A small watershed global water and sediment dynamic monitoring method

Abstract

The application discloses a kind of small watershed global water and sediment dynamic monitoring method, it is related to hydrological monitoring technical field.The high-resolution digital base of small watershed and space-ground integration monitoring network are established to obtain global multi-source monitoring data;Mechanism library containing multiple replaceable water and sediment process mechanism modules is constructed;Based on global multi-source monitoring data and current basin state condition, at least one mechanism module is dynamically selected and combined to assemble into the water and sediment process simulation model for current or forecast scenario;Water and sediment dynamic simulation or forecast result under different situations is generated;According to the difference between actual monitoring data and simulation result, the selection and combination rule of mechanism module is feedback optimized.This scheme solves the core uncertainty problem of water and sediment simulation under complex conditions, and provides reliable technical support for precise management, risk prevention and control and intelligent management of watershed in engineering application.

Description

Technical Field

[0001] This invention relates to the field of hydrological monitoring technology, and in particular to a method for dynamic monitoring of water and sediment across a small watershed. Background Technology

[0002] As the basic geographical unit for the occurrence, development, and regulation of soil and water loss, the accurate monitoring and simulation of water and sediment processes in small watersheds are the cornerstone for supporting the evaluation of soil and water conservation benefits, comprehensive watershed management, and ecological restoration decisions. Traditional water and sediment monitoring in small watersheds mainly relies on setting up hydrological stations at the watershed outlet section and measuring flow and sediment concentration manually or automatically. Although this method can obtain the total water and sediment output of the watershed, it cannot reveal the spatial heterogeneity of water and sediment generation, transport, and deposition processes within the watershed, making it difficult to provide refined process data for systematic management.

[0003] With the development of remote sensing (RS), geographic information systems (GIS), and the Internet of Things (IoT) technologies, integrated "space-air-ground" monitoring networks have become an important means of acquiring spatial information about watersheds. However, the massive amounts of multi-source monitoring data urgently need to be deeply integrated with mechanistic models that can accurately describe physical processes in order to extract knowledge from the data and achieve a leap from monitoring to simulation and early warning. Currently, distributed or semi-distributed hydrological and sediment models, such as SWAT, WEPP, and HEC-HMS, are widely used. These models are based on a fixed set of physical equations (such as the SCS-CN runoff generation method and the MUSLE erosion equation), and the model parameters are adjusted to match the observation data.

[0004] However, under complex underlying surfaces (such as fragmented terrain, patchy land use, and dense soil and water conservation projects) and extreme rainfall conditions, traditional models, employing pre-defined and fixed physical equation structures, may fail to accurately characterize the dominant water and sediment processes (e.g., in watersheds where infiltration-excess runoff and saturation-excess runoff mechanisms coexist, and road concentrating flow interacts with natural slope flow). Model calibration solely through parameter adjustments often leads to the problem of heterogeneous parameters achieving similar effects, meaning multiple sets of parameters can achieve good fits to the outlet section, but the corresponding internal physical processes are drastically different, resulting in a lack of physical consistency and extrapolation reliability in model predictions. Furthermore, traditional model calibration heavily relies on total volume or process line data at the watershed outlet, lacking constraints on key process states such as soil moisture dynamics and the spatial distribution of erosion and deposition within the watershed. Therefore, the model may fit well at the outlet, but the simulated internal hydrological response zones and erosion hotspots may significantly deviate from reality, leading to biases in the design of model-based spatial governance schemes.

[0005] To address the aforementioned issues, recent studies have attempted to couple data-driven models (such as deep learning) to improve fitting accuracy. However, these "black box" models suffer from poor physical interpretability, making them unsuitable for practical mechanism exploration and management decision-making. Therefore, developing a new generation of water and sediment dynamic monitoring and simulation methods capable of adaptively optimizing physical mechanisms, integrating multi-source spatial observation data, and quantifying uncertainties has become an urgent need in this field to overcome technical bottlenecks and achieve accurate simulation and intelligent management. To this end, a method for monitoring water and sediment dynamics across the entire small watershed is proposed. Summary of the Invention

