Method and system for forecasting groundwater levels with rapid kinetics, in order to anticipate the occurrence of flooding due to rising groundwater levels.

Abstract

A method and system for forecasting rapid groundwater levels using artificial intelligence models, in order to anticipate the occurrence of flooding due to rising groundwater levels. The invention essentially consists of a method for forecasting the levels of one or more rapid groundwater aquifers in a defined geographical area, which combines the use of a pre-existing semi-physical forecasting model according to patent FR3082939B1, with two artificial intelligence models at different time steps and forecast horizons (Hz1, Hz2). Figure for the abstract: Fig. 1

Description

Title of the invention: Method and system for predicting groundwater levels with rapid kinetics, in order to anticipate the occurrence of flooding due to rising groundwater levels. technical field

[0001] The present invention relates to the field of forecasting groundwater levels.

[0002] It relates more specifically to rapidly rising groundwater, with a view to anticipating the occurrence of flooding in areas sensitive to the phenomenon known as rising groundwater.

[0003] The risk of flooding due to rising groundwater is present in areas where groundwater is highly responsive to rainfall (in terms of timing and amplitude of groundwater level fluctuations). In these areas, groundwater levels can rise by several tens of meters in less than 24 hours. This type of phenomenon is therefore very rapid and can lead to the flooding of underground structures or exert uplift pressure on the foundations of certain structures.

[0004] The invention therefore aims to anticipate the occurrence of this type of phenomenon by proposing a forecast of groundwater levels in the short or even very short term on a local geographical scale.

[0005] More specifically, the invention aims to be able to predict more accurately the levels reached by a water table within a horizon of 6 to 24 hours and thus to improve decision-making (alert and crisis management) in the event of a suspected or ongoing risk. Previous technique

[0006] In the field of forecasting groundwater rise, methods have already been proposed.

[0007] These methods can be classified into two categories, one based on models based on historical weather data, the other based on models based on the development of forecast scenarios.

[0008] In the first category, reference may be made to patent application US2005 / 038763 which proposes a method for forecasting groundwater levels, based on the history of observations of meteorological conditions and groundwater levels, and a statistical processing of these historical data.

[0009] This method consists of making forecasts based on statistical trends defined on the basis of past data. The forecasts in question are long-term, typically from several months to several years, in a context of change climatic but not suitable for very short-term forecasts, from a few hours to a few days.

[0010] Thus, this method is suitable for aquifers with slow kinetics, such as alluvial aquifers, but is not suitable for aquifers with rapid kinetics, such as karst aquifers.

[0011] With regard to the second category, publication [1] describes a decision support tool which aims to predict groundwater levels and river flows across the French territory, using hypothetical meteorological scenarios.

[0012] Here again, the forecasts in question are long-term, typically several months, and concern slow-moving aquifers, but are not suitable for fast-moving aquifers.

[0013] The applicant has proposed in patent FR3082939B1 a method and system for forecasting groundwater levels in rapidly evolving aquifers (exhibiting strong responsiveness to precipitation), enabling short-term forecasts, typically from 1 to 4 or 5 days, and moreover at a local scale. This method implements a semi-physical numerical simulation model, known as the "rainfall-level model" as described in [2], which is manually calibrated beforehand by an operator using historical data. Since the calibration parameters are obtained iteratively, they may differ from one operator to another, leading to significantly different forecasts, and may not meet optimal requirements.

[0014] Furthermore, the semi-physical model does not accurately represent intermediate levels, i.e., average / high water levels, which are useful indicators for forecasting imminent flood levels. In certain situations, these uncertainties can lead to inappropriate decision-making resulting from an underestimation or overestimation of the actual observed flood risk.

[0015] Many other forecasting solutions for fast-kinetic sheets have been put forward in the literature, particularly in patents.

[0016] Among these, we can cite US7254564B1, CN117172131A, and CN114022305, which relate to groundwater level forecasting solutions that implement artificial neural network models. The major drawback of these solutions is the lack of precise demonstrations and details on the type of aquifer, its kinetics, the time steps and forecast horizons concerned, which makes it impossible to estimate the applicability of these methods to a complex aquifer subject to rapid groundwater rise, such as karst or fractured aquifers.

[0017] Moreover, some of the solutions according to the patents are applied to data exhibiting very different dynamics (levels in mines, pipelines or flow rate at the (source). The models are therefore not directly transposable to the groundwater levels of a complex aquifer.

