational constraints were
confined to simulation of single events and/or hillslopes. This development reflects that it has been
gradually possible to enhance the level of theoretical process descriptions in new emerging PB
model codes.

 Spatial discretisation. The number of spatial grids (N) in model studies has increased significantly
over time from a single unit in the 1960s to more than 10 million units in some recent modelling
studies (Figure 2). This increase in N can clearly be seen both for HRU and PB models, while the

This article is protected by copyright. All rights reserved.

 picture is less obvious for LU models. This opportunity to increase N, has however been exploited
differently for LU, HRU and PB models. The LU and HRU models have kept the same spatial grid size
(Δx) indicating that these model types have limited scope for improving model accuracy by reducing
Δx (Figure 3). Instead, LU and HRU models used the increased N to simulate larger catchments up
to continental models (Table 1). For some PB models the increased N has during the past two
decades been used to decrease Δx and in this way exploring the possibility to reduce some of the
scale problems occurring when spatial heterogeneity of catchment properties cannot be resolved.
Our findings are in agreement with Melsen et al. (2016) who reviewed 192 applications of a single
model code (VIC) published between 1997 and 2014 and found that the spatial resolutions of the
Accepted Article

VIC model grids (Δx) in general decreased while the number of grid cells (N) increased over time.
Based on this, Melsen et al. (2016) argued that the process-based evaluation should keep pace with
the increased spatial resolution, which requires new and spatially distributed data sources for
model evaluation.

 Evaluation. Figure 4 shows how the level of evaluation has gradually increased. During the first
couple of decades, hydrological modelling was dominated by deterministic, single models with
limited focus on uncertainties in data, parameters and model structures. This changed during the
1990s after the call for the hydrological modelling community to seriously consider uncertainty
(Beven, 1989) and when computer technology allowed execution of many model runs. Most recent
model studies in Table 1 perform evaluation at level 2 or higher implying that multiple data types
and/or uncertainties in parameter values and model structures are considered. Among the level
2/3/4 studies the most frequent factor contributing to increasing the evaluation level is multiple
calibration targets (23 studies) followed by multiple parameter values (11 studies) and multiple
model structures (7 studies). This reflects that multiple calibration targets are common in coupled
models (e.g. discharge and groundwater heads) and that parameter uncertainty is generally easier
to handle and better supported by built-in software tools than model structural uncertainty.
Recently, we have, however, seen an increasing attention on model structural uncertainty with
development of software platforms such as MMS (Leavesley, Markstrom, Restrepo & Viger, 2002),
SUMMA (Clark et al., 2015) and RAVEN (Craig et al., 2020). It may be noted that the only level 4
study included in Table 1 deals with modelling of karst hydrology, where uncertainties are
recognized as being very large (Hartmann et al., 2013). As comprehensive evaluation requires some
kind of ensemble modelling it can only be made at the cost of significantly reduced N, which can be
seen in Table 1.

Case study
Outline and scope of case study
Our case study builds upon a national scale coupled groundwater-surface water model based on the MIKE
SHE code (PB-GW). The model is initially set up and calibrated at a grid resolution of 500m for the entire
land phase of Denmark (43,000 km2). The calibration targets are primarily based on 305 discharge stations
and 24,000 groundwater monitoring wells. In order to map shallow groundwater table dynamics and meet
stakeholder requirements, a 100m version of the model was developed in parallel. Therefore, all model
parametrizations of the 500m model are applied in 100m resolution. After model calibration at 500m
resolution, the model is run in 100m resolution with full parameter transfer and model performance is
compared to the 500m model version utilized during parameter optimization. The national model is run in

This article is protected by copyright. All rights reserved.

 parallel for seven subdomains and for the purpose of brevity, just a single, yet with 12,500 km2 the largest,
subdomain (DK5) is presented. The model domain of DK5 is shown in Figure 5 and the aggregated
performance of the two models evaluated at the entire model domain in Table 2. The groundwater head
performance is generally poorer for deeper wells with a mean absolute error of above 4 m, whereas the
errors are smaller for the shallow groundwater system. Discharge is generally simulated at a satisfactory
level with regard to the Kling-Gupta-Efficiency and the water balance, calculated as the ratio of simulated
and observed mean flow with average errors below 5%. Based on the information provided by Table 2, we
conclude that the best process representation, depending on grid scale, cannot be determined by model
evaluation based on traditional data and performance metrics, even when evaluated on a very large
Accepted Article

observational dataset.

In order to investigate if different spatial resolutions affect the simulated hydrological processes, despite
the overall similarity in performance, we show simulated streamflow and its components for an arbitrary
branch in the DK5 model domain (Figure 6). Overall, the simulated discharge is very comparable between
the 100m and the 500m model. However, the streamflow components vary considerably. Baseflow is the
dominating source in both models, but more pronounced in the 500m model. The reduced baseflow of the
100m model is substituted with an increased drain flow and overland flow with respect to the 500m model.
The overland flow is relatively small in the 100m model and two orders of magnitude less in the 500m
model. This again underlines that standard hydrological data, such as discharge observations, cannot
discriminate between the two models even though they provide different process knowledge. The
apparent differences in simulated streamflow components will have implications for solute transport and
e.g. in-stream nitrate loads.

In order to further investigate the spatially explicit hydrological process knowledge provided by the 100m
and 500m models, we show the average simulated water levels in Figure 5. At regional scale, the two
models resemble each other visually. First when zooming at local scale, covering a few km2, differences in
simulated water levels emerge. The uppermost water table, which is represented by the water level maps
in Figure 5, is topography driven. Small scale variations in topography yield water level heterogeneity, such
as surficial water levels in valley bottoms and deeper water levels for smaller topographical features. This
heterogeneity can effectively be simulated by an increase in spatial resolution, from 500m to 100m,
without recalibrating the model parameters.

As traditional hydrological observations cannot discriminate between the two models, alternative
observations need to be consulted for evaluating. For this we processed a land surface temperature (LST)
dataset which can be utilized as a proxy for soil wetness conditions and thereby shallow water levels. The
process of evaporative cooling typically yields a lower LST for wet than for dry soils. The thermal sensors of
the Landsat satellites provide LST estimates at 30m spatial resolution, which can potentially explain
variability of water table depth for areas with homogenous land use. In order to evaluate the hydrological
process knowledge gained by the abovementioned models (500m and 100m) we have generated a long-
term (1997-2019) average LST map using Landsat data obtained from the Analysis Ready Data (ARD)
developed by the Global Land Analysis and Discovery team (GLAD), which is described in full detail by
Potapov et al. (2020).

We clearly observed that grids with a water level close to the surface are characterized with cooler LST,
whereas warmer grids are associated with a deeper water level. This link is only apparent in the 100m
model, while the 500m cannot reflect this. Such local scale differences are not discriminated by the
objective functions used for calibration, or from large scale pattern comparisons, but will have significant
impacts on groundwater surface water interactions and flowpaths. Based on the high-resolution evaluation

This article is protected by copyright. All rights reserved.

 against the surface wetness proxy (LST) from the Landsat data, we rate the 100m model and the resulting
hydrological process knowledge as more trustworthy.

Discussion
Drivers and obstacles for utilizing hydrological process understanding in catchment
models
Accepted Article

Spatial heterogeneity of catchment properties – scale problems

The most fundamental problem in hydrological modelling relates to handling of spatial heterogeneity of
physical properties and hydrological variables. This is done differently in LU and PB models. In LU models
the heterogeneity of catchment properties is built into process equations. A beautiful example of this is the
infiltration equation in the Stanford Watershed model (Crawford & Linsley, 1966), where the infiltration
capacity varies across the catchment in a contributing area approach, so that the fraction of the catchment
contributing to overland flow increases with rainfall intensity. Similarly, Lahmer, Becker, Müller-Wohlfeil &
Pfützner (1999) nicely illustrated how a HRU model can represent heterogeneity of catchment properties
via statistical distribution functions of e.g. water holding capacity.

The process equations in PB models are, in contrary, point scale equations assuming homogeneous
conditions, where catchment heterogeneity only can be represented through variability of parameter
values. This is typically done using field data such as soil types, vegetation classes and geological units and
assign different parameter values to each category. An important limitation of this approach is that use of
effective parameter values for each category does not account for all heterogeneity. It is for instance well
documented that the spatial variability of e.g. soil properties within one standard soil type at field scale is
very high and can significantly influence the water balance and solute transport at this scale (Djuurhus,
Hansen, Schelde & Jacobsen, 1999; Nielsen, Bigger & Erk, 1973). Hence, even with a spatially distributed
approach that enables PB models to explain a large part of the catchment heterogeneity, a considerable
part of the heterogeneity may still not be accounted for, which in some cases leads to performance
problems (Andersen, Refsgaard & Jensen, 2001; Hansen, Refsgaard, Hansen & Ernstsen, 2007). This
obstacle has led to the development of parameter regionalization strategies that link fully distributed
catchment properties, such as soil texture or vegetation indices, to effective model parameters via transfer
functions. The transfer functions have just a few parameters that need to be calibrated and result in
physically meaningful, i.e. seamless, parameter distributions. Such a regionalization is implemented into
mHM (Samaniego et al., 2010), but can in theory be added to any model, which has been demonstrated by
Beck et al. (2016), Mizukami et al., (2017), Demirel et al., (2018).