[0006] The main objective of this invention is to provide a method for dynamic monitoring of water and sediment across a small watershed, which can effectively solve the problems in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for monitoring water and sediment dynamics across a small watershed includes the following steps: S1: Establish a high-resolution digital infrastructure and an integrated air-space-ground monitoring network for small watersheds to obtain multi-source monitoring data across the entire region; S2: Construct a mechanism library containing multiple interchangeable water and sediment process mechanism modules, where each mechanism module represents a verified physical process description; S3: Based on the multi-source monitoring data of the whole domain and the current watershed state conditions, dynamically select and combine at least one mechanism module from the mechanism library to assemble a watershed water and sediment process simulation model for the current or forecast scenario; S4: Run the assembled watershed water and sediment process simulation model to generate water and sediment dynamic simulation or forecast results under different scenarios, and quantify the uncertainty of the results; S5: Based on the differences between the actual monitoring data and simulation results obtained subsequently, the selection and combination rules of the mechanism module are optimized by feedback, and the mapping relationship between the watershed state conditions and the dynamic mechanism of water and sediment is updated.

[0008] Also includes: After each complete hydrological event, a cost function for multiple candidate mechanism combinations is calculated based on the multi-source monitoring data of the entire region; The mechanism combination that minimizes the cost function is selected as the optimal combination for this hydrological event and used to update the condition-mechanism map.

[0009] Also includes: An autoencoder is used to map high-dimensional conditional vectors to a low-dimensional manifold space. Clustering of historical event conditions in the low-dimensional manifold space identifies water and sediment response prototypes. Each water and sediment response prototype is associated with a dominant combination of mechanistic modules, thereby simplifying the mapping and enabling knowledge discovery.

[0010] Furthermore, step S3 specifically includes: The definition includes at least a watershed state condition vector consisting of rainfall characteristics, anterior soil moisture, vegetation status, and topographic attributes; Define and maintain the optimal or high-probability combination of mechanism modules under different watershed state condition vectors to form a condition-mechanism map; Based on the current watershed state condition vector acquired or predicted in real time, the condition-mechanism map is queried to determine the combination of mechanism modules to be used.

[0011] Furthermore, step S4 specifically includes: Multiple possible mechanism combinations and corresponding parameters are sampled from the condition-mechanism map to form an ensemble forecast; Run the model corresponding to each mechanism combination separately to obtain the simulation set of water and sediment processes; Analyzing the statistical characteristics of the simulation set, the total forecast variance is decomposed into structural uncertainty, parameter uncertainty, and forced uncertainty.

[0012] Furthermore, the mechanism module includes at least: The infiltration and runoff generation module is used to simulate the distribution process of precipitation after it reaches the surface, that is, to simulate the amount of water infiltrating into the soil and forming surface runoff. It is the starting point of the hydrological cycle and determines the total amount of runoff and the peak time. The slope erosion module is used to simulate the process of soil particles being stripped from the surface and transported on slopes under the action of rainfall splash and slope runoff, which is the main source of watershed sediment. The channel sediment transport module is used to simulate the transport, deposition and resuspension processes of sediment in a defined channel or river network, and it connects slope sediment production and watershed outlet sediment transport. Road concentrating flow is used to simulate the unique hydrological effects of roads as efficient runoff generation and confluence channels. The engineering measures for sediment trapping module is used to simulate the human intervention and abrupt changes in water and sediment processes caused by soil and water conservation projects such as silt-retaining dams, dams, and sediment traps.

[0013] Furthermore, the conditional-mechanism map is updated through event-driven Bayesian learning, and the update process is represented as an iteration of the probability distribution: ,in, C represents the combination of mechanism modules; C represents the condition vector. Monitoring data representing new hydrological events; It is the likelihood function calculated based on the goodness of fit between simulation and observation, expressed as: Let be the cost function.

[0014] Furthermore, the cost function includes at least an overall fitting error term. Spatial consistency error term and complexity penalty terms Defined as: ,in, All are weighting coefficients, and satisfy the following conditions: .

[0015] Furthermore, the overall fitting error term Used to measure the combination of mechanisms The degree of overall agreement between the model simulation results and the observation data of the watershed outlet section is specifically defined as: These are the goodness-of-fit functions for the flow rate process, the goodness-of-fit function for the sediment concentration process, and the goodness-of-fit function for the total sediment transport, respectively. These are all weighting coefficients, manually set based on data reliability and management needs, and must meet the following requirements: The observation data at the outlet section of the basin include at least flow rate, sediment concentration, and sediment transport.

