Water resource survey data real-time assimilation method based on internet of things and kilometer grid

Abstract

The application discloses a kind of water resource investigation data real-time assimilation methods based on Internet of Things and kilometer grid, comprising: pre-processing to multi-source monitoring data, form kilometer grid data;Quality control is carried out in edge computing node, and hydrological response unit is divided, and state variable is quickly predicted using lightweight machine learning agent model;With DHRU as unit, real-time data assimilation is carried out on edge side;Based on event discrimination, significant change information is uploaded to cloud computing center;Cloud gathers global data and carries out periodic deep assimilation and model parameter global calibration, generates high-precision kilometer grid water resource state product, and the optimized model and parameter are fed back to edge node;Finally, multi-level water resource investigation service product is generated.The application solves the problem that fixed model unit does not match dynamic hydrological process and traditional assimilation calculation efficiency is low, realizes high-precision, high-efficiency real-time sensing and prediction of water resource state.

Description

Technical Field

[0001] This invention relates to the field of hydrological monitoring technology, specifically to a method for real-time assimilation of water resource survey data based on the Internet of Things and kilometer grids. Background Technology

[0002] Currently, the water resources survey and monitoring system relies heavily on sparsely distributed hydrological and meteorological stations for periodic manual or automatic data collection. This results in problems such as insufficient spatiotemporal density of data, long update cycles (usually daily or hourly), and information silos. It is difficult to capture the rapid dynamic changes and spatial heterogeneity of water resources, and cannot meet the needs of modern refined water resources management and emergency response (such as rapid early warning of floods and water pollution incidents).

[0003] While existing data assimilation techniques can integrate observational data into hydrological models to improve simulation accuracy, their application faces two major bottlenecks. First, most assimilation frameworks are based on fixed model spatial units (such as static hydrological response units (HRUs) or regular grids). This static, a priori spatial division cannot accurately characterize the runoff generation and confluence response regions that change in real time with the spatiotemporal distribution of rainfall and underlying surface humidity conditions. This leads to systematic biases in the spatial allocation of observational information, limiting further improvements in assimilation effectiveness. Second, distributed physical hydrological models (such as SWAT and MIKESHE) used to achieve high-precision simulations often have high computational complexity and are extremely time-consuming, making them difficult to embed into real-time assimilation loops requiring minute-level or even second-level responses. Furthermore, excessively simplifying the model in pursuit of efficiency can result in the loss of key physical processes, affecting the reliability and physical consistency of forecasts. Summary of the Invention

[0004] To address this issue, the present invention provides a real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grids, in order to solve the problem of mismatch between fixed model units and dynamic hydrological processes in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for real-time assimilation of water resource survey data based on the Internet of Things and kilometer grids includes the following steps;

[0007] Step 1. Perform collaborative sensing, aggregation, and spatiotemporal standardization processing on multi-source heterogeneous water resource monitoring data to form kilometer grid multi-source data with consistent spatiotemporal benchmarks;

[0008] Step 2. Perform quality control on the multi-source data of the kilometer grid under the jurisdiction of the edge computing node, and dynamically divide the dynamic hydrological response units with similar characteristics based on real-time rainfall, topography and previous soil moisture; use a lightweight machine learning proxy model to quickly predict the key state variables of each unit.

[0009] Step 3. On the edge side, through the set hybrid assimilation engine, with the dynamic hydrological response unit as the basic operation unit and the prediction of the machine learning agent model as the background field, physical constraints are integrated to assimilate real-time observation data into state variables and update the grid state field.

[0010] Step 4. Based on the comparison between the state change and the dynamic threshold or the detection of a specific hydrological event, the edge node is triggered to upload only the state increment, the key parameter adjustment, and the event information to the cloud computing center after compression.

[0011] Step 5. The cloud computing center aggregates all edge incremental and global remote sensing data, runs a high-resolution hydrophysical model, performs periodic deep assimilation and global parameter calibration, and generates a global kilometer grid water resource status benchmark product.

[0012] Step 6. Distribute the calibrated global parameters and updated machine learning agent model from the cloud computing center to the edge nodes to achieve cloud-edge collaborative iterative optimization of the model and parameters;

[0013] Step 7. Based on the benchmark product and the real-time edge status, generate and publish a multi-level water resources survey service product.