Another problem is that spatial discretization of a model domain does not allow a model to resolve
variations of e.g. topography and water levels within a computational unit. As illustrated in our case study
this may adversely affect the realism of simulated hydrological processes. While Refsgaard (1996) reported
that a 500m model calibrated against streamflow and groundwater heads for a neighbouring area showed
a deteriorating performance when applied with larger grid sizes (1000m/2000m/4000m), the 100m model
in our case study at a first glance showed similar performance to the 500m model for which the calibration
was made. This indicates on one hand, that a 500 m spatial resolution in our hydrogeological setting may
be sufficient for reproducing streamflow and groundwater heads from deep wells, and that a calibration of
the 100m model against the same targets as used for the 500m model calibration would not significantly
affect the 100m model performance and simulations. The large differences between the 100m and 500m

This article is protected by copyright. All rights reserved.

 models in partitioning of streamflow components and shallow water table depth, on the other hand, clearly
demonstrate the inadequacy of the present, state-of-the-art, model evaluation strategy. This suggests that
improved evaluation strategies exploiting more field data will be required to further enhance the
hydrological process knowledge in the case study areas.

From theoretical process understanding to enhanced process knowledge in a catchment

Due to lack of sufficient field data any hydrological model remains a hypothesis that needs to be thoroughly
tested before it can be stated to represent catchment conditions. All hydrological models applied beyond
Accepted Article

the point scale require some degree of calibration (Hrachowitz & Clark, 2017).

The possibility to exploit state-of-the-art hydrological process understanding built into a model code to
enhance hydrological process knowledge for a catchment therefore depends not only on model structure
but also on how a model is parameterised, how much field data are available to assess parameter values
and which evaluation methods and calibration techniques are used. Methodologies to support parameter
optimization have developed extensively from the first parameter optimisation attempts using a single
objective function and simple search techniques (Dawdy & O’Donnel, 1965) to gradually more efficient
parameter optimization techniques allowing use of multiple objective functions and evaluation of pareto
fronts between multiple calibration targets (Asadzadeh & Tolson, 2013; Stisen et al., 2018). The use of
multiple objectives originally focused on evaluating several performance metrics from the same
observation data, typically streamflow hydrographs (Yapo, Gupta & Sorooshian, 1998). Recently, more
attention has been given to utilizing multiple types of observational data and exploiting the multitude of
satellite based data available (Madsen, 2003; Nijzink et al., 2018). Particularly, confronting PB models with
spatial pattern information has been subject to development (Dembele, Hrachowitz, Savenije,Mariethoz &
Schaefli, 2020; Koch, Demirel & Stisen, 2018; Zink, Mai, Cuntz & Samaniego, 2018) and have highlighted the
limitations and non-uniqueness of streamflow-based optimization in PB modelling (Demirel et al., 2018).

Due to uncertainties in data, parameter values and model structure as well as mismatch of scales between
observations and models, model simulations will always include uncertainties, which often can be
considerable and crucial for water management practice. Although many practitioners and water managers
struggle with how to cope with uncertainty (Höllermann & Evers, 2019), uncertainty assessment is today a
fundamental element of good practices in both hydrological science and practical water management
(Refsgaard et al., 2013; Poff et al., 2016).

What is the best strategy to enhance hydrological process knowledge in models?

Different research groups are following different strategies to enhance hydrological process knowledge.
One strategy, which we shall denote brute force, is to pursue the PB modelling track outlined by Freeze &
Harlan (1969) in a fundamentally deterministic approach. Another strategy, which we shall denote
comprehensive evaluation, is to explicitly consider uncertainties in data and models and use advanced
model evaluation tools to assess the reliability of parameter values and model simulations. The two
strategies are competing and complementary.

The PB brute force strategy implies further model developments by including more and more detailed
process descriptions and by adopting gradually smaller spatial grids. The basic hypothesis behind this
strategy is that it will lead to models with improved accuracy and predictive capabilities. This was the
dominating strategy until the 1990s and is still used today (e.g. Kollet et al., 2018).

This article is protected by copyright. All rights reserved.

 The comprehensive evaluation strategy shares the views first expressed by Beven (1989) that single
deterministic models are not adequate to reliably assess a model’s predictive capability due to the large
uncertainties in data and models. Modellers adopting this strategy therefore use a considerable part of a
model study resource to run ensembles of models to evaluate parameters and predictions. This strategy
has gained increasing attention in recent years. Hrachowitz & Clark (2017) argue along this line that “the
fundamental problems in catchment modelling do not lie in the type of model used but in the way a model
is applied” implying that model evaluation is crucial to achieve reliable simulations that support improved
process knowledge of a particular catchment.
Accepted Article

We argue that both strategies can provide valuable insights contributing to enhancing hydrological process
knowledge. Although the brute force studies may not result in more reliable models as such they have
moved the borders for model applications with more sophisticated and detailed process descriptions,
larger number of small-scale computational grids and coupling with other domains. The comprehensive
evaluation studies, on the other hand, are required to fully test the hypotheses that may emerge from the
brute force studies, to provide documented reliable models and hence to consolidate the process
knowledge. Brute-force modelling applications are often at the limit of what is possible given the available
computational power. Through technological development, a sophisticated PB model that was once used in
a brute-force strategy can later be applied more comprehensively.

Perspectives for enhancing hydrological process knowledge in future catchment

modelling
The emergence of big data
What is the most important factor constraining the performance of hydrological models? Is it i) the model
codes with their built-in process equations and default parameter values (Cuntz et al., 2016); ii) ineffective
methodologies for parameterisation, testing and use of models; or iii) lack of sufficient, good quality data?
This question has been discussed extensively during the past 60 years (Bergström, 1991; Beven, 2002;
Hrachowitz & Clark, 2017; Refsgaard et al., 2016).

In spite of enormous progress in new measurement techniques such as remote sensing, geophysics, and
instruments for ground‐based measurements, the general perception until recently has been, as argued by
Beven (2002), that “catchment hydrology suffers from a measurement problem,” which “is a basic
limitation on the possibility for characterizing a catchment system.” A commonly accepted view has
therefore been that the development of sophisticated model codes has been ahead of the development of
data acquisition techniques and that the lack of data constituted the most important constraint. This view
was supported by the fact that simple LU-models usually turned out to have the same prediction accuracy
as complex PB models, at least for streamflow simulation (Refsgaard and Knudsen, 1996; Reed et al., 2004).
A logical consequence of this is that the parsimonious principle leads to use of robust LU models
(Bergström, 1991, Perrin et al., 2003).

We would argue that the emergence of huge amounts of new high-resolution spatial data that we witness
in these years, and for which we may only have seen the beginning, indicates a new era with
unprecedented new opportunities for hydrological modelling and enhanced hydrological process
knowledge. This is likely to change the balance between the constraining factors, so that lack of data will
not be as important as previously. During the first decades of model development, especially PB models
were typically a step ahead with respect to data availability. This has completely turned around during the

This article is protected by copyright. All rights reserved.

 past decade. Now, the model codes that were developed in the data-poor decades struggle to make use of
the available data with respect to model parametrization and process descriptions (Best et al., 2015). The
applications on ML models on hydrological data are often stories of success where a ML model
outperforms a conventional hydrological model. Based on this it may be argued that hydrological models
cannot exploit the information contained in the available data and therefore flexible ML models, that are
not constrained by predefined process descriptions yield a better performance (Nearing et al., 2016). At the
same time, it should, however, be recognised that better performance of ML models not necessarily can be
associated with enhanced hydrological knowledge, and that ML models may have relatively poorer
performance outside the range of calibration data. To exploit all the new available data and at the same
Accepted Article

time enhance the hydrological process knowledge, we see a large scope for further developments of
advanced models – and this is likely to be PB and ML-HM models.

According to a quote attributed to Albert Einstein “everything should be made as simple as possible, but no
simpler”. This implies that a model that can be characterised as parsimonious at one point in time may be
considered too simple later on, when more data have become available. We believe that the good old
phrase “we will never have enough data to parameterize our distributed models” might soon be outdated.
This does not imply that additional data may not be useful, but that the benefit of additional data will
decrease. While we consider this realistic for discharge and surface flux aspects of catchment modelling, it
will probably still not apply for contaminant transport studies at small scale nor for studies assessing the
detailed spatial heterogeneity of the subsurface.

