Coastal variable-density pollution migration simulation method based on multi-modal data assimilation

Abstract

The application discloses a coastal variable-density pollution migration simulation method based on multi-modal data assimilation, and comprises the following steps: S101, depicting the heterogeneity of the water-containing medium; according to the hydrogeological drilling data and the hydrogeophysical detection data of a region to be researched, a transfer probability geostatistical simulation method is used to obtain the spatial distribution of the permeability coefficient field of the water-containing medium; the spatial structure characteristics of the permeability coefficient field of the water-containing medium are identified and corrected; S102, constructing a numerical model of pollution migration; a variable-density multiphase reaction migration simulation method is used to construct a numerical model of pollution migration and update parameters and verification; S103, simulating the coastal variable-density pollution migration process based on the numerical model of pollution migration. The method corrects and depicts the heterogeneous structure by the transfer probability geostatistical method, and significantly improves the accuracy and reliability of the pollution migration simulation under the high-salt tidal environment of the coast.

Description

Technical Field

[0001] This invention relates to the field of environmental pollution research technology, specifically to a method for simulating the migration of coastal variable-density pollution based on multimodal data assimilation. Background Technology

[0002] In the research on soil and groundwater pollution, due to the hidden, complex, and long-term evolution of groundwater systems, it is impossible to directly obtain the spatiotemporal migration patterns of pollutants through physical experiments or intuitive observations. Therefore, it is necessary to use simulation methods or models to conduct numerical simulations of groundwater pollution. Numerical simulation involves constructing and solving mathematical models to reproduce water flow, solute transport, and multiphase and multi-component reaction processes in a virtual environment. It can quantitatively predict the diffusion range, concentration changes, and arrival time of pollution plumes at sensitive targets, providing a scientific basis for pollution source identification, risk management plan formulation, and remediation effect evaluation. At the same time, it can significantly reduce the cost of on-site testing and avoid the risk of secondary pollution.

[0003] However, existing numerical simulation techniques for groundwater pollution typically simplify aquifers into homogeneous or layered homogeneous media and use a single parameter to characterize permeability. This makes it difficult to depict the spatial variability of complex heterogeneous structures such as lenses and interlayers in coastal sedimentary environments, leading to difficulties in identifying dominant pollution flow channels and significant biases in migration path prediction. Furthermore, traditional simulation methods often rely on manual parameter tuning using single types of observation data, resulting in low data utilization and an inability to quantify geological structural uncertainties. This makes it difficult to meet the high-precision simulation requirements of coupled processes involving variable density flow, multiphase flow, and multireaction in high-salinity tidal environments in coastal areas. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a method for simulating the migration of coastal variable-density pollution based on multimodal data assimilation.

[0005] A coastal variable-density pollution migration simulation method based on multimodal data assimilation includes the following steps: S101, Characterization of heterogeneity in aqueous media; Based on the hydrogeological borehole data and hydrogeophysical exploration data of the area to be studied, the initial spatial distribution of the permeability coefficient field of the aquifer was obtained by using the transition probability geostatistical simulation method. Based on the initial spatial distribution, the spatial structure characteristics of the permeability coefficient field of the aqueous medium are identified; then, the spatial structure characteristics are corrected by parameter inversion through the water flow and solute transport response information obtained by the field tracer test, and the parameter field of the heterogeneity of the aqueous medium is obtained. S102. Construct a numerical model for pollutant migration; An initial numerical model is established using a variable density multiphase reactive transport simulation method. The input of the initial numerical model is the parameter field of the heterogeneity of the aquifer, and the output is the groundwater level field and the pollutant concentration field. The initial numerical model simulates the spatiotemporal dynamic evolution of pollutant concentration under the combined action of density-driven and hydraulic-driven processes by numerically solving the variable density groundwater flow equation and the multi-component reactive transport convection-dispersion coupling equation. By using the Bayesian data assimilation method and groundwater level, pollutant concentration and tracer test data, the model parameters of the initial numerical model are updated to obtain the pollutant migration numerical model for simulating the migration of variable density pollution in coastal areas. S103. Simulate the coastal variable density pollution migration process based on the pollutant migration numerical model.

[0006] Note: The above method organically integrates geostatistical analysis of transfer probability, coupled numerical simulation of variable density, and Bayesian multimodal data assimilation. It overcomes the shortcomings of traditional methods, such as strong subjectivity in characterizing the heterogeneity of aquifers, neglect of density effects, and inability to dynamically update models. Through the sequential fusion of multi-source data such as borehole, geophysical, and tracer experiments, it achieves the fine construction of heterogeneous parameter fields and dynamic correction of key model parameters, and quantifies the uncertainty of prediction results. This significantly improves the reliability and adaptability of pollution migration simulation under coastal variable density conditions, providing a scientific basis for the prevention and control of coastal groundwater pollution.

