You are on page 1of 4

Advances in WaterResources17 (1994) 197-200

© 1994 Elsevier Science Limited


Printed in Great Britain. All fights reserved
0309-1708/94/$07.00
ELSEVIER

Short Communication

Model definition and model validation


Anton Leijnse & S. Majid Hassanizadeh
National Institute of Public Health and Environmental Protection ( R I V M ) , PO Box 1, 3720 BA, Bilthoven, The Netherlands

(Received 22 March 1994; accepted 4 May 1994)

INTRODUCTION Bredehoefl5 stems from the difference in definitions


and perceptions behind the words 'model' and 'valida-
In recent decades, groundwater models have been used tion'. Contrary to Bredehoeft and Konikow, 1 we believe
in studying more and more complicated situations. The that the word validation does not have a clear and
advanced uses of groundwater models are in the unambiguous meaning, certainly not to the general
framework of studies of groundwater contamination, public and even not within the scientific community (see
nuclear/hazardous waste disposal, thermal energy also Hassanizadeh and Carrera 4 and McCombie and
storage/exploration, etc. In these applications, large McKinley7). We go even one step further and contend
temporal and/or spatial scales are often encountered and that there is no unanimity within the groundwater
many uncertainties enter the model. As a result, the modelling community on what constitutes a model. In
investigation of the validity of a model becomes a very fact, it appears that there is a significant difference in
complicated task and the issue of model validation takes what Konikow and Bredehoefl5 and de Marsily et al. 3
upon a much greater dimension than that commonly understand under the word 'model'. We would like to
associated with groundwater supply problems. Much discuss these points here. Our discussion is strictly
discussion about validation of geo-hydrological models limited to mathematical models; all other kinds of model
has taken place in recent years and, as might be are not considered here. Also, the discussion is mainly
expected, the opinions are diverse. A list of various focussed on mathematical models employed in ground-
studies on model validation is found in Tsang. 8 water studies. However, the ideas brought up here apply
Recently, a two-part special issue of A d v a n c e s in to other geo-hydrological problems as well.
W a t e r R e s o u r c e s (vol. 15, nos 1 and 3) was devoted to
the question of validation of geo-hydrological models.
In an article in that special issue Konikow and WEAK AND STRONG DEFINITIONS OF A
Bredehoeft5 submit that 'groundwater models cannot MODEL
be validated'. This point of view, shared by some other
researchers, has proven to be controversial. For Konikow and Bredehoeft5 define a model as 'a
example, it has been challenged in a commentary by representation of a real system or process'. This
de Marsily et al. 3 who present evidence which they definition includes the basic equations as well as all
consider as proof that their groundwater model has been input parameters needed to describe the system
validated. In a reply to that commentary, Bredehoeft behaviour. De Marsily et al., 3 on the other hand, state
and Konikow 1 reiterate their stand that despite their that 'groundwater flow models rely essentially on two
usefulness, groundwater models are impossible to concepts: (i) mass balance; (ii) Darcy's law'. Obviously,
validate. A similar discussion was carried out in two the model envisaged by de Marsily et al. 3 constitutes
editorials which appeared in the journal G r o u n d W a t e r only a part of a model as defined by Konikow and
(Bredehoeft & Konikow;2 McCombie & McKinley7). Bredehoefl;5 the part that the latter authors call the
The question to be considered here is 'Is there a conceptual model.
contradiction between these two points of view?' We In fact, a survey of the geo-hydrology literature will
believe not. The apparent contradiction between the reveal that the word model has been used (and
views of de Marsily et al. 3 and Konikow and sometimes misused) to refer to many different entities.
197
198 A. Leijnse, S. M . Hassanizadeh

