Catchment Modelling Guidance

Good practise in joint use of monitoring and modelling

Anker Lajer Højberg^a
Jens Christian Refsgaard^a
Lisbeth Flindt Jørgensen^a
Frans van Geer^b
István Zsuffa^c

a: Geological Survey of Denmark and Greenland;
b: TNO; c: VITUKI

Deliverable no. D4-8

August 2007

Good practise in joint use of monitoring and modeling.
A Catchment Modelling Guidance
Anker Lajer Højberg; Jens Christian Refsgaard; Lisbeth Flindt Jørgensen; Frans van Geer & István Zsuffa

Harmoni-CA is a research project supported by the European Commission under the Fifth Framework Programme and contributing to the implementation of the Key Action "Sustainable Management and Quality of Water" within the Energy, Environment and Sustainable Development. EVK1-CT-2002-20003.

Disclaimer:
This report is the sole responsibility of the authors and does not represent the opinion of the European Commission, nor is the European Commission responsible for any use that might be made of the information appearing herein.

In order to support the implementation of the Water Framework Directive, the European Commission has established a cluster on Integrated Catchment Modelling (CatchMod). The objective of this cluster is the development of common harmonised modelling tools and methodologies for integrated water management at river basin or sub-basin scales, including the interface to the coastal zone.

Within the CatchMod cluster, the project Harmoni-CA – Harmonised Modelling Tools for Integrated River Basin Management – has the objective to create a forum for unambiguous communication, information exchange and harmonisation of the use and development of ICT-tools relevant to integrated river basin management, and the implementation of the WFD.

Harmoni-CA is a large-scale concerted action, meaning that it does not carry out research, but provide synthesised available knowledge with the help of knowledge providers such as researchers. Typical actions of Harmoni-CA are meetings and workshops, leading to synthesis reports and guidances.

One of the tasks of the Harmoni-CA project is defined within Harmoni-CA Work Package 4: “Joint use of monitoring and modelling”. The objective of this WP is to develop a guideline on best practise on joint use of the two disciplines, providing an overview of possibilities and a number of examples. To accomplish this task five workshops have been organised in the period 2004 to 2007 under the umbrella “Joint use of monitoring and modelling when implementing the Water Framework Directive”. Important to the workshops have been to facilitate discussions and exchange of experiences between the researchers/developers and practitioners to bring the two communities closer and develop a common understanding on the needs and opportunities. The workshops succeeded to bring almost 80 people together representing 21 of the 27 EC countries besides from Norway Moldova, Ukraine and the US. The weighting between researchers and water managers were equal, a few representing both sides. Besides, a couple of consultants attended the workshops, mostly presenting tangible examples from all around the world. Several attended more than one of the workshops. Experiences, conclusions, recommendations and lessons learned are brought forward in this report.

An earlier version of this document was presented and discussed at the Harmoni-CA review workshop in Cologne 1 March 2007. The reviewers were Faycal Bourai, Morten Sørensen and Bernardas Paukstyks.

1 Summary

What do monitoring data tell us about the environment? How do we formulate programmes of measures based on monitoring data? How confident are we in the decisions we make based on observation data?

In the Water Framework Directive much emphasise is put on monitoring data and the guiding principle is based on uncertainty and risk assessments. Monitoring is a prerequisite for water management but data itself only reflect the current state of the system at the point and time where the measurements were taken. A challenging task is thus to move from the discrete measurements to integrated water resources assessment, where the risk of making the wrong decisions are at an acceptable level.

In accordance to this, the need to support the EU environmental monitoring by new knowledge and scientifically based tools including an enhanced communication is often emphasised by specialists and even by practitioners. However, descriptions of present state-of-the-art in joint use of modelling and monitoring techniques show that most often models are not considered an option when the monitoring obligation in the Water Framework Directive are solved in practise. If modelling is used the interaction is commonly reduced to a single step one-way (data-to-model) communication.

Practical experience from several model studies and research projects have demonstrated that joint use of monitoring and modelling has a much larger potential than what is presently utilised. One of the main arguments for using modelling tools is to bring the monitoring data “alive”. Although monitoring data may be subject to rigid quality assurance protocols during sampling and analysis, many errors in data may only be discovered if they are interpreted based on a conceptual understanding of the system and confronted with other data types, describing the same system only from another “perspective”. To formulate future strategies, such as programmes of measures or redesign of monitoring networks, a thorough understanding of the natural system is a prerequisite. This can only be achieved if we are able to understand all data from the system based on a coherent and plausible conceptual understanding. In this sense, model systems provide a powerful tool to integrate all data and to test and revise our understanding. Additionally, models have the ability to test different scenarios. By this, the effect of different alternative programmes of measures can be tested prior to implementation and may thus aid in the identification of the most effective strategy.

In the present report, some methods and examples are provided on how modelling and monitoring can be used jointly, with emphasis on the tasks required in the WFD monitoring programmes:

The Harmoni-CA workshops on joint use of monitoring and modelling identified several obstacles limiting the present joint use, of which many can be related to historical praxis. However, the workshops also demonstrated that if people from the monitoring community and the modelling community are brought together for open-minded discussions and with free hands they are very open to discuss ideas and suggestions from the other side of the table, and able to see the perspectives in the combined use. To advance the joint use of monitoring and modelling there appears to be a strong need to first of all raise the awareness of how monitoring and modelling can be combined. Additionally, a shift in the perception of models is needed, where models currently often are considered useful only to provide answers to specific questions. This view must be replaced by recognition of models as being valuable tools that should be used continuously and interactively along with monitoring data, a tool that serves many purposes from the design phase of monitoring programmes to the interpretation and quality assurance of the monitoring data.

The WFD imposes many new challenges among those are the requirements to consider an integrated use of all waters and an evaluation of the qualitative status in terms of ecological aspects. To fulfil these requirements development of new tools and models is a necessity. To avoid a future parallel development of monitoring and modelling praxis in the field of integrated and ecological assessments, it is of vital importance that monitoring and modelling are realised to be inter-linked activities and not independent disciplines.

2 Introduction

In the past, monitoring has traditionally been considered an independent discipline, but within the last decades modelling has entered the arena as a supplementary tool to help extracting the information retained in the observation data. In the research communities it is generally accepted that monitoring and modelling are inter-linked activities (Holt et al., 2000; Parr et al., 2003). The guiding principle for monitoring in the WFD is based on uncertainty and risk assessments, which is quite advanced, when considered as guidance to practitioners in water resources management. In full accordance with this line of thinking the guidance document on planning (EC, 2003b) describes the benefits of using models to support the WFD implementation. The WFD therefore creates new challenges on monitoring and modelling and provides new incentives for improving the joint use of monitoring and modelling (Højberg et al., 2007; Zsuffa et al., 2005). The need to support the EU environment monitoring, specific to the groundwater domain, by new knowledge and scientifically based tools including an enhanced communication among scientists, stakeholders and policy-makers is discussed by Quevauviller (2005). Descriptions of present state-of-the-art in joint use of modelling and monitoring techniques, however, show that most often models are not considered an option when the monitoring obligation in the WFD are solved in practise (Kamphorst et al., 2005; Arustiene et al., 2005; Borowski and Hare, 2007).

2.1 Why joint use?

The prospective of models have varied since the early days of the modelling discipline ranging from deep scepticism to overly optimism. Critics often argue that a model is only a speculative picture of the natural system, and as such does only provide speculative information of the system of interest. This argument may hold true if the model’s ability to mimic the system under concern has not been tested. Models may be thought of as sophisticated databases that provide an ordered way to store field data and define the relationships between data, e.g. the temporal development in streamflow as a response to a storm event. These relationships may be based on pure empiricism, statistical relations, or mechanistic processes, but are all generic formulations and models must therefore be adjusted to describe the site-specific conditions. The adjustment of model parameters is commonly accomplished through calibration, or history matching, where model parameter values are varied until the model reproduces the response of the reality within some predefined accuracy. If no site-specific data are available, the models can not provide insight in the system of interest, but are limited to provide some general insight in the system behaviour, e.g. how different processes interact. Using such models to predict the state of a site-specific natural system and its response to stresses will be guesswork and monitoring data are therefore a prerequisite to site-specific modelling.