[0007] Existing technologies employ step-by-step, serial calculations using software such as T-PROGS and TOUGHREACT, with data conversion relying on manual export. This invention constructs a simulation workflow of "transfer probability geostatistics-variable density reactive transport," based on borehole lithology coding and geophysical exploration soft data. It utilizes the transfer probability matrix and sequential indicators to simulate and generate a permeability field, and inversely corrects spatial structural characteristics (especially the location and connectivity of dominant flow channels) using water flow and solute response information from field tracer experiments. Then, within a unified grid framework, the corrected parameter field directly drives the coupled numerical solution of the variable density groundwater flow equation and the multi-component reactive transport convection-dispersion equation. Simultaneously, observational data such as groundwater level, pollutant concentration, and tracer breakthrough curves are sequentially absorbed into the model using a Bayesian data assimilation method, dynamically updating key parameters such as dispersion and reaction rate, achieving a closed-loop iteration of heterogeneity characterization, forward simulation, and parameter inversion. This avoids manual data conversion between external software, provides good interpretability, and the sequential assimilation strategy effectively reduces computational redundancy.

[0008] Traditional Bayesian data assimilation often employs static prior assumptions or updates parameters based solely on a single type of observation data, frequently neglecting the quantification of uncertainty in the posterior distribution. This makes it difficult to address the joint constraints of multi-source heterogeneous observation data (water level, concentration, tracer data) in coastal areas. This invention employs a sequential Bayesian data assimilation framework, using prior numerical simulation results as the background field. It sequentially absorbs measured groundwater level data, measured pollutant concentration data, and tracer experiment breakthrough curve data. Monte Carlo sampling (such as the Markov chain Monte Carlo method) is used to randomly sample the posterior distribution of the model parameters, obtaining a parameter set. Then, forward simulation is used to obtain the confidence interval and variance of the prediction results, achieving full-chain uncertainty quantification from parameters to prediction. During the assimilation process, there is no need to make strict distribution assumptions about the observation errors beforehand; instead, posterior sampling is driven by actual observation data, making parameter updates more consistent with the characteristics of the field data. Finally, independent observation data not involved in the assimilation are used to cross-validate the updated model, calculating accuracy indicators such as root mean square error, Nash efficiency coefficient, and prediction interval coverage to ensure that the model's prediction accuracy meets requirements. This framework can still output simulation results with reliable uncertainty ranges even when the amount of data is limited or there is observation noise, providing a scientific basis for decision-making on the migration of coastal variable density pollution.

[0009] Furthermore, based on hydrogeological borehole data and hydrogeophysical exploration data of the area under study, the spatial distribution of the permeability coefficient field of the aquifer is obtained using the transition probability geostatistical simulation method; this includes: The hydrogeological borehole data are classified and coded for lithology, and the lithological description of each borehole is transformed into a sequence of lithological types along the borehole depth direction; Based on the lithological type sequence, the transition probabilities of different lithological types in each direction of three-dimensional space are statistically analyzed, and a transition probability matrix is ​​constructed. Using the aforementioned transition probability matrix as the spatial structure driver, a sequential indicator simulation method is employed to simulate the spatial distribution of lithology types point by point on the three-dimensional grid nodes, generating a lithology type field. During the simulation process, the aforementioned hydrogeophysical exploration data is used as soft data, and the simulation results are constrained through a Bayesian update method, so that the spatial distribution of the lithology type field simultaneously conforms to the response characteristics of both borehole hard data and geophysical soft data. Based on the pre-established conversion relationship between lithology type and permeability coefficient, the lithology type field is converted into a permeability coefficient field, thus obtaining the spatial distribution of the permeability coefficient field of the water-bearing medium.

[0010] Explanation: The above method uses lithological types as a bridge and utilizes the transition probability matrix to objectively characterize the transformation laws of different lithologies in various directions in three-dimensional space. This more realistically reflects the spatial distribution of complex structures such as lenses and interlayers in heterogeneous aquifers. At the same time, geophysical exploration data is used as soft data to constrain the simulation results using a Bayesian update method. This ensures that the final permeability coefficient field takes into account the accuracy of borehole hard data while making full use of the advantage of the continuous lateral coverage of geophysical data. This significantly improves the rationality and reliability of the parameter field construction and provides a more accurate geological model input for subsequent numerical simulation of pollution migration.

[0011] Furthermore, the spatial structural features include the spatial variation law of the permeability coefficient, the anisotropic direction, the distribution pattern of high-permeability regions, and the development location and connectivity of dominant flow channels.

[0012] Furthermore, through on-site tracing experiments, the development location and connectivity of the dominant flow channels in the aforementioned spatial structural features were corrected.

[0013] Note: The above method focuses on dominant flow channels that control solute transport. It uses actual water flow and solute migration response information provided by tracer experiments to independently invert and correct the parameters. This ensures that the final parameter field retains the ability of geostatistical methods to characterize large-scale structural features, while also compensating for the shortcomings of static geostatistical methods in connectivity characterization through dynamic experimental data. This significantly improves the accuracy of heterogeneous parameter fields in characterizing dominant pollutant migration paths and provides a more reliable parameter basis for subsequent variable-density pollution migration simulation.