[0016] Furthermore, the spatial consistency error term This is used to penalize combinations of mechanisms that fit well only at the watershed outlet but whose internal spatial processes do not match observed distribution patterns, thereby forcing the model to select mechanisms that conform to actual spatial physical processes. This is calculated by comparing simulated and observed spatial variation maps of soil moisture content or erosion and deposition distribution maps, and is specifically defined as: These are error terms based on the spatial pattern of soil moisture content variation, the spatial distribution of erosion / deposition, and the spatial pattern of water and sediment process pathways or responses. These are all weighting coefficients used to reflect the reliability and importance of different spatial data sources, and satisfy the following conditions: .

[0017] Furthermore, the complexity penalty term This approach, while ensuring good model fit, follows Occam's razor principle, favoring simpler, fewer-parameter, and physically more stable mechanism combinations to enhance the model's generalization ability and interpretability. Specifically, it is defined as follows: ,in, The penalty term based on information criteria, used to penalize the number of parameters, is defined as follows: When using mechanism combination At that time, the total number of effective parameters of the model, The penalty coefficient is set by the individual.

[0018] Furthermore, The entropy penalty term, used to encourage the sparsity or concentration of weights within the mechanism combination, is defined as: When using mechanism combination At time, the dynamic weight of the j-th mechanism module, where m is the total number of mechanism modules. To prevent the occurrence And an artificially set extremely small positive number.

[0019] Furthermore, The penalty term, based on prior knowledge, is used to encourage the selection of more reliable or commonly used mechanistic modules and is defined as follows: To adopt mechanism combination At that time, the reliability score of the j-th mechanism module is obtained from expert evaluation or historical performance statistics. Let be the indicator function, representing the dynamic weight of the j-th mechanistic module. Exceeding the preset threshold Effective immediately; Furthermore, These are all regularization coefficients, used to balance the strength of penalty terms with different complexities, and satisfy the following conditions: .

[0020] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for dynamic monitoring of water and sediment across a small watershed.

[0021] The present invention has the following beneficial effects: Compared with existing technologies, this solution dynamically selects and combines mechanism modules, such as automatically selecting Green-Ampt or SCS-CN flow generation formulas under different conditions, enabling the model architecture itself to have adaptive capabilities. This fundamentally avoids systematic deviations caused by using incorrect physical representations under complex conditions and significantly reduces structural uncertainties caused by model structural defects.

[0022] Compared with existing technologies, this scheme introduces a spatial consistency error term as the core constraint for model selection and optimization. It forces the selected mechanism combination to not only fit the water and sediment process line at the watershed outlet, but also to ensure that the simulated internal spatial processes, such as soil moisture change patterns and erosion and deposition distribution, are consistent with the "air-space-ground" monitoring data. This ensures that the model can truly reflect the internal hydrological response and erosion and sediment production spatial patterns, and greatly improves the physical reliability of the simulation.

[0023] Compared with existing technologies, this solution establishes a condition-mechanism map. When encountering extreme conditions close to historical experience, the system can directly call the verified optimal mechanism combination for high-precision simulation. For unprecedented extreme combinations, the system can also assemble a relatively reasonable model based on similarity reasoning. Its extrapolation prediction robustness far exceeds that of traditional models with fixed structures.

[0024] Compared with existing technologies, this scheme, through ensemble forecasting, can clearly decompose the total forecast uncertainty into structural uncertainty, parameter uncertainty, and forced uncertainty. It can not only obtain the forecast range, but also understand the main sources of uncertainty, thereby providing a clear decision-making basis for taking targeted measures.

[0025] Compared with existing technologies, this solution encodes Occam's Razor principle and the physical cognition of domain experts into the automatic optimization process by setting complexity penalty terms and prior knowledge penalty terms. This makes the system tend to select simpler and more physically grounded mechanism combinations, avoiding the generation of complex black box models with ambiguous physical meanings. The final output not only includes prediction results, but also a clear explanation of the dominant physical mechanism, which greatly improves the acceptability and practical value of the results for researchers and managers.