[0018] There is therefore a need to improve the forecasting of groundwater levels in an aquifer, even a complex one, for horizons of 24 hours or less, by eliminating the bias of the human operator who calibrates the model and, as far as possible, the rainfall forecasts which are often the main source of uncertainty or error, in order to better constrain the measures to be taken in the event of a crisis, while having sufficient time to prevent and allow the intervention of competent personnel.

[0019] The aim of the invention is to meet at least part of this need. Description of the invention

[0020] To this end, the invention relates, in one of its aspects, to a method for forecasting the levels of one or more rapidly evolving groundwater aquifers in a defined geographical area, comprising the following steps:

[0021] a / real-time acquisition of piezometric level(s) data within the determined area;

[0022] b / real-time acquisition of meteorological data by means of one or more meteorological stations within the determined area;

[0023] c / calibration of a numerical simulation model, called "rain-level model", adapted to groundwater with rapid kinetics, to determine groundwater levels in the determined area, from the data acquired in steps a / and b / ;

[0024] d / carrying out a numerical simulation of the rainfall-level model calibrated according to step c / , so as to obtain a forecast of the level(s) of the groundwater(s) in the determined area at a given time horizon (Hzo);

[0025] e / comparison of the groundwater level(s) (H0) with an alert threshold (S), and

[0026] - if the alert threshold is not exceeded in step d, then repeat steps d to d / ;

[0027] - if the alert threshold is exceeded at step e / , then:

[0028] fl / reiteration of steps d and b / with a time step kl;

[0029] f2 / design of a model of a first artificial neural network (ANN) to starting from the data acquired at step fl / with the time step kl;

[0030] f3 / implementation of a numerical simulation of the first neural network model artificial according to step f2 / , so as to obtain a forecast of the level(s) of the aquifer(s) in the determined area at a given time horizon (Hz1) lower than the given time horizon (Hzo) of step d / ;

[0031] f4 / comparison of the groundwater level(s) (Hl) with an alert threshold (S), and

[0032] - if the alert threshold is not exceeded at step f4 / , then repeat steps a / to d / ;

[0033] - if the alert threshold is exceeded at step f4 / , then:

[0034] g 1 / reiteration of steps a / and b / with a time step k2;

[0035] g2 / design of a model of a second artificial neural network (ANN) to starting from the data acquired in step g 1 / with the time step k2;

[0036] g3 / implementation of a numerical simulation of the second network model of artificial neurons according to step g2 / , so as to obtain a prediction of the level(s) of the aquifer(s) in the determined area at a given time horizon (Hz2) lower than the given time horizon (Hz1) of step f3 / ;

[0037] g4 / comparison of the groundwater level(s) (H2) with an alert threshold (S), and

[0038] - if the alert threshold is not exceeded at step g4 / , then repeat steps a / to d / ;

[0039] - if the alert threshold is exceeded at step g4 / , then an alarm is triggered and / or the implementation of active pumping methods.

[0040] By "calibration", we mean the usual technological meaning, namely an adjustment of the numerical values ​​attributed to the parameters of the rainfall-level model, so that the calculated values ​​of a groundwater level are as close as possible to the observed values ​​of that level at a time t.

[0041] The invention therefore essentially consists of a method for forecasting the levels of one or more groundwater aquifers with rapid kinetics in a determined geographical area which combines the use of a semi-physical (pre-existing) forecasting model according to patent FR3082939B1, with two artificial intelligence models at different time steps and forecast horizons (Hz1, Hz2).

[0042] The advantages of the solution are numerous, among which we can mention:

[0043] - to overcome the uncertainties associated with the current semi-physical model retained in patent FR3082939B1, in particular related to the choice of parameters during manual calibration which can differ significantly depending on the operator and lead to a bias in the calculated outputs;

[0044] - improve the accuracy of the levels obtained for each forecast horizon by using two artificial neural network models, previously optimized on the entire past database, with an objective criterion, and for each target horizon;

[0045] - benefit from a short-term forecast (Hz2) which does not use the forecast of rain, which is generally the main source of uncertainty for this type of forecast, nor any hypothetical scenario about rainfall, such as the assumption of constant rain.

[0046] The groundwater levels obtained by the process according to the invention are thus more precise and more robust than the prior art: they make it possible to improve decision-making in the event of a supposed or actual risk of rising groundwater.