[0014] Furthermore, the method of simulating multiphase reaction transport using variable density is used to establish an initial numerical model, including: Based on the parameter field of the heterogeneity of the water-bearing medium, the area to be studied is divided into multiple units, and each unit is assigned parameters of permeability coefficient, porosity and dispersion. The initial and boundary conditions of the variable density groundwater flow equation and the initial and boundary conditions of the component reaction transport convection-dispersion equation are set. Based on the initial and boundary conditions of the variable density groundwater flow equation and the component reaction transport convection-dispersion equation, the variable density groundwater flow equation and the multi-component reaction transport convection-dispersion equation are coupled numerically solved to obtain the initial a priori groundwater level field and the a priori pollutant concentration field, thus completing the establishment of the initial numerical model.

[0015] Note: The above method clarifies the complete construction process from a heterogeneous parameter field to an initial numerical model, including spatial discretization and parameter assignment, boundary / initial condition setting for the variable density flow equation and the reaction transport equation, and coupling numerical solution to obtain the prior water level field and concentration field. Unlike the empirical processing of artificial partitioning and assignment in traditional methods, this method enables the model to truly reflect the heterogeneous characteristics of the aquifer at the spatial discretization level. At the same time, through the coupling solution of the variable density groundwater flow equation and the multi-component reaction transport equation, the synergistic effect of density-driven and hydraulic-driven processes is incorporated for the first time in the initial modeling stage, which is more consistent with the actual physical process of seawater intrusion and pollution plume interaction in coastal areas.

[0016] Furthermore, it also includes: Using the prior groundwater level field and prior pollutant concentration field as background fields, the Bayesian data assimilation method is used to sequentially absorb the measured groundwater level data, the measured pollutant concentration data and the tracer test breakthrough curve data to sequentially update the key parameters such as dispersion and reaction rate in the initial model, thus obtaining the assimilated and updated pollutant migration numerical model. The uncertainty of the model parameters and prediction results is quantified based on the Bayesian posterior distribution, and the prediction accuracy of the updated model is cross-validated using independent observation data that has not been assimilated. Finally, a numerical model of variable density pollution migration with uncertainty interval is output.

[0017] Description: This study achieves sequential fusion of multi-source heterogeneous data and dynamic correction of model parameters, enabling the model to adaptively adjust based on actual observation information. It also introduces Bayesian posterior distribution to quantify parameters and uncertainties in prediction results, and uses independent observation data that has not participated in assimilation for cross-validation, effectively avoiding overfitting problems during the data assimilation process. This ensures that the updated model achieves a balance between prediction accuracy and reliability, providing a quantifiable confidence assessment system for simulating the migration of variable-density pollution in coastal areas.

[0018] Furthermore, the uncertainty of the model parameters and prediction results quantified based on the Bayesian posterior distribution includes: The Monte Carlo sampling method is used to randomly sample the posterior distribution of the model parameters in the pollution migration numerical model to generate a parameter set containing multiple sets of parameter samples. Each set of parameter samples in the parameter set is then input into the pollutant migration numerical model for forward simulation to obtain multiple sets of corresponding output results. Statistical analysis is performed on the multiple sets of output results to calculate the confidence interval and variance of the output results at each time and spatial location, which serve as a quantitative indicator of the uncertainty of the model prediction results.

[0019] Furthermore, the verification of the pollutant migration numerical model includes: The updated numerical model for pollutant migration was validated using groundwater level, pollutant concentration, and tracer test data. Calculate the prediction accuracy index, which includes the root mean square error, Nash efficiency coefficient, and prediction interval coverage. The prediction accuracy index is compared with a preset threshold. If the prediction accuracy index meets the preset requirements, the updated pollutant migration numerical model and its parameter uncertainty quantification results are output. If not, the hyperparameters of the Bayesian data assimilation framework are adjusted or Markov chain Monte Carlo sampling is performed again until the model prediction accuracy meets the requirements.

[0020] The present invention also provides a coastal variable density pollution migration simulation system based on multimodal data assimilation, which is used to realize the coastal variable density pollution migration simulation method based on multimodal data assimilation, including a data acquisition and preprocessing module, a water-bearing medium heterogeneity characterization module, a numerical model construction module, a data assimilation and parameter update module, and a model verification and optimization module. The data acquisition and preprocessing module is used to acquire hydrogeological borehole data and hydrogeophysical exploration data, groundwater level, pollutant concentration and tracer test data of the area to be studied. The heterogeneity characterization module for aquifer media is used to obtain the spatial distribution of the permeability coefficient field of the aquifer media by using the transition probability geostatistical simulation method based on the hydrogeological borehole data and hydrogeophysical exploration data of the area under study; based on the spatial distribution of the permeability coefficient field of the aquifer media, the spatial structural features of the permeability coefficient field of the aquifer media are identified; and then the spatial structural features are corrected by on-site tracer experiments to obtain the corrected spatial structural features. The numerical model construction module is used to construct a pollutant migration numerical model based on the spatial distribution of the permeability coefficient field of the aqueous medium and the corrected spatial structure characteristics, using a variable density multiphase reaction transport simulation method. The data assimilation and parameter update module is used to update the model parameters of the pollutant migration numerical model and quantify the uncertainty by using the Bayesian data assimilation method with groundwater level, pollutant concentration and tracer test data. The model validation and optimization module is used to validate the prediction accuracy of the updated pollutant migration numerical model and calculate the prediction accuracy index. When the prediction accuracy index meets the preset threshold, the updated pollutant migration numerical model and its uncertainty quantification results are output. When the prediction accuracy index does not meet the preset threshold, the hyperparameters of the Bayesian data assimilation framework are adjusted, triggering the data assimilation and parameter update module to re-execute the parameter update until the model prediction accuracy meets the requirements.

