Experiences from a case study of multi-model application to assess the behaviour of pollutants in the Dnieper–Bug Estuary

Experiences from a case study of multi-model application to assess the behaviour of pollutants in the Dnieper–Bug Estuary Luigi Montea,∗, Lars H ˚akansonb, Raul Peria ˜nezc, Gennady Laptevd, Mark Zheleznyake, Vladimir Maderichf, Giacomo Angelig, Vladimir Koshebutskyh a ENEA, CR Casaccia, via P. Anguillarese, 301, 00100 Rome, Italy b Institute of Earth Sciences, Uppsala University, Sweden c Departamento Fisica Aplicada, University of Sevilla, Spain d Ukrainian Institute for Hydrometeorology, Ukraine e Institute of Mathematical Machines and System Problems, Ukraine f Institute of Mathematical Machines and System Problems, Ukraine g ENEA, CR Casaccia, Italy h Institute of Mathematical Machines and System Problems, Ukraine Keywords: Environmental models Coastal areas Contamination Multi-model approach abstract The present paper describes the results of the application of four state-of-the-art models to predict the concentrations of pollutants in the abiotic components of the Dnieper–Bug Estuary (Ukraine). The estuary was contaminated by the radioactive substances introduced in the environment following the Chernobyl accident. The scope, the methodological approaches and the theoretical foundations underpinning the examined models are presented and compared. The model performances were assessed by comparison with available empirical data of water contamination. The main factors influencing the inherent uncertainty of the models were examined: incomplete knowledge, paucity of extensive data sets relevant to some environmental quantities, the vagueness and the ambiguity of certain information about environmental processes that can be hardly parameterised in quantitative way, etc. Model performances reflect the intrinsic uncertainty of knowledge concerning the quantitative behaviour of the involved environmental process, the ambiguity of interpretation and parameterisation of such processes, the inherent variability of environmental quantities, etc. The difficulties in selecting the “best performance” model and the benefits arising from a multi-model approach to afford complex environmental problems are presented and discussed. Multi-model approach helps to get an insight into complex problems of environmental management, to promote co-operation among modellers and to profit by the different perspectives of the models. ∗Corresponding author. Tel.: +39 06 30484645; fax: +39 06 30486716. E-mail address: [email protected] (L. Monte). 1. Introduction The present paper, although dealing with the behaviour of radioactive contaminants in coastal areas, affords some problems of general nature concerning the application of models to complex environmental systems. It should be recognised that the accident at the Chernobyl nuclear power plant in 1986 occurred when, owing to a great deal of previous theoretical and experimental studies, some disciplines, like radioecology, had reached mature stages of development. Therefore, the number of international projects launched for the assessment of model applications to the environmental contamination from the abovementioned accident (BIOMOVS, 1991; IAEA, 2000) offered the unique opportunity of learning more about problems of general relevance concerning the validity, the usage and the usefulness of environmental modelling for solving complex problems of environmental management. This work will try to discuss and analyse such lessons. Due to the biological value of the coastal zone and the different demands on the utilisation of coastal waters, it is easy to understand why so much interest concerns this ecosystem and, more generally, the marine environment (Baxter et al., 1998; Aarkrog, 1998; Kryshev and Sazykina, 1995). In particular, estuaries are the object of research and environmental management in view of their high biological productivity, of their economic exploitation and of their importance for the whole biological and ecological cycles of many living species. However, in the framework of international projects on model validations in radioecology, much more efforts have been focused on lakes and rivers than on coastal areas and seas. Recently, some reviews of state-of-the-art models for the management of lakes, catchments and rivers polluted by radioactive substances have been presented (Monte et al., 2003, 2004, 2005a). These reviews were done within the thematic network EVANET-HYDRA (“Evaluation and Network of EC-Decision Support Systems in the field of Hydrological Dispersion Models and of Aquatic Radioecological Research”; http://info. casaccia.enea.it/evanet-hydra) financed by the European Commission. In spite of the attention that coastal environment deserves, it seems that no similar studies are available for models aimed at predicting the behaviour of radionuclides in estuarine systems. One of the aims of the present work is to bridge this gap. We will outline an assessment of the methodologies used by stateof-the-art models that can be applied to the coastal environment. A particular attention will be devoted to the assessment of the results of an exercise of the application of different coastal models to a specific estuarine scenario, the Dnieper–Bug Estuary (DBE). The benefits from a multi-modelling approach will be analysed and assessed in view of the difficulties, such as the shortage of input data and information, encountered by modellers when dealing with complex environmental problems. 