Dynamic visual analysis method and system for spatial and temporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a river basin

Abstract

This invention discloses a method and system for dynamic visual analysis of oxygen-consuming pollutants across multiple interfaces in watersheds. Addressing bottlenecks in large-scale rendering of complex watershed hydrology, such as GPU memory overflow, lack of cross-interface flux representation, and difficulty in perceiving prediction errors, this invention acquires a multi-source heterogeneous spatiotemporal sequence set of water quality data; constructs a topological directed acyclic graph (DAG) based on Strahler hierarchy in memory, and performs online hierarchical temporal aggregation algorithm in the GPU for dynamic LOD memory scheduling; generates a three-dimensional cross-medium particle streamline flux layer using Runge-Kutta numerical integration and pollution degradation constant constraints; and extracts the prediction variance field using a regularized multi-decoder scene representation network (RMDSRN), mapping it to the physical geometric roughness and fogging effects of the water surface mesh. This invention achieves hardware-level isomorphism between geographical laws and low-level memory scheduling, intuitively presenting hidden physical flux interactions and deep AI-predicted risks, significantly improving the scientific rigor and hardware / software efficiency of watershed management.

Description

Technical Field

[0001] This invention relates to the field of data analysis and visualization technology, specifically to a dynamic visual analysis method and system for the spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a watershed. Background Technology

[0002] With the continuous development of modern environmental engineering and watershed management, the dynamic evolution of oxygen-consuming pollutants (such as biochemical oxygen demand (BOD), chemical oxygen demand (COD), and dissolved oxygen (DO)) concentrations across multiple scales within watersheds is crucial to the stability of aquatic ecosystems and biodiversity, and is a key indicator for assessing the self-purification capacity and health status of water bodies. Currently, continuous monitoring of watershed water quality relies heavily on a multi-source heterogeneous big data matrix comprised of physical sensor networks, satellite multispectral remote sensing, and numerical soft measurement models.

[0003] However, transforming these high-dimensional, multi-scale, and cross-media physical and biochemical data into intuitive visual representations with significant decision-making guidance still faces serious technical bottlenecks imposed by traditional computer graphics architectures and data processing mechanisms. Specifically, existing technologies suffer from the following three major shortcomings: First, there is the visual fragmentation of multi-scale spatiotemporal features and the bottleneck of underlying hardware memory. Accurate water quality analysis must capture multi-scale spatiotemporal patterns ranging from minutes to years. However, when processing massive geographic and hydrological grid data of large-scale watersheds, the traditional Euclidean spatial gridding method leads to extremely high concurrent draw calls in the computer rendering pipeline, resulting in severe GPU memory (VRAM) overflow and stuttering during panoramic roaming. Although some existing technologies have proposed fast encoding methods for watersheds and rivers based on the Strahler hierarchy, they are limited to spatial retrieval and topological localization of static digital river networks and have not extended them interdisciplinaryly to the LOD (Level of Detail) memory scheduling mechanism of the underlying graphics hardware. Therefore, they cannot fundamentally solve the hardware performance bottleneck when roaming online in three dimensions with large-scale hydrological spatiotemporal data.

[0004] Second, there is a lack of three-dimensional vector representation of multi-interface physical flux processes. In complex physical interaction spaces such as the convergence zone of surface water and groundwater (GW-SW) and river sediment-water, the transfer flux of chemical substances across interfaces is three-dimensional and dynamic. Existing mainstream visualization systems are often limited to two-dimensional scalar color mapping mechanisms, which can only reflect the static concentration distribution of surface water and cannot intuitively represent the direction and magnitude of three-dimensional vector flux exchange between invisible interfaces. Although some existing technologies involve pollution propagation visualization analysis based on time-series particle tracking, they only remain at the macroscopic level of two-dimensional trajectory clustering and lack flux mapping mechanisms across complex multi-media three-dimensional interfaces. More importantly, existing graphics rendering technologies have failed to achieve deep coupling mapping between hydrodynamic numerical integration and biogeochemical reaction dynamics (such as the chemical degradation rate constant of specific pollutants) in the underlying computer shader.