[0021] The beneficial effects of this invention are: This invention addresses the problems of existing technologies that simplify aquifers as homogeneous media, rely on manual parameter adjustment based on single data, leading to difficulties in identifying dominant flow channels, large prediction biases, and the inability to quantify uncertainties. By fusing borehole and geophysical data through probabilistic geostatistics and combining tracer experiments to characterize heterogeneous structures, a variable-density multiphase reactive transport model is constructed. Furthermore, Bayesian methods are used to assimilate multimodal observation data to achieve parameter inversion and uncertainty quantification. After independent data closed-loop verification, the accuracy and reliability of pollution migration simulation in coastal high-salinity tidal environments are significantly improved. Attached Figure Description

[0022] Figure 1 This is a schematic diagram simulating the evolution of the pollutant concentration field over time according to an embodiment of the present invention. Detailed Implementation

[0023] To further illustrate the methods and effects of this invention, the technical solution of this invention will be clearly and completely described below in conjunction with experiments.

[0024] Example 1: A method for simulating the migration of coastal variable-density pollution based on multimodal data assimilation, comprising the following steps: S101, Characterization of heterogeneity in aqueous media; Based on the hydrogeological borehole data and hydrogeophysical exploration data of the area to be studied, the initial spatial distribution of the permeability coefficient field of the aquifer was obtained by using the transition probability geostatistical simulation method. Based on the initial spatial distribution, the spatial structure characteristics of the permeability coefficient field of the aqueous medium are identified; then, the spatial structure characteristics are corrected by parameter inversion through the water flow and solute transport response information obtained by the field tracer test, and the parameter field of the heterogeneity of the aqueous medium is obtained. The spatial distribution of the permeability coefficient field of the aquifer medium is obtained by using the transition probability geostatistical simulation method based on hydrogeological borehole data and hydrogeophysical exploration data of the area under study; including: The hydrogeological borehole data are classified and coded for lithology, and the lithological description of each borehole is transformed into a sequence of lithological types along the borehole depth direction; Based on the lithological type sequence, the transition probabilities of different lithological types in each direction of three-dimensional space are statistically analyzed, and a transition probability matrix is ​​constructed. Using the aforementioned transition probability matrix as the spatial structure driver, a sequential indicator simulation method is employed to simulate the spatial distribution of lithology types point by point on the three-dimensional grid nodes, generating a lithology type field. During the simulation process, the aforementioned hydrogeophysical exploration data is used as soft data, and the simulation results are constrained through a Bayesian update method, so that the spatial distribution of the lithology type field simultaneously conforms to the response characteristics of both borehole hard data and geophysical soft data. Based on the pre-established conversion relationship between lithology type and permeability coefficient, the lithology type field is converted into a permeability coefficient field, thus obtaining the spatial distribution of the permeability coefficient field of the water-bearing medium.

[0025] The spatial structural features include the spatial variation law of the permeability coefficient, the direction of anisotropy, the distribution pattern of high-permeability regions, and the development location and connectivity of dominant flow channels.

[0026] Through on-site tracing experiments, the development location and connectivity of the dominant flow channels in the aforementioned spatial structural features were corrected.