Data alone do, on the other hand, only provide information of the current state of the environment, e.g. a chemical concentration, at a specific location and at the time at which the data was collected. This information is, in itself, of very little use. If more measures of the compound have been taken a time series can be constructed and obvious trends in the concentration levels may be revealed. The data do, however, not explain the cause for this trend. To extract this kind of information from the data set one needs to interpret the data based on an understanding of the physical system, that is, we formulate a conceptual understanding (or conceptual model) of the system from which we explain the data. As such we add knowledge to the data, which is necessary if we want to formulate future management strategies based on the observations. Natural systems are, however, very complex with several processes occurring simultaneously and interacting. It is therefore very difficult to fully comprehend and separate the different processes, their interaction and their consequences by interpretation of field data alone. In this context, models can be seen as a formal description of knowledge of the hydrological system studied, describing the spatial and temporal coherence in the system.

Presently, the main and often only interaction between monitoring and modelling is in the phases of constructing, calibrating and validating models. Once these tasks have been completed, the monitoring data are, at best, only used to quantify the uncertainty of the models, by some expression of the deviation between the model prediction and observed variables. The interaction is thus merely a single step one-way (data-to-model) link. Much improvement in models may, however, be obtained if the quantified uncertainty is used as basis for correcting and up-dating the model and model predictions. An example of such procedure is data assimilation, which is nearly only applied in flood forecasting but may be used in other domains as well for model updating.

Monitoring data and models provide complementary information of the system. The model captures physical knowledge about the systems behaviour and monitoring provides information from the actual system including phenomena which are not included in the model. It may be argued that “modelling without monitoring data is guesswork” and “monitoring without models is a waste of money”.

2.2 What are the problems?

Several obstacles have been brought forward for the joint use of monitoring and modelling in the practical world. From the workshops the following impediments were identified:

Common to all the obstacles are that they are not related to technical difficulties. The scientific basis to improve the combined use is well developed and new methods are continuously developed.

2.3 Objectives and target audience of this document

The obstacles listed in the previous section need to be solved at different levels and by different means. Most of the obstacles may be tackled at an institutional level, and different solutions strategies may be necessary to solve the problem at different institutions. But before a solution is sought the motivation for using models must exist – is it worthwhile? The first obstacle to tackle is thus to raise awareness of the mutual benefits. This was the primary objective of the workshops and here it was experienced that if people from the monitoring community and from the modelling community are brought together for open-minded discussions and with free hands they are very open to discuss ideas and suggestions from the other side of the table. Monitoring people ask for tangible examples showing benefits and limitations and modellers asks for information on data and for being involved in possible redesigning monitoring programmes.

The objective of the present report is thus to demonstrate some of the possibilities on how to combine modelling and monitoring and to illustrate the advantages thereof. Many advantages of the combined use are natural spin offs from a model study, which do not require additional work beyond the modelling process. For more comprehensive interaction it may be necessary to employ different methods, which vary widely from very simple tasks that can be carried out without much additional effort, to more complex methods where the workload is substantial. However, the present development and practical experiences offers a wide range of opportunities to combining monitoring data and modelling in a more intelligent way than is commonly praxis.

The combined use of monitoring and modelling may be beneficial in wide range of studies independent of the study objectives. In the present report only aspects related to the implementation of the WFD are discussed, however, the discussions and methods presented are not limited to the implementation of the WFD but are generic, and may thus apply to any study.

The report is not a technical guideline on specific methods that can be employed in the effort of combining monitoring and modelling. Instead, the focus is to illustrate the benefits of the combined use, through a short description on the basic principles and benefits of different topics, where the joint use already are or easily can be included, as well as some methods that are still in the developing phase, but in the near future may be operational.

The target audience is policy makers, water managers and professionals involved in implementation of the Water Framework Directive.

3 Monitoring and modelling within the WFD

The starting point for discussing the potentials of joint use of monitoring and modelling in the implementation of the WFD is an identification of the different monitoring programmes and their specific objectives. Based on the specific objectives and requirements for a given study different modelling approaches may be appropriate.

3.1 WFD monitoring programmes

The overall objective of the Water Framework Directive is to achieve a good quantitative and qualitative status of all waters by 2015. To fulfil the goals of the WFD the EU member states are obliged to prevent further deterioration, protect and enhance the status of the water resources. The WFD thus prescribes both an improvement for waters already negatively affected by anthropogenic activities, as well as their protection with respect to future activities.

To accomplish the overall aim of the WFD a number of tasks have to be carried out. Firstly, all the water bodies must be described, delineated and classified with respect to the quantitative and qualitative status. This classification includes the detection of possible trends that may cause the water body to be at risk of failing to achieve the good status in the future. Next, the cause of the trend has to be assessed by separating the anthropogenic activities from natural variation, and the deviation of observed conditions to those that would be found under reference conditions must be evaluated. Finally, pressures have to be identified and evaluated upon the possible risk they may pose to the water bodies. With an overview of the present status and the possible future development, programmes of measures have to be established and the effect thereof evaluated. The backbone to resolve these issues within the WFD is the establishment of comprehensive monitoring networks. Four types of monitoring has been defined, with different purposes as described in the monitoring guidance (EC, 2003a):

The relation between the different monitoring programmes and the requirements in the WFD (Annex II analysis, Article 5 and 8) is shown schematically for surface water and groundwater monitoring in Figure 1 and 2, respectively.

The overall purpose of the monitoring programme is to establish a coherent and comprehensive overview of the water status within each river basin district. This forms the basis for assessing whether the ecological status of a water body is in accordance with the predefined objectives. The specific objectives and requirements of monitoring in the WFD are described in a guidance document (EC, 2003a). Here some general guidance are provided on where, what and when to monitor. These recommendations do not, however, specify the levels of precision and confidence required by the monitoring programmes, but instead state a key principle: “the actual precision and confidence levels achieved should enable meaningful assessments of status in time and space to be made. Member States will have to quote these levels in River Basin Management Plans and will thus be open to scrutiny and comment by others”. The acceptable level of precision and confidence is thus a subjective quantity that depends on the socio-economic interests that are at stake and of the risk strategy of the decision makers. In this context it should be borne in mind that the cost of measures to improve the water status are orders of magnitude greater than the costs of monitoring. In general, the lower the desired risk of misclassification, the more monitoring (and hence costs) is required. Therefore, there is a balance between the costs of monitoring against the risk of water bodies being misclassified.

Figure 1 Schematic diagram illustrating the relationship between the WFD requirements and the surface water monitoring programmes (from EC, 2003a).

Figure 2 Schematic diagram illustrating the relationship between the WFD requirements and the groundwater monitoring programmes (from EC, 2003a).

3.2 Modelling categories

Numerous model systems have been developed in the past tailored to describe different parts of the hydrological system in different details. Dependent on the model approach the data requirements vary greatly both with respect to the need for time-series and system data, like soil maps and land use, for more discussion on data availability and accessibility see Refsgaard et al. (2007). Here we differentiate between four different model complexities, which may be applicable for different model purposes:

The different model concepts provide different degree of insight in the physical system and have different requirements with respect to system data and data for model calibration and validation. While the empirical models only describe an observed relation between the variables of interest, the physical-based models can, theoretically, describe all state variables of the entire model domain, and thus hold the potential of providing the most detailed and correct representation of the physical system. Most distributed models are, however, a combination of distributed and lumped descriptions, where less important areas/processes are described by a lumped approach, which lowers the data requirements for the model application.

It must be emphasised that this classification is schematic and that many model codes or applications do not fit exactly. Some model codes fall naturally into one of the above four boxes, while others may be placed in more than one.

Some examples of different model systems appropriate for the different modelling domains are provided in Table 1.

Table 1 Types of model codes and examples of codes suitable for different domains and different levels of model complexity

In general, the more complex model systems may provide more detailed results and thus be used in studies where the discrimination between different processes is critical. However, the more complex the model system becomes, the more data is usually required and more parameter values have to be estimated. It is not possible to provide a finite relation between model complexity, data availability and the model performance. One common illustration of the conceptual relationship is, however, shown in Figure 3. For a given model complexity the predictive performance increases with data availability up to a certain limit, beyond which more data do not improve the predictive performance. This limit is reached first for the less complex models, and only the very complex models are able to make full use of large data availability. It also illustrates, however, that for limited data availability the moderately complex models may often perform better than the very complex models. The reason for this is that the more complex models may be able to describe the physical system more correctly, as the simple models rely on more assumptions and thus have a higher uncertainty in the model structure, i.e. incorrect representation of the system. On the other hand, the more complex models do generally require more parameters. If these parameters cannot be measured or are otherwise inaccessible, they must be estimated, which introduce higher uncertainty due to the parameter uncertainty. Thus, for a given data availability the advantages gained in using the more complex models may be counteracted by the increase in parameter uncertainty.