In general, two categories of model definition can be It must be emphasized that we do not attempt to
recognized: weak definitions and strong definitions. A give a unique definition of weak and strong models.
weak definition model is one which is intended to We recognize that there is an overlap in some elements
simulate only the processes of interest in a given system of the two model definitions. We simply want to point
and does not characterize that particular system. out that there are two extremes. At one extreme, a
According to the weak definition, a groundwater model model is seen as a process simulator, consisting only
consists of the equation of mass balance, Darcy's of conceptual description and corresponding mathe-
equation, the equation of state, and all simplifying matical equations. At the other extreme, a model
assumptions; it does not incorporate the input data. The is seen as a system simulator which includes any item
development of a model is considered to be accom- and any factor that affect the quantitative character-
plished as soon as the modeller has demonstrated that ization of the system and future events. Some people
the processes of interest are correctly simulated by the may even go to the extreme of considering the 'modeller'
model. Of course, in order to use the model, one has to as part of the model too! Obviously, between these
assign values to the parameters appearing in the two extremes many intermediate definitions are possible.
equations and choose initial and boundary conditions. For example, one may choose to define a model
However, the fact that the values of such parameters, or consisting of basic equations plus some of the input
their spatial and temporal variations, may not be known data which either remain invariant during future
for a given groundwater system is not considered to events, or their variations can be determined with an
have any bearing on the utility, applicability, or validity acceptable degree of certainty. Typical examples of
of the model. Commonly, although not necessarily, a such data are the geometrical structure, permeability,
model defined according to the weak definition is not a and porosity of a porous medium. In general, how-
site-specific model. ever, all parameter values are subject to temporal and
According to the strong definition, a model is a spatial variations and the magnitude of these vari-
substitute for the whole system. Thus, it includes not ations and the associated uncertainties are dependent
only the basic equations, but also all parameter values on the spatial and temporal scales of processes being
and their spatial and temporal variations, initial and considered.
boundary conditions, system geometry, sources, and With regard to the foregoing discussion, two
sinks, and all other data which characterize a given questions arise: (i) Is the difference of opinions about
system. A model defined according to a strong definition what constitutes a model simply a matter of semantics?
is almost always a site-specific model. If chosen (ii) Should one of the model definitions be selected as the
parameter values appear to be in error, the applicability most appropriate definition? We believe that the answers
of the model is brought under question. to both of these questions are negative. The distinction
There are arguments both against and in favour of among various model definitions is not a matter of
each definition. For example, de Marsily et al., 3 who semantics; it is a matter of having consistency and
invoke the weak definition, comment that 'a ground- clarity in the meaning of a word which is being used very
water flow model does not claim that it can predict extensively. The point is that the same word is being
climate variability in the future'. Indeed, input data are used to refer to different entities, and this can obviously
often provided by geological, meteorological, agricul- be confusing, to say the least. Concerning the second
tural and/or industrial models and are not prepared by question, we believe that the weak and strong model
the groundwater hydrologists. Why should the ground- definitions are both useful; they simply serve different
water modeller be held responsible for possible short- purposes. Also, the adjectives 'weak' and 'strong', by no
comings in input data!? means refer to the quality of a model. A weak definition
On the other hand, those in favor of invoking the model is of a 'strong' quality if it is shown to be a valid
strong definition of a model argue that modelling is a model of the processes it is simulating. On the other
subjective activity and the selection of input data is an hand, a strong definition model may be proven to be of
integral part of modelling. Although groundwater flow a 'weak' quality if it makes incorrect prediction of the
is distinguished from other processes such as climatic system behaviour.
changes and geological events, there is a coupling
through boundaries, sources and sinks, etc. The
groundwater modeller has to decide where to put the 'ANALYSIS MODELS' AND 'PREDICTION
boundaries and which type of boundary conditions to MODELS'
employ. Such decisions are based on the modeller's
conceptual understandings of the system and its The question of what constitutes a model is closely
interaction with the surroundings. Hence, it can be related to the intended purpose and usage of the model.
argued that the input data are an integral part of the In general, two major categories of models can be
model and a correct choice of the data is the identified: models used as analytical tools, and models
responsibility of the groundwater modeUer. used as predictive tools. In the analytical mode, a model
M o d e l definition and model validation 199