[0027] Specifically, this step generates a permeability coefficient parameter field characterizing the heterogeneity of the aquifer by fusing borehole data, geophysical exploration data, and tracer test data. Borehole data provides discrete vertical lithological descriptions, geophysical data (such as resistivity tomography) provides soft information with continuous lateral coverage, and tracer test data reflects actual water flow and solute transport responses. The fusion of these three data types employs a phased strategy. The aforementioned data processing methods are all existing technologies, and their specific processes will not be elaborated upon. In practice, the lithological descriptions of all hydrogeological boreholes in the study area are first systematically organized and categorized, grouped into several representative lithological types (such as coarse sand, medium sand, silt, silty clay, clay, etc.), transforming the lithological descriptions of each borehole into a lithological type sequence along the depth direction. Subsequently, the transition probabilities of each lithological type in each direction of three-dimensional space are statistically analyzed to construct a transition probability matrix. The vertical transition probability is directly obtained from the borehole sequence statistics, while the horizontal transition probability is derived from the vertical statistical results based on Walther's phase rule, resulting in the transition probability matrix for each direction of three-dimensional space. Driven by the transition probability matrix as the spatial structure, a sequential indicator simulation method was employed to simulate the spatial distribution of lithology types point-by-point on a three-dimensional grid. During the simulation, hydrogeophysical survey data was used as soft data. Based on the lithology-geophysical response relationship, this data was converted into lithology probabilistic information and incorporated into the conditional probability estimation of the sequential indicator simulation using a Bayesian update method. This constrained the simulation results, ensuring that the final generated lithology type field simultaneously met the accuracy of borehole hard data and the lateral distribution characteristics reflected by the geophysical soft data. Finally, based on the lithology-permeability coefficient conversion relationship, the lithology type field was converted into a permeability coefficient field, yielding the initial spatial distribution of the permeability coefficient field. Based on this initial permeability coefficient field, its spatial structural characteristics were identified, including the spatial variation law of permeability coefficients, anisotropy direction, distribution morphology of high-permeability areas, and the development location and connectivity of dominant flow channels. These characteristics were quantitatively extracted using variogram analysis and spatial clustering methods. Subsequently, a field tracer experiment was conducted, injecting tracers and monitoring their concentration changes over time to obtain breakthrough curves. The penetration curve is compared with the simulated penetration curve under the initial parameter field. With the goal of minimizing the deviation between the two, a genetic algorithm is used to perform local parameter inversion on the permeability coefficient within the dominant flow channel region, correcting its development location and connectivity characterization, thus obtaining the final parameter field for the heterogeneity of the aquifer. This parameter field provides the core input for establishing the initial numerical model in step S102.

[0028] S102. Construct a numerical model for pollutant migration; An initial numerical model is established using a variable density multiphase reactive transport simulation method. The input of the initial numerical model is the parameter field of the heterogeneity of the aquifer, and the output is the groundwater level field and the pollutant concentration field. The initial numerical model simulates the spatiotemporal dynamic evolution of pollutant concentration under the combined action of density-driven and hydraulic-driven processes by numerically solving the variable density groundwater flow equation and the multi-component reactive transport convection-dispersion coupling equation. By using Bayesian data assimilation methods and utilizing groundwater level, pollutant concentration, and tracer test data, the model parameters of the initial numerical model are updated to obtain the pollutant migration numerical model used to simulate the migration of variable-density pollution in coastal areas.

[0029] The method employing variable density multiphase reaction transport simulation establishes an initial numerical model, including: Based on the parameter field of the heterogeneity of the water-bearing medium, the area to be studied is divided into multiple units, and each unit is assigned parameters of permeability coefficient, porosity and dispersion. The initial and boundary conditions of the variable density groundwater flow equation and the initial and boundary conditions of the component reaction transport convection-dispersion equation are set. Based on the initial and boundary conditions of the variable density groundwater flow equation and the component reaction transport convection-dispersion equation, the variable density groundwater flow equation and the multi-component reaction transport convection-dispersion equation are coupled numerically solved to obtain the initial a priori groundwater level field and the a priori pollutant concentration field, thus completing the establishment of the initial numerical model.

[0030] Also includes: Using the prior groundwater level field and prior pollutant concentration field as background fields, the Bayesian data assimilation method is used to sequentially absorb the measured groundwater level data, the measured pollutant concentration data and the tracer test breakthrough curve data to sequentially update the key parameters such as dispersion and reaction rate in the initial model, thus obtaining the assimilated and updated pollutant migration numerical model. The uncertainty of the model parameters and prediction results is quantified based on the Bayesian posterior distribution, and the prediction accuracy of the updated model is cross-validated using independent observation data that has not been assimilated. Finally, a numerical model of variable density pollution migration with uncertainty interval is output.

[0031] The uncertainty of the model parameters and prediction results based on the Bayesian posterior distribution quantification includes: The Monte Carlo sampling method is used to randomly sample the posterior distribution of the model parameters in the pollution migration numerical model to generate a parameter set containing multiple sets of parameter samples. Each set of parameter samples in the parameter set is then input into the pollutant migration numerical model for forward simulation to obtain multiple sets of corresponding output results. Statistical analysis is performed on the multiple sets of output results to calculate the confidence interval and variance of the output results at each time and spatial location, which serve as a quantitative indicator of the uncertainty of the model prediction results.

[0032] The validation pollutant migration numerical model includes: The updated numerical model for pollutant migration was validated using groundwater level, pollutant concentration, and tracer test data. Calculate the prediction accuracy index, which includes the root mean square error, Nash efficiency coefficient, and prediction interval coverage. The prediction accuracy index is compared with a preset threshold. If the prediction accuracy index meets the preset requirements, the updated pollutant migration numerical model and its parameter uncertainty quantification results are output. If not, the hyperparameters of the Bayesian data assimilation framework are adjusted or Markov chain Monte Carlo sampling is performed again until the model prediction accuracy meets the requirements.