[0005] Third, there is a blind spot in the statistical uncertainty of deep learning prediction results in visual perception. While mainstream deep learning prediction systems (such as Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRUs) are widely used in water environment prediction, their outputs are often single, deterministic values ​​stripped of statistical distribution attributes. Even with the introduction of Scene Representation Networks (SRNs) for spatial science data compression and visualization, traditional frameworks typically cannot express the quality or uncertainty boundaries of the predictions. This makes it difficult for environmental managers and decision-makers to intuitively perceive the underlying statistical risks and error ranges behind the data in a three-dimensional physical view when the system outputs high-risk water quality evolution warnings, significantly reducing the scientific rigor of auxiliary control decisions. Summary of the Invention

[0006] To address the aforementioned deficiencies in existing technologies, the present invention aims to provide a dynamic visual analysis method and system for the spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in watersheds, thereby solving the following core technical problems: First, it solves the problem of graphics processor (GPU) memory (VRAM) overflow and rendering stuttering caused by the extremely high concurrent rendering calls when using traditional Euclidean spatial grid division to process panoramic roaming of large-scale hydrological spatiotemporal data. Secondly, it addresses the problem that traditional two-dimensional static color mapping cannot intuitively represent the vector flux direction, magnitude, and biochemical degradation process of material transfer across complex three-dimensional physical interfaces such as surface water and groundwater. Third, it addresses the problem that deep learning water quality evolution prediction models output single deterministic values, leading to a lack of intuitive visual perception of the underlying statistical errors and confidence levels of the models in three-dimensional space for decision-makers.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A dynamic visual analysis method for the spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a watershed, characterized by comprising the following steps: S1. Obtain a multi-source heterogeneous water quality spatiotemporal sequence set containing physical sensor monitoring data and numerical model simulation data; S2. Construct a directed acyclic graph (DAG) data structure of watershed spatiotemporal topology based on the Strahler river classification algorithm in computer physical memory, and perform online hierarchical temporal aggregation (HTA) algorithm in the graphics processing unit (GPU) rendering pipeline for dynamic scheduling of level of detail (LOD). S3. Through three-dimensional flow field reconstruction and volume rendering algorithms, a three-dimensional particle streamline flux rendering layer is generated to characterize the transfer process of pollutants at multiple physical interfaces such as water-air and water-sediment. S4. Execute the Regularized Multi-Decoder Scene Representation Network (RMDSRN) in the deep learning computing engine to generate a spatial statistical variance field, and map the variance field to the physical geometric roughness and fogging halo effect of the three-dimensional water surface mesh. S5 responds to user actions in the visual interactive interface, and integrates and renders multi-dimensional dynamic features in real time.

[0008] Furthermore, S1 includes S11 and S12; S11, deploying monitoring programs and hardware devices in the watershed monitoring network to dynamically acquire multimodal water quality data; The dynamic acquisition of multimodal water quality data uses in-situ optical and acoustic physical sensors to acquire proxy variable data such as turbidity, temperature and conductivity at a set high-frequency sampling rate. A soft measurement model combining a long short-term memory network and a gated cyclic unit (LSTM-GRU) is used as input to invert and estimate oxygen-consuming pollutant parameters such as biochemical oxygen demand (BOD) and chemical oxygen demand (COD) that cannot be directly measured at high frequencies. S12. Preprocess and spatially align the acquired proxy variable data and inversion estimation parameters to obtain a standardized spatiotemporal water quality sequence set; The preprocessing and spatial alignment are achieved by fusing satellite multispectral remote sensing data, using cubic spline interpolation algorithm to fill in missing values ​​of the sensor network in wide-area spatial resolution, and performing normalization to eliminate the influence of dimensions. The aligned data are then packaged into a high-dimensional multi-source heterogeneous water quality spatiotemporal sequence set.

[0009] Furthermore, S2 includes S21 and S22; S21. Construct a directed acyclic graph (DAG) of watershed spatiotemporal topology based on the Strahler hierarchical algorithm in computer physical memory; The construction mechanism of the DAG is as follows: the outer river section of the source without tributary inflow is set as level 1 and mapped as the bottom child node; when two rivers of the same level converge, the downstream river section increases in order and is mapped as a higher-level macroscopic parent node; object-oriented encapsulation is adopted in each node, and the encapsulation covers the feature vector matrix of oxygen-consuming pollutant concentration scalar, transfer flux vector and timestamp. S22. Execute online hierarchical temporal aggregation (HTA) and LOD detail scheduling mechanism in the GPU rendering pipeline; When the user's viewpoint camera zooms out in the 3D scene and triggers the preset Level of Detail (LOD) spatial threshold, the vertex shader and fragment shader of the system rendering engine are forced to automatically remove high-frequency meshes and feature data of the bottom child nodes, releasing the video memory (VRAM) occupied; and instead call the macroscopic parent node data that has been aggregated using spatial convolution and temporal smoothing interpolation algorithms of graph neural networks in the video memory.