[0026] Compared with existing technologies, this solution upgrades the dynamic monitoring of water and sediment in small watersheds from the traditional "data acquisition + fixed model post-processing" model to an integrated closed-loop intelligent model of "intelligent sensing - dynamic modeling - co-evolution - decision support". This not only scientifically solves the core uncertainty problem of water and sediment simulation under complex conditions, but also provides a reliable, transparent, and adaptive next-generation technological foundation for precise watershed governance, risk prevention and control, and intelligent management in engineering applications, possessing significant theoretical value and broad prospects for industrial application. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a method for monitoring water and sediment dynamics across a small watershed according to the present invention. Figure 2 This is a schematic diagram of the structure of a small watershed-wide dynamic water and sediment monitoring system according to the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0029] Example 1: See Figure 1 The flowchart shown is a method for monitoring water and sediment dynamics across a small watershed according to the present invention, which specifically includes the following stages and steps: Phase 1: Infrastructure Development and Environment Deployment Step 1: Construct a high-resolution digital infrastructure for the target watershed, specifically: Data collection: UAV LiDAR / oblique photography was used to acquire centimeter-level DEM (Digital Elevation Model) and DOM (Digital Orthophoto).

[0030] Collect or generate land use / vegetation cover maps (resolution ≤ 1 meter) through multispectral / hyperspectral remote sensing interpretation.

[0031] A spatial database of soil attributes such as soil type, texture, and organic matter content was constructed through field sampling and spatial interpolation.

[0032] Field survey and digitization: Accurately map and digitize the spatial location, geometric parameters, and operational status of engineering measures such as terraces, silt-retaining dams, grain dams, and roads.

[0033] Data processing and fusion: Unify the spatial coordinate system and resolution of all data.

[0034] A "small watershed digital baseline" integrating topography, landforms, soil, vegetation, and engineering is generated as the spatial benchmark for all simulations and analyses.

[0035] Step 2: Deploy an integrated air-space-ground monitoring network, specifically as follows: Three-dimensional network deployment: "Sky" base (satellite): Access data streams from a series of satellites such as Sentinel and GF, and regularly acquire multispectral and radar images.

[0036] "Air" base (drones): Deploy fixed-wing or multi-rotor drones, and plan routine patrol routes and take-off and landing points for emergency response to rainstorms.

[0037] "Ground" foundation (Internet of Things): Superstation at the watershed outlet: Equipped with ADCP flow meter, laser / optical sand analyzer, high-definition video monitoring, and automatic water quality sampler.

[0038] Internal key node stations: Integrated stations of water level gauges, rain gauges, turbidity meters, and soil temperature and humidity sensors are deployed at the outlets of major sub-basins and key geomorphic locations.

[0039] Distributed sensor network: Deploy soil moisture sensor networks and automatic weather stations on typical slopes and different land use plots.

[0040] Data platform construction: Establish a unified spatiotemporal data management platform to realize the automatic reception, quality control, storage and fusion of multi-source heterogeneous data.

[0041] Step 3: Establish a standardized mechanism module library The mechanism module includes at least the following: The infiltration and runoff generation module is used to simulate the distribution process of precipitation after it reaches the surface, that is, to simulate the amount of water infiltrating into the soil and forming surface runoff. It is the starting point of the hydrological cycle and determines the total amount of runoff and the peak time. The slope erosion module is used to simulate the process of soil particles being stripped from the surface and transported on slopes under the action of rainfall splash and slope runoff, which is the main source of watershed sediment. The channel sediment transport module is used to simulate the transport, deposition and resuspension processes of sediment in a defined channel or river network, and it connects slope sediment production and watershed outlet sediment transport. Road concentrating flow is used to simulate the unique hydrological effects of roads as efficient runoff generation and confluence channels. The engineering measures for sediment trapping module is used to simulate the human intervention and abrupt changes in water and sediment processes caused by soil and water conservation projects such as silt-retaining dams, dams, and sediment traps.

[0042] Develop or package core modules for infiltration and runoff generation (such as Green-Ampt, Horton), slope erosion (such as MUSLE, RUSLE), and gully sediment transport (such as Engelund-Hansen).