[0033] Based on the content of S102 above, the specific steps include: First, dividing the study area into a three-dimensional structured grid. Based on the final parameter field obtained in S101, assigning permeability, porosity, and dispersion parameters to each grid cell. Initial water level conditions and boundary conditions (including constant head boundary, impermeable boundary, or constant flow boundary) are set for the variable density groundwater flow equation, and initial concentration conditions and boundary conditions are also set for the multi-component reactive transport convection-dispersion equation. On this basis, the finite difference or finite element method is used to numerically solve the variable density groundwater flow equation and the multi-component reactive transport convection-dispersion equation—the former describes groundwater flow under the combined action of density-driven and hydraulic-driven forces, while the latter describes the convection, dispersion, and chemical reaction processes of each component. The two are bidirectionally coupled through Darcy velocity and density coupling terms. The above numerical solution techniques are existing technologies and will not be elaborated further. After the solution is completed, the a priori groundwater level field and a priori pollutant concentration field are output, completing the establishment of the initial numerical model. Then, using the prior groundwater level field and prior pollutant concentration field as background fields, and based on Bayesian data assimilation methods (such as ensemble Kalman filtering or particle filtering), the measured groundwater level data, measured pollutant concentration data, and tracer experiment breakthrough curve data are sequentially absorbed to sequentially update key parameters such as dispersion and reaction rate in the initial model. During the data absorption process, the parameter state is updated once after each set of observation data is absorbed, until all multimodal data are absorbed, resulting in the assimilated and updated numerical model. Finally, the Markov chain Monte Carlo method is used to sample the posterior distribution of the updated model parameters, generating multiple sets of parameter samples. These samples are then substituted into the model for forward simulation to obtain multiple sets of prediction results. Based on these results, the confidence interval and variance of the prediction results at each spatiotemporal node are calculated as uncertainty quantification indicators. Simultaneously, cross-validation is performed using independent observational data not involved in assimilation. Accuracy indicators such as root mean square error, Nash efficiency coefficient, and prediction interval coverage are calculated and compared with preset thresholds. If the requirements are not met, the hyperparameters of the assimilation framework are adjusted or resampling is performed until the accuracy meets the standards. Finally, a numerical model of variable density pollution migration with an uncertainty interval is output. The simulation results are as follows: Figure 1 As shown.

[0034] S103. Simulate the coastal variable density pollution migration process based on the pollutant migration numerical model.

[0035] Existing technologies employ step-by-step, serial calculations using software such as T-PROGS and TOUGHREACT, with data conversion relying on manual export. This invention constructs a simulation workflow of "transfer probability geostatistics-variable density reactive transport," based on borehole lithology coding and geophysical exploration soft data. It utilizes the transfer probability matrix and sequential indicators to simulate and generate a permeability field, and inversely corrects spatial structural characteristics (especially the location and connectivity of dominant flow channels) using water flow and solute response information from field tracer experiments. Then, within a unified grid framework, the corrected parameter field directly drives the coupled numerical solution of the variable density groundwater flow equation and the multi-component reactive transport convection-dispersion equation. Simultaneously, observational data such as groundwater level, pollutant concentration, and tracer breakthrough curves are sequentially absorbed into the model using a Bayesian data assimilation method, dynamically updating key parameters such as dispersion and reaction rate, achieving a closed-loop iteration of heterogeneity characterization, forward simulation, and parameter inversion. This avoids manual data conversion between external software, provides good interpretability, and the sequential assimilation strategy effectively reduces computational redundancy.

[0036] Traditional Bayesian data assimilation often employs static prior assumptions or updates parameters based solely on a single type of observation data, frequently neglecting the quantification of uncertainty in the posterior distribution. This makes it difficult to address the joint constraints of multi-source heterogeneous observation data (water level, concentration, tracer data) in coastal areas. This invention employs a sequential Bayesian data assimilation framework, using prior numerical simulation results as the background field. It sequentially absorbs measured groundwater level data, measured pollutant concentration data, and tracer experiment breakthrough curve data. Monte Carlo sampling (such as the Markov chain Monte Carlo method) is used to randomly sample the posterior distribution of the model parameters, obtaining a parameter set. Then, forward simulation is used to obtain the confidence interval and variance of the prediction results, achieving full-chain uncertainty quantification from parameters to prediction. During the assimilation process, there is no need to make strict distribution assumptions about the observation errors beforehand; instead, posterior sampling is driven by actual observation data, making parameter updates more consistent with the characteristics of the field data. Finally, independent observation data not involved in the assimilation are used to cross-validate the updated model, calculating accuracy indicators such as root mean square error, Nash efficiency coefficient, and prediction interval coverage to ensure that the model's prediction accuracy meets requirements. This framework can still output simulation results with reliable uncertainty ranges even when the amount of data is limited or there is observation noise, providing a scientific basis for decision-making on the migration of coastal variable density pollution.