[0014] Furthermore, the specific content of step 1 is as follows: real-time collection of water level, flow rate, rainfall, water quality and soil moisture parameters through ground IoT sensor nodes deployed in the target area; synchronous access to aerospace remote sensing inversion data obtained from satellite and UAV platforms; aggregation of all sensing data through communication network, time synchronization and geographic registration, and unified mapping to the preset standard kilometer grid spatial coordinate system.

[0015] Further: In step 2, the method for dividing the dynamic hydrological response unit is as follows: within each standard kilometer grid, extract the real-time rainfall intensity, topographic humidity index, previous soil moisture and topographic gradient features at the sub-pixel level; adopt a dynamic clustering algorithm that considers topographic gradient weighting, with the goal of minimizing the differences in hydrological response within the cluster, dynamically identify and delineate multiple continuous areas within each grid as the dynamic hydrological response unit at the current moment.

[0016] Furthermore: In step 2, the lightweight machine learning agent model is a long short-term memory network, graph neural network, or gradient boosting decision tree model pre-trained in the cloud; the key state variables include soil moisture content and surface water storage depth.

[0017] Furthermore, in step 3, the hybrid assimilation engine includes: a machine learning proxy model prediction module for providing a fast background field, a physical constraint module for embedding physical conservation laws, and an ensemble Kalman filter algorithm module for performing state updates.

[0018] Furthermore: In step 3, during the assimilation process using ensemble Kalman filtering, the state set of each dynamic hydrological response unit is used as the background field. The Kalman gain matrix is ​​calculated using real-time observation data, and the state set of each unit is analyzed and updated. The updated analysis field is further processed through physical constraint rules to ensure that the assimilation result conforms to the laws of mass conservation or energy conservation.

[0019] Furthermore, in step 4, the event determination specifically involves: calculating the relative rate of change of the state variable through a sliding window, and comparing the rate of change with a dynamic threshold that is adaptively adjusted according to the season and weather conditions; when the rate of change exceeds the dynamic threshold, or when the rate of water level rise exceeds the preset flood threshold or the rate of change of water quality parameters exceeds the preset pollution threshold, it is determined as an event that needs to trigger incremental synchronization.

[0020] Furthermore: In step 7, the water resources survey service products include real-time monitoring distribution maps, water resources statistical reports, hydrological forecasting and early warning products, and development and utilization analysis and evaluation reports; the products are released in the following ways: providing standardized geographic information services, visual display through web portals or mobile applications, generating and pushing standardized reports, or integrating with business systems through API interfaces.

[0021] The present invention has the following advantages: By introducing the Dynamic Hydrological Response Unit (DHRU) as the basic operation unit for assimilation, the present invention effectively solves the problem of mismatch between fixed model units and dynamic hydrological processes, enabling the data assimilation process to match the dynamic spatial heterogeneity of actual hydrological processes, and significantly improving the spatial allocation rationality and state update accuracy of observation information.

[0022] Meanwhile, this invention constructs a cloud-edge collaborative computing architecture, deploys a lightweight machine learning agent model on the edge to achieve rapid inference, and combines it with efficient assimilation algorithms (such as ensemble Kalman filtering) to overcome the bottleneck of time-consuming computation of high-complexity physical models. It reduces the assimilation computation time of the entire process from several hours in traditional methods to seconds, meeting the needs of high-frequency real-time perception of water resource status and rapid response to emergencies.

[0023] Other features and advantages of the present invention will be set forth in the following description. Attached Figure Description

[0024] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0025] Figure 1 This is a flowchart illustrating the implementation of the real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grids provided in this application embodiment. Detailed Implementation

[0026] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these embodiments are merely for further explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Those skilled in the art can make some non-essential improvements and adjustments to the present invention based on the above-described content.

[0027] Please see Figure 1 A real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grids includes the following steps:

[0028] Step 1. Perform collaborative sensing, aggregation, and spatiotemporal standardization processing on multi-source heterogeneous water resource monitoring data.

[0029] By deploying various IoT sensor nodes in the target area, water resource status parameters are collected in real time at preset frequencies or event-triggered modes. At the same time, data sources such as satellite remote sensing and drone aerial photography are accessed. All sensing data are uniformly aggregated to the corresponding edge computing nodes and cloud computing centers through IoT gateways or communication networks, and undergo unified spatiotemporal benchmark processing to ensure that all data are mapped to a standard kilometer grid coordinate system and the same time benchmark.