2. Overview of modelling approaches 2.1. General remarks In this section, we will briefly outline the general, theoretical foundations underpinning the models for assessing the behaviour of pollutants in estuarine systems. More detailed descriptions of the models object of the present study are reported in Appendix A. In principle, most approaches for predicting the migration of radionuclides through estuaries are similar to the ones implemented in models for lakes and rivers (Monte et al., 2003, 2005a). It is, however, necessary to account for the particular environmental conditions and processes that significantly influence the behaviour of contaminants in coastal environments. Generally speaking, the overall structure of models for predicting the behaviour of radionuclides in estuaries comprises the following sub-models: (a) modules for predicting the physical, hydrological, the hydraulics and the biotic processes that occur in the estuarine system (such as water temperature and salinity profiles, water fluxes and current velocities, erosionsedimentation processes and dynamics of suspended matter in water, growth rates of organisms, tidal cycles, etc.) and are supposed to influence the contaminant migration; (b) modules for predicting the radionuclide transfer: 1. to the estuary from its catchment; 2. through the abiotic components of the estuary and 3. from the abiotic components to the biota. As for the other water bodies, the main processes that control the migration of contaminants in estuaries are, basically, the diffusion of dissolved substances due to water turbulent motion (eddy diffusion), the transport due to the water current, the interaction of dissolved pollutant with suspended matter and bottom sediment, the mixing processes between different layers of water and, finally, the migration of pollutants from the water to the bottom sediments (sedimentation) and from the bottom sediment to the water (re-suspension). An example of sub-model for predicting hydrodynamic processes is described in Appendix A (University of Seville model). In general, several different configurations may be adopted for this kind of models. The most general equations are the full 3D hydrodynamic equations including baroclinic terms (density differences). However, “2D vertical depth-averaged” models are often used when a vertical mixing of the water column can be hypothesised (Peria ˜ nez et al., 1996). These models provide predictions of the water current. Suspended sediment transport is also described by transport/diffusion equation accounting for water velocity and erosion and deposition processes (Eisma, 1993; Peria ˜ nez, 2002). It is obvious that not all the models make use of hydrological/hydraulic sub-models for assessing the quantitative behaviour of the environmental processes influencing the migration of pollutants through the components of an estu- ary ecosystem. Some models make use of empirical values of those parameters that relate the migration of pollutants to the environmental processes and that are averaged over finite regions of space and intervals of time (for instance, monthly averages of time-dependent quantities, average water fluxes among different sections of a water body, etc.). The sub-models for predicting the interaction of radionuclide in the water column with suspended matter and bottom sediments show different degrees of complexity ranging from simple kd-based assumptions (it is supposed that a reversible equilibrium between dissolved and particulate phases of radionuclide is quickly achieved) to complex multistage interactions (Peria ˜ nez, 2004; Ciffroy et al., 2001). These sub-models assess the complicated problem of radionuclide interaction with sediment particles from various perspectives emphasising, for instance, the stochastic nature of such a process (Børretzen and Salbu, 2002), the variety of physical and chemical characteristics that influence its variability (Abril and Fraga, 1996; Matsunaga et al., 2004; Smith and Comans, 1996) or the different modelling approaches (Monte et al., 2005b). As many of the mentioned aspects were described in previous papers (Monte et al., 2003, 2005a), we do not deem necessary to repeat the results of assessments that have been discussed in the scientific literature. Some particular environmental conditions and processes are typical of the estuarine ecosystem and significantly influence the migration of toxic substances. For instance, the different density between sea and fresh waters generates a vertical stratification that affects the diffusion of pollutants through the water column. Moreover, tidal cycle is a further factor that should be considered and modelled for assessing both the hydrodynamics of estuaries and the dispersion of contaminants through estuaries (Dyer, 1980). The mentioned principles and approaches underpin the great deal of models developed to predict the dispersion of radionuclide through the marine and the coastal environment (Prandle, 1984; Breton and Salomon, 1995; Salomon et al., 1995; Schonfeld, 1995; Aldridge et al., 2003; Aldridge, 1998; Goshawk et al., 2003; Smith et al., 2003; Abril and Abdel-Aal, 2000; Harms, 1997; H˚ akanson, 1999, 2000; Cetina et al., 2000; Fisher et al., 1999). Most methodologies of radioecological modelling are similar to the ones used to predict the distribution and the effect of other kinds of toxicants, such as heavy metals, in water Table1–Mainfeaturesofthemodels that participated in the exercise Model Developer Main model features Horizon CoastMab Uppsala University, Sweden Generic, process-based, dynamic, high emphasis on ecological aspects. TRP = monthly averages, SRP = the entire coastal system Compartment model to predict monthly average values of radionuclide concentrations in water and fish accounting for prevailing ecological and environmental processes when radionuclide concentrations in sea and direct radionuclide deposition over the coastal area are available (for the specific applications also direct radionuclide flux from rivers are supplied as input data) U. Sevilla University of Sevilla, Spain 2D (depth-averaged) based on advection-diffusion equation, high emphasis on hydrodynamic processes TRP and SRP: in principle the model can supply very detailed information 2D model. The hydrological characteristics of the system that influence the migration of pollutant (water current velocity field) are modelled from fundamental equations accounting for meteorological conditions ENEA ENEA, Italy Generic, process-based, dynamic, emphasis on radionuclide flux balance and migration to sediment TRP = monthly averages SRP = for the present application the coastal system is subdivided in three sectors Compartment model to predict monthly average values of radionuclide concentrations in deep and surface water accounting for prevailing radionuclide fluxes when radionuclide concentrations in sea and direct radionuclide deposition over the coastal area are available. Balance of radionuclide in the system is evaluated using, as input data (a) the radionuclide fluxes from the rivers flowing into the coastal system; or (b) the deposition of radionuclide over the whole Dnieper catchment (regional model for 90Sr) THREETOX IMMSP, Ukraine 3D model based on advection-diffusion equation, high emphasis on hydrodynamic processes TRP and SRP: in principle the model can supply very detailed information 3D model for predicting concentration in water when radionuclide input into the system is known. The hydrodynamics is simulated on the base of 3D, time-dependent, free surface, primitive equation model The hydrological characteristics of the system that influence the migration of pollutant (water current velocity field) are modelled from fundamental equations accounting for meteorological conditions. The model account for the processes governing the exchange of radionuclide between water and suspended/bottom sediment TRP: time resolving power, SRP: spatial resolving power. systems (Jørgensen, 1979; Ciffroy et al., 2000). On the other hand, the environmental processes, such as sedimentation, sediment re-suspension, tidal dynamics, water stratification, etc., that control the behaviour of a contaminant in water, are common to both radioactive and non-radioactive substances (H˚ akanson et al., 2004). Nevertheless, particular chemical processes affecting the behaviour of specific contaminants need to be considered and appropriately modelled (mercury is an obvious example; Rajar et al., 2004). The situation is even more complicated for pollutants, like organic compounds, whose fate is intimately linked to the cycling of organic matter in the aquatic ecosystem (DeBruyn and Gobas, 2004; Gobas, 1993; Jimenez-Montealegre et al., 2002) or is controlled by processes of degradation like photolysis and hydrolysis (Mossman and AlMulki, 1996) that do not affect the behaviour of radionclides or heavy metals. Today, many user-friendly software tools are available to implement models in computer codes and many models have been developed and tested. Modellers can exchange data and information assessing and comparing their results in a very dynamic and interactive environment dominated by the information technologies. How can we take advantage of all that? 2.2. Outline of model features The main features of the models used for the present exercise are described in Appendix A and summarised in Table 1.Itcan be useful to classify models according to their horizon and to their time and spatial resolving powers (TRP and SRP) in view of their scopes and aims. Horizon, the range of information (knowledge obtained from investigation and study, instructions and requested input data) that the model is meant to process and of the outcomes that is meant to predict. The horizon defines the starting and the end points of a model. For instance, a model can use as input data the deposition of radionuclide onto the water surface and the catchment area of the estuary. The model horizon is different when the input data are the deposition onto the water surface and the empirical values of radionuclide flowing from the tributary rivers. Obviously in the first example, the model aim is more ambitious and the model is more informative as it does not require information (input of contaminant from rivers) that can be difficult or impossible to obtain for some scenarios (for instance, in case of an accident). Nevertheless, a larger horizon generally implies a higher uncertainty. The resolving power of a model is a measure of the level of detail of its predictions. The “time resolving power” (TRP) is the ability of a model to predict differences in the system behaviour over a given interval of time. In other words, the model is supposed to describe the average behaviour of the system over a defined time interval. Example of factors affecting the TRP are the intervals of time necessary to assure the damping of some transient processes which are not intended to be modelled in a sufficient detailed time scale (for instance, the time necessary to approach a homogeneous distribution of the contaminant in the water column, a condition that, obviously, is not immediately achieved, when box models are used; the sorption and de-sorption processes of radionuclide on suspended matter when the model is based on the hypothesis of an “instantaneous” equilibrium between the dissolved and the particulate contaminant phases, etc.). Similarly the spatial resolving power (SRP) is the ability of a model to predict differences in the system behaviour over a given spatial grid. For instance, the SRP of box models for predicting the spatial distribution of a substance in water is the size of the boxes. The horizon, the TRP and the SRP are “a priori” characteristics of a model. They are planned and set up by model developers according to their expert judgement. It is quite obvious that the aim is to properly identify such model attributes to maximise the predictive power of the model. 