A fuzzy classification of the hydrodynamic forcings of the Rhone River plume: An application in case of accidental release of radionuclides

Highlights

• Analyze of 10 years records of discharge and wind conditions at Rhone River mouth.

• Fuzzy clustering performed to identify six different hydroclimatological trends.

• These trends induce different spreading of the Rhone river plume and contaminants.

• Plume shapes identification allows faster decision in case of accidental releases.

Abstract

Assessing and modelling the coastal plume dispersion of nuclearized rivers is strategic in case of accidental releases, but taking into account the variation of main hydrodynamic forcings is challenging. This study uses fuzzy c-mean clustering of a 10 years series of discharge and wind speed at the Rhone River estuary (France) in order to explain the variability of its plume. The method allows to classify the data into 6 scenarios of hydrodynamic forcings that were related to different spatial extensions of the plume, as well as to surface currents measured in-situ. These scenarios were used to simulate the extension and dilution of a radioactive release issued from the river. Based on threshold values of the forcings, a decisional tree is proposed to provide a quick decision tool identifying, in real time, which climatological scenario occurs at the river mouth and the potential plume pattern.

Introduction

The Rhone River catchment extends over 98,000 km² and covers one fifth of the French metropolitan territory. It is the main source of particles and freshwater for the Gulf of Lion in the North Western Mediterranean sea (Durrieu De Madron et al., 2000), and all together one of the most important input to the Mediterranean sea (Ludwig et al., 2009). The Rhone valley also hosts the largest concentration of nuclear power plants in Europe with 4 nuclear power plants in process and a spent fuel reprocessing center, under dismantlement since 1997. Eyrolle et al. (2020) recently synthesized the studies showing that this river carries artificial radionuclides from decades, resulting from authorized releases of low level radioactive liquid wastes and from the export of atmospheric deposits on watersheds consequently to nuclear weapons testing and Chernobyl accident.

France is presently ranked second in the world for the production of nuclear energy, and the total electricity production in the combined regions of Northern, Western and Southern Europe is projected to increase by 2050 ((IAEA, 2019)). Also, the risk of incident on any kind of nuclear installations is still of concern in France and must be taken into account. As for any river, the transport of artificial radionuclides in case of accidental release occurs both in dissolved and particulate form, depending on the amount of suspended particulate matter and on the chemical properties of the radionuclides, and particularly their distribution coefficient (Tomczak et al., 2019). For the Rhone River, the prediction of dissolved vs particulate fluxes and the associated time scale for transit can be evaluated through numerical modeling (Launay et al., 2019), but the behavior of radionuclides once at sea is clearly less constrained, because it will primarily depend on the forcings governing the shape of the Rhone River plume.

The area of the Rhone river mouth is characterized by a very small tidal amplitude about 30 cm inducing the formation of a sedimentary delta. As usual in this case, the freshwater input forms a thin stratified plume of low salinity water (and higher turbidity) overlying the seawater and extending between 4 and 1000 km² (Estournel et al., 1997; Gangloff et al., 2017) with a thickness decreasing seaward (Pairaud et al., 2011; Gangloff et al., 2017). It is preferentially deflected westward in a clockwise orientation running East to West (Reffray et al., 2004) due to the general circulation induced by the Northern Current along the continental slope. Under north-northwest winds, the plume extends offshore towards the southwest, whereas it is pushed to the coast west of the river inlet in case of southeastern winds. Satellite and modelling results have also shown that the plume size increases with river discharge (Fraysse et al., 2014; Gangloff et al., 2017). More episodic processes impact the plume pattern such as dense water formation and cascading (Ulses et al., 2008), upwelling cells and marine storms (Millot, 1990, 1999). As a result, this plume extends far beyond the coastal areas and may covers a large area in the GoL extending from the vicinity of the mouth up to the Cap de Creus at the french-spanish border (Sanchez-Cabeza et al., 1992). It can also reach the Gulf of Fos (Gontier et al., 1992; Charmasson et al., 1999) or the Bay of Marseille (Pairaud et al., 2011; Fraysse et al., 2014) on the eastern side of river mouth. The Gulf of Fos is an important economic area with one of the biggest commercial port in Europe and a large shellfish area, and Marseille is one of the biggest Mediterranean coastal city with one million inhabitants. Due to the oligotrophic nature of the Mediterranean Sea, the region of freshwater influence (ROFI) of the Rhone river has a major influence on the distribution of plankton groups (Diaz et al., 2019) and thus pelagic catches on the GoL. Obviously, the inputs of chemical contaminants from the Rhone River can greatly affect the fishery activity.