The view illustrated in Figure 3 originates from statistical theoretical considerations assuming that the number of free parameters requiring calibration increases with model complexity. This is, however, not always the case. For model codes with comprehensive application records there are often experience based recommendations for setting default parameter values, so that typically only a handful of parameters are adjusted by calibration in the particular catchment. In this way secondary data sources (all previous model applications in other locations) are used. This would correspond to moving left on the data availability axis towards larger data availability.

Figure 3 Schematic diagram of the relationship between model complexity, data availability and predictive performance (modified from Grayson and Blöschl, 2000)

3.3 Balancing precision, data and model concept

In the monitoring guidance document (EC, 2003a) some general guidance is provided on where, what and when to monitor in the different monitoring programmes. Most detailed information is provided on which variables must be included in the monitoring programmes, while the required temporal and spatial resolutions in general are linked to the chosen level of confidence and precision, where a key principle is stated: “the actual precision and confidence levels achieved should enable meaningful assessments of status in time and space to be made. Member States will have to quote these levels in River Basin Management Plans and will thus be open to scrutiny and comment by others”. The acceptable level of precision and confidence is thus a subjective quantity that depends on the socio-economic interests that are at stake and of the risk strategy of the decision makers.

An optimal model approach to support the monitoring obligation does thus depend on both technical aspects, such as type of problem to solve and the problem/system complexity as well as the required confidence. Simple models are therefore likely to be appropriate in a first step evaluation, where all existing data are combined to obtain a first overview of possible areas at risk. But, as the uncertainty in these models are high, they do generally not provide the sufficient basis for specific action strategies, such as detailed design or redesign of monitoring networks. In a socio-economic context the formulation of the programmes of measures is probably the most challenging, as it may have a direct impact on individuals. Here the confidence in the decision making is likely to be high and more complex model systems are needed. More guidance on how to select an appropriate model system based on the problem at hand can be found in the Benchmark Models for Water Framework Directive (BMW) project (http://www.rbm-toolbox.net/bmw/index.php, more discussion on the data availability issue is provided by Refsgaard et al. (2007)).

4 Joint use of monitoring and modelling

Although the joint use of modelling and monitoring has not been common praxis in the implementation of the WFD to date, the literature is abundant with case studies where the use of models have contributed significantly in the analysis of the monitoring data, and provided information not easily extractable from the monitoring data alone. Based on such studies five tasks may be identified, where modelling has proved beneficial for objectives similar to those outlined for the WFD monitoring programmes, Figure 1 and 2,:

Table 2 provides an overview on how these five tasks relate to the different purposes of the monitoring programmes. For these tasks several operational methods have been developed and the use of models to support the tasks is therefore relatively straight forward and can be included on a routinely basis. A more detailed discussion of the tasks is given in the next section.

In addition to the operational methods which may readily be used for the five tasks listed above, the research community continuously develops methods to improve the joint use of monitoring and modelling. While many of these methods are still under development or too complicated to use in standard model projects, they are likely to be operational in near future and expected to be valuable in a WFD context. A few examples of such methods are provided in section 4.2

Table 2 Overview of how the described model uses support the different WFD monitoring requirements. Requirements are listed for monitoring related to groundwater (italic) and surface water (bold) and both (italic and bold).

4.1 Operational modelling support for monitoring

4.1.1 Quality assurance

High quality data is a prerequisite to a sound management of the water resources. Quality assurance is specifically mentioned as an important activity in the WFD guidance document on monitoring (EC, 2003a). However, quality assurance is here restricted to sampling and laboratory procedures. While this is a very important step it is not sufficient to ensure a good data quality. Beside the laboratory analysis, data may be erroneous due to many sources, such as bias or drift in the monitoring equipment, lack of representativeness of the data, malpractice in sampling, erroneous storing of data in databases etc.

A common approach to evaluate monitoring data is by graphical analyses, either constructing time-series or 2D/3D plots, and identify possibly “out-liers” that are likely to be erroneous. While this may capture obvious errors it may not always be easy to discriminate between errors and system response. The use of models, describing the system responses to input stresses, may here have provided a much better basis for the detection of errors, Figure 4.

Similarly, if a monitoring device is located wrong, e.g. expected to be in the vadose zone but actually placed in the saturated zone, or there is a drift or bias in the monitoring device, it may not be detectable from graphical plots. Yet another limitation in simple data analysis is that different data types are evaluated independently and does thus not consider possible linkages between the data types, based on physical/chemical laws. Data therefore need to be used actively to identify possibly errors (besides laboratory analysis) and evaluated if the data are representative.

May often be based on simple graphical plots comparing model results and observed data from which suspicious spatial/temporal regions are identified for further analyses.

Quality assurance of data by use of models is highly relevant and recommendable for all types of monitoring. This can be seen as a basic analysis of monitoring data and a supplement to the quality assurance on sampling and laboratory procedures described in EC (2003a).

Procedures and supporting tools for quality assurance in modelling have been developed in the HarmoniQuA project: Refsgaard et al., 2005; Scholten et al., 2005. Studies reporting the use of models for Quality assurance may be found in Styczen, 2002; Styczen et al., 2004.

4.1.2 Interpolation and extrapolation in time and space

Monitoring data are discrete in time and space and some interpolation technique is therefore required to transform these discrete points into a temporally and spatially continuous image of the variables. We thus have reliable observations at few discrete points, whereas the reliability outside the measurement points is strongly dependent on the ability of the interpolation routine to capture the variability of the natural system. Sophisticated interpolation schemes are available, which, e.g., allow overall trends and anisotropy to be included, but they will always be limited by their generality and underlying assumptions. In contrast interpolation by models has the ability to include the system characteristics and stresses and is, as such, based on physical quantities. This is a very important advantage, especially in spatial and/or temporal heterogeneous systems, where the sampling resolution is insufficient to resolve the heterogeneity. An example is monitoring of river systems which are very dynamic and heavy rainfall of short duration (possibly shorter than an hour) results in a sudden change in the discharge. If data is measured at a lower temporal resolution an estimated discharge based on simple interpolation techniques may be significantly in error. Spatial interpolation is important in groundwater management, but generalised interpolation routines do not consider non-stationarity (an aquifer may be homogenous in some parts and very heterogeneous in other parts) but this may be built into the model. The models therefore provide a much stronger interpolation routine, which, in addition, can be improved by adjusting the conceptual model on the basis of new data from field observations.

Extrapolation is specifically relevant in the WFD to establish reference conditions in surface waters. Where possible this reference condition should be based on data from a reference site network, i.e. sites that are unaffected by anthropogenic activities. Where such reference sites do not exist the use of models is often the only available option to extrapolate reference conditions, and the WFD explicitly mentions modelling as a suitable method (EC, 2000) for this purpose.

Interpolation and extrapolation are the actual results from model simulations and predictions. In distributed models spatial interpolation is obtained directly in the resolution used in the model. Temporal interpolation may be discrete or near continuous depending on the model approach (stationary or dynamic), and temporal extrapolation is often accomplished by the formulation of scenarios describing different possible changes in external stresses.

Model based interpolation of monitoring data is highly relevant and recommendable for all types of monitoring. This can be seen as a basic analysis of monitoring data.

Nilsen et al. (2003) provides an example on extrapolation of reference conditions

4.1.3 Conceptual model

Monitoring data alone provide only a specific measure like a chemical concentration, river discharge or a groundwater level. This information is, however, of little value for the formulation of management strategies without a thorough understanding of the system with respect to its dynamics and the cause-effect relations. A conceptual model is therefore needed not only to interpret the monitoring data but also to formulation of management strategies. Construction of a conceptual model is therefore a very important but also often a very complicated task, due to the complexity and heterogeneity of physical systems.