3. Model applications 3.1. Introductory remarks: the ideal case The main aim of the present work is to analyse the performance of different models for application to real coastal systems, accounting for the available empirical information and input data and in view of the benefit from a multi-modelling approach. Therefore, our scope is somewhat more than a simple validation study. It can be useful to summarise the possible options of the problem-solving process in view of environmental modelling applications. The simplest kind of exercise that can be performed is what we can call a “school-boy problem”. Candidates are requested to answer a specific question on the basis of a complete set of preliminary data and information. The answer is univocal and the exercise has only two possible outcomes: the answer can be right or wrong. A more elaborate exercise consists in formulating a problem that supplies not only the necessary input data and information but also some more data that are unnecessary to answer the problem itself. It is quite obvious that the previous examples of problems are typical of academic exercises. There are only two kinds of actors playing different roles in these exercises: an “omniscient” referee and one or more candidates. It is quite obvious that, in real circumstances, the situation is more complicated. In relation to complex, natural systems, we can summarise a traditional perspective as follows: a theoretical “ideal model” inherent to the nature does exist: the “MODEL”. The aim of science is to reveal such a “MODEL”. Due to objective difficulties, only approximate realisations of such a “MODEL” are available although these can be incessantly improved. The different realisations of the “MODEL” can be temporarily accepted or definitely rejected by “verification–falsification” procedures. If we accept the above point of view, we should conclude that, when performing blind test exercises of model validation and comparison, the modellers have to prove that the performances of the models they developed are close to the ideal one by showing that model results fit empirical data. The modeller that got the results more close to the empirical outcome wins the game. Moreover, in general, a “rule” exists to decide whether the model is “right” or “wrong”. Such a rule is based on statistical tests of significance to ascertain if the model output are in agreement with the empirical data when the uncertainty of the model outcome and of the measurements are accounted for. It is out of the scope of this paper to debate about the existence of such an “ideal” model, nevertheless it is quite obvious that the assessment of the correctness of the realisation of the “ideal” model is based on several assumptions: experiments can be performed and repeated “ad libitum” (reproducibility); uncertainties of empirical data can be reduced at will by repeated measurements (accuracy); it is possible to perform “crucial experiments” having two possible outcomes, false or true, for assessing the validity of a model (falsification). The above assumptions straightforwardly imply that ever more accurate empirical data can be compared with ever more reliable model output (it is sufficient that modellers and experimentalists are clever enough to achieve high quality results in their respective fields). From a theoretical point of view the assessment of the uncertainty of the model output has a sound mathematical foundation. According to the most general formulation, any model can be defined as follows: L(m1,...,m n)X=Y(1) where Lis an operator acting on vector X(the unknown quantities that should be predicted and that depends on time tand space position x), vector Yis the set of input functions and m1,...,m nare a set of parameters. Together with Eq. (1) initial/boundary conditions should be considered. Model parameter values, initial/boundary conditions and input functions are obtained from experimental evaluations corresponding, respectively, to the empirical characteristics of the examined environmental system, to the initial status of environmental system and of the contamination and to the “forcing functions” corresponding to the pollution of the system from external sources. Similarly, repeated experiments, model test, hypothesis verifications give the opportunity of deepening our knowledge of the complex set of quantitative rules and laws that underlie the many processes occurring in the environment. This corresponds to the improvement of information about the structure and the parameters of operator L.Inthe ideal case, initial/boundary conditions, model parameters and input functions are characterised by expected values and the relevant probability distributions. When these distributions are known, it is possible to determine the average values and the distribution of the model output by analytical mathematical techniques or by numeric procedures such as Monte Carlo routines. More optimistically, repeated experimental activities and the improvement of measurement methods can allow one to reduce the “dispersion” of the empirical values associated with the model parameter and the initial/boundary conditions. Thus, for a given input, the “dispersion” (uncertainty) of the output values can be lowered at will. This is the most traditional and reassuring recipe for managing model uncertainty. 3.2. The practical case: applications of state-of-the-art models Unfortunately, we have to cope with less optimistic circumstances: (a) the available values of model parameters (chiefly the transfer parameters) for many environmental systems are scanty, consequently it is almost impossible to obtain their statistical distributions (it is preferable to say that, in general, only few empirical evaluations of many parameters are available and these should be considered as “reference” values); (b) input functions can be affected by significant uncertainty that cannot be rigorously quantified chiefly in connection with the emergency phase of an accident; (c) the structure itself of the model (the operator L) is uncertain. Traditional approaches for assessing model validity are not applicable to environmental modelling. Most knowledge is often obtained by accidental events (like the Chernobyl incident), whereas laboratory experiments can never reproduce the complex environmental processes and situations occurring in real circumstances (Peters, 1986). It is, in general, impossible to perform ad hoc experiments to reduce uncertainties and to falsify a model. Therefore, reproducibility, accuracy and falsification principles can be difficult to apply. Generally, different modellers make use of different experimental data and information to develop, to test and to calibrate their models. All that can make it difficult