[0047] In other words, thanks to the invention, it is possible to predict more precisely the levels reached by a water table over a horizon of 6 to 24 hours and thus improve the decision-making for alert and / or crisis management, in the event of an identified or ongoing risk.

[0048] In the end, the invention makes it possible to anticipate as best as possible a risk of flooding by upwelling of groundwater of the "flash" type.

[0049] Advantageously, the first and / or second artificial neural network is / are a multilayer perceptron. This type of model exhibits universal approximation and parsimony properties that are advantageous within the scope of the invention.

[0050] Advantageously still, step f2 / is carried out with a rain forecast, step g2 / being carried out without a rain forecast.

[0051] Preferably, the actual time of steps a / and b / is less than 1 hour, preferably still between 10 and 30 minutes.

[0052] Preferably, the time step kl is equal to 24 hours.

[0053] Preferably, the time horizon kl is between 12 and 24h.

[0054] Preferably, the time step k2 is equal to 1 hour.

[0055] Preferably, the time horizon k2 is between 6 and 12 hours.

[0056] According to an advantageous embodiment, the piezometric and meteorological data are sent by teletransmission to a central server in which the numerical models are hosted.

[0057] The invention also relates to a system for implementing the method just described, comprising:

[0058] - a central server hosting the data and digital and network models of artificial neurons (RNA);

[0059] - one or more piezometers equipped with a pressure sensor in the area geographically defined;

[0060] - one or more meteorological stations arranged in the geographical area determined;

[0061] in which the piezometer(s) and the meteorological station(s) are connected to the central server in order to transmit their measurement data in real time.

[0062] According to an advantageous variant, the piezometer(s) and the meteorological station(s) are equipped with teletransmission transmitters to transmit the data in real time to the central server.

[0063] According to an advantageous embodiment, the system includes an alarm and / or active pumping means, the control unit of which is connected to the central server.

[0064] The invention also relates to the use of the method described above and / or the system described above, for operational decision support in crisis situations related to a rapid rise in groundwater levels, particularly in karst and fractured aquifers.

[0065] Other advantages and features of the invention will become clearer from the detailed description of examples of implementation of the invention given by way of illustration and not limitation with reference to the following figures. Brief description of the drawings

[0066] [Fig-1] [Fig.1] is a flowchart of the groundwater forecasting method according to the invention.

[0067] [Fig.2] [Fig.2] illustrates the architecture of a multilayer perceptron implemented as an artificial neural network, in the method according to the invention.

[0068] [Fig.3] [Fig.3] illustrates in the form of curves an example of chronicles respectively of measured rainfall heights and groundwater levels as well as the groundwater levels obtained by simulation according to the invention and comparatively according to patent FR3082939B1, with identical 24h rainfall forecasts.

[0069] [Fig.4] [Fig.4] illustrates in the form of curves an example of chronicles respectively of measured rainfall heights and groundwater levels as well as the groundwater levels obtained by the simulation according to the invention, at an hourly time step with 6 hours of anticipation. Detailed description

[0070] Fig. 1 illustrates the sequence of different steps implemented in the method for forecasting groundwater levels in a local area, according to the invention.

[0071] Groundwater levels are measured by so-called reference piezometers (boreholes that conform to and are representative of the hydrodynamic behavior of the monitored aquifer) installed in areas considered most susceptible to flooding due to rising groundwater. These piezometers are equipped with a pressure sensor that records variations in groundwater levels (at a time interval adapted to the aquifer's responsiveness, typically 30 minutes) and a remote transmission station that allows for real-time recording and transmission of the measurements.

[0072] Furthermore, one or more local weather stations make it possible to record in real time and at 10-minute time steps certain parameters such as the amount of precipitation, the ground temperature and the relative meteorological humidity which are necessary for the implementation of the method.

[0073] The first step of the method therefore consists of acquiring meteorological and piezometric measurement data from the various aforementioned equipment, to apply a treatment (homogenization and correction), and to store them in the system to then feed the rainfall-level model.

[0074] The rainfall-level model chosen is that of a model adapted to groundwater with rapid kinetics.

[0075] The inventors chose the model described in publication [2].

[0076] This rainfall / level model makes it possible to reconstruct the rapid kinetic groundwater levels from meteorological data.

[0077] The purpose of this step is to determine the calibration parameters that allow the piezometric data measured from the meteorological data to be reconstructed as accurately as possible.

[0078] The calibration of the rainfall / level model is carried out manually by the operator, by iterations, as described in [2].