[0010] Furthermore, S3 includes S31 and S32; S31. Render a static scalar field using a ray casting algorithm; The rendering of the static scalar field maps the concentration scalar field of oxygen-consuming pollutants into a three-dimensional spatial color and transparency gradient field that follows the optical transfer function. S32. Render the cross-interface flux vector field using the fourth-order Runge-Kutta numerical integration scheme and biochemical reaction kinetics mechanism; The local three-dimensional velocity component matrix at the boundary of the multi-physics interface is extracted by the fluid dynamics inversion model, and the fourth-order Runge-Kutta numerical integration algorithm is used in the computation shader to perform high-precision incremental calculation of the advection trajectory of the graphic particles in order to generate the motion trajectory of dynamic particles. Simultaneously, the alpha channel transparency decay function of the generated particles is forcibly bound to the chemical degradation rate constant k of a specific oxygen-consuming pollutant in the water body, and the particle lifetime decay equation is: This ensures that the visual disappearance process of graphic particles maintains consistency with the actual physical and biochemical decomposition process on a mathematical time scale.

[0011] Furthermore, S4 includes S41 and S42; S41. Execute the Regularized Multi-Decoder Scene Representation Network (RMDSRN) to generate a spatial statistical variance field; The spatial semantics of water quality data are extracted by a single shared feature grid encoder and multiple parallel multilayer perceptrons (MLPs) are connected as parallel decoders. Each decoder outputs a set of predictions for the same spatial coordinates. The system calculates and extracts the spatial statistical variance field representing the prediction confidence based on this set of predictions. S42. Perform physical geometry space mapping of the variance field in the graphics rendering pipeline; In the geometric processing stage of the GPU rendering pipeline, the spatial statistical variance field is directly parameterized into the vertex normal perturbation vector of the three-dimensional water surface mesh, generating physical geometric roughness distortion that is positively correlated with the statistical error; and simultaneously, the coordinates of high variance regions are extracted in the fragment shader to render and overlay a semi-transparent volumetric fog confidence halo effect.

[0012] Furthermore, S5 includes S51 and S52; S51, Response Spatiotemporal Multidimensional Slicer Interactive Command Rendering Evolution Heatmap; The front-end component receives the surface segmentation operation drawn by the user in the three-dimensional panoramic watershed model. The system automatically extracts the profile data matrix along the surface and calculates and renders the thermal distribution of oxygen-consuming pollutants in both depth and time dimensions. S52, Adaptive high-frequency resampling closed-loop feedback mechanism based on error variance field triggering; When the local spatial statistical variance field calculated in step S4 continues to be higher than the preset confidence safety threshold, the system automatically generates an adaptive high-frequency resampling command based on the coordinates of the high variance region, and transmits it back to the corresponding in-situ physical sensor network in step S1 through the IoT gateway, so as to control the underlying physical sensor to increase the sampling frequency.

[0013] The present invention also provides a dynamic visual analysis system for the spatiotemporal evolution and flux interaction of oxygen-consuming pollutants at multiple interfaces in a watershed, applicable to the method described in any of the above, characterized in that it includes a hybrid data access module, a DAG graph calculation engine module, a three-dimensional hybrid rendering engine module, an uncertainty assessment module, and a comprehensive interactive feedback module. The hybrid data access module acquires a standardized spatiotemporal water quality sequence set by deploying physical sensors in the monitoring network, collecting high-frequency proxy variables, and combining soft measurement models with multispectral remote sensing data for inversion preprocessing. The DAG graph calculation engine module is based on a standardized water quality spatiotemporal sequence set. It constructs a watershed spatiotemporal topology data structure based on the Strahler hierarchical algorithm in computer memory and executes an online hierarchical time aggregation algorithm in the GPU to perform LOD level of detail dynamic memory scheduling. The three-dimensional hybrid rendering engine module utilizes volume rendering and the fourth-order Runge-Kutta numerical integration algorithm to control the attenuation of the alpha channel transparency of particles based on the degradation constant, and calculates and generates a three-dimensional particle streamline rendering layer that characterizes the direction and flux of material transfer at multi-physics interfaces. The uncertainty assessment module executes a regularized multi-decoder scene representation network to extract a prediction set to generate a spatial statistical variance field, and maps this variance field in the shader as the normal perturbation displacement and geometric roughness of the water surface mesh. The integrated interactive feedback module responds to the front-end slicing interactive command to draw a three-dimensional profile heat map, and sends an adaptive resampling closed-loop control command to the physical sensor network based on the spatial statistical variance field.