to parameterise in a quantitative and univocal way the key processes and mechanisms regulating the value of a target variable. In principle, it is, therefore, hard to select a unique, optimal model (the model closest to the “ideal” one). Several authors have discussed these problems (Hornberger and Spear, 1981). According to the “equifinality” principle, it was claimed that, given a certain level of process-understanding, different model structures and parameter values can be equally acceptable in assessing the behaviour of complex environmental systems (Beven and Freer, 2001). In other words, we are usually dealing with many plausible models whose results are expert estimates entailing certain commitment based on scientific knowledge (Giles, 1981). There are several other difficulties that modellers face for the application of environmental models such as vagueness, ambiguity and incompleteness of information and input data relevant to the examined environmental scenario. Input data demands from modellers rarely correspond to the input data offered by experimentalists. In spite of all that, we have the urge of getting the best from the available information for the most effective exploitation of existing knowledge in view of practical applications. Therefore, we should consider a complex scenario that includes several actors: firstly the “customers” (those who are formulating questions to which modellers should answer); secondly, the data/information suppliers (“scenario developers”); and thirdly, the modellers. In this respect, the present exercise is not simply aimed at evaluating the assessed models in order to rank the quality of their performances. In this work, we are concerned with the evaluation of potential advantages of a multi-modelling approach for the management of a complex environmental problem when input data are insufficient and non-univocal. In such conditions, the so-called “expert judgment” is an essential factor. The main question is: can a multi-model approach contribute to achieve an expert-based consensus concerning the prediction of the evolution of the considered environmental system? 3.3. Description of the Dnieper–Bug Estuary case study The Dnieper–Bug Estuary (DBE) (Fig. 1) is the largest of all the Black Sea Estuaries (surface area = 1006.3 km2, volume = 4.24 km3). The DBE water system consists of the Dnieper Estuary and the Bug Estuary. The length of the DBE is 63 km with a width of up to 15 km. The DBE is connected with the Black Sea through the Strait of Kirnburn. The average depth of the DBE is 4.4 m. There is a narrow, 10–12m deep channel suitable for shipping along the estuary to the Black Sea. The bottom is covered mainly by clay (50%) and sand. The main factor affecting the regime of the system is the process of mixing fresh river waters with saline marine waters. This forms the saline wedge in the estuary, which in the summer months can reach Kherson city. Stratification in the estuary ranges from almost none in the eastern part at the Dnieper mouth to a defined two-layer system in the western marine part of the DBE. These processes are highly dependent on season. The regime of this drowned-river estuary varies from stratified to partially mixed. The average discharge of the Dnieper ranges from about 400 to about 6000 m3s−1 in spring, whereas average discharge of the Southern Bug ranges from 80 to 1000m3s−1. The average mean water retention time in the estuary can be estimated of the order of 1 month. The Dnieper discharge, unlike the Southern Bug, is not simple seasonal because it is regulated from Kakhovka reservoir dam placed at 70km from Dnieper mouth. Therefore, in the summer the saline wedge penetrates much further into the estuary than in the spring. Also, the salinity of the upper stratum in the summer is much higher than in the spring. In addition to the fresh water input, other key factors governing the transport of contaminants are wind and Fig. 1 – Outline of the Dnieper–Bug Estuary. The boxes correspond to the sectors for which empirical radiological data were available: data relevant to West DBE (Table 5); data relevant to Southern Bug (Table 3); data relevant of East DBE (Table 6); data relevant to Dnieper mouth (Table 2); data relevant to Black Sea (Table 4). sea level variability. The estuary is ice covered in January to February and wind surges cause short-term excursions of salt wedge into the river mouths. All these factors force complicate 3D time-dependent stratified flows in DBE (Kostyanitsyn, 1964). The following input data were available for modellers: •Bathymetry of DBE and coastal area of the Black Sea were provided (2 km grid). •Daily water discharges of Dnieper River and S. Bug River in 1984–1987. •Sea level in Kinbourn Strait (Ochakiv) 1986–1987. •Daily sea level in Kinbourn Strait (Ochakiv) 1986–1987. •Daily temperature and salinity (ppt) at water surface. •Three-hour values of wind, air temperature, relative humidity and cloudiness (0–10). •Survey data in 1986 of temperature and salinity measurements. •Survey data in 1987 of temperature, salinity and velocity. Other parameters of DBE hydrology were not available: temperature in the Dnieper River and S. Bug River, temperature and salinity profiles in the Kinbourn Strait. Information was supplied on their seasonal changes. 