More advanced model evaluation strategies

Our case study clearly demonstrated the inadequacy of the existing model evaluation strategy and
illustrated the need for developing improved evaluation strategies exploiting more field data.

In light of the increased process descriptions in PB models and availability of new data types more
advanced model evaluation strategies are emerging. One important element is the increased attention on
evaluating the spatial pattern performance of distributed PB models. Historically, even complex and
distributed PB models have been evaluated solely against catchment aggregated observations (stream
flow) or point observations such as eddy flux data, soil moisture or groundwater levels. This evaluation
approach neglects the ability of the models to reproduce spatial patterns in hydrological states and fluxes.
Improved satellite remote sensing data and the increased focus on seamless model parametrization has
facilitated the development of methods that exploit spatial pattern information in model evaluation (Koch,
Siemann, Stisen & Sheffield, 2016; Stisen et al, 2018) and have highlighted the limitations and non-
uniqueness of streamflow-based optimization in PB modelling (Demirel et al., 2018). Out of this
development a new set of spatial pattern-oriented model evaluation metrics have emerged (Zink et al.,
2018; Koch et al., 2018; Dembele et al., 2020) and applied also for large scale (Koch et al., 2020) and
operational modeling systems (Soltani et al., 2021) as well as for improving process representation and
understanding (Hulsman, Savanije & Hrachowitz, 2021). Another model evaluation approach facilitated by
improved model complexity and data availability is the tracer-aided models (Birkel & Soulsby, 2015;
Sprenger, Tetzlaff, Buttle, Laudon & Soulsby., 2018) which exploit the increasing availability of isotope
tracer data to close the information gap on runoff generation processes. This approach has enabled a move
beyond simple hydrograph separation and runoff calibration to gain insight into flow partitioning and
transit times at catchment scale (Smith, Tetzlaff, Kleine, Maneta & Soulsby, 2021). Coupling this to
integrated groundwater surface water models seems to be an interesting avenue for future developments
and research (Sterte, Lidman, Lindborg, Sjöberg & Laudon, 2021).

This article is protected by copyright. All rights reserved.

 Incorporation of machine learning techniques
In our model classification we introduce hybrid models (ML-HM), as a recent reoccurrence and
advancement of the initial BB models. Given the nowadays availability of hydrological big data and the
technological developments of machine learning in general, such data-driven applications have become
increasingly popular. There is an abundance of successful stand-alone machine learning studies in the
hydrological literature in the recent years. However, in order to relate to the theme of our review we like to
limit our discussion on the class of hybrid models that to some degree exploits hydrological process
Accepted Article

understanding. Due to the limitations of state-of-the-art hydrological models to exploit available big data
sources we see a clear need to further the advancement of ML-HM. While different strategies have been
reported (Koch et al., 2019; Read et al., 2019; Konapala et al., 2020; Kraft et al., 2020; Hoedt et al., 2021),
formalized approaches to introduce process understanding into ML models have not yet been developed.

Recent literature indicates that ML-HM leverage hydrological predictions at improved spatial scales and
overall accuracy with respect to state-of-the-art hydrological models. However, it remains unresolved to
what degree ML-HM can generate novel hydrological process understanding. So far ML-HM are typically
trained to predict a single variable of the hydrological cycle, which is a drawback when compared to most
conventional hydrological models that simulate a broad range of hydrological states and fluxes jointly. Also,
it remains unresolved to what degree ML-HM can be utilized for hypothesis testing and scenario analysis,
since the capability of ML-HM models to extrapolate well beyond the variability contained in the training
data has not yet been properly examined.

Opposed to the abovementioned, where hydrological process understanding is imposed on a ML model,

the application of ML concepts to provide insights into process descriptions and hydrological models in
general presents an alternative approach. Here, generating a prediction with the highest possible accuracy
is not the prime motivation, instead the aim is to e.g. assess the information content of models via
information theory principles (Nearing et al. 2020).

Improved models – enhanced credibility

The benefit of a model in practical water management is not only dependent on performance of the model.
It also depends on the credibility of the model as perceived by stakeholders and policy makers. If a model is
not perceived credible it will not be used as a basis for decision making. Even if a model has a performance
that is well documented by rigorous tests, it may still be perceived as unreliable by some stakeholders, if
they cannot see how the model reflects the local catchment conditions as they see them. Many
stakeholders have for instance problems understanding how lumping of processes and effective parameter
values at larger computational units can represent the heterogeneity of soils, land use, topography, and
river network that they see locally in the field.

As an example, our experience when moving from 2D to 3D groundwater modelling was that it improved
the communication with water managers who could recognize geological data from boreholes in the model
representation. Therefore, they considered the 3D model to be more reliable than a 2D model, even if the
test results showed similar performance. Our general understanding from two decades of work with a
national water resources model in Denmark is that non-modellers perceive a model as being more credible
if they can find their own data in the model (Højberg et al., 2013). The possibility to recognise own data can
sometimes be seen as a test to pass before stakeholders open their mind and become willing to learn from

This article is protected by copyright. All rights reserved.

 a model. In this respect, the possibility to build models that can utilise the new high-resolution good quality
data together with improved process understanding, holds a great potential for enhancing the credibility
and hence the application of hydrological models to support future water management.

Conclusions
Much of the hydrological process knowledge, both in terms of theoretical knowledge on processes and
specific field data, has been derived from studies at small (point or plot) spatial scales. Hydrological
Accepted Article

catchment models have until now not been able to operate at sufficiently small spatial scales to exploit this
knowledge. The vast difference in hydrological simulations of flow components between the 100m and the
500m models in our case study, is a clear example of this barriere. The two main obstacles in this regard
have during the past decades been the limitations in computer power and the lack of high-resolution data
for many catchment properties and hydrological variables.

The emergence of huge amounts of high-resolution good quality spatial data in these years combined with
the ongoing rapidly increase of computational power, are in our view likely within the next decade to
remove this historical barrier and bridge the gap between local scale hydrological process knowledge and
catchment modelling.

To make this happen and in this way enhance hydrological process knowledge, there is a need to develop
both improved hydrological model codes and improved evaluation strategies. We argue that model codes
with best possibilities to exploit all the new data would be spatially high-resolution process-based models
combined with machine learning techniques. In addition, new evaluation strategies need to include
calibration methodologies to better simulate spatial patterns and multiple hydrological variables.

Acknowledgements
We appreciate the valuable review comments and suggestions from James Kirchner and Hoshin Gupta.

ORCID
Jens Christian Refsgaard - http://orcid.org/0000-0003-0420-349X

Simon Stisen - https://orcid.org/0000-0001-6695-8412

Julian Koch - https://orcid.org/0000-0002-7732-3436

References
Abbaspour, K.C., Rouholahnejad, E., Vaghefi, S., Srinivasan, R., Yang, H., & Kløve, B. (2015). A continental-scale
hydrology and water quality model for Europe: Calibration and uncertainty of a high-resolution large-scale SWAT
model. Journal of Hydrology, 524, 733-752.
Abbott, M.B., Bathurst, J.C., Cunge, J.A., O'Connel, P.E., & Rasmussen, J. (1986a). An introduction to the European
Hydrological System - Systeme Hydrologique Européen "SHE", 1: History and philosophy of a physically-based
distributed modelling system. Journal of Hydrology, 87, 45-59.

This article is protected by copyright. All rights reserved.

 Abbott, M.B., Bathurst, J.C., Cunge, J.A., O'Connel, P.E., & Rasmussen, J. (1986b). An introduction to the European
Hydrological System - Systeme Hydrologique Européen "SHE", 2: Structure of a physically-based distributed
modelling system. Journal of Hydrology, 87, 61-77.
Addor, N., & Melsen, L.A. (2019). Legacy, rather than adequacy, drives the selection of hydrological models. Water
Resources Research, 55, 378–390.
Andersen, J., Refsgaard, J.C., & Jensen, K.H. (2001). Distributed hydrological modelling of the Senegal River Basin –
model construction and validation. Journal of Hydrology, 247, 200-214.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., & Williams, J.R. (1998). Large area hydrologic modeling and assessment -
Part 1: Model development. Journal of the American Water Resources Association, 34, 73-89.
Arnold, J.G., Muttiah, R.S., Srinivasan, R., & Allen, P.M. (2000). Regional estimation of base flow and groundwater
Accepted Article