[0014] Compared with the prior art, the present invention has the following significant progress and beneficial effects: (1) This invention achieves cross-disciplinary hardware-level isomorphism between geographical laws and computer low-level memory scheduling. It breaks through the limitation of traditional GIS systems that only use Strahler hierarchies for static retrieval, creatively integrating them directly into the GPU's LOD control pipeline. By forcibly removing high-frequency grid data from massive source river sections from a macroscopic viewpoint using vertex shaders, the spatial computational complexity of the system decreases exponentially when performing large-scale 3D online watershed roaming, completely avoiding the VRAM overflow problem.

[0015] (2) A fluid dynamics-based graphical mapping mechanism constrained by biochemical reaction kinetic equations was constructed. This invention no longer relies on simple animation effects to simulate the physical interface, but directly uses the degradation constants of specific pollutants in environmental engineering as the core constraint parameters controlling the extinction of graphical particles (Alpha channel decay) in computer shading languages. Combined with fourth-order Runge-Kutta integrals, the "invisible" cross-interface physicochemical generation and consumption fluxes are transformed into a three-dimensional visual expression that conforms to absolute mathematical logic, which is extremely difficult to be replaced by existing technology combinations.

[0016] (3) It fills the gap in physical perception of statistical uncertainty in deep learning environment prediction. This invention uses an RMDRN multi-decoder architecture to capture prediction variance at low cost, and instead of processing it into conventional statistical charts, it directly reduces the dimension and maps it into machine instructions that control the light emission and pixel arrangement of physical display hardware (i.e., vertex normal perturbation and geometric roughness variation). This method of transforming abstract AI risks into physical geometric distortions that can be intuitively perceived by human subconsciousness greatly reduces the cognitive load of managers and constructs an "information-physical-vision" intelligent prevention and control closed loop based on variance-driven sensors. Attached Figure Description

[0017] Figure 1 This is an overall flowchart of the method provided in the embodiments of the present invention; Figure 2 The system architecture and logic module diagram provided for embodiments of the present invention; Figure 3 This is a schematic diagram of a multi-scale watershed topology directed graph data structure provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the principle of multi-interface hybrid particle tracking flux rendering provided in an embodiment of the present invention; Figure 5 A comparison chart of the uncertainty rendering effect of error perception provided in the embodiments of the present invention.

[0018] Explanation of reference numerals in the attached figures: 100 - Hybrid Data Access Module; 110 - Physical Sensor; 120 - Satellite Remote Sensing Receiver; 130 - Inversion Inference Engine; 200 - DAG graph calculation engine module; 210 - Strahler hierarchical module; 220 - Online hierarchical time aggregation module; 300 - 3D Hybrid Rendering Engine Module; 310 - Ray Projector Painter; 320 - Runge-Kutta Particle Tracker; 400 - Uncertainty assessment module; 410 - Feature mesh encoder; 420 - MLP decoder; 430 - Variance calculation node; 500-Comprehensive Interactive Feedback Module; 601 - Bottom-level child node; 602 - Macro-level parent node; 603 - Eigenvector matrix; 701 - Air layer; 702 - Surface water layer; 703 - Sediment layer; 704 - Groundwater layer; 705 - Dynamic graphic particles; 706 - Legend box; 801 - Smooth mesh region; 802 - Rough distortion region; 803 - Vertex displacement; 804 - Fog halo. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0020] Example 1 Please see Figure 1 This invention provides a dynamic visual analysis method for the spatiotemporal evolution and flux interaction of oxygen-consuming pollutants at multiple interfaces in a watershed, comprising the following steps: S1. Obtain a multi-source heterogeneous water quality spatiotemporal sequence set containing physical sensor monitoring data and numerical model simulation data; S2. Construct a directed acyclic graph (DAG) data structure of watershed spatiotemporal topology based on the Strahler river classification algorithm in computer physical memory, and perform online hierarchical temporal aggregation (HTA) algorithm in the graphics processing unit (GPU) rendering pipeline for dynamic scheduling of level of detail (LOD). S3. Through three-dimensional flow field reconstruction and volume rendering algorithms, a three-dimensional particle streamline flux rendering layer is generated to characterize the transfer process of pollutants at multiple physical interfaces such as water-air and water-sediment. S4. Execute the Regularized Multi-Decoder Scene Representation Network (RMDSRN) in the deep learning computing engine to generate a spatial statistical variance field, and map the variance field to the physical geometric roughness and fogging halo effect of the three-dimensional water surface mesh. S5 responds to user actions in the visual interactive interface, and integrates and renders multi-dimensional dynamic features in real time.