3.4. Radionuclide data The Dnieper–Bug Estuary was contaminated by radionuclides introduced in the environment following the accident occurred at the Chernobyl nuclear power plant. The contamination was caused by direct deposition onto the estuary and by the radioactive substances transported by River Dnieper and, to a lesser extent, by River Bug. Whereas very much data relevant to the morphometry, the hydrology and the meteorological conditions of DBE were available, radiological input data were more difficult to find. Existing data were gathered by two different Ukrainian Institutes (Institute of Mathematical Machines and System Problems and Ukrainian Institute for Hydrometeorology) on the basis of experimental campaigns carried out by several laboratories (Kanivets et al., 1997; Katrich et al., 1993). It should be noticed that a significant effort of selection and evaluation of the empirical data were performed to assure the quality of these data sets. The input data were the time dependent radionuclide concentrations in Black Sea, in River Dnieper and in southern Bug outlet. Table 2 shows radionuclide concentrations in River Dnieper, the main tributary to the coastal area. Tables 3 and 4 show the concentrations of radionuclides in Southern Bug outlet and North west Black Sea (input data). Tables 5 and 6 show data of measured radionuclide concentrations in water of the West and East part of DBE (data for the assessment of the model performances). Abbreviations in the tables indicate the Institutes that collected samples and performed the radiological measurements: •IBSS, Institute for Biology of the Southern Seas, Sevastopol. •IEM, Institute of Experimental Meteorology (Typhoon), Obninsk. •MHI, Marine Hydrophysical Institute, Sevastopol. Table 2 – Fluxes of 137Cs and 90Sr from River Dnieper to the estuary Date Data gathered by institute of mathematical machines and system problems Data gathered by ukrainian institute for hydrometeorology 137Cs (dissolved) (Bqm−3)137Cs (suspended) (Bqm−3)90Sr (Bq m−3)137Cs (total) (Bq m−3)90Sr (Bq m−3) Pre-accident 3.0 23–33 May 1986 27.8 20.4 18.5 76.0 June 1986 12.2–14.1 14.8 37 4–11/16/7 26–92 July 1986 7.8 12.2 100 7.0 58–61 August 1986 7.4 12.2 78 September 1986 6.7 4.4 56 October 1986 7.0 4.1 63–89 November 1986 7.4 3.7 59–104 10.0 37–55 December 1986 7.8 3.7 89 January 1987 February 1987 14.1 48. March 1987 April 1987 3–11/5.2 660–470 May 1987 June 1987 5.6–19.2 3.7–6.7 40–278 July 1987 August 1987 September 1987 8.1–14.8 5.2 October 1987 November 1987 December 1987 300–330 Table 3 – Concentration of radionuclides in the Southern Bug outlet (data gathered by Ukrainian Institute for Hydrometeorology) Date 137Cs (Bq m−3)90Sr (Bq m−3) Pre-accident 3.5–7 (NOSS) 7–10.5 (NOSS) June 1986 16 (UCME) October 1986 10 (UCME) December 1987 3.5 (UCME) In parentheses the acronyms of the Institutes that carried out sampling and measurements. •UCME, Ukrainian Centre for Marine Ecology, Odessa. •NOSS, Nikolaevskaya Oblast Sanitary Station, Nikolaev. •CGO, Central Geophysical Observatory, Kiev. 3.5. The exercise The main goal of the exercise is to simulate the response of a multi-actor system comprised of modellers and data suppliers in the occasion of an environmental emergency. Therefore, the actors are urged to carry out a cooperative effort rather than hampered by a conflicting competition. Modellers were asked to supply predictions of both 137Cs and 90Sr in water on the basis of the supplied data and/or using further information from other sources they deemed useful and trustworthy. Among the input data not supplied by the “scenario developers”, modellers deemed important the depositions of radionuclides over the estuary. These data were obtained from the literature (De Cort et al., 1998) for 137Cs. Data of deposition of 90Sr of Chernobyl origin were estimated on the basis of the relative low mobility in atmosphere of this radionuclide compared with 137Cs (a negligible or very low 90Sr deposition was the common hypothesis of the modellers). 4. Discussion As previously stated many models for predicting the migration of radionuclides through large estuaries require the preliminary assessment of hydrological characteristics such as the water current field, the water temperature and salinity profiles, etc. Examples of results of such kind of models are reported in Figs. 2 and 3. Comparisons of model output Table 4 – Concentration of radionuclide in the North-west Black Sea (data gathered by Ukrainian Institute for Hydrometeorology) Date 137Cs (Bq m−3)90Sr (Bq m−3) Pre-accident 18 (IEM), 15 (IBSS) 22 (IEM), 18-20 (MHI) June 1986 55–80 (IEM), 125–185 (IBSS) 26–33 (IEM), 30–130 (IBSS) October 1986 110 (UCME) April 1987 33–140 (IBSS) 19–63 (IBSS) June 1987 52 (UCME) In parentheses the acronyms of the Institutes that carried out sampling and measurements. Table 5 – Concentration of radionuclides in West DBE (validation data; data gathered by Ukrainian Institute for Hydrometeorology) Date 137Cs (Bq m−3)90Sr (Bq m−3) Pre-accident 3.5–7 (NOSS) 7–10.5 (NOSS), 28±5 (MHI) June 1986 12–34 (UCME) 26–41 (IBSS) October 1986 14–18 (UCME) April 1987 14–30 (UCME) 220 (IBSS) December 1987 220 (IBSS) In parentheses the acronyms of the Institutes that carried out sampling and measurements. Table 6 – Concentration of radionuclides in the East DBE (validation data; the data were gathered by Ukrainian Institute for Hydrometeorology) Date 137Cs (Bq m−3)90Sr (Bq m−3) Pre-accident 3.5–7 (NOSS) 7–10.5 (NOSS), 28±5 (MHI) June 1986 12–19 (UCME) 26–41 (IBSS) October 1986 14–18 (UCME) April 1987 3–10 (UCME) 400 (IBSS) December 1987 260 (IBSS) In parentheses the acronyms of the Institutes that carried out sampling and measurements. and empirical data of radionuclide concentration in water are reported in Figs. 4 and 5. From the assessment of many models (Monte et al., 2005a) it seems quite obvious that a main factor of uncertainty is represented by the difficulties for predicting quantitatively the complex interaction with bottom sediment of the pollutant in water (contaminant diffusion from the water column to sediment, sedimentation and re-mobilisation). Nevertheless in the present exercise, due to the fast dynamic of the water within the estuary and the relatively low interaction of considered radionuclide with sediment in sea water, this difficulty did not significantly affect the model performances. Indeed, in Fig. 2 – A graphical output showing the results of the hydrological model developed by the University of Sevilla. spite of the different approaches, parameter values and algorithms used by the models (one of these has quite neglected the interaction of radionuclides with sediments), there are no significant differences in model performances that can be attributed to the methodologies employed for predicting the contaminant removal from