recharge in the Upper Mississippi river basin. Journal of Hydrology, 227, 21–40.
Auken, E., Boesen, T., & Christiansen, A.V. (2017). A review of airborne electromagnetic methods with focus on
geotechnical and hydrological applications from 2007 to 2017. Advances in Geophysics, 58, 47-93.
Asadzadeh, M., & Tolson, B. (2013). Pareto archived dynamically dimensioned search with hypervolume-based
selection for multi-objective optimization. Engineering Optimization, 45(12), 1489-1509.
Bartosova, A., Capell, R., Olesen, J.E., Jabloun, M., Refsgaard, J.C., Donelly, C., Hyytiäinen, K., Pihlainen, S., Zandersen,
M., & Arheimer, B. (2019). Future socioeconomic conditions may have a larger impact than climate change on
nutrient loads to the Baltic Sea. Ambio, 48(11), 1325-1336.
Bathurst, J.C. (1986). Physically-based distributed modelling of an upland catchment using the Systeme Hydrologique
Europeén. Journal of Hydrology, 87, 103-123.
Beck, H.E., van Dijk, A.I.J.M., de Roo, A., Miralles, D.G., McVicar, T.R., Schellekens, J., & Bruijnzeel, L.A. (2016). Global-
scale regionalization of hydrologic model parameters. Water Resources Research, 52, 3599–3622.
Becker, A. (1977). The integrated hydrological catchment model EGMO. Hydrological Sciences Journal, 22(1), 145-151.
Bergström, S. (1991). Principles and confidence in hydrological modelling. Nordic Hydrology, 22, 123-136.
Bergström, S., Forsman, A. (1973). Development of a conceptual deterministic rainfall-runoff model, Nordic Hydrology,
4, 147-170.
Best, M.J., Abramowitz, G., Johnson, H.R., Pitman, A.J., Balsamo, G., Boone, A., Cuntz, M., Decharme, B., Dirmeyer,
P.A., Dong, J., Ek, M., Guo, Z., Haverd, V., Van Den Hurk, B.J.J., Nearing, G.S., Pak, B., Peters-Lidard, C., Santanello
Jr, J.A., Stevens, J.L., & Vuichardi, N. (2015). The Plumbing of Land Surface Models: Benchmarking Model
Performance. Journal of Hydrometeorology, 16, 1425-1442.
Beven, K. (1989). Changing ideas in hydrology – the case of physically based models. Journal of Hydrology, 105, 157-
172.
Beven, K. (1995). Linking parameters across scales: subgrid parameterizations and scale dependent hydrological
models. Hydrological Processes, 9, 507-525.
Beven, K. (2002). Towards an alternative blueprint for a physically based digitally simulated hydrologic response
modelling system. Hydrological Processes, 16(2), 189–206.
Beven, K. (2006). A manifesto for the equifinality thesis. Journal of Hydrology, 320, 18–36.
Beven, K., Calver, A., & Morris, E.M. (1987). The Institute of Hydrology distributed model. Institute of Hydrology
Report 98, Wallingford UK.
Beven, K.J., Kirkby, M.J. (1979). A physically based, variable contributing area model of basin hydrology. Hydrological
Sciences Bulletin, 24, 43-69.
Birkel, C., Soulsby, C. (2015). Advancing tracer-aided rainfall–runoff modelling: a review of progress, problems and
unrealised potential. Hydrological Processes, 29(25), 5227-5240.
Bixio, A.C., Gambolati, G., Paniconi, C., Putti, M., Shestopalov, V.M., Bublias, V.N., Bohuslavsky, A.S., Kasteltseva, N.B.,
& Rudenko, Y.F. (2002). Modeling groundwater-surface water interactions including effects of morphogenetic
depressions in the Chernobyl exclusion zone, Environmental Geology, 42(2–3), 162–177.
Blöschl, G., & Sivapalan, M. (1995). Scale issues in hydrological modelling: a review. Hydrological Processes, 9, 251-
290.
Box, G.E.P., & Jenkins, G.M. (1970). Time series analysis, forecasting and control. Holden-Day Inc., San Francisco.

This article is protected by copyright. All rights reserved.

 Burnash, R.F.J., Ferral, R.L., & McGuire, R.A. (1973). A general streamflow simulation system – conceptual modelling
for digital computers. US Department of Commerce, National Weather Service and State of California,
Department of Water Resources, Sacramento, 204 pp.
Butts, M.B., Payne, J.T., Kristensen, M., & Madsen, H. (2004). An evaluation of the impact of model structure on
hydrological modelling uncertainty for streamflow simulation. Journal of Hydrology, 298, 242-266.
Camporese, M., Paniconi, C., Putti, M., & Orlandini, S. (2010). Surface-subsurface flow modeling with path-based
runoff routing, boundary condition-based coupling, and assimilation of multisource observation data. Water
Resources Research, 46, W02512.
Cao, G., Zheng, C., Scanlon, B., Liu, J., & Li, W. (2013). Use of flow modeling to assess sustainability of groundwater
resources in the north China Plain. Water Resources Research, 49, 159-175.
Accepted Article

Chaney, N.W., Metcalfe, P., & Wood, E.F. (2016). HydroBlocks: a field-scale resolving land-surface model for
application over continental extents. Hydrological Processes, 30, 3543-3559.
Clark, M.P., Slater, A.G., Rupp, D.E., Woods, R.A., Vrugt, J.A., Gupta, H.V., Wagener, T., & Hay, L.E. (2008). Framework
for Understanding Structural Errors (FUSE): A modular framework to diagnose differences between hydrological
models. Water Resources Research, 44, W00B02.
Clark, M.P., Nijssen, B., Lundquist, J.D., Kavetski, D., Rupp, D.E., Woods, R.A., Freer, J.E., Gutmann, E.D., Wood, A.W.,
Brekke, L.D., Arnold, J.R., Gochis, D.J., & Rasmussen, R.M. (2015) A unified approach for process-based
hydrological modeling: 1. modeling concept. Water Resources Research, 51, 2498–2514.
Craig, J.R., Brown, G., Chlumsky, R., Jenkinson, R.W., Jost, G., Lee, K., Mai, J., Serrer, M., Sgro, N., Shafii, M., Snowdon,
A.P., & Tolson, B.A. (2020). Flexible watershed simulation with the Raven hydrological modelling framework.
Environmental Modelling & Software, 129, 104728.
Crawford, N.H., Linsley, R.K. (1966). Digital simulation in hydrology, Stanford Watershed Model IV, Department of Civil
Engineering, Stanford University, Technical Report 39.
Cuntz, M., Mai, J., Samaniego, L., Clark, M., Wulfmeyer, V., Branch, O., Attinger, S., & Thober, S. (2016). The impact of
standard and hard-coded parameters on the hydrologic fluxes in the Noah-MP land surface model. Journal of
Geophysical Research: Atmospheres, 121, 10,676–10,700.
Das, N.N., Mohanty, B.P., Cosh, M.H., & Jackson, T.J. (2008). Modeling and assimilation of root zone soil moisture
using remote sensing observations in Walnut Gulch watershed during SMEX04. Remote Sensing of Environment,
112, 415-429.
Dawdy, D.R., O’Donnel, T. (1965) Mathematical models of catchment behaviour. Journal of the Hydraulics Division,
ASCE, 91(4), 123-137.
De Lange, W.J., Prinsen, G.F., Hoogewoud, J.C., Veldhuizen, A.A., Verkaik, J., Oude Essink, G.H.P., van Walsum, P.E.V.,
Delsman, J.R., Hunink, J.C., Massop, H.T.L., & Kroon, T. (2014). An operational, multi-scale, multi-model system
for consensus-based, integrated water management and policy analysis: The Netherlands Hydrological
Instrument. Environmental Modelling & Software, 59, 94-108.
De Schepper, G., Therrien, R., Refsgaard, J.C., He, X., Kjaergaard, C., & Iversen, B.V. (2017). Simulating seasonal
variations of tile drainage discharge in an agricultural catchment. Water Resources Research, 53(5), 3896-3920.
Dembélé, M., Hrachowitz, M., Savenije, H.H.G., Mariéthoz, G., & Schaefli, B. (2020). Improving the Predictive Skill of a
Distributed Hydrological Model by Calibration on Spatial Patterns With Multiple Satellite Data Sets. Water
Resources Research, 56(1), e2019WR026085.
Demirel, M.C., Mai, J., Mendiguren, G., Koch, J., Samaniego, L., & Stisen, S. (2018). Combining satellite data and
appropriate objective functions for improved spatial pattern performance of a distributed hydrologic model.
Hydrology and Earth System Sciences, 22, 1299–1315.
Djuurhus, J., Hansen, S., Schelde, K., & Jacobsen, O.H. (1999). Modelling mean nitrate leaching from spatially variable
fields using effective parameters. Geoderma, 87, 261-279.
Ehlers, L.B., Refsgaard, J.C., & Sonnenborg, T.O. (2019). Observational and predictive uncertainties for multiple
variables in a spatially distributed hydrological model. Hydrological Processes, 33, 833-848.
Fang, X., Pomeroy, J.W., Westbrook, C.J., Guo, X., Minke, A.G., & Brown, T. (2010). Prediction of snowmelt derived
streamflow in a wetland dominated prairie basin. Hydrology and Earth System Sciences, 14, 991–1006.