[0043] For specific processes, we independently developed a road centralized flow module and an engineering measure sand-blocking module.

[0044] Interface standardization: Define a unified interface specification for input variables, output variables and parameters for all modules to ensure pluggability and interchangeability between modules.

[0045] Establish a mechanism library index: Add metadata to each module to describe its physical principles, applicable conditions, parameter meanings, etc.

[0046] Phase Two: System Initialization and First Learning Step 4: Define condition vectors and initialize the graph Define condition vector C: Determine the specific indices that constitute the condition vector, including: Rainfall characteristics: previous effective rainfall, cumulative rainfall of this event, maximum rainfall intensity .

[0047] Watershed status: Spatial mean soil moisture content, vegetation index (NDVI) for different land uses.

[0048] Static properties: unit slope, aspect, soil erodibility K value, etc.

[0049] Initialization conditions - Mechanism diagram: Based on literature and expert experience, an initial, broad probability distribution of mechanism combinations is pre-defined for typical conditions (such as "drought-high intensity" and "humid-long duration"). Alternatively, it can be set as a no-information prior (uniform distribution).

[0050] Step 5: Capture the first complete hydrological event and fuse the data. Event Triggering and Enhanced Monitoring: Based on meteorological warnings, activate enhanced monitoring mode, coordinate satellite transit plans, emergency drone flights, and high-frequency data collection from ground sensors.

[0051] Full-element data acquisition: Simultaneously collect the spatiotemporal distribution of rainfall, the process line of outflow and sediment content, and the dynamic change sequence of internal soil moisture throughout the entire event process.

[0052] Spatial change detection: Using UAV imagery before and after the event, a high-precision DEM differential map (DoD) is generated through photogrammetry to quantify the spatial distribution of erosion / deposition.

[0053] Step 6: Perform initial mechanism diagnosis and atlas update Run candidate mechanism combinations: based on current conditions Select multiple reasonable combinations of candidate mechanisms from the mechanism library. Run the simulations separately.

[0054] According to the formula: Calculate the cost function For each combination, calculate its overall fitting error. Spatial consistency error and complexity penalty ,in: Overall Fitting Error Term Used to measure the combination of mechanisms The degree of overall agreement between the model simulation results and the observation data of the watershed outlet section is specifically defined as: These are the goodness-of-fit functions for the flow rate process, the goodness-of-fit function for the sediment concentration process, and the goodness-of-fit function for the total sediment transport, respectively. These are all weighting coefficients, manually set based on data reliability and management needs, and must meet the following requirements: The observation data at the outlet section of the basin should include at least the flow rate, sediment concentration, and sediment transport. Spatial consistency error term This is used to penalize combinations of mechanisms that fit well only at the watershed outlet but whose internal spatial processes do not match observed distribution patterns, thereby forcing the model to select mechanisms that conform to actual spatial physical processes. This is calculated by comparing simulated and observed spatial variation maps of soil moisture content or erosion and deposition distribution maps, and is specifically defined as: These are error terms based on the spatial pattern of soil moisture content variation, the spatial distribution of erosion / deposition, and the spatial pattern of water and sediment process pathways or responses. These are all weighting coefficients used to reflect the reliability and importance of different spatial data sources, and satisfy the following conditions: ; Complexity penalty This approach, while ensuring good model fit, follows Occam's razor principle, favoring simpler, fewer-parameter, and physically more stable mechanism combinations to enhance the model's generalization ability and interpretability. Specifically, it is defined as follows: ,in, The penalty term based on information criteria, used to penalize the number of parameters, is defined as follows: When using mechanism combination At that time, the total number of effective parameters of the model, A penalty coefficient set manually; The entropy penalty term, used to encourage the sparsity or concentration of weights within the mechanism combination, is defined as: When using mechanism combination At time, the dynamic weight of the j-th mechanism module, where m is the total number of mechanism modules. To prevent the occurrence And artificially set extremely small positive numbers; The penalty term, based on prior knowledge, is used to encourage the selection of more reliable or commonly used mechanistic modules and is defined as follows: To adopt mechanism combination At that time, the reliability score of the j-th mechanism module is obtained from expert evaluation or historical performance statistics. Let be the indicator function, representing the dynamic weight of the j-th mechanistic module. Exceeding the preset threshold Effective immediately; These are all regularization coefficients, used to balance the strength of penalty terms with different complexities, and satisfy the following conditions: .