A thorough examination of the conceptual understanding must be based on all relevant data types as all data contain (different) information of the same system, and the conceptual model must thus be consistent with all data. In this context, models may be thought of as sophisticated databases that provide an ordered way to store field data in which the relationships between data are defined so that the consistency of the conceptual model can be checked. Confronting the model results with monitoring data may result in either an increased reliability in the conceptual model, or a falsification hereof. In the latter case the monitoring data reveal new information on the system that is not consistent with the present understanding, and the conceptual model must be revised. More monitoring data are then usually required to reformulate the conceptual model.

An erroneous conceptual model may have a crucial impact on the management strategies by misclassification of the water bodies, by the design of ineffective programmes of measures and by non-optimal design of monitoring programmes. A conceptual model will, however, always be a simplified version of the nature and the conceptual model is therefore inherently uncertain. Risks assessment is part of an optimal water management and models may be used in this context to assess the importance of being wrong on the conceptual model. This can be achieved by the formulation of multiple plausible conceptual understandings, which are implemented in the model followed by model simulations and an analysis of the model results.

Test of a conceptual model is often addressed in several stages of a model study, initially at the model construction and later, if applicable for the model system, during the calibration and validation of the model.

A thorough conceptual understanding of the physical system is vital to all monitoring programmes and is explicitly required in the WFD for the characterisation of the water districts and the assessment of the status and possibly grouping of water bodies. But a conceptual understanding is furthermore a prerequisite for the required risk and impact assessments. The conceptual model must be tested and validated by confronting the model with new monitoring data. If the monitoring data reveal surprises not foreseen by the conceptual model, the model must be revised and it may often be required to make a more comprehensive dedicated monitoring for this purpose, namely investigative monitoring.

4.1.4 Assess effects of anthropogenic activities

Monitoring data are used to identify trends in the ecological status and to assess whether an implemented programme of measures has had the expected effects. In this regard the natural climate variability induces variability in the variables to be analysed. Natural variability, together with measurement errors, will act as noise that may hide the signals from and make it very difficult to identify the effects of the anthropogenic activities. This is especially problematic in cases where the full effect of the intervention is slow, and the changes are modest initially and small compared to natural variations. Models may here be useful, if they are able to explain some of the natural variability and thereby enhance the signal from the anthropogenic activity. The key idea is to focus on the differences (residuals) between the model results and the observation data. If the model is able to describe the natural variations the residual does not include the part of the natural variability that the model is able to explain and the variability in the residuals will therefore be less than the variability in the raw observations. A reduced variance in the test series strengthens the tests and makes it easier to identify the signal from the anthropogenic activities. Tests can be made both for (linear) trends and for sudden shifts in conditions.

This approach can be used to filter out both temporal and spatial natural variations, see also section 4.1.5 below and case study 7.4.

Statistical tests on residuals, e.g.: student t-test and a Wilcoxon Rank-Sum test.

Identification of trends is a key objective of surveillance monitoring. Furthermore, analysis of the effects of implemented programmes of measures is a key objective of operational monitoring. In both cases it may be difficult to distinguish between the effects of anthropogenic activity and climate variation when analyses are made on the raw observation data alone. As illustrated in the examples above, this can be done much more efficient by joint use of modelling and monitoring data.

4.1.5 Design of monitoring programme

There are many different objectives to design and operate a monitoring system. Because the operation of monitoring programmes is expensive, the question is how we can assess the balance between the information content and the cost of the monitoring programme. State variables (pressure, fluxes, concentration, etc.) show variability is space and time. This variability appears as noise, when estimating the characteristics of the state variables. Consequently, estimates of the characteristics are uncertain. In principle the more variability in the state variables, the more measurements are required to arrive at the same level of uncertainty. For the design of a monitoring programme we have to assess the relation between the measurement effort (number of locations and measurement frequency) and the reduction of the uncertainty.

A basic principle in the application of models-based monitoring design is the link of model simulations to optimisation schemes by which the best monitoring strategy under specified constraints can be identified. Commonly, the purpose of the optimisation is to make the monitoring programme cost-effective, but the optimisation may have multiple objectives. The advantage of a model-based design is that the complexity of the physical system can be taken into account in the evaluation of alternative monitoring designs. In particular, physically based models that give a spatially continuous prediction of the state variables, account for a part of the natural variability. Subtracting the model predictions and observation data the part of the natural variability, described by the model, is filtered out from the observations. The noise, or uncertainty, in the monitoring data is consequently reduced, and the monitoring programme can be designed according to the reduced uncertainty. The better the model describes the natural variability, the less monitoring data points are required to arrive at the same level of uncertainty. In the design of an optimal monitoring network, a model, combined with appropriate techniques, can be used to evaluate the information a specific observation holds on the natural system. This can be done for any existing or imaginary sampling point or data type. The model can thus be used to define the number and location of sampling points required to achieve a certain level of uncertainty. Conversely, the model can be used to quantify how much each potential observation will reduce the uncertainty in the model predictions.

The analyses are based on the residuals (observation data minus model predictions) which, due to the models ability to describe the natural variability, have a smaller variance than the observations. With a reduced variance fewer points are required to obtain the same accuracy in an interpolation scheme. Kriging may, e.g., be chosen for interpolation, which can provide an estimate for the accuracy of the interpolation (standard deviation), and the monitoring network can be designed to support a predefined accuracy in the Kriging interpolation. Others approaches are based on filter techniques like Kalman Filtering or representers.

Optimal design of monitoring network is relevant for all monitoring programmes. Specific for the surveillance monitoring is that data acquired here should provide information for an efficient and effective design of future monitoring programmes.

4.2 Perspectives for future modelling support for monitoring

As previously stated, the link between data and modelling in practical applications is at present a one way relation – data to model, after which the model is run to study a specific problem. A continuous update of the model by utilisation of new monitoring data, commonly referred to as data assimilation, is presently almost restricted to meteorological and flood forecasting applications. But this approach may similarly be used for other domains and has received a growing attention in the research community. Especially the potential of remote sensing has been studied, due to high temporal resolution of satellite images and relatively low cost, but traditional monitoring data have also been applied in data assimilation. Besides the potentials of data assimilation, the joint use of modelling and monitoring is beneficial in an early stage, where the monitoring programmes are designed. This elaborated above where models calibrated to a specific system can be utilised to optimise the monitoring programme. Models may, however, also be applied in the design of monitoring programmes even before sufficient data are available for calibration, which is accomplished in, e.g., optimal experimental designs, where synthetic model produced observations are employed in the design phase.

Good examples on integration of monitoring and models do exists – as the early warning systems for floods and accidents with chemicals spills as developed in the Rhine and in the Tisza River, but are generally not common in praxis. Below short descriptions are provided for a few methods where monitoring and modelling are closely linked. This illustrates some future perspectives of joint modelling and modelling. The descriptions are only provided as examples and should not be considered a comprehensive list.

4.2.1 Model development

The WFD brings two important challenges: 1) water must be managed in an integrated way, i.e. including all uses of water and all domains, such as surface- and groundwater, and 2) the qualitative status is related not only to the chemical substances measured in the water, but should also be evaluated based on the ecological status. The integration of all domains and the assessments of the chemical and ecological status pose major challenges to both monitoring designing and praxis, see e.g. the Chemical Monitoring Activity (Quevauviller, 2006) and A combined monitoring-based and modelling-based priority setting (COMMPS) (http://ec.europa.eu/environment/water/water-framework/preparation_priority_list.htm), as well as to modelling, e.g. Wasson et al. (2003). In present years much effort is put in developing new models and methods to deal especially with the integrated and ecological aspects. In this regard, much is gained if monitoring and modelling tools and methodologies are developed with a joint application in mind. Most of the obstacles in joint use of monitoring and modelling listed in section 2.2 can be traced back to the parallel development of monitoring and modelling praxis, with separate focuses. To avoid a similar situation in the future, it is essential that there is a strong collaboration between the two communities, to learn the requirements, advantages and limitations in monitoring and modelling.

4.2.2 Uncertainty reduction in models

Monitoring systems are generally not designed to provide information to design or calibrate models. Therefore, often the data are not available from locations and times that are optimal for the model. The available data influence the models in two ways:

Reduction of model uncertainty can be achieved by quantifying the relation between the uncertainty in the model and the monitoring system. Not only is the impact of existing monitoring locations on the uncertainty reduction considered, but also the impact of potential locations. Monitoring locations and measurement times can then be ranked, according to the impact on the model uncertainty. The monitoring locations and measurement times are selected based on the highest impact on the reduction of uncertainty. So, the design criterion for the monitoring system is the model uncertainty.