The combination of meteorological and hydrodynamic forcings with the dynamics of the Rhone River discharge results in a large spatio-temporal variability in freshwater and associated pollutants delivery to the GoL (Martin et al., 2019). If an accidental release occurs in the Rhone River, the dissolved radionuclides may reach the estuary within 48 h h to few days depending on the source location and water discharge [unpublished results]. Once at sea, the different shapes that the plume may present will depend on hydrodynamic and weather conditions and will lead to contaminate different areas. Since one goal of radioprotection is to predict the transfer of radionuclides in the environment, there is a need to anticipate their dispersion at any time and in any kind of meteorological and hydrodynamical conditions.

Different numerical hydrodynamic models have been set up in the GoL, including the river mouth (Pairaud et al., 2011; Duffa et al., 2016), and they could be used actually in case of accidental release in order to predict the behavior of the freshwater input. However, the delay necessary for their implementation will range from few hours to few days, whereas very quick and concise information should be provided to experts and decision-makers as a first picture of the local issues. Alsothe potentially impacted zones will be better defined by performing a fine spatial scale simulation adequately centered, compared to a large-scale simulation.

As a result, a preliminary study embraces all possible plume patterns is necessary, and a first step is to target the general behavior of the estuarine-plume system. Bárcena et al. (2015) explained that two approaches may be conducted for that: simulating several scenarios using constant conditions of hydrodynamic forcings or simulating few scenarios using the most frequent or extreme real hydrodynamic forcings during short-medium term periods (month to year).

These authors demonstrated that the first approach is not complete because real forcings cannot be deduced from the combination of simple idealized scenarios. The second approach relies on a subjective selection of scenarios by an expert and it will have an expensive computational cost for simulations if the need is to get on overview of the different kind of realistic responses of the estuarine-plume mean behavior. In this case, and to minimize subjectivity, a methodology based on data mining should be able to select the most relevant condensed hydrodynamic scenarios, taking into account the time evolution and the occurrence probability of the forcings.

Plume classifications based on satellite observations or hydrodynamic model output have been defined in several river-sea systems using Empirical Orthogonal Function or Self-Organizing-Map (Falcieri et al., 2014; Xu et al., 2019). Such classification method deals with large spatial scale but implies a heavy data pretreatment like « masking » to treat the satellite data or for the computation of the model. In addition, the need of long-term environmental databases (e.g., 10–20 years) to assess probabilities implies significant computational costs as well as long and multiple series of data to be used as boundary conditions and climatic forcings. Another approach is to classify the main hydrodynamics drivers by looking for example at the catchment discharge and the winds intensities and directions (Kaufmann and Whiteman 1999; Zhang et al., 2011). Since the plume response to these forcings can be longer than 24h (Demarcq and Wald, 1984; Estournel et al., 1997), the classification should work observation by observation but must also keep consistency over longer temporal scales of few days in order to be accurate. Clustering performed on temporal series helps to assess the consistency of a trend over time, and a fuzzy clustering algorithm provides a continuous cluster membership function allowing to spot significant trend changes.

In this context, this paper presents a methodology based on statistical analysis and numerical modelling that was developed to address the limitations of the previously mentioned approaches. Firstly, we used a fuzzy c-mean algorithm to identify and classify combinations of winds and discharge at the mouth of the Rhone river in order to define “model scenarios” of realistic forcings. Secondly, the consequences for sea surface currents will be assessed and the resulting plume pattern will be modelled for each scenario, as well as the distribution of dissolved radionuclides due to a hypothetical and episodic release on the Rhone River. These plumes scenario can be used as a support for operational tools improvement and decision.