[0055] Determine the optimal combination based on the computational structure. The determining principle is: choose the option that makes the cost function... The smallest combination of mechanisms.

[0056] The conditional-mechanism graph is updated through event-driven Bayesian learning, and the update process is represented as an iteration of the probability distribution: ,in, Representative mechanism module combination; Represents the condition vector; Monitoring data representing new hydrological events; It is the likelihood function calculated based on the goodness of fit between simulation and observation, expressed as: And will share this experience Integrate into the atlas and complete the initial learning. Generate atlas V1.0.

[0057] Phase Three: Continuous Learning and Dynamic Simulation, specifically: Step 7: Routine monitoring and forecast triggering Lightweight daily operation: The monitoring network continuously collects data while operating with low power consumption.

[0058] Forecast scenario triggering: Receive meteorological grid forecast data.

[0059] By combining real-time monitoring data, the watershed state condition vector at the forecast time is calculated. .

[0060] Query the current condition - mechanism map, and sample or select the most likely combination of mechanisms based on the probability distribution. .

[0061] Dynamically assemble the forecast model, run the ensemble forecast, and generate future water and sediment processes and the range of uncertainties.

[0062] Step 8: Post-event analysis and graph evolution Repeat steps 5 and 6 after each actual hydrological event occurs.

[0063] Utilizing new data And newly determined Based on the event-driven Bayesian learning formula described above, the graph is iteratively updated, enabling continuous evolution of the graph from V1.0 → V2.0 → … → Vn.0.

[0064] Step 9: Graph Self-Organization and Knowledge Extraction Perform regularly (e.g., quarterly): Dimensionality reduction clustering: Use an autoencoder to reduce the dimensionality of the conditional vector {C} of all historical events and perform clustering analysis in a low-dimensional manifold space.

[0065] Identify response prototypes: Each cluster represents a typical combination of "watershed state-rainfall characteristics", that is, a water and sediment response prototype.

[0066] Correlation of dominant mechanisms: Analyze the historical best combination of mechanisms for each prototype to identify its dominant mechanism module.

[0067] Simplified diagram: "If the condition belongs to prototype A, then the mechanism combination should be used preferentially." The rule simplifies complex probability graphs and improves query efficiency and interpretability.

[0068] Phase 4: Advanced Applications and System Maintenance Step 10: Uncertainty Quantification and Decision Support Ensemble forecast generation: During the forecasting in step 7, multiple high-probability mechanism combinations and parameters are sampled from the spectrum to form a forecast ensemble.

[0069] Variance decomposition: Statistical analysis of ensemble forecast results decomposes the total forecast variance into three types of uncertainty contributions: structural uncertainty, parameter uncertainty, and forced uncertainty.

[0070] Product generation: Provide users with probability forecast products and uncertainty source tracing reports to support risk decision-making.

[0071] Step 11: System Performance Evaluation and Module Library Maintenance Periodic evaluation: Compare the simulation accuracy, forecast lead time, and rationality of mechanism explanation of the system of this invention with traditional fixed models in historical extreme events.

[0072] Mechanism module performance optimization: The performance of each module is tracked over a long period under different conditions. Modules with consistently poor performance are downgraded or flagged; when the system frequently identifies unexplained errors, the development of a new mechanism module is triggered.

[0073] Parameter library update: Utilize accumulated data to update the posterior distribution of parameters in each module, thereby improving the rationality of parameter values.

[0074] Phase 5: Expansion and Application Step 12: Transfer Learning and Rapid Deployment in New Watersheds Select a source watershed: Select a watershed where the system is already running successfully as the knowledge source.

[0075] Computational transferability: By comparing the similarities between the target watershed and the source watershed in terms of topography, climate, underlying surface, etc., the transferability of "condition-mechanism" knowledge is assessed.

[0076] Map migration and fine-tuning: Using the map of the source watershed as the initial map of the target watershed, and using a small amount of initial monitoring data from the target watershed for rapid fine-tuning, the "learning" cycle of the new watershed system is significantly shortened.