[0037] This embodiment also provides a coastal variable density pollution migration simulation system based on multimodal data assimilation, which is used to simulate the migration of coastal variable density pollution. It includes a data acquisition and preprocessing module, a water-containing medium heterogeneity characterization module, a numerical model construction module, a data assimilation and parameter update module, and a model verification and optimization module. The data acquisition and preprocessing module is used to acquire hydrogeological borehole data and hydrogeophysical exploration data, groundwater level, pollutant concentration and tracer test data of the area to be studied. The heterogeneity characterization module for aquifer media is used to obtain the spatial distribution of the permeability coefficient field of the aquifer media by using the transition probability geostatistical simulation method based on the hydrogeological borehole data and hydrogeophysical exploration data of the area under study; based on the spatial distribution of the permeability coefficient field of the aquifer media, the spatial structural features of the permeability coefficient field of the aquifer media are identified; and then the spatial structural features are corrected by on-site tracer experiments to obtain the corrected spatial structural features. The numerical model construction module is used to construct a pollutant migration numerical model based on the spatial distribution of the permeability coefficient field of the aqueous medium and the corrected spatial structure characteristics, using a variable density multiphase reaction transport simulation method. The data assimilation and parameter update module is used to update the model parameters of the pollutant migration numerical model and quantify the uncertainty by using the Bayesian data assimilation method with groundwater level, pollutant concentration and tracer test data. The model validation and optimization module is used to validate the prediction accuracy of the updated pollutant migration numerical model and calculate the prediction accuracy index. When the prediction accuracy index meets the preset threshold, the updated pollutant migration numerical model and its uncertainty quantification results are output. When the prediction accuracy index does not meet the preset threshold, the hyperparameters of the Bayesian data assimilation framework are adjusted, triggering the data assimilation and parameter update module to re-execute the parameter update until the model prediction accuracy meets the requirements.

Claims

1. A coastal variable-density pollution migration simulation method based on multimodal data assimilation, characterized in that, Includes the following steps: S101, Characterization of heterogeneity in aqueous media; Based on the hydrogeological borehole data and hydrogeophysical exploration data of the area to be studied, the initial spatial distribution of the permeability coefficient field of the aquifer was obtained by using the transition probability geostatistical simulation method. Based on the initial spatial distribution, the spatial structure characteristics of the permeability coefficient field of the water-bearing medium were identified; Then, by using the water flow and solute transport response information obtained from the field tracer experiment, the spatial structure characteristics are parameter inversion correction to obtain the parameter field of the heterogeneity of the water-containing medium; S102. Construct a numerical model for pollutant migration; An initial numerical model is established using a variable density multiphase reactive transport simulation method. The input of the initial numerical model is the parameter field of the heterogeneity of the aquifer, and the output is the groundwater level field and the pollutant concentration field. The initial numerical model simulates the spatiotemporal dynamic evolution of pollutant concentration under the combined action of density-driven and hydraulic-driven processes by numerically solving the variable density groundwater flow equation and the multi-component reactive transport convection-dispersion coupling equation. By using the Bayesian data assimilation method and groundwater level, pollutant concentration and tracer test data, the model parameters of the initial numerical model are updated to obtain the pollutant migration numerical model for simulating the migration of variable density pollution in coastal areas. S103. Simulate the coastal variable density pollution migration process based on the pollutant migration numerical model.

2. The coastal variable-density pollution migration simulation method based on multimodal data assimilation as described in claim 1, characterized in that, The spatial distribution of the permeability coefficient field of the aquifer medium is obtained by using the transition probability geostatistical simulation method based on hydrogeological borehole data and hydrogeophysical exploration data of the area under study; including: The hydrogeological borehole data are classified and coded for lithology, and the lithological description of each borehole is transformed into a sequence of lithological types along the borehole depth direction; Based on the lithological type sequence, the transition probabilities of different lithological types in each direction of three-dimensional space are statistically analyzed, and a transition probability matrix is ​​constructed. Using the aforementioned transition probability matrix as the spatial structure driver, a sequential indicator simulation method is employed to simulate the spatial distribution of lithology types point by point on the three-dimensional grid nodes, generating a lithology type field. During the simulation process, the aforementioned hydrogeophysical exploration data is used as soft data, and the simulation results are constrained through a Bayesian update method, so that the spatial distribution of the lithology type field simultaneously conforms to the response characteristics of both borehole hard data and geophysical soft data. Based on the pre-established conversion relationship between lithology type and permeability coefficient, the lithology type field is converted into a permeability coefficient field, thus obtaining the spatial distribution of the permeability coefficient field of the water-bearing medium.

3. The coastal variable-density pollution migration simulation method based on multimodal data assimilation as described in claim 1, characterized in that, The spatial structural features include the spatial variation law of the permeability coefficient, the direction of anisotropy, the distribution pattern of high-permeability regions, and the development location and connectivity of dominant flow channels.

4. The coastal variable-density pollution migration simulation method based on multimodal data assimilation as described in claim 3, characterized in that, Through on-site tracing experiments, the development location and connectivity of the dominant flow channels in the aforementioned spatial structural features were corrected.