the water column due to the mentioned processes. The modellers made use of different values of radionuclide deposition onto the estuary. An estimated value of 2000–5000 Bq m−2of 137Cs initial deposition onto the DBE was obtained from graphical data reported by De Cort et al. (1998). The differences among the output of the models at initial time reflect the uncertainty of the assumed 137Cs initial deposition. Unfortunately, empirical evaluations of 90Sr deposition are not available. Although several evidences indicate negligible contamination levels of the environment due to the initial fall-out of 90Sr in areas far from the Chernobyl power plant, one of the modellers used a cautionary, conservative value of 1000 Bq m−2. This gives reason of the predicted initial peak of 90Sr contamination in water (Fig. 5). Fig. 5 shows the results from an application of the ENEA model at a regional scale. The input data of this ”regional model” was the deposition of 90Sr onto the catchment of the Dnieper system around the Chernobyl power plant. The contribution of contaminant to the estuary from River Dnieper was evaluated by an application of model MARTE (see the ENEA model description in Appendix A) to the whole basin of the river. In spite of the larger horizon of this particular model application, the predictions of the time behaviour of radionuclide concentration in the estuary were within the range of the results obtained by the other models. This is a further evidence that the transport of pollutants characterised by weak adsorption on bottom sediment can be predicted with a sufficient accuracy although migration occurs over large distances. An important question is whether or not all the approaches used can be integrated in a single model. Although this is considered categorical in many traditional fields, in principle, it is not so obvious in case of environmental models. The perspectives, the horizons, the features, the structures of the models are often not fully consistent. The integration of these Fig. 3 – Temperature and salinity along DBE sections in summer simulated by model THREETOX. Fig. 4 – Comparison of the results of the model output with empirical data of 137Cs concentration in water. different models in a comprehensive “super-model” can be, therefore, hard to achieve. The examined models make use of significantly different values of similar parameters (Table 7). As sufficient information on the statistical distributions of the empirical values of these parameters are not available, the results of the models cannot be framed in a probabilistic perspective for instance in terms of mean values and confidence levels. As a matter of fact, it seems more proper to state that the output of each model is one of the possible outcomes that the community of experts deems worthy of commitment. Nothing is more instructive than a comparative graph, like Figs. 4 and 5, to show the information available for the “customers”. From the figures it is quite clear that, despite the abovementioned difficulties, there is an apparent consensus among modellers concerning the time behaviour of the radionuclide in water. Figures make intuitive what information obtained from the models can be perceived worthy of consensus and in which measure model output should be considered illustrative of the empirical outcomes. For instance, a delay of radionuclide concentration peak in water is predicted for 90Sr, whereas for 137Cs the concentration in water shows a clear decline on time. The range of variability of model output is comparable with the range of variability of the empirical concentrations. This clearly confirm that model performances reflect the intrinsic uncertainty of knowledge concerning the quantitative behaviour of the involved environmental process, the ambiguity of interpretation and parameterisation of such processes, the inherent variability of environmental quantities, etc. When the exercise was carried out, data of radionuclide concentration in the estuary compartments were available from the scientific literature. This occurrence made impossible a blind validation of the assessed models. Consequently, modellers were asked to run their models avoiding preliminary calibration. It is very difficult to appreciate how much the availability of contamination data influenced the model appliTable 7 – Selection of parameter values used by the models Parameter Model Value (m2s−1) Vertical mixing coefficient (surface-deep waters) CoastMab 10−6(order of magnitude) ENEA 1.9 ×10−7–1.9 ×10−6(depending on the distance from the tributary river) Horizontal diffusion coefficient U. Sevilla 0.58 Mixing coefficient sea water-coastal waters CoastMab 23 ENEA 116 Harms, I.H., 1997. Modelling the dispersion of 137Cs and 239Pu released from dumped waste in the Kara Sea. J. Marine Syst. 13, 1–19. Hornberger, G.M., Spear, R.C., 1981. An approach to the preliminary analysis of environmental systems. J. Environ. Manage. 12, 7–18. IAEA, 2000. Modelling of the transfer of radiocaesium from deposition to lake ecosystems. Report of the VAMP Aquatic Working Grou IAEA-TECDOC-1143. Vienna. Jimenez-Montealegre, R., Verdegem, M.C.J., van Dam, A., Verreth, J.A.J., 2002. Conceptualisation and validation of a dynamic model for the simulation of nitrogen transformations and fluxes in fish ponds. Ecol. Model. 147, 123–152. Jørgensen, S.E., 1979. Modelling the distribution and effect of toxic substances. Ecol. Model. 6, 181. Kanivets, V.V., Voitsekhovich, O.V., Simov, 1997. Radioactive contamination of the Black and Azov Seas. In: Voitsekhovich, O.V. (Ed.), Radioecology of Water Objects of the Chernobyl NPP Accident Impact Area. Chernobylinterinform, Kiev, pp. 127–151 (in Russian). Katrich, I.Yu., Nikitin, A.I., Medinets, V.I., Lepeshkin, V.I., Kabanov, A.I., Semko, N.N., Bazhanov, V.N., 1993. Dynamics of the radioactive contamination caused by CNPP accident on observed data 1986–1990. In: Borzilov, V.A., Kryshev, I.I. (Eds.), Ecological and Hydrophysical Consequences of the Nuclear Accidents. Hydrometerological Publ, Moscow, pp. 57–61 (in Russian). Koshebutsky, V., Maderich, V., Nesterov, O., Heling, R., 2004. Modelling of heat dispersion in inland waters and coastal areas of sea. Appl. Hydromech. 