4.2.3 Flood forecasting

Flood forecasting tools are driven by hydrological model systems that calculate future discharges and water levels by routing runoff from the sites of precipitation events, through the different runoff paths (surface runoff, subsurface runoff and channel flow), down to the sections where forecasts are demanded. Boundary conditions for forecasting consist of measured and forecasted hydro-meteorological data coming from gauging stations as well as from (other) forecasting systems. Flood forecasting tools are operated on continuous or event basis using real-time hydro-meteorological boundary data. Future discharges and water levels are calculated at each forecasting station with an appropriate lead time.

Flood forecasting systems fully integrate monitoring and modelling by use of data assimilation. This implies that the monitoring data are used in real-time to update (i.e. correct) the hydrological model.

5 Discussion and Conclusions

Monitoring data have always been requested by modellers to convert a generic model into a site-specific model. Modellers have thus long acknowledged the worth of monitoring data and been concerned in how monitoring programmes can be designed to improve the data foundation for modelling. Monitoring people have, on the other hand, not been dependent on models to design, implement and complete a monitoring programme. Traditionally, a good monitoring data set has been equalled long time-series, and the use of models to interpret and evaluate a monitoring programme is rarely seen in practice. While modelling may not necessarily be required for all monitoring activities at any site-specific location, the methods and examples presented here clearly demonstrate the potentials of using models to utilise the monitoring data in a much more efficient way.

The present report focuses only on the combined use of modelling and monitoring, potential use of models in the WFD is, however, not limited to the monitoring obligations. As described in Hattermann et al. (in prep.), models may be beneficial and – in a variety of circumstances – indispensable when implementing the WFD, e.g. for impact prediction and what-if analysis. They may help to understand and optimise the use of resources and can in particular be useful in the following steps of the WFD, Figure 5.

Figure 5 The WFD cycle displaying the sequence and relation between the WFD activities

The Harmoni-CA workshops on joint use of modelling and monitoring showed that most often models are not considered an option when the monitoring obligation in the WFD are solved in practise. A number of obstacles were identified to why this is so. Most of these obstacles are settled in historical, cultural and institutional aspects, but the workshops illustrated that if people from the monitoring community and from the modelling community are brought together for open-minded discussions and with free hands they are very open to discuss ideas and suggestions from the other side of the table, and able to see the perspectives in the combined use. To advance the joint use of monitoring and modelling there appears to be a strong need to first of all raise the awareness of how monitoring and modelling can be combined. Additionally, a shift in the perception of models is needed, where models currently often are considered useful only to provide answers to specific questions. This view must be replaced by recognition of models as being valuable tools that should be used continuously and interactively along with monitoring data, a tool that serves many purposes from the design phase of monitoring programmes to the interpretation and quality assurance of the monitoring data.

The overall objective of the Water Framework Directive is to achieve a good quantitative and qualitative status of all waters by 2015. The WFD brings two important challenges: 1) water must be managed in an integrated way, thus including all uses of water and all domains, such as surface- and groundwater, and 2) the qualitative status is related not only to the chemical substances measured in the water, but should also be evaluated based on the ecological status. To fulfil these requirements development of new tools and models is a necessity. To avoid a future parallel development of monitoring and modelling praxis in the field of integrated and ecological assessments, it is of vital importance that monitoring and modelling are realised to be inter-linked activities and not independent disciplines.

6 References

Arustiene J, Vaitiekuniene J and Jørgensen LF (Eds.) (2005) Joint use of modelling and monitoring for implementing the Water Framework Directive, workshop report. Harmoni-CA. Available on http://www.harmoni-ca.info

Borowski I, Hare M (2007) Exploring the gap between water managers and researchers: Difficulties of model-based tools to support practical water management, Water Resources Management, 21(7), 1049-1074

Brugnach M, Tagg A, Keil F, de Lange WJ (2007) Uncertainty Matters: Computer Models at the Science–Policy Interface, Water Resources Management, 21(7), 1075-1090

EC (2000) Directive 2000/60/EC of the European Parliament and of the Council of October 23 2000 establishing a framework for Community action in the field of water policy. Official Journal of the European Communities, L327/1–L327/72, 22.12.2000.

J. Grath, A. Scheidleder, S. Uhlig, K. Weber, M. Kralik, T. Keimel, D. Gruber (2001): "The EU Water Framework Directive: Statistical aspects of the identification of groundwater pollution trends, and aggregation of monitoring results". Final Report. Austrian Federal Ministry of Agriculture and Forestry, Environment and Water Management (Ref.: 41.046/01-IV1/00 and GZ 16 2500/2-I/6/00), European Commission (Grant Agreement Ref.: Subv 99/130794), in kind contributions by project partners. Vienna.

Grayson R and Blöschl G (2000) Spatial Modelling of Catchment Dynamics. In: Grayson and Blöschl (eds) Spatial Patterns in Catchment Hydrology: Observations and Modelling. Chapter 3, 51-81. Cambridge University Press.

Hatterman F, and Kundzewicz ZW (Eds.) (In Prep.) Model-supported implementation of the Water Framework Directive – A water managers guide. Harmoni-CA

Holt MS, Fox K, Grießbach E, Johnsen S, Kinnunen J, Lecloux A, Murray-Smith R, Peterson DR, Schröder R, Silvani M, ten Berge WFJ, Toy RJ, Feijtel TCM (2000), Monitoring, modelling and environmental exposure assessment of industrial chemicals in the aquatic environment. Chemosphere, 41, 1799-1808.

Højberg AL, Refsgaard JC (2005) Model uncertainty-parameter uncertainty versus conceptual models. Water Sci Technol 52(6):177–186

Højberg AL, Refsgaard JC, van Geer F, Jørgensen LF and Zsuffa I (2007) Use of Models to Support the Monitoring Requirements in the Water Framework Directive, Water Resources Management. http://dx.doi.org/10.1007/s11269-006-9119-y

Kamphorst E, Jørgensen LF, van Griensven A and Vanrolleghem PP (Eds.) (2005) Joint use of modelling and monitoring for implementing the Water Framework Directive, workshop report, 1st. workshop: State of the art on existing monitoring programmes around Europe. Harmoni-CA. Available on http://www.harmoni-ca.info

Lørup JK Refsgaard JC and Mazvimavi D (1998) Assessing the effect of land use change on catchment runoff by combined use of statistical tests and hydrological modelling: Case studies from Zimbabwe. J Hydrol, 205, 147-163.

Neuman, SP and Wierenga PJ (2003) A comprehensive strategy of hydrogeologic modeling and uncertainty analysis for nuclear facilities and sites. University of Arizona, Report NUREG/CR-6805.

Nilsen K, Somod B, Ellegaard C and Krause-Jensen D (2003) Assessing reference conditions according to the European Water Framework Directive using modelling and analysis of historical data: An example from Randers fjord, Denmark. AMBIO 32(4), 287-294.

Parr TW, Sier ARJ, Battarbee RW, Mackay A and Burgess J. (2003) Detecting environmental change: science and society – perspectives on long-term research and monitoring in the 21st century. Sci Total Environ, 310, 1-8.

Quevauviller P (2005) Groundwater monitoring in the context of EU legislation: reality and integration needs. J Environ Monitor, 7, 89-102.

Quevauviller P (2006) Chemical monitoring activity under the common implementation strategy of the WFD. J Soils & Sediments, 6 1), 2-3.

Reed PM, Minsker BS (2004) Striking the balance: Long-term groundwater monitoring design for conflicting objectives. J Water Resour Plan Manage-ASCE 130(2):140–149

Refsgaard JC, Alley WM and Vuglinsky VS. (1989) Methodology for Distinguishing Between Man's Influence and Climatic Effects on the Hydrological Cycle. IHP-III Project 6.3. Technical Documents in Hydrology, UNESCO.

Refsgaard JC and Henriksen HJ (2004) Modelling guidelines – terminology and guiding principles. Adv Water Resour, 27, 71-82.