[0030] Specifically, ground-based IoT sensing involves the dense deployment or integration of various existing smart sensor nodes, including radar or pressure level gauges, ultrasonic or Doppler flow meters, tipping bucket or weighing rain gauges, multi-parameter water quality monitors (measuring pH, dissolved oxygen, conductivity, turbidity, ammonia nitrogen, COD, etc.), soil moisture and temperature profile sensors, groundwater pressure level gauges, etc. These smart sensor nodes automatically collect raw data at a preset fixed frequency or by events (such as the start of rainfall or sudden changes in water level).

[0031] Space-based remote sensing involves using data interfaces to access data from satellite remote sensing data sources (such as Landsat, Sentinel-1 / 2 / 3, MODIS, GF series, etc.) in real time or near real time to retrieve information such as surface water area, soil moisture, chlorophyll concentration, and surface temperature. In addition, during critical periods or in key areas, UAV-borne hyperspectral, multispectral, or lidar sensors can be deployed to acquire detailed monitoring data with centimeter- to meter-level resolution.

[0032] All sensed data is transmitted to the data aggregation point via IoT-specific protocols (such as MQTT, CoAP) or communication networks (4G / 5G, LoRa); each data aggregation point is equipped with an edge computing node, and each data aggregation point is typically associated with one or more kilometer grids.

[0033] Then, all raw data are synchronized in time and registered geographically. All point observation data are mapped to a predefined standard kilometer grid spatial coordinate system through spatial interpolation (such as Kriging or inverse distance weighting) or resampling of areal data. At the same time, all data are stamped with a unified timestamp (UTC) to form a multi-source data cube with consistent spatiotemporal reference.

[0034] Step 2. Perform data quality control, intelligent partitioning of dynamic hydrological response units, and rapid inference of lightweight proxy models at edge computing nodes.

[0035] Edge computing nodes receive multi-source data from their managed kilometer grid and neighboring areas. First, they perform data quality control, including outlier detection, missing value imputation, and consistency verification. Then, the dynamic hydrological response unit engine dynamically identifies and divides the smallest computing units with similar runoff generation and confluence characteristics within the standard kilometer grid based on the real-time spatial distribution of rainfall intensity, digital elevation model, and previous soil moisture conditions. These are called dynamic hydrological response units (DHRUs). Each edge node loads and runs a lightweight machine learning agent model, which is pre-trained in the cloud and can quickly predict the key state variables of each DHRU within its responsible grid based on current and recent meteorological driving data.

[0036] The specific content of data preprocessing and quality control (QC) includes: extreme value checking based on physical range, sudden value detection and smoothing based on time series, cross-validation based on multi-sensor consistency, and lightweight imputation for missing data, such as neighbor interpolation or simple linear regression. Among them, the data that passes quality control is marked as trustworthy and enters the subsequent process.

[0037] The dynamic hydrological response unit engine can receive real-time minute-level rainfall grid data (generated by interpolation from IoT rain gauges), high-precision digital elevation models (DEMs) and their derived topographic indices (such as the topographic moisture index TWI), as well as the previous soil moisture field from the previous assimilation period.

[0038] The Dynamic Hydrological Response Unit (DHRU) engine runs a lightweight clustering or segmentation algorithm to identify contiguous regions with similar hydrological response characteristics (such as runoff potential) within each standard kilometer grid. These regions are defined as Dynamic Hydrological Response Units (DHRUs). For example, in a rainfall event, only those sub-regions where the rainfall intensity exceeds the infiltration capacity or the soil is close to saturation will be classified as active DHRUs and participate in the runoff calculation.

[0039] In this embodiment, the K-means algorithm is used to dynamically determine the optimal number of clusters k (based on the contour coefficient criterion) to avoid the boundary effects caused by a fixed grid.

[0040] Assuming each kilometer grid is discretized into m×m sub-pixels, and then the feature vector Z is extracted, then... Where P represents real-time minute-level rainfall grid data; For high-precision digital elevation models, The topographic humidity index derived from the DEM; This represents the soil moisture field from the previous cycle.