[0021] This embodiment, through the coordinated implementation of the above steps, achieves accurate identification and intuitive presentation of cross-media physical flux interactions of oxygen-consuming pollutants in complex watersheds without changing the existing hardware architecture, and empowers decision-makers with an intuitive perception of risks predicted by deep learning, thereby constructing a closed-loop solution from risk identification to strategy response.

[0022] Example 2 Please see Figure 1 Specifically: S1 includes S11 and S12; S11. Deploy physical sensors 110 and soft measurement inversion models in the watershed monitoring network to dynamically acquire multimodal water quality data; The dynamic acquisition of multimodal water quality data specifically involves: acquiring high-frequency and easily measurable surrogate variable data such as turbidity, temperature, and conductivity through the deployed in-situ optical and acoustic physical sensors 110; and simultaneously using an inversion inference engine 130 combining a long short-term memory network and a gated recurrent unit (LSTM-GRU) as input to invert and estimate oxygen-consuming pollutant parameters such as biochemical oxygen demand (BOD) that are difficult to measure directly.

[0023] S12. The acquired proxy variable data and inversion estimation parameters are preprocessed and spatially aligned to obtain a standardized water quality spatiotemporal sequence set. The preprocessing and spatial alignment are performed by fusing satellite multispectral remote sensing data through satellite remote sensing receiver 120, using spatiotemporal interpolation algorithms to fill in monitoring gaps, eliminate the influence of dimensions, and construct a high-dimensional multi-source heterogeneous water quality spatiotemporal sequence set.

[0024] This embodiment utilizes soft measurement technology to solve the physical blind spot problem of high-frequency direct measurement equipment in water quality monitoring, ensuring that the subsequent rendering engine and deep learning evaluation model have high-quality and complete data input.

[0025] Example 3 Please see Figure 1 and Figure 3 Specifically: S2 includes S21 and S22; S21. Construct a directed acyclic graph (DAG) of watershed spatiotemporal topology based on the Strahler hierarchical algorithm in computer physical memory; the construction mechanism is as follows: the outer river segment of the source without tributary inflow is set as level 1 and mapped as the bottom-level child node 601; when two rivers of the same level converge, the downstream river segment increases in order and is mapped as a higher-level macroscopic parent node 602; and each node is encapsulated with an eigenvector matrix 603 covering concentration scalar, transfer flux vector and timestamp.

[0026] S22. Execute online hierarchical temporal aggregation (HTA) and LOD detail scheduling mechanism in the GPU rendering pipeline; when the user view camera zooms out in the 3D scene and triggers the preset LOD space threshold, the vertex shader of the rendering engine is forced to automatically remove the high-frequency mesh and feature data of the bottom child node 601; and instead call the data of the macroscopic parent node 602 that has been aggregated using the spatiotemporal graph neural network in the video memory.

[0027] This embodiment integrates geographical principles with the underlying memory scheduling of computers, enabling dynamic optimization of memory usage during large-scale roaming and avoiding overflow problems caused by massive mesh rendering.

[0028] Example 4 Please see Figure 1 and Figure 4 Specifically: S3 includes S31 and S32; S31. Render a static scalar field using a ray projection rendering algorithm; that is, map the concentration scalar field of oxygen-consuming pollutants into a three-dimensional transparency and color gradient field that follows the optical transfer function.