This article is protected by copyright. All rights reserved.

 Fersch, B., Senatore, A., Adler, B., Arnault, J., Mauder, M., Schneider, K., Völksch, I., & Kunstmann, H. (2020). High-
resolution fully coupled atmosphere-hydrological modeling: a cross-compartment regional water and energy
cycle evaluation. Hydrology and Earth System Sciences, 24, 2457–2481.
Flügel, W.A. (1995). Delineating hydrological response units by geographical information system analyses for regional
hydrological modelling using PRMS/MMS in the drainage basin of the River Bröl, Germany. Hydrological
Processes, 9(3-4), 423-426.
Freer, J., Beven, K., & Ambroise, B. (1996). Bayesian estimation of uncertainty in runoff prediction and the value of
data: An application of the GLUE approach. Water Resources Research, 32(7), 2161-2173.
Freeze, R.A. (1971). Three-dimensional transient saturated-unsaturated flow in a groundwater basin. Water Resources
Research, 7(2), 347-366.
Accepted Article

Freeze, R.A., Harlan, R.L. (1969). Blueprint for a physically-based digitally-simulated hydrologic response model.
Journal of Hydrology, 9, 237-258.
Goderniaux, P., Brouyère, S., Fowler, H.J., Blenkinsop, S., Therrien, R., Orban, P., & Dassargues, A. (2009). Large scale
surface-subsurface hydrological model to assess climate change impacts on groundwater reserves. Journal of
Hydrology, 373, 122–138.
Graham, D.N., Butts, M.B. (2005). Flexible integrated watershed modelling with MIKE SHE. In: Singh VP, Frevert DK
(Eds) Watershed Models. CRC Press, Chapter 10.
Grayson, R.B., Moore, I.D., & McHahon, T.A. (1992). Physically based hydrologic modelling, 1. A terrain-based model
for investigative purposes. Water Resources Research, 28(10), 2639-2658.
Hansen, J.R., Refsgaard, J.C., Hansen, S., & Ernstsen, V. (2007). Problems with heterogeneity in physically based
agricultural catchment models. Journal of Hydrology, 342 (1-2), 1-16.
Hartmann, A., Wagener, T., Rimmer, A., Lange, J., Brielmann, H., & Weiler, M. (2013). Testing the realism of model
structures to identify karst system processes using water quality and quantity signatures. Water Resources
Research, 49, 3345–3358.
Henriksen, H.J., Troldborg, L., Nyegaard, P., Sonnenborg, T.O., Refsgaard, J.C., & Madsen, B. (2003). Methodology for
construction, calibration and validation of a national hydrological model for Denmark. Journal of Hydrology, 280,
52-71.
Henriksen, H.J., Kragh, S.J., Gotfredsen, J., Ondracek, M., van Til, M., Jakobsen, A., Schneider, R.J.M., Koch, J.,
Troldborg, L., Rasmussen, P., Pasten-Zapata, E., & Stisen, S. (2020). Udvikling af landsdækkende
modelberegninger af terrænnære hydrologiske forhold i 100m grid ved anvendelse af DK-modellen
(Development of nationwide model simulations of shallow hydrogeology in a 100 m grid with the DK model).
Geological Survey of Denmark and Greenland.
Hoedt, J.P., Kratzert, F., Klotz, D., Halmich, C., Holzleitner, M., Nearing, G., Hochreiter, S., & Klambauer, G. (2021). MC-
LSTM: Mass-conserving LSTM. arXiv:2101.05186v2.
Höllermann, B., & Evers, M. (2019). Coping with uncertainty in water management: Qualitative system analysis as a
vehicle to visualize the plurality of practitioners' uncertainty handling routines. Journal of Environmental
Management, 235, 213-223.
Hrachowitz, M., & Clark, M.P. (2017). HESS Opinions: The complementary merits of competing modelling philosophies
in hydrology. Hydrology and Earth System Sciences, 21, 3953-3873.
Hulsman, P., Savenije, H.H.G., & Hrachowitz, M. (2021). Learning from satellite observations: increased understanding
of catchment processes through stepwise model improvement. Hydrology and Earth System Sciences, 25, 957-
982.
Højberg, A.L., Troldborg, L., Stisen, S., Christensen, B.B.S., & Henriksen, H.J. (2013). Stakeholder driven update and
improvement of a national water resources model. Environmental Modelling & Software, 40, 202–213.
Jain, S.K., Storm, B., Bathurst, J.C., Refsgaard, J.C., & Singh, R.D. (1992). Application of the SHE to Catchments in India -
Part 2: Field Experiments and Simulation Studies with the SHE on the Kolar Subcatchment of the Narmada River.
Journal of Hydrology, 140, 25-47.
Jakeman, A.J., & Hornberger, G.M. (1993). How much complexity is warranted in a rainfall-runoff model? Water
Resources Research, 29(8), 2637-2649.
Jakeman, A.J., Letcher, R.A., & Norton, J.P. (2006). Ten iterative steps in development and evaluation of environmental
models. Environmental Modelling & Software, 21, 602–614.

This article is protected by copyright. All rights reserved.

 Jing, M., Hesse, F., Kumar, R., Wang, W., Fischer, T., Walther, M., Zink, M., Zech, A., Samamiego, L., Kolditz, O., &
Attinger, S. (2018). Improved regional-scale groundwater representation by the coupling of the mesoscale
hydrologic model (mHM v5.7) to the groundwater model OpenGeoSys (OGS). Geoscientific Model Development,
11, 1989–2007.
Jønch-Clausen, T., & Refsgaard, J.C. (1984). A Mathematical Modelling System for Flood Forecasting. Nordic Hydrology,
15, 307-318.
Kirchner, J.W. (2006). Getting the right answers for the right reasons: Linking measurements, analyses, and models to
advance the science of hydrology. Water Resources Research, 42, W03S04
Koch, J., Siemann, A., Stisen, S., & Sheffield, J. (2016). Spatial validation of large-scale land surface models against
monthly land surface temperature patterns using innovative performance metrics. Journal of Geophysical
Accepted Article

Research, 121 (10), 5430–5452.

Koch, J., Demirel, M.C., & Stisen, S. (2018). The SPAtial EFficiency metric (SPAEF): multiple-component evaluation of
spatial patterns for optimization of hydrological models. Geoscientific Model Development, 11, 1873-1886.
Koch, J., Berger, H., Henriksen, H.J., & Sonnenborg, T.O. (2019). Modelling of the shallow water table at high spatial
resolution using random forest. Hydrology and Earth System Sciences, 23, 4603-4619.
Koch, J., Zhang, W., Martinsen, G., He, X., & Stisen, S. (2020). Estimating Net Irrigation Across the North China Plain
Through Dual Modeling of Evapotranspiration. Water Resources Research, 56(12), e2020WR027413.
Kollet, S.J., & Maxwell, R.M. (2006). Integrated surface-groundwater flow modeling: A free-surface overland flow
boundary condition in a parallel groundwater flow model. Advances in Water Resources. 29(7), 945–958.
Kollet, S.J., & Maxwell, R.M. (2008). Capturing the influence of groundwater dynamics on land surface processes using
an integrated, distributed watershed model. Water Resources Research, 44, W02402.
Kollet, S.J., Maxwell, R.M., Woodward, C.S., Smith, S., Vanderborgh, J., Vereecken, H., & Simmer, C. (2010). Proof of
concept of regional scale hydrologic simulations of hydrologic resolution utilizing massively parallel computer
resources. Water Resources Research, 46, W04201.
Kollet, S.J., Gasper, F., Brdar, S., Goergen, K., Hendricks-Franssen, H.J., Keune, J., Kurtz, W., Küll, V., Pappenberger, F.,
Poll, S., Trömel, S., Shrestha, P., Simmer, C., & Sulis, M. (2018). Introduction of an experimental terrestrial
forecasting/monitoring system at regional to continental scales bases on the Terrestrial System Modeling
Platform (v1.1.0). Water, 10, 1697.
Konapala, G., Kao, S.C., Painter, S.L., & Lu, D. (2020). Machine learning assisted hybrid models can improve streamflow
simulation in diverse catchments across the conterminous US. Environmental Research Letters, 15, 104022.
Kraft, B., Jung, M., Körner, M., & Reichstein, M. (2020). Hybrid modeling: Fusin of a deep learning approach and a
physics-based model for global hydrological modeling. The International Archives of the Photogrammetry,
Remote Sensing and Spatial Information Sciences, Volume XLIII-B2-2020, 2020 XXIV ISPRS Congress (2020
edition).
Kratzert, F., Klotz, D., Shalev, G., Klambauer, G., Hochreiter, S., & Nearing, G. (2019). Towards learning universal,
regional, and local hydrological behaviours via machine learning applied to large-sample datasets. Hydrology and
Earth System Sciences, 23, 5089-5110.
Kumar, R., Samaniego, L., & Attinger, S. (2013). Implications of distributed hydrologic model parameterization on
water fluxes at multiple scales and locations. Water Resources Research, 49(1), 360–379.
Lahmer, W., Becker, A., Müller-Wohlfeil, D.I., & Pfützner, B. (1999). A GIS based approach for regional hydrological
modelling. IAHS Publication no. 254, 33-43.
Le Coz, J., Patalano, A., Collins, D., Guillén, N.F., Garciá, C.M., Smart, G.M., Bind, J., Chiaverini, A., Le Bourisicaus, R.,
Dramais, G., & Braud, I. (2016). Crowdsourced data for flood hydrology: Feedback from recent citizen science
projects in Argentina, France and New Zealand. Journal of Hydrology, 541, 766-777.
Leavesley, G.H., Markstrom, S.L., Restrepo, P.J., & Viger, R.J. (2002). A modular approach to addressing model design,
scale, and parameter estimation issues in distributed hydrological modelling. Hydrological Processes, 16, 173–
187.
Lindström, G., Johansson, B., Persson, M., Gardelin, M., & Bergström, S. (1997). Development and test of the
distributed HBV-96 hydrological model. Journal of Hydrology, 201, 272-288.