[0041] Since the objective of DHRU partitioning is to minimize intra-cluster hydrological response differences while maximizing inter-cluster differences, the weighted K-means objective function is:

[0042] ;

[0043] in, For physical weights, and ; For terrain sensitivity parameters, For TWI gradient; denoted as feature distance; k is the number of DHRUs;

[0044] Select initial centers using K-means++ The formula for the allocation step is:

[0045] ;

[0046] The formula for the update step is: ;

[0047] when If the maximum number of iterations is reached, the iteration update will terminate.

[0048] After each iteration, the water balance is verified, when a certain cluster of Then split the cluster;

[0049] Where Z is the sub-pixel feature vector; Let i be the point set of the i-th DHRU; The feature center of DHRU; Physical weights; This is the convergence threshold; This is the saturation threshold.

[0050] Each edge computing node is preloaded with one or more lightweight ML agent models pre-trained in the cloud, such as a simplified LSTM network, a graph neural network (GNN), or a gradient boosting tree model. The lightweight ML agent model learns a fast mapping relationship from a simplified set of meteorological driving variables to key state variables of DHRU within the grid. Key state variables include soil moisture profile and surface water depth.

[0051] When new driving data is received, the lightweight ML agent model calculates the background field or prior estimate of the state variables for each DHRU within seconds, which serves as a fast prediction starting point for subsequent assimilation.

[0052] Step 3. Perform real-time data assimilation at the edge based on the hybrid assimilation engine.

[0053] Within each edge computing node, a hybrid assimilation engine is constructed. This hybrid assimilation engine uses the DHRU partitioned in step 2 as the basic assimilation unit, the prediction results of the aforementioned lightweight ML proxy model as the background field, and integrates physical constraints, such as water balance equations.

[0054] Specifically, the hybrid assimilation engine integrates three core components: an ML surrogate model prediction module that provides a fast background field; a physical constraint module that embeds core physical laws such as mass conservation and energy conservation as soft constraints or post-processing rules to prevent physically unreasonable assimilation results; and an optimized assimilation algorithm module that preferentially selects Ensemble Kalman Filter (EnKF) or its variants (such as Localized EnKF). EnKF represents uncertainty by maintaining a set of model states, is computationally efficient, and is very suitable for handling nonlinear problems.

[0055] When an edge node receives new valid observation data, such as the latest water level of a certain water level station, it immediately triggers the assimilation process. Assimilation uses DHRU as the basic operation unit. The EnKF algorithm takes the state set provided by the ML surrogate model for each DHRU as the background field, takes the point or area observation data (transformed into the model space by the observation operator) as input, calculates the Kalman gain matrix, and then updates the state set of each DHRU to obtain the analytical field or posterior estimate of the state variables.

[0056] In this embodiment, using DHRU as the unit, the core update equation for the EnKF analysis step is:

[0057] ;

[0058] Where the Kalman gain K is:

[0059] ;

[0060] To ensure that the assimilated state satisfies the physical conservation laws, the post-processing rules for physical constraints are as follows:

[0061] ;

[0062] in, is the set matrix of the analysis state; is the set matrix of the background state; K is the Kalman gain matrix; H is the observation operator matrix; The background error covariance; Let y be the set of perturbed observations; y be the vector of actual observations. These are the projection coefficients for physical constraints; The physical projection matrix;

[0063] The assimilated and updated DHRU state (analysis field) is integrated back into its respective kilometer grid, updating the overall state field of that grid; for example, after updating the soil moisture content of each DHRU, the average soil moisture content of that kilometer grid can be obtained by area-weighted averaging.

[0064] Step 4. Efficient data synchronization based on event discrimination and incremental compression mechanism.

[0065] The edge node's hybrid assimilation engine has a built-in state change monitoring and event discrimination module. It compares the state variables updated in step 3 with the state of the previous period. When the change in the state variables exceeds a preset threshold, or when a specific event such as the arrival of a flood wave front or a sudden change in water quality is detected, the edge node is triggered. It only compresses and encapsulates the incremental data of the state variables, the key parameters after assimilation, and the event markers, and uploads them to the cloud computing center. When the change is gradual, it only updates the state field in the local area, which greatly reduces the network data transmission load.

[0066] Specifically, the built-in intelligent discrimination module compares the change with a preset dynamic threshold, which can be adaptively adjusted according to the season and weather conditions. At the same time, the intelligent discrimination module can also detect specific hydrological event signals, such as the rate of water level rise exceeding the warning value or the sudden deterioration of water quality parameters.