[0029] S32. Rendering the cross-interface flux vector field using a fourth-order Runge-Kutta numerical integration scheme and biochemical reaction kinetics; such as... Figure 4 As shown, at the multi-physical interface boundary of air layer 701, surface water layer 702, sediment layer 703, and groundwater layer 704, local three-dimensional velocity component matrices are extracted. The RK4 algorithm is used to perform high-precision incremental calculations of the advection trajectory of particles to generate dynamic graphic particles 705; simultaneously, as... Figure 4 As shown in the legend box 706, the alpha channel transparency decay function of the dynamic graphic particle 705 is forcibly bound to the chemical degradation rate constant k of the specific oxygen-consuming pollutant in that water area. The life cycle decay equation is defined as: .

[0030] This embodiment achieves a rigorous mathematical consistency between the visual particle extinction process and the mineralization and decomposition process of matter in the physical environment by directly using the pollution degradation constant as the core parameter for controlling the alpha extinction of the graphics shader.

[0031] Example 5 Please see Figure 1 and Figure 5 Specifically: S4 includes S41 and S42; S41. Execute the Regularized Multi-Decoder Scene Representation Network (RMDSRN) to generate the spatial statistical variance field; For example... Figure 2 As shown, a single shared feature grid encoder 410 extracts multidimensional spatiotemporal semantics of water quality, connects multiple parallel multilayer perceptrons (MLP) decoders 420 to output prediction sets, and converges them to a variance calculation node 430, thereby calculating the spatial statistical variance field representing the uncertainty of the model.

[0032] S42. Perform physical geometric space mapping of the variance field during the geometry processing stage of the rendering pipeline; such as... Figure 5 As shown, compared to the smooth mesh region 801 with low error on the left, the system directly parameterizes the spatial statistical variance field into the vertex normal perturbation vector of the three-dimensional water surface mesh, generating the physical geometric roughness distortion as shown in the rough distortion region 802 on the right. The calculation formula for its vertex displacement 803 is as follows: Simultaneously, a semi-transparent, foggy halo effect (804 effect) is rendered in the fragment shader.

[0033] This embodiment greatly reduces the cognitive load on environmental managers by reducing the dimensionality of abstract neural network statistical biases into geometric distortions that are perceptible to the visual subconscious.

[0034] Example 6 Please see Figure 1 Specifically: S5 includes S51 and S52; S51, responds to spatiotemporal multidimensional slicer interaction commands to render evolution heatmap; the front-end component receives the surface segmentation operation drawn by the user in the three-dimensional panoramic watershed model, the system automatically extracts the corresponding profile data of the surface, and renders the evolution heatmap in both depth and time dimensions.

[0035] S52. Trigger the adaptive high-frequency resampling closed-loop feedback mechanism; when the local spatial statistical variance field calculated in step S4 is continuously higher than the safety threshold, the system generates an adaptive resampling command based on the coordinates of the high variance region and sends it back to the physical sensor 110 corresponding to step S1, forming a closed-loop control.

[0036] This embodiment realizes an intelligent closed loop of "information-physical-vision", enabling the system to use AI errors to guide the working status of physical hardware.

[0037] Example 7 Please see Figure 2 A dynamic visual analysis system for spatiotemporal evolution and flux interaction of oxygen-consuming pollutants at multiple interfaces in a watershed, applied to the method described in any one of claims 1-6, characterized in that it includes a hybrid data access module 100, a DAG graph calculation engine module 200, a three-dimensional hybrid rendering engine module 300, an uncertainty assessment module 400, and a comprehensive interactive feedback module 500. The hybrid data access module 100 includes a physical sensor 110, a satellite remote sensing receiver 120, and an inversion inference engine 130, which is used to collect high-frequency proxy variables and combine them with a soft measurement model to invert and obtain a standardized water quality spatiotemporal sequence set. The DAG graph computation engine module 200 includes a Strahler hierarchical module 210 and an online hierarchical temporal aggregation module 220, which are used to construct the watershed spatiotemporal topology in memory and perform aggregation scheduling in the GPU. The 3D hybrid rendering engine module 300 includes a ray projector painter 310 and a Runge-Kutta particle tracker 320, which renders a 3D particle streamline flux layer using numerical integration algorithms and degradation constant constraints. The uncertainty assessment module 400 includes a feature mesh encoder 410, an MLP decoder 420, and a variance calculation node 430. It executes an RMDRN network to generate a variance field and maps it to the normal perturbation displacement and geometric roughness of the water surface mesh. The integrated interactive feedback module 500 responds to front-end interactive commands, draws a heat map, and issues resampling closed-loop control commands.