[0077] Example 2: This invention also provides a system structure for implementing the above-mentioned method for monitoring water and sediment dynamics across a small watershed, see [link to relevant documentation]. Figure 2 The system architecture diagram shown includes: The data sensing and fusion module is used to establish and connect to the integrated air-space-ground monitoring network of the small watershed, acquire and fuse the multi-source monitoring data collected by the monitoring network, and construct a high-resolution digital foundation for the small watershed. The mechanism library module stores multiple water and sediment process mechanism modules with standardized input and output interfaces. Each mechanism module represents a verified physical process description. The dynamic model assembly and simulation engine, connected to the data perception and fusion module and the mechanism library module, is configured as follows: Extract the current watershed status conditions based on the currently acquired multi-source monitoring data across the entire region; Based on the current watershed conditions, at least one mechanism module is dynamically selected and combined from the mechanism library module to assemble a watershed water and sediment process simulation model adapted to the current conditions. Run the assembled watershed water and sediment process simulation model to generate dynamic simulation or forecast results of water and sediment; The learning and optimization module, connected to the dynamic model assembly and simulation engine and the data awareness and fusion module, is configured as follows: Based on the differences between the simulation results generated by the watershed water and sediment process simulation model and the subsequent actual monitoring data, feedback optimization is performed on the selection and combination rules of mechanism modules in the dynamic model assembly and simulation engine, and the mapping relationship between watershed state conditions and mechanism module combinations is updated. The uncertainty quantification and output module, connected to the dynamic model assembly and simulation engine, is used to quantify the uncertainty of water and sediment dynamic simulation or forecast results and output visualization results with uncertainty information.

[0078] Example 3:

[0079] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the above-described process steps of a method for dynamic monitoring of water and sediment across a small watershed.

[0080] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring water and sediment dynamics across a small watershed, characterized in that, Includes the following steps: S1: Establish a high-resolution digital infrastructure and an integrated air-space-ground monitoring network for small watersheds to obtain multi-source monitoring data across the entire region; S2: Construct a mechanism library containing multiple interchangeable water and sediment process mechanism modules, where each mechanism module represents a verified physical process description; S3: Based on the multi-source monitoring data of the whole domain and the current watershed state conditions, dynamically select and combine at least one mechanism module from the mechanism library to assemble a watershed water and sediment process simulation model for the current or forecast scenario; S4: Run the assembled watershed water and sediment process simulation model to generate water and sediment dynamic simulation or forecast results under different scenarios, and quantify the uncertainty of the results; S5: Based on the differences between the actual monitoring data and simulation results obtained subsequently, the selection and combination rules of the mechanism module are optimized by feedback, and the mapping relationship between the watershed state conditions and the dynamic mechanism of water and sediment is updated.

2. The method for monitoring water and sediment dynamics across a small watershed according to claim 1, characterized in that, Step S3 specifically includes: The definition includes at least a watershed state condition vector consisting of rainfall characteristics, anterior soil moisture, vegetation status, and topographic attributes; Define and maintain the optimal or high-probability combination of mechanism modules under different watershed state condition vectors to form a condition-mechanism map; Based on the current watershed state condition vector acquired or predicted in real time, the condition-mechanism map is queried to determine the combination of mechanism modules to be used.

3. The method for monitoring water and sediment dynamics across a small watershed according to claim 2, characterized in that, The conditional-mechanism map is updated through event-driven Bayesian learning, and the update process is represented by an iteration of the probability distribution: ,in, C represents the combination of mechanism modules; C represents the condition vector. Monitoring data representing new hydrological events; It is the likelihood function calculated based on the goodness of fit between simulation and observation, expressed as: Let be the cost function.

4. The method for monitoring water and sediment dynamics across a small watershed according to claim 3, characterized in that, Also includes: After each complete hydrological event, a cost function for multiple candidate mechanism combinations is calculated based on the multi-source monitoring data of the entire region; The mechanism combination that minimizes the cost function is selected as the optimal combination for this hydrological event and used to update the condition-mechanism map.