is used as a tool to analyse a system of interest and to WEAK AND STRONG VALIDATION
increase understanding of its behaviour. For example, a
model may be used in site investigation studies, in Parallel with the above definitions of a model, one may
analysing pumping test data, in investigating the speak of validation in the weak sense and in the strong
effect of heterogeneities on flow and transport proper- sense. Validation in the weak sense refers to the validity
ties, in determining how the variations in boundary of the 'conceptual part of a model'. That is, one
conditions or hydrological stresses affect the system attempts to determine the dominant processes and the
behaviour, and in scenario comparison studies. range of applicability of the basic equations describing
We refer to this category of models as 'analysis models'. these processes. The concept of weak validation is
An analysis model is primarily used in those modelling applicable to analysis models. We believe that the
activities which do not constitute quantitative prediction validation of analysis models can be achieved with a
of future events. The weak definition of a model reasonable degree of certainty, for a given range of
discussed above is appropriate for this category of temporal and spatial scales. In other words, it is possible
models. An analysis model must simulate internal to show that the basic equations are valid and applicable
processes of interest occurring in the system, correctly. to a large variety of situations, regardless of parameter
Input data are not necessarily part of such a model values and boundary conditions. An example of model
type. Although data are always required to perform validation in the weak sense is that in groundwater flow
analysis of any real or generic system, it is not strictly problems, equations of mass balance, Darcy, and state
necessary to have the model sufficiently calibrated and/ are generally accepted as validated. When de Marsily
or to have an accurate and detailed knowledge of et al. 3 state that 'we claim that this is a "validation" of
parameter values and hydrological stresses. In fact, the model', they refer primarily to the conceptual part of
one may consider many realizations of the hetero- their model: Darcy's law and equation of mass balance.
geneities or the temporal variations of parameter Thus, they imply a weak validation of 'their model'.
values and stresses (perhaps with a given set of statistical Validation in the strong sense refers to the validity of
properties). The results of calculations with an analysis the model of a given system as a whole including all
model do not have to correspond to any future input data. This validity is related to the predictive
measurements of the system response. Although ability of a model. When Konikow and Bredehoeft5
analysis models are not intended to make predictions, state that 'groundwater models cannot be validated',
they can be very helpful in increasing our understanding they refer to validation in the strong sense and they have
of the behaviour of a system, in designing field 'prediction models' in mind. We agree with Konikow
investigations, in estimating parameter values, and in and Bredehoeft5 that validation of groundwater models
determining the range of validity of a given set of basic in the strong sense is almost impossible, because, for
equations. example, future changes in geo-hydrological stresses are
In the predictive mode, a model is used as a tool often unknown.
to predict the system behaviour. A groundwater Thus, we submit that both Konikow and Bredehoeft5
model is expected to provide the spatial distribution and de Marsily et al. 3 are partially correct; 'prediction
of head or flow rates and their future variations, at models' cannot be validated (in the strong sense), but
least in a stochastic sense. We refer to this category of 'analysis models' can be validated (in the weak sense).
models as 'prediction models'. A prediction model is
primarily used in those modelling activities which
lead to a quantitative prediction of future events. A SUMMARY
prediction model basically replaces the real system.
It contains both causes and effects. It gives a quanti- Two major definitions of a model are possible: a weak
tative picture of various processes of interest and definition and a strong definition, with the main difference
their future development. For a prediction model to being whether input data should be considered as part of
have these capabilities, it is not enough to know the the model or not. According to the weak definition, a
basic equations with certainty. It is also necessary to model consists only of the basic equations and input
have a good knowledge of all parameter values and their data are considered to be external elements immaterial
spatial and temporal variations for the system being to the applicability of the model. The strong definition
modelled. In other words, input values become an considers a model to be composed of input data as well
integral part of the model. The strong definition of a as the basic equations. Models defined according to the
model dis-cussed above is appropriate for this category weak definition are not suitable for predicting the
of models. It should be noted that a prediction model system behaviour and are called analysis models.
can be used as an analysis tool too. But, an analysis Models defined according to the strong definition may
model has to be augmented with much additional be used for prediction purposes and are, therefore,
information and input before it can be used as a called prediction models. Parallel with these definitions,
prediction tool. the concepts of 'weak validation' and 'strong validation'
200 A. Leijnse, S. M. Hassanizadeh

may be introduced. Validation of a model in the weak gained during their participation in the international
sense implies that the processes of interest are correctly model validation studies H Y D R O C O I N and INTRA-
described by the basic equations incorporated in the VAL. 6 Constructive remarks by William G. Gray, which
model. Thus, the concept of weak validation applies to resulted in improving the presentation of certain ideas in
'analysis models' only. The (weak) validation of analysis this manuscript, are gratefully acknowledged.
models is a viable task. On the other hand, validation of
a model in the strong sense implies that the model truly
represents the system of interest. Thus, the concept of REFERENCES
strong validation applies to 'prediction models'. We
believe that the (strong) validation of prediction models 1. Bredehoeft, J. D. & Konikow, L. F., Reply to comment.
is an almost impossible task in typical geo-hydrological Adv. Water Resour., 15 (1992) 371-2.
applications. 2. Bredehoeft, J. D. & Konikow, L. F., Groundwater models:
Validate or invalidate. Ground Water, 31 (1993) 178-9.
Finally, note that the foregoing discussion does not 3. de Marsily, G., Combes, P. & Goblet, P., Comment on
support abandoning the use of the term 'model 'Ground-water models cannot be validated', by L. F.
validation'. It only makes it apparent that the modellers Konikow and J. D. Bredehoeft. Adv. Water Resour., 15
have to carefully specify what their models constitute (1992) 367-9.
and what is being validated. In this regard, we are in full 4. Hassanizadeh, S. M. & Carrera, J., Editorial of the Special
Issue on Validation of Geo-hydrological Models. Adv.
agreement with McCombie and McKinley7 that, if Water Resour., 15 (1992) 1-3.
confusion arises about validation, 'use of the term 5. Konikow, L. F. & Bredehoeft, J. D., Ground-water models
validation is much less to blame than poor science and cannot be validated. Adv. Water Resour., 15 (1992) 75-83.
sloppy documentation'. 6. Larson, A., The international projects INTRACOIN,
HYDROCOIN and INTRAVAL. Adv. Water Resour.,
15 (1992) 85-7.
7. McCombie, C. & McKinley, I., Validation -- Another
ACKNOWLEDGEMENT perspective. Ground Water, 31 (1993) 530-1.
8. Tsang, C., The modelling process and model validation.
The authors have benefited greatly from the experience Ground Water, 29 (1991) 825-31.

You might also like