This article is protected by copyright. All rights reserved.

 Lindström, G., Pers, C., Rosberg, J., Strömqvist, J., & Arheimer, B. (2010). Development and testing of the HYPE
(Hydrological Predictions for the Environment) water quality model for different spatial scales. Hydrology
Research, 41(3-4), 295-319.
Liu, Y., & Gupta, H.V. (2007). Uncertainty in hydrologic modeling: Toward an integrated data assimilation framework.
Water Resources Research, 43, W07401.
Lundquist, D. (1978). Practical use of a simple conceptual model. Nordic Hydrological Conference, Helsinki, 11 pages.
Madsen, H. (2003). Parameter estimation in distributed hydrological catchment modelling using automatic calibration
with multiple objectives. Advances in Water Resources, 26, 205-216
Mango, L.M., Melesse, A.M., McClain, M.E., Gann, D., & Setegn, S.G. (2011). Land use and climate change impacts on
the hydrology of the upper Mara River Basin, Kenya: results of a modeling study to support better resource
Accepted Article

management. Hydrology and Earth System Sciences, 15, 2245–2258.

Markstrom, S.L., Niswonger, R.G., Regan, R.S., Prudic, D.E., & Barlow, P.M. (2008). GSFLOW—Coupled ground-water
and surface-water flow model based on the integration of the Precipitation-Runoff Modeling System (PRMS) and
the Modular Ground-Water Flow Model (MODFLOW-2005): U.S. Geological Survey Techniques and Methods 6-
D1, 240 p. https://pubs.usgs.gov/tm/tm6d1/pdf/tm6d1.pdf
Maxwell, R.M., Lundquist, J.K., Mirocha, J.D., Smith, S.G., Woodward, C.S., & Tompson, A.F.B. (2011) Development of a
coupled groundwater-atmosphere model. Monthly Weather Review, 139, 96-116.
Maxwell, R.M., Condon, L.E., & Kollet, S.J. (2015). A high-resolution simulation of groundwater and surface water over
most of the continental US with the integrated hydrologic model ParFlow v3. Geoscientific Model Development,
8, 923–937.
McCabe, M.F., Rodell, M., Alsdorf, D.E., Miralles, D.G., Uijlenhoet, R., Wagner, W., Lucieer, A., Houborg, R., Verhoest,
N.E.C., Franz, T.E., Shi, J., Gao, H., & Wood, E.F. (2017). The future of Earth observation in hydrology. Hydrology
and Earth System Sciences, 21, 3879–3914.
McMichael, C.E., Hope, A.S., & Loaiciga, H.A. (2006). Distributed hydrologic modelling in California semi-arid
shrublands: MIKE SHE model calibration and uncertainty estimation. Journal of Hydrology, 317, 307–324.
Melsen, L.A., Teuling, A.J., Torfs, P.J.J.F., Uijlenhoet, R., Mizukami, N., & Clark, M.P. (2016). HESS Opinions: The need
for process-based evaluation of large-domain hyper-resolution models. Hydrology and Earth System Sciences, 20,
1069–1079.
Mizukami, N., Clark, M.P., Newman, A.J., Wood, A.W., Gutmann, E.D., Nijssen, B., Rakovec, O., & Samaniego, L. (2017).
Towards seamless large-domain parameter estimation for hydrologic models. Water Resources Research, 53,
8020– 8040.
Mulvany, T.J. (1850). On the use of self registering rain and flood gauges. Institution of Civil Engineering Proceedings
(Dublin), 4, 1-8.
Nash, J.E. (1959). Systematic determination of unit hydrograph parameters using method og moments. Journal of
Geophysical Research, 64, 111-115.
Nearing, G.S., Mocko, D.M., Peters-Lidard, C.D., Kumar, S.V., & Xia, Y. (2016). Benchmarking NLDAS-2 Soil Moisture
and Evapotranspiration to Separate Uncertainty Contributions. Journal of Hydrometeorology, 17(3), 745-759.
Nearing, G.S., Ruddell, B.L., Bennett, A.R., Prieto, C., & Gupta, H.V. (2020). Does information theory provide a new
paradigm for earth science? Hypothesis testing. Water Resources Research, 56(2).
Nearing, G.S., Kratzert, F., Sampson, A.K., Pelissier, C.S., Klotz, D., Frame, J.M., Prieto, C., & Gupta, H.V. (2021). What
role does hydrological science play in the age of machine learning? Water Resources Research, 57(3),
e2020WR028091.
Nielsen, D.R., Bigger, J.W., & Erk, K.T. (1973). Spatial variability of field measured soil water properties. Hilgardia, 42,
215-259.
Nielsen, S.A., & Hansen, E. (1973). Numerical simulation of the rainfall-runoff process on a daily basis. Nordic
Hydrology, 4, 171-190.
Nijzink, R.C., Almeida, S., Pechlivanidis, I.G., Capell, R., Gustafssons, D., Arheimer, B., Parajka, J., Freer, J., Han, D.,
Wagener, T., van Nooijen, R.R.P., Savenije, H.H.G., & Hrachowitz, M. (2018). Constraining conceptual hydrological
models with multiple information sources. Water Resources Research, 54, 8332–8362.
Panday, S., Brown, N., Foreman, T., Bedekar, V., Kaur, J., & Huyakorn, P.S. (2009). Simulating Dynamic Water Supply
Systems in a Fully Integrated Surface–Subsurface Flow and Transport Model. Vadose Zone Journal, 8(4), 858–872.

This article is protected by copyright. All rights reserved.

 Paniconi, C., & Putti, M. (2015). Physically based modeling in catchment hydrology at 50: Survey and outlook. Water
Resources Research, 51, 7090-7129.
Pechlivanidis, I.G., Jackson, B.M., McIntyre, N.R., & Wheater, H.S. (2011). Catchment scale hydrological modelling: A
review of model types, calibration approaches and uncertainty analysis methods in the context of recent
developments in technology and applications. Global NEST Journal, 13(3), 193-214.
Perrin, C., Michel, C., & Andréassian, V. (2001). Does a large number of parameters enhance model performance?
Comparative assessment of common catchment model structures on 429 catchments. Journal of Hydrology, 242,
275-301.
Perrin, C., Michel, C., & Andreassian, V. (2003). Improvement of a parsimonious model for streamflow simulation.
Journal of Hydrology, 279, 275-289.
Accepted Article