[0079] A comparison of H0 is then performed with at least one alert threshold S. This threshold can be determined based on several parameters intrinsic to the local area, such as the elevation of foundations, buried structures, the location and availability of pumping / drainage devices, ...

[0080] Preferably, this comparison step is carried out from the highest of the predicted groundwater levels.

[0081] In this operation, a 24-hour forecast is established every day at midnight, up to three days in advance (prediction horizon Hz0), taking into account rainfall forecasts.

[0082] However, no information is provided within the 24-hour interval. As the approach of the rainfall event draws near, it is crucial for the risk manager to have the most reliable and accurate information possible. Therefore, the fact that forecasts can only be updated once a day is a significant limitation.

[0083] Also, in order to improve the performance of the previous model, the inventors have further implemented two complementary models applied cumulatively and sequentially in the event that the threshold S is exceeded by the value H0:

[0084] - a first artificial neural network model, called the IAJ model, which uses daily piezometric and meteorological measurements to calculate a forecast over a Hzl horizon of between 12 and 24h, with Hzl < Hz0. Unlike the aforementioned rainfall-level model, the training of the first model by artificial neural network is carried out on the entire database, except for a test set, and with the minimization of the mean squared error as the objective criterion.

[0085] - a second model using an artificial neural network with an hourly time step, called the IAH model, which improves the temporal resolution of forecasts. This second artificial neural network model, defined with a forecast horizon of Hz2 between 6am and 12pm, with Hz2 <Hzl < HzO, sans prévision de pluie, supprime également de fait l’incertitude sur les prévisions de pluie qui est la principale incertitude affectant les hauteurs d’eau.

[0086] Finally, as illustrated in [Fig.1], if the threshold is exceeded by the value H2 of the second model, an alert can be sent.

[0087] The cumulative and sequential application of the two IAJ and IAH models makes it possible, in particular, to identify the variables relevant for the forecast of groundwater(s).

[0088] Thus, the humidity of a water basin is represented prior to the triggering of the rainfall event using information on previous rainfall, and the process state variables using previous water levels. The model's complexity is minimized through the application of cross-validation. Generalization performance is evaluated using the cross-validation score and the graphical representation of water levels over time, visualized on the test set, which is the most severe flood event in the database.

[0089] For both the IAJ and IAH models, the inventors chose the multilayer perceptron (MLP), which is a type of artificial neural network organized into several layers, all of whose inputs are connected to each of the neurons in a hidden layer of neurons up to the output neuron. Information flows only from the input layer to the output layer: it is therefore a forward-propagating network. The neurons in the outermost layer are the outputs of the overall system.

[0090] The architecture of a multilayer perceptron, as implemented for the two IAJ and IAH models, is schematically represented in [Fig. 2]. In this [Fig. 2], neurons are represented by circles and input variables by rectangles. In this diagram, uk is an input vector corresponding to exogenous variables, i.e., observed meteorological data (precipitation, temperatures, etc.); yp corresponds to the output variable, here groundwater levels; w and r are the sliding time windows of the time series; k corresponds to discrete time; and hp is the forecast horizon.

[0091] The inventors judiciously chose this model within the framework of the advantage because it exhibits universal approximation and parsimony properties that are particularly advantageous within the framework of the invention.

[0092] The universal approximation is the ability of neural networks to approximate any differentiable bounded-valued nonlinear function: [3]. This property therefore allows networks to identify, and thus model, nonlinear processes such as the hydrodynamic behavior of complex aquifers.

[0093] The parsimony property means that neural models, which depend non-linearly on their parameters, are more parsimonious than an approximator whose output signal depends linearly on the latter: [4]. In the method according to the invention, the models are directed, that is to say, the observed variables of the simulated process are used as input: a state noise assumption is made.

[0094] The methods for selecting input variables and hyperparameters of network structures, as well as the learning, generalization and regularization methods used, have already been presented in several studies: [5],[6].

[0095] Thus, once the neural network models are designed at the horizons Hz1, with rain forecast, and Hz2, without rain forecast, with the adapted time steps, they are implemented cumulatively in the aforementioned rain-level model and sequentially.

[0096] The improvements of the invention can be illustrated by means of indicator tables and two graphs as follows: - [Fig.3] presents the forecast made by the rainfall-level model as per patent FR3082939B1 for the entire wet season of 2019, as well as the forecast of the IAJ model; - [Fig.4] presents a time zoom on the result of the IAH model forecast on the most intense event of October-November 2019 (without rain forecast), together with the forecast made according to patent FR3082939B1 at a time step of 24h with the rain forecast.