[0067] In this embodiment, a sliding window is used to calculate the rate of change, and fuzzy logic is combined to determine flood / pollution events;

[0068] The time-based discrimination condition is:

[0069] ;

[0070] in, ;

[0071] For example, event signals:

[0072] ;

[0073] ;

[0074] Incremental data packets are ;

[0075] in, It is a relative change; For dynamic thresholds; The baseline threshold; v represents the seasonal decline rate; v represents the water level rise rate. The flood warning threshold; is the pollution warning threshold; c is the change in COD.

[0076] When the change exceeds the threshold or a specific event is detected, it is judged as a "significant change / event" and incremental synchronization is triggered. The edge node only compresses and encapsulates the change (increment) of the state variable, the key local parameters adjusted during the assimilation process (such as the penetration rate correction coefficient of a certain DHRU), and the detailed event description tags, and uploads them to the cloud computing center.

[0077] If the change is below the threshold and there are no events, it is judged as a "stable change". The original data or complete state field is not synchronized, and the updated state is only stored locally. The cloud computing center can know that the node is running normally through heartbeat messages.

[0078] Step 5. Perform periodic global deep data assimilation at the cloud computing center, and simultaneously perform high-precision model parameter calibration.

[0079] The cloud computing center aggregates incremental data and event information from all edge nodes and combines them with big data such as remote sensing inversion products from the entire domain. The cloud computing center runs high spatiotemporal resolution hydrological-hydrodynamic physical models (such as SWAT, MIKESHE, WRF-Hydro, etc.) and performs periodic (such as daily) deep data assimilation at the standard kilometer grid scale. This process utilizes more comprehensive observation information to optimally calibrate the global parameters of the physical model (such as permeability coefficient and Manning roughness) and generate kilometer grid water resource status products that are consistent across the entire domain and have higher accuracy.

[0080] Specifically, the cloud computing center receives incremental data packets and event information from all edge nodes; combined with the low- and medium-frequency remote sensing data products of the entire domain, the cloud computing center first uses the incremental data and the full state of each edge node that was last synchronized to reconstruct the background field of the water resources status of the entire domain at the current moment with high precision.

[0081] On a cloud server cluster, a complete high spatiotemporal resolution distributed hydrophysical model is run. The reconstructed background field is used as a better initial condition to drive the physical model to simulate. Subsequently, using all the observation data collected in the past cycle, including high-frequency edge data and low-frequency satellite data, a more complex and comprehensive data assimilation is performed at the kilometer grid scale. For example, four-dimensional variational assimilation 4D-Var or more complex particle filtering algorithms can be used.

[0082] After deep assimilation and calibration, a set of physically consistent, kilometer-grid water resource status products covering the entire region at the current moment is generated as the Best Estimate for this period.

[0083] Step 6. Optimization, feedback, and iterative model updates for cloud-edge collaboration.

[0084] The cloud computing center dynamically distributes the latest global optimal parameter set of the physical model obtained from step 5, as well as the lightweight ML agent model retrained and updated based on long-term series data of the entire region, to the corresponding edge computing nodes. After receiving the update, the edge nodes immediately load the new model and parameters to replace the old version, so that their real-time prediction and assimilation in the next cycle are based on a more accurate foundation.

[0085] After receiving the update package from the cloud, the edge node can seamlessly switch to the new physical parameters and ML agent model during off-peak hours or as instructed. This ensures that the next real-time prediction and assimilation by the edge node is based on a more accurate model that has been optimized with global data.

[0086] Step 7. Generate multi-level, multi-objective water resources survey service products.

[0087] Based on the optimized kilometer grid water resource status field generated in step 5, and the edge real-time status generated in step 3, the cloud computing center and edge nodes collaborate to generate water resource survey service products that meet different needs.

[0088] Water resources survey service products include testing products, statistical products, forecasting and early warning products, and analysis and evaluation products.

[0089] Monitoring products include real-time / near real-time soil moisture distribution maps, groundwater isohyetal maps, spatial distribution of eutrophication index of water bodies, and flow process lines of important cross sections.

[0090] Statistical products include daily / monthly / annual reports on total water resources calculated by administrative region and river basin, statistics on changes in groundwater storage, and analysis of water quality compliance rates.

[0091] Forecast and early warning products are based on the optimal state after assimilation to start model forecasting, generating short-term flood warnings for the next 6 hours, soil drought forecasts for the next 7 days, and watershed runoff prospects for the next month.