Poff, N.L., Brown, C.M., Grantham, T., Matthews, J.H., Palmer, M.A., Spence, C.M., Wilby, R.L., Haasnoot, M.,
Mendoza, G.F., Dominique, K.C., & Baeza, A. (2016). Sustainable water management under future uncertainty
with eco-engineering decision scaling. Nature Climate Change, 6, 25–34.
Potapov, P., Hansen, M.C., Kommareddy, I., Kommareddy, A., Turubanova, S., Pickens, A., Adusei, B., Tyukavina, A., &
Ying, Q. (2020) Landsat Analysis Ready Data for Global Land Cover and Land Cover Change Mapping. Remote
Sensing, 12(3), 426.
Porter, J.W., & McHahon, T.W. (1971). A model for simulation of streamflow data from climatic records. Journal of
Hydrology, 13, 297-324.
Read, J.S., Jia, X., Willard, J., Appling, A.P., Zwart, J.A., Oliver, S.K., Karpatne, A., Hansen, G.J.A., Hanson, P.C., Watkins,
W., Steinbach, M., & Kumar, V. (2019). Process-guided deep learning predictions of lake water temperature.
Water Resources Research, 55, 9173-9190.
Reed, S., Koren, V., Smith, M., Zhang, Z., Moreda, F., & Seo, D.J. (2004). Overall distributed model intercomparison
project results. Journal of Hydrology, 298, 27-60.
Refsgaard, J.C. (1996). Terminology, modelling protocol and classification of hydrological model codes. In: Abbott, MB
and Refsgaard JC (Eds.) Distributed Hydrological Modelling. Kluwer Academics Publishers, 17-39.
Refsgaard, J.C., & Hansen, E. (1982). A distributed groundwater/surface water model for the Suså catchment. Part 1:
Model description. Nordic Hydrology, 13, 299-310.
Refsgaard, J.C., & Knudsen, J. (1996). Operational validation and intercomparison of different types of hydrological
models. Water Resources Research, 32 (7), 2189-2202.
Refsgaard, J.C., Sørensen, H.R., Mucha, I., Rodak, D., Hlavaty, Z., Bansky, L., Klucovska, J., Topolska, J., Takac, J., Kosc, V.,
Enggrob, H.G., Engesgaard, P., Jensen, J.K., Fiselier, J., Griffioen, J., & Hansen, S. (1998). An integrated model for the
Danubian Lowland – methodology and applications. Water Resources Management, 12, 433-465.
Refsgaard, J.C., Henriksen, H.J. (2004). Modelling guidelines – terminology and guiding principles. Advances in Water
Resources, 27(1), 71-82.
Refsgaard, J.C., van der Sluijs, J.P., Højberg, A.L., & Vanrolleghem, P.A. (2007). Uncertainty in the environmental
modelling process – A framework and guidance. Environmental Modelling & Software, 22, 1543-1556.
Refsgaard, J.C., Arnbjerg-Nielsen, K., Drews, M., Halsnæs, K., Jeppesen, E., Madsen, H., Markandya, A., Olesen, J.E.,
Porter, J.R., & Christensen, J.H. (2013). The role of uncertainty in climate change adaptation strategies – A Danish
water management example. Mitigation and Adaptation Strategies for Global Change, 18(3), 337-359.
Refsgaard, J.C., Højberg, A.L., He, X., Hansen, A.L., Rasmussen, S.H., & Stisen, S. (2016). Where are the limits of model
predictive capabilities? Hydrological Processes, Keith Beven Tribute, 30(26), 4956-4965.
Rigon, R., Bertoldi, G., & Over, T.M. (2006). GEOtop: A Distributed Hydrological Model with Coupled Water and Energy
Budgets. Journal of Hydrometeorology, 7, 371-388.
Rockwood, D.M. (1964). Streamflow synthesis and reservoir regulation. US Army Engineer Division, North Pacific,
Portland Oregon. Engineering Studies Project 171, Technical Bulletin No. 22
Samaniego, L., Kumar, R., & Attinger, S. (2010). Multiscale parameter regionalization of a grid-based hydrologic model at
the mesoscale. Water Resources Research, 46, W05523.
Searcy, J.K. (1960). Graphical correlation of gaging-station records. Manual of hydrology: Part 1. General surface water
techniques. Geological Survey Water-Supply Paper 1541-C. USGS.
Shen, C. (2018). A Transdisciplinary Review of Deep Learning Research and Its Relevance for Water Resources
Scientists. Water Resources Research, 54, 8558-8593.

This article is protected by copyright. All rights reserved.

 Sherman, L.K. (1932). Stream flow from rainfall by unitgraph method. English News Record, 1008, 501-505.
Singh, V.P. (1995). Watershed modeling. In Singh VJ (Ed) Computer models in watershed hydrology. Water Resources
Publication, 1-22.
Smith, A., Tetzlaff, D., Kleine, L., Maneta, M., & Soulsby, C. (2021). Quantifying the effects of land use and model scale
on water partitioning and water ages using tracer-aided ecohydrological models. Hydrology and Earth System
Sciences, 25, 2239–2259.
Soltani, M., Bjerre, E., Koch, J., & Stisen, S. (2021) Integrating remote sensing data in optimization of a national water
resources model to improve the spatial pattern performance of evapotranspiration. Journal of Hydrology, 603(B),
127026.
Sophocleous, M.A., Koelliker, J.K., Govindaraju, R.S., Birdie, T., Ramireddygari, S.R., & Perkins, S.P. (1999). Integrated
Accepted Article

numerical modeling for basin-wide water management: The case of the Rattlesnake Creek basin in south-central
Kansas. Journal of Hydrology, 214, 179-196.
Speers, D.D. (1995). SSARR model. In Singh VJ (Ed) Computer models in watershed hydrology. Water Resources
Publication, 367-394.
Sprenger, M., Tetzlaff, D., Buttle, J., Laudon, H., & Soulsby, C. (2018). EcH2O-iso 1.0: water isotopes and age tracking in
a process-based, distributed ecohydrological model. Geoscientific Model Development, 11, 3045-3069.
Srinivasan, R., Ramanarayanan, T.S., Arnold, J.G., & Bednarz, S.T. (1998). Large area hydrologic modeling and
assessment - Part 2: Model application. Journal of the American Water Resources Association, 34, 91-101.
Stephenson, G.R., & Freeze, R.A. (1974). Mathematical simulation of subsurface flow contributions to snowmelt
runoff, Reynolds Creek watershed, Idaho. Water Resources Research, 10(2), 284-294.
Sterte, E.J., Lidman, F., Lindborg, E., Sjöberg, Y., & Laudon, H. (2021). How catchment characteristics influence
hydrological pathways and travel times in a boreal landscape. Hydrology and Earth System Sciences, 25, 2133–
2158.
Stisen, S., Koch, J., Sonnenborg, T.O., Refsgaard, J.C., Bircher, S., Ringgaard, R., & Jensen, K.H. (2018) Moving beyond
runoff calibration - Multi-variable optimization of a surface-subsurface-atmosphere model. Hydrological
Processes, 32(17), 2654-2668.
Sugawara, M. (1967). The flood forecasting by a series storage type model. IAHS Publication no. 85, 1-6.
Therrien, R., McLaren, R.G., Sudicky, E.A., & Park, Y.J. (2010). HydroGeoSphere: A Three-Dimensional Numerical Model
Describing Fully-Integrated Subsurface and Surface Flow and Solute Transport, 453 pp., Aquanty Inc., Waterloo,
Ontario, Canada.
Todini, E. (2007). Hydrological catchment modelling: past, present and future. Hydrological Earth System Sciences,
11(1), 468-482.
Tyralis, H., Papacharalampous, G., & Langousis, A. (2019). A Brief Review of Random Forests for Water Scientists and
Practitioners and Their Recent History in Water Resources. Water, 11, 910.
Van Der Knijff, J.M., Younis, J., De Roo, A.P.J. (2010). LISFLOOD: a GIS‐based distributed model for river basin scale
water balance and flood simulation. International Journal of Geographical Information Science, 24(2), 189-212.
VanderKwark, J.E., & Logue, K. (2001). Hydrologic-response simulations for the R-5 catchment with a comprehensive
physics-based model. Water Resources Research, 37(4), 999-1013.
Vazquez, R.F., Beven, K., & Feyen, J. (2009). GLUE based assessment on the overall predictions of a MIKE SHE application.
Water Resources Management, 23, 1325–1349.
Vrugt, J.A., Diks, C.G.H., Gupta, H.V., Bouten, W., & Verstraten, J.M. (2005). Improved treatment of uncertainty in
hydrologic modeling: Combining the strengths of global optimization and data assimilation. Water Resources
Research, 41, W01017.
Western, A.W., Grayson, R.B., & Green, T.R. (1999). The Tarrawarra project: high resolution spatial measurement,
modelling and analysis of soil moisture and hydrological response. Hydrological Processes, 13, 633-652.
WMO. (1975). Intercomparison of conceptual models used in operational hydrological forecasting. WMO Operational
Hydrology Report No 7, WMO No 429, World Meteorological Organization, Geneva.
WMO. (1988). Intercomparison of models for snowmelt runoff. WMO Operational Hydrology Report No 23, WMO No
646, World Meteorological Organization, Geneva.
WMO. (1992). Simulated real-time intercomparison of hydrological models. WMO Operational Hydrology Report No
38, WMO No 779, World Meteorological Organization, Geneva.

This article is protected by copyright. All rights reserved.