Refsgaard JC, Henriksen HJ, Harrar WG, Scholten H and Kassahun A.(2005a). Quality assurance in model based water management – Review of existing practice and outline of new approaches. Environ Modell Softw, 20, 1201-1215.

Refsgaard JC, Jørgensen LF and Højberg AL (2007). Data availability and accessibility - State of the art on existing data required for modelling for research purposes and for the implementation of the Water Framework Directive, synthesis report. Deliverable D4-6

Scholten H, Kassahun A, Refsgaard JC, Kargas T, Gavardinas C and Beulens AJM (2005). A methodology to support multidisciplinary model-based water management. Environ Modell Softw, 20(10), 2101-1215.

Snepvangers, JJJC, Berendrecht WL and Valstar JR (2005) Model and measument based optimisation of quantitative monitoring networks: a comparative study. Modelcare 2005. Prepublished proceedings pp 145-152.

Styczen M (2002) Development of a tool for estimation of pesticide occurrence in surface water under Danish conditions. Int J Environ Anal Chem, 82 (8-9), 611-630.

Styczen M, Petersen S, Kristensen M, Jessen OZ, Rasmussen D, Andersen MB and Sørensen PB (2004) Calibration of Models Describing Pesticide Fate and Transport in Lillebæk and Odder Bæk Catchment. Danish Environmental Protection Agency, Pesticide Research no 62, 217 pp.

Troldborg L (2004) The influence of conceptual geological models in the simulation of flow and transport in Quaternary aquifer systems, PhD. thesis, Geological Survey of Denmark and Greenland Report 2004/107, 120 pp.

Wasson JG, Tusseau-Vuillemin MH, Andréassian V, Perrin C, Fauer JB, Barreteau O, Bousquet M and Chastan B (2003) What kind of water models are needed for the implementation of the European Water Framework Directive? Examples from France. Intl. J. River Basin Management, 1, 125-135.

7 Appendix - Case studies

In this section a few examples on the joint use of monitoring and modelling are provided for different domains and with different purposes of the study. The examples includes some of the tasks described in the previous sections as indicated in Table 3.

7.1 The national water resources model for Denmark (DK-model)

7.1.1 DK-model structure and purpose

The quality and quantity of the groundwater has long been of major concern in Denmark where groundwater constitutes 99% of the drinking water supply. To improve the quantitative estimate of the exploitable groundwater resource a national water resource model (DK-model) was established, covering the entire 43,000 km² of Denmark (Henriksen et al., 2003). The experiences gained from the work formed the basis for specifying modelling to become a supportive element in the coming WFD monitoring in Denmark. The aspects to which modelling contributes are illustrated below.

7.1.2 DK-model and monitoring

Climatic data, i.e. precipitation and evaporation, were obtained from two national research institutes, Danish Meteorological Institute and Danish Institute of Agricultural Sciences, respectively. Both institutes have extensive quality assurance on data sampling and use a scientific based state-of-the-art approach to correct the precipitation data (accounting for wind effects and wetting) and estimate evaporation (modified Penman equation). From a scientific standpoint the data resembled thus the best and most reliable data set. But when data were applied in the national water resources model and model outputs compared to measured groundwater head and river discharge data, significant water balance errors were observed. This error could only be explained by inconsistency in the climatic input data. While the precipitation and evaporation data were sufficiently accurate when only the precipitation or the evaporation was of interest, the methods used were not adequate when the entire water balance was the target. As such the inconsistency was crucial for the assessment of the quantitative status of the water resource, and certainly crucial for the water management at local and regional scales. These findings resulted in the collaboration of four national institutes involved in the assessment of the water resource. It was concluded that the water balance problem could not be fully solved based on the available knowledge, but a consensus was agreed (Plauborg et al., 2002) providing recommendations for the correction of precipitation and estimation of the evaporation, which are now used nationwide for estimation of net precipitation.

By simulating the entire country the model is used effectively as a spatial and temporal interpolation tool. For example, continuous images of the groundwater heads in different geological layers are obtained. This is of interest for identifying the areas where the groundwater head is critically low, such as areas with phreatic conditions in chalk formations, which may result in oxidation and release of nickel. Complete images of the groundwater level are similarly required to quantify the lowering of the water table by abstractions. Besides the groundwater head, the model provides three-dimensional interpolation of the groundwater recharge at different depths, so that the sustainability of the groundwater exploitation can be assessed by comparing the total abstraction in an aquifer to the total net recharge to the aquifer.

The establishment of the DK-model prompted the construction of a nationwide conceptual model on the hydrological system, including the subsurface geology, overland flow processes, unsaturated flow, saturated flow and groundwater-river interactions. Calibration and validation of the model improved the conceptual understanding in several instances. Examples include the observation of abstraction wells that went dry in the simulations, although the correct abstraction rates were applied. This is a typical indication of an erroneous hydrogeological model where the aquifer containing the abstraction well is misinterpreted and the simulation thus provided feed-back to the geological interpretation at the local scale.

The present DK-model provides an initial nationwide conceptual model. Since the model construction, the Danish counties, who act as the regional water authorities, have put much effort in updating the geological and hydrological understanding, which has been required for the regional water management by national legislation. The DK-model is currently being updated as a joint project between the regional water managers and the Geological Survey of Denmark and Greenland, where especially the conceptual model is revised. The numerical model provides a logical framework for this update, where the present conceptual understanding is confronted with the new knowledge, which is accommodated by the overlay of the two conceptual models and illustrated graphically. Differences in the two conceptual models are then discussed, and the initial conceptual model is adjusted as agreed upon.

The construction of nationwide conceptual model has recently proved valuable, as the national water resources model has been employed in the revision of groundwater bodies in Denmark, to form a consistent basis for the delineation in large parts of Denmark.

The overall objective of the DK-model is to provide an estimate of the exploitable resource. An acceptable exploitation is clearly linked to the level of impact on the physical system we accept as a consequence of human activities. Groundwater abstraction is the only anthropogenic activity considered explicitly in the model, and two critical aspects have been identified constraining the acceptable exploitation: 1) induced groundwater recharge to the deeper aquifer as a response to abstraction, this is critical as the younger water are generally polluted, and 2) reduction of discharge in the stream, which lowers the potential for achieving a good ecological status in the stream. The impact by water abstractions have been evaluated in a series of model simulations in which the abstraction rates were varied. Furthermore, the robustness to climate change was assessed by making simulations for dry and wet climate conditions.

To date the national water resources model has not been used in a monitoring network design phase. However, the national level monitoring network is to be evaluated and redesigned, and it is the intention that this will be aided by the use of the national water resources model.

Sophisticated data assimilation is not planned to be included in the national water resource model at the present stage. However, intensive geological characterisation is still ongoing in Denmark and it is the ambitions that the updated (hydro) geological knowledge is utilised to update the national water resource model.

7.1.3 References

Henriksen HJ, Troldborg L, Nyegaard P, Sonnenborg TO, Refsgaard JC and Madsen B (2003) Methodology for construction, calibration and validation of a national hydrological model for Denmark. J Hydrol, 280, 52-71.

Plauborg F, Refsgaard JC, Henriksen HJ, Blicher-Mathiesen G. and Kern-Hansen C. (2002) Vandbalance på mark- og oplandsskala. DJF rapport 70. Danmarks JordbrugsForskning. 45 pp. (in Danish)

7.2 Matsalu River Catchment

7.2.1 Modelling area and purpose

Modelling studies are commonly expected to require a vast amount of data to setup and run the models. Such data may not be readily available and thus hinder the use of models without intensive data collection prior to the study. Although the reliability of the model results generally are proportional to the amount of data, valuable model studies may be carried out with less data, which, in turn, may identify locations and data types that should be collected to improve the model and thereby the reliability of the results and future decisions. Such model exercise was undertaken for the Matsalu River Catchment in Estonia, the example is based on the example by Jensen (2005).

The catchment size is 4,870 km² and inhabited by 61,315 people (year 2002), of which 28,150 are served by public wastewater treatment plants. The model system MIKE BASIN was used to provide an overview of the anthropogenic load of nutrients (Nitrogen and Phosphorus) and their spatial distribution within the system. The existing monitoring programme included only 7 monitoring stations, Figure 6. The stations are primarily located in the downstream part of the river catchment.