6 (78), No. 4 (in Russian). Kostyanitsyn, M.N., 1964. Hydrology of the Dnieper and Southern Bug Estuary. M. Gidrometeoizdat, 336 p. (in Russian). Koziy, L., Maderich, V., Margvelashvili, N., Zheleznyak, M., 1998. Three-dimensional model of radionuclide dispersion in the estuaries and shelf seas. J. Environ. Model. Software 13 (5–6), 413–420. Koziy, L., Maderich, V., Margvelashvili, N., Zheleznyak, M., 2000. Numerical modelling of seasonal dynamics and radionuclide transport in the Kara Sea. In: Oceanic fronts and Related Phenomena. Konstantin Fedorov Int. Memorial Symp., IOC Workshop Rep. Series, N 159, UNESCO’2000, pp. 296– 301. Krone R.B., 1962. Flume studies of the transport of sediment in estuarial processes. Final Report. Hydraulic Engineering Laboratory and Sanitary Engineering Research Laboratory, University of California, Berkeley, 120 p. Kryshev, I.I., Sazykina, T.G., 1995. Radiological consequences of radioactive contamination of the Kara and Barents Seas. J. Environ. Radioact. 29, 213–223. Margvelashvili, N., Maderich, V., Zheleznyak, M., 1997. THREETOX: a computer code to simulate three-dimensional dispersion of radionuclides in stratified water bodies. Radiat. Protect. Dosimetr. 73, 177–180. Margvelashvili, N., Maderich, V., Zheleznyak, M., 1999. Simulation of radionuclide fluxes from the Dnieper–Bug Estuary into the Black Sea. J. Environ. Radioact. 43, 157– 171. Margvelashvili, N., Maderich, V., Yuschenko, S., Zheleznyak, M., 2002. 3D modelling of the mud and radionuclide transport in Chernobyl cooling pond and Dnieper–Bug Estuary. In: Winterwerp, J.C., Kranenburg, C. (Eds.), Fine Sediments Dynamics in the Marine Environment Proceedings of INTERCOH-2000. Elsevier, pp. 595– 610. Matsunaga, T., Nagao, S., Ueno, T., Takeda, S., Amano, Hikaru, Tkachenko, Y., 2004. Association of dissolved radionuclides released by the Chernobyl accident with colloidal materials in surface water. Appl. Geochem. 19, 1581–1599. Monte, L., Salvatori, L., Spezzano, P., 1995. Modelling radionuclide behaviour in deep lakes of the Italian Alpine region: seasonality effects and comparison with deep volcanic lakes of Central Italy. Technical report ENEA, RT/AMB/95/12, Rome. Monte, L., 2001. A generic model for assessing the effects of countermeasures to reduce the radionuclide contamination levels in abiotic components of fresh water systems and complex catchments. Environ. Model. Software 16, 669– 690. Monte, L., Brittain, J.E., H˚ akanson, L., Heling, R., Smith, J.T., Zheleznyak, M., 2003. Review and assessment of models used to predict the fate of radionuclides in lakes. J. Environ. Radioact. 69, 177–205. Monte, L., Brittain, J.E., H˚ akanson, L., Smith, T.J., van der Perk, M., 2004. Review and assessment of models for predicting the migration of radionuclides from catchments. J. Environ. Radioact. 75, 83–103. Monte, L., Boyer, P., Brittain, J.E., H˚ akanson, L., Lepicard, S., Smith, J.T., 2005a. Review and assessment of models for predicting the migration of radionuclides through rivers. J. Environ. Radioact. 79, 273–296. Monte, L., Desideri, D., Grimani, C., Angeli, G., 2005b. Modelling the long-term behaviour of radiocaesium and radiostrontium in two Italian lakes. J. Environ. Radioact. 80, 105–123. Mossman, D.J., AlMulki, N., 1996. One-dimensional unsteady flow and unsteady pesticide transport in a reservoir. Ecol. Model. 89, 259–267. Partheniades, E., 1965. Erosion and deposition of cohesive soil. J. Hydr. Div. ASCE. 91, 105–139. Peria ˜ nez, R., 2002. Modelling the suspended matter dynamics in a marine environment using a three-dimensional s coordinate models: application to the eastern Irish Sea. Appl. Math. Model. 26, 583–601. Peria ˜ nez, R., 2004. Testing the behaviour of different kinetic models for up-take/release of radionuclides between water and sediments when implemented in a marine dispersion model. J. Environ. Radioact. 71, 243– 259. Peria ˜ nez, R., Abril, J.M., Garc´ ıa-Le´ on, M., 1996. Modelling the dispersion of non-conservative radionuclides in tidal waters. Part 1: conceptual and mathematical model. J. Environ. Radioact. 31, 127–141. Peters, R.H., 1986. The role of prediction in limnology. Limnol. Oceanogr. 31, 1143–1159. Polikarpov, G.G., Timoschuk, V.I., Kulebakina, L.G., 1988. Concentration of 90Sr in the aquatic environment of lower Dnieper toward the Black Sea. In: Dopovidi (Proceedings) of National Academy of Sciences of Ukraine, ser. B, N3, pp. 75–76. Polikarpov, G.G., Livingston, H.D., Kulebakina, L.G., Buesseler, K.O., Stokozov, N.A., Casso, S.A., 1992. Inflow of Chernobyl 90Sr to the Black Sea from the Dniepr River. J. Estuarine, Coastal Shelf Sci., V 34, 315–320. Prandle, D., 1984. A Modellino Study of the Mixing of 137Cs in the seas of the European continental shelf, vol. A310. Philosophical Transactions of the Royal Society, London, pp. 407–436. Rajar, R., Zagar, D., Cetina, M., Akagi, H., Yano, S., Tomiyasu, T., Horvat, M., 2004. Application of three-dimensional mercury cycling model to coastal seas. Ecol. Model. 171, 139– 155. Salomon, J.C., Breton, M., Guegueniat, P., 1995. A 2D long term advection dispersion model for the Channel and southern North Sea. Part B. Transit time and transfer function from Cap de La Hague. J. Marine Syst. 6, 515–527. Schonfeld, W., 1995. Numerical simulation of the dispersion of artificial radionuclides in the English Channel and the North Sea. J. Marine Syst. 6, 529–544. Smith, C.N., Goshawk, J.A., Charles, K., McDonald, P., Leonard, K., McCubbin, D., 2003. Mead (Part 2), predictions of radioactivity concentrations in the Irish Sea. J. Environ. Radioact. 68, 193–241. Smith, J.T., Comans, R.N.J., 1996. Modelling the diffusive transport and remobilisation of 137Cs in sediments: the effects of sorption kinetics and reversibility. Geochimica et Cosmochimica Acta 60, 995–1004. Van Rijn, L.C., 1984. Sediment transport. Part II. Suspended load transport. J. Hyd. Engr. 110, 1613–1641.