 WMO. (1994). Guide to hydrological practices. WMO-No. 1968. Fifth edition. World Meteorological
Wood, E.F., Lettenmaier, D., Liang, X., Nijssen, B., & Wetzel, S.W. (1997). Hydrological modeling of continental-scale
basins. Annual Review of Earth and Planetary Sciences, 25, 279-300.
Woolhiser, D.A. (1973). Hydrologic and watershed modeling – State of the art. Transactions of the ASAE, 16, 533-559.
Xevi, E., Christiaens, K., Espino, A., Sewnandan, W., Mallants, D., Sørensen, H., & Feyen, J. (1997). Calibration,
validation and sensitivity analysis of the MIKE-SHE model using the Neuenkirchen Catchment as case study.
Water Resources Management, 11, 219–242.
Yapo, P.O., Gupta, H.V., & Sorooshian, S. (1998). Multi-objective global optimization for hydrological models. Journal
of Hydrology, 204, 83–97.
Zhao, R.J., Zhuang, Y.L., Fang, L.R., Liu, X.R., & Zhang, Q.S. (1980) The Xinanjiang model. Hydrological Forecasting
Accepted Article

Proceedings Oxford Symposium, IAHS 129, 351-356.

Zink, M., Mai, J., Cuntz, M., & Samaniego, L. (2018). Conditioning a hydrologic model using patterns of remotely
sensed land surface temperature. Water Resources Research, 54(4), 2976 – 2998.

This article is protected by copyright. All rights reserved.

HYP_14463_Figure 1.tiff
HYP_14463_Fig_2-fig_grids_number.tif
HYP_14463_Fig_3-fig_grids_size.tiff
HYP_14463_Fig_4-fig_eval_level.tiff
HYP_14463_Fig_6-500_100_flow.png
ID# Year Study Model code Model Model Catchment Area [km2] Grid size Grids total N
Type Evaluation Δx [km]
Level
1 1966 Crawford and Linsley (1966) Stanford IV LU 1 French Broad 758 27.5 1
River, US
2 1971 Porter and McHahon (1971) Monash LU 1 Dandenong Creek, 269 16.4 1
AU
3 1974 Stephenson and Freeze (1974) Freeze (1971) PB-event 1 Hillslope in Hilslope 0.0008 7,476
Reynolds Creek, US
4 1979 Beven and Kirkby (1979) TOPMODEL LU 1 Crimple Beck 8 0.59 23
5 1982 Refsgaard and Hansen (1982) SUSAA HRU 2 Susaa, DK 1,000 2.99 448
6 1984 Jønch-Clausen and Refsgaard NAM LU 1 Damodar, IN 19,550 38.8 13
(1984)
7 1986 Bathurst (1986) SHE PB-event 1 Wye, UK 11 0.25 6,591
8 1987 Beven et al. (1987) IHDM PB-event 1 Wye, UK Hilslope 0.049 117
9 1992 Jain et al. (1992) SHE PB 1 Kolar, IN 820 2.00 2,050
10 1992 Grayson et al. (1992) THALES PB-event 1 Wagga Wagga, 0.07 0.006 1,854
AUS
11 1995 Flügel (1995) PRMS HRU 1 River Bröl, DE 216 3.06 23
12 1996 Freer et al. (1996) TOPMODEL LU 2 Ringelbach, FR 0.36 0.005 14,400
13 1997 Lindström et al. (1997) HBV-96 HRU 1 Suorva, SE 4,688 4.97 190
14 1997 Wood et al. (1997) VIC-2L LU 1 Missouri river, US 1,370,000 97.2 145
15 1997 Xevi et al. (1997) MIKE SHE PB 1 Neuenkirchen, DE 1 0.025 1,600
16 1998 Refsgaard et al. (1998) MIKE SHE PB-GW 2 Danubian Lowland, 3,000 0.50 48,000
SK
17 1998 Srinivasan et al. (1998) SWAT HRU-WQ 1 Richland and 5,080 2.91 600
Chambers Creek,
US
18 1999 Western et al. (1999) THALES PB 3 Tarrawarra, AUS 0.11 0.012 750
18 1999 Western et al. (1999) VIC LU 3 Tarrawarra, AUS 0.11 0.32 1
19 2000 Arnold et al. (2000) SWAT HRU 1 Upper Missisipi 491,700 31.5 496
Basin, US
20 2001 Andersen et al. (2001) MIKE SHE PB 1 Senegal river, West 375,000 4.00 23,438
Africa
21 2001 VanderKwaak and Loague (2001) InHM PB-event 1 R-5, US 0.10 0.011 42,416
22 2002 Bixio et al. (2002) CATHY (pre) PB-GW 1 Chernobyl, UA 112 0.035 406,944
23 2003 Perrin et al. (2003) GR4J LU 1 429 catchments [1-50,600] - 1
24 2003 Henriksen et al. (2003) MIKE SHE PB-GW 2 Nordjylland, DK 5,478 1.00 82,170
25 2004 Butts et al. (2004) MIKE SHE PB 3 Blue River, US 1,760 4.00 110
26 2005 Vrugt et al. (2005) HYMOD LU 2 Leaf River, US 1,950 44.2 1
27 2006 McMichael et al. (2006) MIKE SHE PB 2 Jameson 34 0.27 466
catchment, US
28 2006 Ricon et al. (2006) Geotop PB-event 2 Little Washita, US 603 0.20 75,375
29 2008 Clark et al. (2008) 79 Model LU 2 French Broad 2,448 2,448 1
structures River, US
30 2008 Markstrom et al. (2008) GSFLOW HRU-GW 2 Sagehen Creek, US 27 0.46 11,954
31 2008 Kollet and Maxwell (2008) ParFlow-CLM PB-GW 1 Little Washita, US 600 1.00 561,600
32 2008 Das et al. (2008) SWAP PB 2 Walnut Gulch, US 150 0.80 11,200
33 2009 Vazquez et al. (2009) MIKE SHE PB-GW 3 Gete, BE 586 0.60 29,322
34 2009 Goderniaux et al. (2009) HydroGeoSphere PB-GW 2 Geer, BE 480 0.78 10,205
35 2009 Panday et al. (2009) MODHMS PB-GW 2 Santa Clarity Vally, 1,083 0.22 117,439
US
36 2010 Fang et al. (2010) CRHM HRU 2 Smith Creek 400 3.38 35
37 2010 Lindström et al. (2010) HYPE HRU-WQ 2 River Vindåen, SE 440 0.70 887
38 2010 Samaniego et al. (2010) mHm PB 1 Upper Neckar 4,000 2.00 1,000
Basin, DE
39 2010 Van Der Knijff et al. (2010) LISFLOOD LU 1 Elbe, CZ+DE 148,000 1.00 148,000
40 2011 Maxwell et al. (2011) WRF-CLM- PB-ATM- 1 Little Washita, US 600 1.00 633,600
Parflow event
41 2012 Stisen et al. (2012) MIKE SHE PB-GW 2 Midtjylland, DK 12,501 0.50 750,060
42 2013 Cao et al. (2013) MODFLOW PB-GW 1 North China Plain, 144,000 2.00 108,000
CHN
43 2013 Hartmann et al. (2013) 4 model LU 4 Mt. Hernan, 783 783 1
structures Middle East
44 2014 De Lange et al. (2014) NHI (MODFLOW PB-GW 2 The Netherlands 37,726 0.25 4,225,340
etc.)
45 2015 Abbaspour et al. (2015) SWAT HRU-WQ 3 Europe 10,000,000 12.9 60,012
46 2015 Maxwell et al. (2015) ParFlow PB-GW 1 North America 6,300.000 1.00 31,548,480
47 2016 Chaney et al. (2016) HydroBlocks PB 3 Little Washita, US 610 0.03 677,778
47 2016 Chaney et al. (2016) HydroBlocks HRU 3 Little Washita, US 610 0.78 1,000
48 2017 De Schepper et al. (2017) HydroGeoSphere PB-GW 2 Fensholt, DK 6 0.02 205,758
49 2018 Jing et al. (2018) mHm-GW PB-GW 2 Nägelstadt,DE 850 0.25 411,400
50 2018 Kollet et al. (2018) ParFlow-CLM- PB-ATM 3 Northrein- 22,500 0.50 4,725,000
COSMO Westfalia, DE
51 2019 Bartosova et al. (2019) E-HYPE HRU-WQ 2 Baltic Sea Drainage 1,800,000 15.9 7,145
basin
52 2019 Ehlers et al. (2019) MIKE SHE PB-GW 3 Skjern River, DK 1,055 0.50 92,840
53 2020 Fersch et al. (2020) WRF-Hydro PB-ATM 1 Ammer, DE 600 0.10 90,600
54 2020,45 Henriksen et al. (2020) MIKE SHE PB-GW 2 Denmark 42,700 0.10 46,970,000
 Variable Metric DK5 100m DK5 500m
MAE sh. [m] 3.57 3.13
(1990-1999) (2000-2010)
Validation Calibration

H
MAE de. [m] 4.43 4.76
KGE [-] 0.64 0.64
Q
WBE [%] 4.31 2.49
MAE sh. [m] 2.99 3.32
H
MAE de. [m] 4.01 4.32
KGE [-] 0.66 0.67
Q
WBE [%] 3.69 -0.43