[0038] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0039] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity; those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both, and such implementations should not be considered beyond the scope of protection of this invention.

Claims

1. A dynamic visual analysis method for the spatiotemporal evolution and flux interaction of oxygen-consuming pollutants at multiple interfaces in a watershed, characterized by: Includes the following steps: S1. Acquire a multi-source heterogeneous spatiotemporal sequence set of water quality data, including physical sensor monitoring data and numerical model simulation data; S2. Construct a watershed spatiotemporal topological directed acyclic graph (DAG) data structure based on the Strahler river classification algorithm in the computer's physical memory, and perform online hierarchical time aggregation (HTA) algorithm for dynamic scheduling of level of detail (LOD) in the graphics processing unit (GPU) rendering pipeline; S3. Generate a 3D particle streamline flux rendering layer characterizing the transfer process of pollutants at multiple physical interfaces such as water-air and water-sediment through 3D flow field reconstruction and volume rendering algorithms; S4. Execute a regularized multi-decoder scene representation network (RMDSRN) in the deep learning computing engine to generate a spatial statistical variance field, and map this variance field to the physical geometric roughness and fogging halo effect of the 3D water surface mesh; S5. Respond to user operations in the visualization interface, and fuse and render multi-dimensional dynamic features in real time.

2. The dynamic visual analysis method for spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a watershed, as described in claim 1, is characterized in that: S1 includes S11 and S12; S11, deploying monitoring programs and hardware devices in the watershed monitoring network to dynamically acquire multimodal water quality data; the dynamic acquisition of multimodal water quality data uses in-situ optical and acoustic physical sensors to acquire proxy variable data such as turbidity, temperature, and conductivity at a set high-frequency sampling rate, and uses a soft measurement model combining a long short-term memory network and a gated cyclic unit (LSTM-GRU) as input to invert and estimate oxygen-consuming pollutant parameters such as biochemical oxygen demand (BOD) and chemical oxygen demand (COD) that cannot be directly measured at high frequencies; S12, preprocessing and spatially aligning the acquired proxy variable data and inverted estimated parameters to obtain a standardized spatiotemporal water quality sequence set; the preprocessing and spatial alignment uses satellite multispectral remote sensing data, cubic spline interpolation algorithm to fill in missing values ​​in the sensor network at wide-area spatial resolution, and normalization processing to eliminate dimensional influences, and then packages the aligned data into a high-dimensional multi-source heterogeneous spatiotemporal water quality sequence set.

3. The dynamic visual analysis method for spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a watershed, as described in claim 2, is characterized in that: S2 includes S21 and S22; S21: Construct a directed acyclic graph (DAG) of the watershed spatiotemporal topology based on the Strahler hierarchical algorithm in the computer's physical memory; the construction mechanism of the DAG is as follows: the outer river section of the source without tributary inflow is set as level 1 and mapped as the bottom-level child node; when two rivers of the same level converge, the downstream river section increases in order and is mapped as a higher-level macroscopic parent node; object-oriented encapsulation is adopted in each node, and the encapsulation covers the feature vector matrix of oxygen-consuming pollutant concentration scalar, transfer flux vector and timestamp; S22: Execute online hierarchical temporal aggregation (HTA) and LOD detail scheduling mechanism in the GPU rendering pipeline; when the user's viewpoint camera is raised in the 3D scene and triggers the preset LOD spatial threshold, the vertex shader and fragment shader of the system rendering engine are forced to automatically remove the high-frequency mesh and feature data of the bottom-level child nodes, releasing the video memory (VRAM) occupation; and then call the macroscopic parent node data that has been aggregated by the spatial convolution and temporal smoothing interpolation algorithm of the graph neural network in the video memory.