Figure 6 Existing monitoring station and water quality classification (1999-2002) of the Matsalu River Catchment (from Jensen, 2005).

7.2.2 Matsalu model

The MIKE BASIN model was setup by utilising standard data such as digital elevation model (DEM) and precipitation. From information on land use (soil types, crops, use of fertilisers, livestock population) the agricultural non-point pollution was estimated for the different parts of the catchment using the GIS facilities of the tool. The transport of pollutants (BOD, Nitrogen and Phosphorus) was estimated taking into account a distance dependant (source to river distance) decay rate in the individual catchments. Pollution input from domestic sources was estimated based on information about human population density, coverage by sewer system and types of treatment plants. Furthermore, other point sources e.g. industrial input were included in the model set-up.

Based on this information, the pollution input from each part of the river system was calculated. The model was calibrated to the possible extent based on data from the existing seven monitoring stations. As a part of the calibration procedure the turnover of pollutants within the sub-catchment (from source to river) as well as within the river system was adjusted. The result of this calibration process is illustrated in Figure 7 with respect to nitrogen and phosphorous. Based on this characterisation water bodies of different quality could be delineated and monitoring stations could be identified with the purpose of following the trend in water quality at the most critical location.

Figure 7 Classification of the quality condition in the river systems based on the modelled annual average total N (left) and total P (right) concentration (present condition): Colour code describes the ecological status (Blue: High; Green: Good; Yellow: moderate; Orange: poor; Red: bad) (from Jensen, 2005).

The GIS-based modelling tool MIKE BASIN combines GIS-coverages on the relevant characteristics of the system to achieve a combined effect of the physical properties (e.g. river configuration, runoff properties, soil types etc.) and the anthropogenic activities (land use, crops, fertilizer etc.). The main advantages of such modelling approaches are that they generally require less data than the more detailed model systems and are easier to setup and run. However, as for all models the reliability of the results are highly dependent on the quality of the input data and the observations available to test (calibrate and validate) the model, as well as the underlying model assumptions. When data is sparse the model results may be associated by significant uncertainties, and it may not be recommendable to formulate expensive programmes of measures based such model studies. Nevertheless, a model study as presented for the Matsalu River Basin are very valuable to identify areas that are most problematic which should be examined in further detail.

7.2.3 Matsalu model and monitoring

In addition to the few monitoring points, data input for the Matsalu catchment were sparse, and to a large extend estimated based on statistics, e.g., land use, live-stock and population densities and location of point sources. By using a model describing the main processes in combination with the available monitoring data, it was possible to test the reliability of the estimated input data, and to interpolate the effect of the pressures for the entire catchment. The final model results depict river stretches and water bodies most likely to meet a good ecological status, or most likely at risk of not meeting the requirements. Using the model in an uncertainty assessment, the reliability of the model results can be addressed, and used to identify areas where more detailed input data and possible more monitoring data and detailed modelling are required to achieve a sufficient level of confidence in the assessment.

7.2.4 References

Jensen JK (2005) Modelling-A tool for interpretation and optimisation of monitoring programmes. In: Arustiene J, Vaitiekuniene J, Jørgensen LF (eds) Joint use of modelling and monitoring for implementing the Water Framework Directive, workshop report. Harmoni-CA. Available on http://www.harmoni-ca.info

7.3 National Hydrological Forecasting System of Hungary (NHFS)

Hungary is one of the most flood-exposed countries in Europe. About 1/4 of its territory is endangered by floods of the Danube, Tisza and Drava rivers. In spite of the strong flood control dike systems, states of extreme emergencies frequently develop, which could easily turn into disasters unless appropriate flood control measures are taken. Just during the past decade six extreme, record-breaking floods occured on the Danube and the Tisza, and it was only a matter of luck and of painstaking and efficient flood control measures that no catastrophes have happened, although the chance at certain instances was rather high.

One of the corner-stones of efficient flood control is the accurate and timely forecasting of flood events, which enables the defenders and authorities to take the necessary measures. For this purpose the Environmental Protection and Water Management Research Institute (VITUKI) developed the National Hydrological Forecasting System (NHFS) (Bartha et al., 1983; Szöllősi-Nagy, 1989; Bálint et al., 1996), which has been being operated and periodically upgraded by the institute since 1983. The overall objective of NHFS is to forecast continuously the hydrological regime of major rivers of Hungary with special regard to flood events.

7.3.1 Structure and operation of the NHFS

Forecasts made by NHFS are based on various meteorological and hydrological input data collected real time from the river basins. Measured precipitation, snow, temperature and water level data are collected from upstream gauging stations on daily basis. The frequency of these measurements ranges from hourly, through daily to weekly, depending on the temporal dynamism of variables. Meteorological data are collected from 96 Hungarian and 410 foreign (German, Austrian, Slovakian, etc.) stations, while hydrological data are gathered from 170 Hungarian and 210 foreign stations. Measured meteorological and hydrological time series are extended into the future by appending forecasted data generated by meteorological forecasting systems as well as by hydrological forecasting systems operated by upstream countries.

The collected and assimilated hydro-meteorological data are input to the forecasting model system (Bálint et. al., 1996; Gauzer & Bartha, 1999), which consists of rainfall-runoff and channel routing model blocks representing the different sub-catchments and river reaches, Figure 8.

Rainfall-runoff models are complex model systems simulating runoff from the mountainous parts of the basins. The following processes are simulated by these models: snow accumulation and ablation, soil frost, rainfall interception, distribution of runoff quantities among the different paths (surface runoff, interflow, base-flow) and finally the actual processes of surface and subsurface runoffs. The different runoff processes are modelled with kinetic wave, continuous cascade or Discrete Linear Cascade (DLCM) models. The two latter are based on the principles of the Nash cascade model.

For routing runoff further in the river channels, DLCM models are applied. Each river reach (reach between two major gauges) are represented by a DLCM cascade (see Figure 6). Discharges forecasted by the model system are corrected by an Auto-Regressive Moving Average (ARMA) based statistical error correction technique. Since DLCM models calculate discharges only, water levels have to be derived subsequently with the help of Q-H rating curves. In those river reaches however where no such rating curves are applicable (e.g. reaches upstream of barrages or tributary mouths), a hydraulic backwater model, similar to the one applied by Todini and Bossi (1986), is used for calculating water levels.

The forecasting model is run on daily basis using updated hydro-meteorological data as boundary conditions. Future discharges and water levels are calculated at each forecasting gauging stations (see Figure 8.) with a lead time of 6 days.

The different modelling modules of NHFS need to be periodically recalibrated and re-verified. For this purpose the NHFS is turned into simulation mode and calibration / validation runs are implemented using historical hydro-meteorological data.

7.3.2 NHFS and monitoring

The primary role of NHFS is hydrological forecasting, which means that monitoring is considered only as a source of real time boundary data. Other important aspects of joint use of monitoring and modelling are taken into consideration only to a limited extend, even though there are considerable potentials in these fields.

For the time being filtering of gross measuring errors is taking place by comparing daily measurements with forecasts from the day before. More accurate error checking could be achieved if measured values were compared with simulated, and not with forecasted values. It is thus recommended to implement regular quality assurance simulations with the help of NHFS.

Running NHFS in simulation mode also enables temporal interpolations between measured water level and derived discharge data. Spatial interpolation on the other hand is not applicable as it is not possible to associate sections within a river reach with components of the cascade model from where calculated discharges can be derived. From this point of view cascade models behaves like lumped models.

Developing conceptual models and NHFS model systems for the major rivers of Hungary were interlinked, mutually supporting processes. Original concepts, based on historical data and expert experiences, were first implemented and tested in NHFS that provided feedback on how to improve the conceptual model, which in turn formed the basis for developing further the models themselves. This iterative process has been going on since the introduction of NHFS.

Runoff from a river basin can be classified into three general classes: surface runoff, subsurface runoff and river channel runoff. Hydraulic principles of these runoff types are quite different requiring distinct modelling approaches. An essential task of conceptual model development is thus to determine the dominant runoff type(s) in the different subsystems of the basin. Based on such subdivision, models can be associated with the subsystems, from which the complex model system can be built up, Figure 8. Accordingly rainfall-runoff models have been assigned to the mountainous sub-catchments where surface and subsurface runoffs are dominating, while channel routing models have been associated with the lowland parts of the basin where runoff is concentrated in river channels, Figure 9.