5. The coastal variable-density pollution migration simulation method based on multimodal data assimilation as described in claim 1, characterized in that, The method employing variable density multiphase reaction transport simulation establishes an initial numerical model, including: Based on the parameter field of the heterogeneity of the water-bearing medium, the area to be studied is divided into multiple units, and each unit is assigned parameters of permeability coefficient, porosity and dispersion. The initial and boundary conditions of the variable density groundwater flow equation and the initial and boundary conditions of the component reaction transport convection-dispersion equation are set. Based on the initial and boundary conditions of the variable density groundwater flow equation and the component reaction transport convection-dispersion equation, the variable density groundwater flow equation and the multi-component reaction transport convection-dispersion equation are coupled numerically solved to obtain the initial a priori groundwater level field and the a priori pollutant concentration field, thus completing the establishment of the initial numerical model.

6. The coastal variable-density pollution migration simulation method based on multimodal data assimilation as described in claim 5, characterized in that, Also includes: Using the prior groundwater level field and prior pollutant concentration field as background fields, the Bayesian data assimilation method is used to sequentially absorb the measured groundwater level data, the measured pollutant concentration data and the tracer test breakthrough curve data to sequentially update the key parameters such as dispersion and reaction rate in the initial model, thus obtaining the assimilated and updated pollutant migration numerical model. The uncertainty of the model parameters and prediction results is quantified based on the Bayesian posterior distribution, and the prediction accuracy of the updated model is cross-validated using independent observation data that has not been assimilated. Finally, a numerical model of variable density pollution migration with uncertainty interval is output.

7. The coastal variable-density pollution migration simulation method based on multimodal data assimilation as described in claim 5, characterized in that, The uncertainty of the model parameters and prediction results based on the Bayesian posterior distribution quantification includes: The Monte Carlo sampling method is used to randomly sample the posterior distribution of the model parameters in the pollution migration numerical model to generate a parameter set containing multiple sets of parameter samples. Each set of parameter samples in the parameter set is then input into the pollutant migration numerical model for forward simulation to obtain multiple sets of corresponding output results. Statistical analysis is performed on the multiple sets of output results to calculate the confidence interval and variance of the output results at each time and spatial location, which serve as a quantitative indicator of the uncertainty of the model prediction results.

8. The coastal variable-density pollution migration simulation method based on multimodal data assimilation as described in claim 5, characterized in that, The validation pollutant migration numerical model includes: The updated numerical model for pollutant migration was validated using groundwater level, pollutant concentration, and tracer test data. Calculate the prediction accuracy index, which includes the root mean square error, Nash efficiency coefficient, and prediction interval coverage. The prediction accuracy index is compared with a preset threshold. If the prediction accuracy index meets the preset requirements, the updated pollutant migration numerical model and its parameter uncertainty quantification results are output. If not, the hyperparameters of the Bayesian data assimilation framework are adjusted or Markov chain Monte Carlo sampling is performed again until the model prediction accuracy meets the requirements.

9. A coastal variable-density pollution migration simulation system based on multimodal data assimilation, used to implement the coastal variable-density pollution migration simulation method based on multimodal data assimilation as described in any one of claims 1 to 5, characterized in that, It includes modules for data acquisition and preprocessing, characterization of heterogeneity of aqueous media, numerical model construction, data assimilation and parameter updating, and model validation and optimization. The data acquisition and preprocessing module is used to acquire hydrogeological borehole data and hydrogeophysical exploration data, groundwater level, pollutant concentration and tracer test data of the area to be studied. The heterogeneity characterization module for aquifer media is used to obtain the spatial distribution of the permeability coefficient field of aquifer media by using the transition probability geostatistical simulation method based on the hydrogeological borehole data and hydrogeophysical exploration data of the area to be studied; and to identify the spatial structural characteristics of the permeability coefficient field of aquifer media based on the spatial distribution of the permeability coefficient field. Then, the spatial structure features are corrected through on-site tracer experiments to obtain the corrected spatial structure features; The numerical model construction module is used to construct a pollutant migration numerical model based on the spatial distribution of the permeability coefficient field of the aqueous medium and the corrected spatial structure characteristics, using a variable density multiphase reaction transport simulation method. The data assimilation and parameter update module is used to update the model parameters of the pollutant migration numerical model and quantify the uncertainty by using the Bayesian data assimilation method with groundwater level, pollutant concentration and tracer test data. The model verification and optimization module is used to verify the prediction accuracy of the updated pollutant migration numerical model and calculate the prediction accuracy index; when the prediction accuracy index meets the preset threshold, it outputs the updated pollutant migration numerical model and its uncertainty quantification results. When the prediction accuracy does not meet the preset threshold, the hyperparameters of the Bayesian data assimilation framework are adjusted, triggering the data assimilation and parameter update module to re-execute the parameter update until the model prediction accuracy meets the requirements.

Images