[0097] From [Fig.3], it appears that the prediction of the IAJ model is significantly better than that according to patent FR3082939B1 in several respects:

[0098] - the rise of the flood is better reproduced by IAJ: it is initially slightly delayed on the observed ascent, then crosses it and becomes slightly ahead.

[0099] - the maxima of the peaks are very well represented by IAJ with 2.7 m error at first peak (4.7 m according to patent FR3082939B1) for a rise in the water table of 37 m, i.e. 7% error (respectively 13%).

[0100] - recessions are very well represented by IAJ, unlike that according to the patent FR3082939B1.

[0101] On the other hand, the model according to patent FR3082939B1 tends to show peaks that are a little more synchronous with those of the observed value than IAJ.

[0102] Figure 4 illustrates the results of the hourly forecast model at Hz2, without rainfall forecast, for the rise of the first peak of autumn 2019. It is clear from Figure 4 that the HAI forecast at the hourly time step and at the Hz2 forecast horizon is very accurate for the most significant rise of the 2019 wet season. This forecast therefore provides highly relevant insights for the manager of critical facilities at sub-daily timescales. It should be noted here that the days The chosen data, although dating from November 2019, are particularly significant because during this period, an exceptional rainfall event occurred in the local area, resulting in the highest rise in groundwater levels ever observed in said area.

[0103] Furthermore, the numerical indicators allow for an assessment of the quality of the forecast across the entire database. These indicators are listed in Tables 1 and 2 below.

[0104] [Table 1]: Nash Model Nash Peak Persistence Persistence Peak AHI, 6h 0.996 0.98 0.72 0.75 AHI, 24h 0.97 0.91 0.71 0.76

[0105] Table 1 shows the persistence and Nash (coefficient of determination) scores calculated on the entire database using cross-validation. These scores represent the model's generalization performance to unknown data. The closer these scores are to 1, the better they are. Therefore, the scores obtained by the AHI and JIA models are very good.

[0106] [Table 2]: Nash Model Nash Peak Persistence Persistence Peak AHI, 6h 0.999 0.991 0.840 0.866 IAJ, 24h 0.97 0.89 0.51 0.49 According to patent FR3082939B1 0.62 - -22.98 -

[0107] Table 1 shows the scores for the test event represented in [Fig. 3]. It should be noted that the AHI and LHI scores are also very good and of the same order of magnitude as the cross-validation scores; this means that the performance shown in [Fig. 4] can be generalized to the rest of the database. It should also be noted in the last row that the scores according to patent FR3082939B1 are lower than those of the AI ​​models implemented in the invention.

[0108] Thus, the solution according to patent FR3082939B1, which makes it possible to anticipate the occurrence of a damaging rainfall event for sensitive installations up to 3 days in advance, thanks to rainfall forecasts, is significantly improved by the invention. This improvement is based in particular on the use of two new multilayer perceptron models, IAJ and IAH, which offer the advantage of making more accurate piezometric level forecasts over horizons of between 6 and 24 hours, while the IAH model does not rely on rainfall forecasts.

[0109] The invention is not limited to the examples just described; in particular, features of the illustrated examples can be combined in unillustrated variants.

[0110] Other variants and embodiments may be envisaged without departing from the scope of the invention. List of cited references [YES] [1]: H. Bessière, B. Mougin, Y. Vigier, J. Nicolas, S. Loigerot, 2017. “MétéEau des nappes: a decision making tool to characterize in almost real-time groundwater quantitative State”. Revue “Géologues” number 195 - December 2017;

[0112] [2]: K. Najib, H. Luge, S. Pistre, 2008. “A methodology for extreme groundwater surge predetermination in carbonate aquifers: Groundwater flood frequency analysis”.

[0113] [3]: Hornik K, Stinchcombe M, White H (1989) “Multilayer feedforward networks are universal approxûnators”. Neural Networks 2:359-366. https: / / doi.org / 10.10 16 / 0893-6080(89)90020-8

[0114] [4]: ​​Barron AR (1993) "Universal approximation bounds for superpositions of a sigmoidal function”. IEEE Transactions on Information Theory 39:930-945. https: / / doi.org / 10.1109 / 18.256500

[0115] [5]: Kong A Siou, Line. Modeling of floods in karst basins using networks of neurons. Case study of the Lez basin (Hérault). Montpellier: 2011. University of Montpellier 2: doctoral thesis, Continental Waters and Society, supervised by Séverin Pistre. https: / / ged.scdi-montpellier.fr / florabium45 / / jsp / nnt.jsp?nnt=2011M0N 20070

[0116] [6]: Kong A Siou, L., Johannet, A., Valérie, BE et al. “Optimization of the generalization capability for rainfall-runoff modeling by neural networks: the case of the Lez aquifer (southern France')”. Environ Earth Sci 65, 2365-2375 (2012).