With respect of mountainous sub-catchments the development of conceptual model begins with the identification of determining processes such as snow accumulation and ablation, interception, surface runoff, interflow and base-flow. This is followed by the exploration of interactions between these processes. The rainfall-runoff modules of NHFS have been providing continuous and valuable support through testing the different concepts of catchment hydrological processes.

Figure 9 Identification of rainfall-runoff sub-catchments and channel systems in the Upper Tisza basin

As indicated above, the conceptual models of the Danube, Tisza and Drava basins, as well as their realizations in NHFS have been continuously improved as gradually more information has been gained (especially during floods) about the behaviour of the systems. The most important milestones of this development were as follows (see also: Gauzer & Bartha, 1999; Bálint et. al., 2005):

In spite of the significant model developments there are certain aspects of the state-of-the-art conceptual models that exceed the modelling capacity of NHFS - only fundamental changes in the model system would make it possible to cope with these aspects. One such aspect is downstream influence on discharges, which the DLCM is unable to take into consideration. Downstream influence plays a determining role on the lowland reaches of the River Tisza, where the impounding effects of the Körös and Maros tributaries as well as that of the Danube influence not just water levels but also the discharges of the river. The ultimate solution for this problem would be the replacement of DLCM with a more sophisticated hydraulic model.

Real time integration of different hydro-meteorological boundary data into a hydrological modelling system makes NHFS an ideal data assimilation framework.

Meteorological data, coming from monitoring stations as well as from meteorological forecasting models, are interpolated and calculated at the nodes of the meteorological grid in order to generate a homogeneous input dataset of appropriate temporal and spatial resolutions (Bálint et. al., 2005). For the time being meteorological data from the following sources are assimilated in this way:

Similar assimilation is taking place in case of hydrological input data too. Measured water level and derived (from Q-H curves) discharge data from the gauging stations are supplemented with values forecasted by hydrological forecasting systems operated by Germany and Austria. The so-assimilated hydrological datasets, along with the assimilated meteorological datasets, form the boundary conditions for NHFS.

Hydrological data assimilation could further be extended with the help of NHFS through repeated temporal interpolations implemented in simulation mode.

7.3.3 References

Bálint, G., A, Csík, P. Bartha, B. Gauzer and I. Bonta (2005) Application of meteorological ensembles for Danube flood forecasting and warning. In: Transboundary floods: Reducing risks through flood management Eds: J. Marsalek,G. Stancalie and G. Balint NATO Science Series, IV. Earth and Environmental Sciences. , Dordrecht, The Netherlands – Vol. 72 pp. 57-68

Bálint, G., Pándi, G. and Illés, L. (1996) Analyses of floods on the Upper-Tisza with special emphasis on hydrological forecasting. In: Proceedings 18th (XVIII) Conference of the Danube Countries on Hydrological Forecasting and Hydrological Bases of water Management, Schriftenreihe zur Wasserwirtschaft, Vol. 19/1, Technische Universität Graz, Graz - Austria, August 1996 pp. B77-82.

Bartha,P; Szöllõsi-Nagy,A; and Harkányi, K. (1983) Hidrológiai adatgyűjtő és előrejelző rendszer. A Duna. (Hydrological data collection and forecasting system, The Danube), Vízügyi Közlemények Vol. 65. (LXV), 3., pp.373-388

Gauzer, B., Bartha, P. (1999) Az 1970. és 1998. évi felső-tiszai árhullámok összehasonlítása, árvízi szimulációs vizsgálatok. (Comparative simulation analysis of the 1970 and 1008 floods of the Upper Tisza river) Vízügyi Közlemények, Budapest 1999/3. pp 354-390.

Szöllősi-Nagy, A. (1989) A mederbeli lefolyás real-time előrejelzése dinamikus strukturális-stochasztikus modellekkel. (Real-time forecasting of channel runoff with the help of dynamic structural-stochastic models) VITUKI, Budapest.

Todini E. and Bossi A. (1986) PAB (Parabolic and Backwater) an unconditionally stable flood routing scheme particularly suited for real time forecasting and control. Journal of Hydraulic Research, Vol. 24, n. 5, (1986) pp. 405-424.

7.4 Groundwater model in monitoring network design (Monitoring Design)

This example is based on a study carried out in the province of Utrecht, in the centre of The Netherlands. The province operates a groundwater head monitoring network and the objective of the study was to optimise the existing monitoring network by identifying the number and locations of observation wells required to achieve the desired accuracy. The objectives of the monitoring network were that it should provide information about the characteristics of the groundwater system, such as the spatial distribution of the time-average head and the average high groundwater head (AHG) defined as the 87.5 % percentile. The accuracy was formulated in terms of standard deviation of the spatial interpolation of the AHG, i.e. the spatially interpolated AHG should represent the true AHG within a specified deviation.

A traditional approach to estimate the spatial distribution of the AHG would be interpolation by standard interpolation routines, such as Kriging. Kriging can provide both an interpolated AHG as well as and estimate of the accuracy of the interpolation (Kriging standard deviation), which is dependent on the data density and the geostatistical properties of the variable (variance and correlation lengths). As an alternative to traditional interpolation, a groundwater model was constructed for the area, which where then used to filter out the regional variation of the AHG. Kriging was thus used to interpolate the deviation between the observations and the model results, rather than AHG itself. Because the model accounts for the regional pattern in the groundwater head, the variability in the residuals between observations and the model is much less than the variation in the observed AHG, Figure 10. In the ideal case, with a perfect model, the residuals are small and spatially uncorrelated. Therefore, the better the model describes reality, the smaller the residuals, and consequently the smaller number of observation wells necessary to meet the required maximum Kriging standard deviation.

Figure 10 Illustration of the reduction in variance when the residual between observations and model results are considered instead of observations.

The basis for Kriging interpolation is the variogram, in particular the sill (indicating the total variability) and the range (a measure of the spatial correlation length). For the province of Utrecht we used 139 existing observation wells to calculate the variogram. In Figure 11a the variogram is given of the observed AHG without using the model and Figure 11b gives the variogram of the residuals. The sill in figure 11a is over 22 m², whereas in the sill in figure 11b is 0.32 m². This means that the model accounts for about 90% of the total spatial variability.

The network lay-out was optimised by varying the number of observation wells and their location with the constraint that the Kriging standard deviation should be less than a given threshold. Using the model, with 40 monitoring wells, the Kriging standard deviation is less than 35 cm. If the model would not be used, the number of monitoring wells would be over 100 to reach the same Kriging standard deviation. Note that the model is used as a fixed reference. Once the models ability to describe the regional pattern in the groundwater level has been mapped it can be used also in the interpolation for new monitoring periods, without running the model with new data.

Complexity of code (Data Requirements)	Water Resources Assessment	Floods	Surface water quality and ecosystems	Agricultural management, including non-point pollution	Groundwater quality	Socio-economic cost assessment
Simple models *(Low)*	GIS based water balance tools, MIKE BASIN, HEC-HMS	Cascade Models (Discrete Linear Cascade Model)	Regression models	Regression models, Export coefficient models	Overlay/index methods (DRASTIC, SINTAC, etc.)	Cost effectiveness
Intermediate models – *(Intermediate)*	Sacramento, HBV, MIKE 11		DYNDIS	USLE, HBV-N, SENSMOD		Hedonic pricing method, Contingent valuation method, Cost-benefit
Comprehensive models – (High)	MIKE SHE, MODFLOW	MIKE 11, SOBEK, ISIS HEC-RAS	DELFT3D-ECO	SWAT, DAISY, SOIL-N, ANIMO	PHREAQC, HST3D, MODFLOW, MIKE SHE
Process studies – (Very high)	Comprehensive models + tailor made research codes	Comprehensive models + tailor made research codes	Comprehensive models + tailor made research codes	Comprehensive models + tailor made research codes	Comprehensive models + tailor made research codes	Comprehensive models + tailor made research codes

Task/Case	DK-model	Matsalu River	NHFS	MonitoringDesign
Quality assurance	X		X
Interpolation/Extrapolation	X	X	X	X
Conceptual model	X		X
Anthropogenic effects	X	X
Design of monitoring programmes	(X)	X		X
Data assimilation	(X)		X