5. The method for monitoring water and sediment dynamics across a small watershed according to claim 4, characterized in that, Step S4 specifically includes: Multiple possible mechanism combinations and corresponding parameters are sampled from the condition-mechanism map to form an ensemble forecast; Run the model corresponding to each mechanism combination separately to obtain the simulation set of water and sediment processes; Analyzing the statistical characteristics of the simulation set, the total forecast variance is decomposed into structural uncertainty, parameter uncertainty, and forced uncertainty.

6. The method for monitoring water and sediment dynamics across a small watershed according to claim 4, characterized in that, The cost function includes at least an overall fitting error term. Spatial consistency error term and complexity penalty terms Defined as: ,in, All are weighting coefficients, and satisfy the following conditions: .

7. The method for monitoring water and sediment dynamics across a small watershed according to claim 6, characterized in that, The overall fitting error term Used to measure the combination of mechanisms The degree of overall agreement between the model simulation results and the observation data of the watershed outlet section is specifically defined as: These are the goodness-of-fit functions for the flow rate process, the goodness-of-fit function for the sediment concentration process, and the goodness-of-fit function for the total sediment transport, respectively. These are all weighting coefficients, manually set based on data reliability and management needs, and satisfying... The observation data at the outlet section of the basin include at least flow rate, sediment concentration, and sediment transport. The spatial consistency error term This is used to penalize combinations of mechanisms that fit well only at the watershed outlet but whose internal spatial processes do not match observed distribution patterns, thereby forcing the model to select mechanisms that conform to actual spatial physical processes. This is calculated by comparing simulated and observed spatial variation maps of soil moisture content or erosion and deposition distribution maps, and is specifically defined as: , These are error terms based on the spatial pattern of soil moisture content variation, the spatial distribution of erosion / deposition, and the spatial pattern of water and sediment process pathways or responses. These are all weighting coefficients used to reflect the reliability and importance of different spatial data sources, and satisfy the following conditions: ; The complexity penalty term This approach, while ensuring good model fit, adheres to Occam's razor principle, favoring simpler, fewer-parameter, and physically more stable mechanism combinations to enhance the model's generalization ability and interpretability. Specifically, it is defined as follows: ,in, The penalty term based on information criteria, used to penalize the number of parameters, is defined as follows: When using mechanism combination At that time, the total number of effective parameters of the model, A penalty coefficient set manually; The entropy penalty term, used to encourage the sparsity or concentration of weights within the mechanism combination, is defined as: When using mechanism combination At that time, the first The dynamic weights of each mechanistic module, where m is the total number of mechanistic modules. To prevent the occurrence And artificially set extremely small positive numbers; The penalty term, based on prior knowledge, is used to encourage the selection of more reliable or commonly used mechanistic modules and is defined as follows: To adopt mechanism combination At that time, the reliability score of the j-th mechanism module is obtained from expert evaluation or historical performance statistics. Let be the indicator function, representing the dynamic weight of the j-th mechanistic module. Exceeding the preset threshold Effective immediately; These are all regularization coefficients, used to balance the strength of penalty terms with different complexities, and satisfy the following conditions: .

8. The method for monitoring water and sediment dynamics across a small watershed according to claim 1, characterized in that, The mechanism module includes at least: The infiltration and precipitation generation module is used to simulate the distribution process of precipitation after it reaches the earth's surface. The slope erosion module is used to simulate the process of soil particles being stripped from the surface and transported on slopes under the action of rainfall splash and slope runoff. The channel sediment transport module is used to simulate the transport, deposition, and resuspension processes of sediment in a defined channel or river network; Road concentrating flow is used to simulate the unique hydrological effects of roads as efficient runoff generation and confluence channels. The engineering measures for sediment retention module is used to simulate the effects of human intervention and abrupt changes in water and sediment processes caused by soil and water conservation projects; Furthermore, each of the aforementioned mechanism modules has a standardized input / output interface.

9. The method for monitoring water and sediment dynamics across a small watershed according to claim 1, characterized in that, The method further includes: An autoencoder is used to map high-dimensional conditional vectors to a low-dimensional manifold space. Clustering of historical event conditions in the low-dimensional manifold space identifies water and sediment response prototypes. Each water and sediment response prototype is associated with a dominant combination of mechanistic modules, thereby simplifying the mapping and enabling knowledge discovery.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method for dynamic monitoring of water and sediment across a small watershed as described in any one of claims 1-9.

Images