Claims

1. Demands A method for forecasting the levels of one or more rapidly evolving groundwater aquifers in a defined geographical area, comprising the following steps: a / real-time acquisition of piezometric level(s) data within the determined area; b / real-time acquisition of meteorological data using one or more meteorological stations within the determined area; c / calibration of a numerical simulation model, called "rainfall-level model", adapted to groundwater with rapid kinetics, to determine groundwater levels in the determined area, from the data acquired in steps a / and b / ; d / carrying out a numerical simulation of the rainfall-level model calibrated according to step c / , so as to obtain a forecast of the level(s) of the groundwater(s) in the determined area at a given time horizon (Hzo); e / comparison of the groundwater level(s) (HO) with an alert threshold (S), and - if no exceedance of the alert threshold at step d, then repeat steps a / to d / ; - if the alert threshold is exceeded at step e / , then: fl / reiteration of steps a / and b / with a time step kl; f2 / design of a model of a first artificial neural network (ANN) from the data acquired in step fl / with the time step kl; f3 / carrying out a numerical simulation of the model of the first artificial neural network according to step f2 / , so as to obtain a prediction of the level(s) of the aquifer(s) in the determined area at a given time horizon (Hz1) lower than the given time horizon (Hzo) of step d / ; f4 / comparison of the groundwater level(s) (Hl) with an alert threshold (S), and - if no exceedance of the alert threshold at step f4 / , then repeat steps a / to d / ; - if the alert threshold is exceeded at step f4 / , then: g 1 / reiteration of steps a / and b / with a time step k2; g2 / design of a model of a second artificial neural network (ANN) from the data acquired in step g 1 / with the time step k2; g3 / carrying out a numerical simulation of the model of the second artificial neural network according to step g2 / , so as to obtain a prediction of the level(s) of the aquifer(s) in the determined area at a given time horizon (Hz2) lower than the given time horizon (Hz1) of step f3 / ; g4 / comparison of the groundwater level(s) (H2) with an alert threshold (S), and

2.

3.

4.

5.

6.

7.

8.

9.

10. - if no exceedance of the alert threshold at step g4 / , then repeat steps a / to d / ; - if the alert threshold is exceeded at step g4 / , then an alarm will be triggered and / or active pumping means will be implemented. Method according to claim 1, the first and / or second artificial neural network being a multilayer perceptron. A method according to claim 1 or 2, wherein step f2 / is performed with a rain forecast, and step g2 / is performed without a rain forecast. A method according to any one of the preceding claims, wherein the actual acquisition time of steps a / and b / is less than 1 hour, preferably between 10 and 30 minutes. A method according to any one of the preceding claims, the time step kl being equal to 24 hours. A method according to any one of the preceding claims, the time horizon Hzl being between 12 and 24h. A method according to any one of the preceding claims, the time step k2 being equal to 1 hour. A method according to any one of the preceding claims, the time horizon Hz2 being between 6 and 12h. A method according to any one of the preceding claims, the piezometric and meteorological data being sent by teletransmission to a central server in which the numerical models are hosted. A system for implementing the method according to any one of claims 1 to 9, comprising: - a central server hosting the data and digital models and artificial neural networks; - one or more piezometers equipped with a pressure sensor in the determined geographical area; - one or more meteorological station(s) arranged in the determined geographical area; in which the piezometer(s) and the meteorological station(s) are connected to the central server in order to transmit their measurement data in real time.

11. System according to claim 10, the piezometer(s) and the meteorological station(s) being equipped with teletransmission transmitters to transmit data in real time to the central server.

12. System according to claim 10 or 11, comprising an alarm and / or active pumping means, the control unit of which is connected to the central server.

13. Use of the method according to any one of claims 1 to 9 and / or of the system according to any one of claims 10 to 12, for operational decision support in crisis situations related to a rapid rise in groundwater levels, particularly in karst and fractured aquifers.

Images