4. The dynamic visual analysis method for spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a watershed, as described in claim 3, is characterized in that: S3 includes S31 and S32; S31, rendering a static scalar field using a ray casting algorithm; the rendering of the static scalar field maps the concentration scalar field of oxygen-consuming pollutants into a three-dimensional spatial color and transparency gradient field following the optical transfer function; S32, rendering a cross-interface flux vector field using a fourth-order Runge-Kutta numerical integration scheme and a biochemical reaction kinetic mechanism; extracting the local three-dimensional velocity component matrix at the multi-physical interface boundary through a hydrodynamic inversion model, and using a fourth-order Runge-Kutta numerical integration algorithm in the computation shader to perform high-precision incremental calculations on the advection trajectory of graphic particles to generate the motion trajectory of dynamic particles; simultaneously, forcibly binding the alpha channel transparency decay function of the generated particles strictly to the chemical degradation rate constant k of the specific oxygen-consuming pollutants in the water area, and the particle life cycle decay equation is... This ensures that the visual disappearance process of graphic particles maintains consistency with the actual physical and biochemical decomposition process on a mathematical time scale.

5. A dynamic visual analysis method for the spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a watershed, as described in claim 4, is characterized in that: S4 includes S41 and S42; S41, Execute a regularized multi-decoder scene representation network (RMDSRN) to generate a spatial statistical variance field; Extract the spatial semantics of water quality data through a single shared feature grid encoder, and connect multiple parallel multilayer perceptrons (MLPs) as parallel decoders. Each decoder outputs a prediction set for the same spatial coordinates. The system calculates and extracts a spatial statistical variance field representing the prediction confidence based on this prediction set. S42. Perform physical geometry space mapping of the variance field in the graphics rendering pipeline; In the geometric processing stage of the GPU rendering pipeline, the spatial statistical variance field is directly parameterized into the vertex normal perturbation vector of the three-dimensional water surface mesh, generating physical geometric roughness distortion that is positively correlated with the statistical error. Simultaneously, the coordinates of high-variance regions are extracted in the fragment shader, and a semi-transparent volumetric fog confidence halo effect is rendered and superimposed.

6. A dynamic visual analysis method for the spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a watershed, as described in claim 5, is characterized in that: S5 includes S51 and S52; S51, rendering an evolutionary heatmap in response to the spatiotemporal multidimensional slicer interaction command; the front-end component receives the surface segmentation operation drawn by the user in the three-dimensional panoramic watershed model, the system automatically extracts the profile data matrix along the surface, and calculates and renders the thermal distribution of oxygen-consuming pollutants in both depth and time dimensions; S52, triggering an adaptive high-frequency resampling closed-loop feedback mechanism based on the error variance field; when the local spatial statistical variance field calculated in step S4 is continuously higher than the preset confidence safety threshold, the system automatically generates an adaptive high-frequency resampling command based on the coordinates of the high variance region, and transmits it back to the corresponding in-situ physical sensor network in step S1 through the IoT gateway, controlling the underlying physical sensor to increase the sampling frequency.

7. A dynamic visual analysis system for the spatiotemporal evolution and flux interaction of multi-interface oxygen-consuming pollutants in a watershed, characterized in that: The system includes a hybrid data access module, a DAG graph calculation engine module, a 3D hybrid rendering engine module, an uncertainty assessment module, and a comprehensive interactive feedback module. The hybrid data access module collects high-frequency proxy variables by deploying physical sensors in the monitoring network and performs inversion preprocessing using a soft-sensor model and multispectral remote sensing data to obtain a standardized water quality spatiotemporal sequence set. The DAG graph calculation engine module, based on the standardized water quality spatiotemporal sequence set, constructs a watershed spatiotemporal topology data structure in computer memory based on the Strahler hierarchical algorithm and executes an online hierarchical time aggregation algorithm in the GPU for LOD level of detail dynamic memory scheduling. The 3D hybrid rendering engine module utilizes volume rendering and the fourth-order Runge-Kutta numerical integration algorithm, based on the degradation constant controlling the alpha channel transparency decay of particles, to calculate and generate a 3D particle streamline rendering layer characterizing the direction and flux of material transfer at multi-physical interfaces. The uncertainty assessment module executes a regularized multi-decoder scene representation network to extract a prediction set to generate a spatial statistical variance field, and maps this variance field in the shader as the normal perturbation displacement and geometric roughness of the water surface mesh. The integrated interactive feedback module responds to the front-end slicing interactive command to draw a three-dimensional profile heat map, and sends an adaptive resampling closed-loop control command to the physical sensor network based on the spatial statistical variance field.

Images