[0092] Analysis and evaluation products include spatial evaluation of water resource development and utilization intensity, estimation of non-point source pollution load, and assessment of the degree of ecological water demand satisfaction.

[0093] The aforementioned products are released in various ways; for example, they form standardized geographic information services (WMS / WFS) for professional GIS systems to call; they are visualized in the form of charts and maps through web portals and mobile apps; fixed-format reports are automatically pushed to the management personnel's office systems; and they are deeply integrated with the business systems of departments such as water affairs and environmental protection through API interfaces.

[0094] The present invention, by introducing Dynamic Hydrological Response Unit (DHRU), fundamentally solves the problem of insufficient representativeness of observation data in fixed model units, enabling the assimilation algorithm to operate on spatial units that better reflect actual hydrological physical processes, thus greatly improving the efficiency of observation information utilization and the accuracy of state updates; at the same time, combined with the physical constraint module, it ensures that the assimilation results strictly follow the basic physical laws.

[0095] In addition, the computationally intensive data assimilation task is decomposed through a cloud-edge collaborative architecture. On the edge side, a lightweight ML proxy model is used to replace most of the heavy physical model integration calculations. Combined with the efficient EnKF algorithm, a real-time assimilation response at the second level is achieved, which can promptly capture sudden hydrological events (such as urban flooding and pollution plume migration). The cloud focuses on periodic deep calibration to provide continuous optimization support for the edge model.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time assimilation of water resource survey data based on the Internet of Things and kilometer grids, characterized in that, Includes the following steps; Step 1. Perform collaborative sensing, aggregation, and spatiotemporal standardization processing on multi-source heterogeneous water resource monitoring data to form kilometer-grid multi-source data with consistent spatiotemporal benchmarks; Step 2. Perform quality control on the multi-source data of the kilometer grid obtained in Step 1 at the edge computing node, and divide the dynamic hydrological response units based on real-time rainfall, topography and previous soil moisture; use a lightweight machine learning proxy model to quickly predict the key state variables of each unit; Step 3. On the edge side, through the set hybrid assimilation engine, the dynamic hydrological response units divided in Step 2 are used as the basic operation units; the key state variables predicted in Step 2 are used as the background field; physical constraints are integrated to assimilate real-time observation data into the state variables and update the grid state field. Step 4. Based on the comparison between the updated state change in Step 3 and the dynamic threshold or the detection of a specific hydrological event, the edge node is triggered to upload only the state increment, key parameter adjustment amount and event information to the cloud computing center after compression. Step 5. The cloud computing center aggregates all edge incremental and global remote sensing data uploaded in Step 4, runs a high-resolution hydrophysical model, performs periodic deep assimilation and global parameter calibration, and generates a global kilometer grid water resource status benchmark product. Step 6. Distribute the calibrated global parameters and updated machine learning agent model from Step 5 to the edge nodes to achieve cloud-edge collaborative iterative optimization of the model and parameters; Step 7. Based on the baseline product generated in Step 5 and the edge real-time status updated in Step 3, generate and publish a multi-level water resources survey service product.

2. The real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grid as described in claim 1, characterized in that, The specific content of step 1 is as follows: Real-time collection of water level, flow rate, rainfall, water quality and soil moisture parameters through ground IoT sensor nodes deployed in the target area; synchronous access to aerospace remote sensing inversion data obtained from satellite and UAV platforms; aggregation of all sensing data through communication network, time synchronization and geographic registration, and unified mapping to the preset standard kilometer grid spatial coordinate system.

3. The real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grid as described in claim 1, characterized in that, In step 2, the method for dividing the dynamic hydrological response units is as follows: within each standard kilometer grid, extract the real-time rainfall intensity, topographic humidity index, previous soil moisture and topographic gradient features at the sub-pixel level; adopt a dynamic clustering algorithm that considers topographic gradient weighting, with the goal of minimizing the differences in hydrological response within the cluster, dynamically identify and delineate multiple continuous areas within each grid as the dynamic hydrological response units at the current moment.

4. The real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grid as described in claim 1, characterized in that, In step 2, the lightweight machine learning agent model is a long short-term memory network, graph neural network, or gradient boosting decision tree model that has been pre-trained in the cloud; the key state variables include soil moisture content and surface water storage depth.

5. The real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grid as described in claim 1, characterized in that, In step 3, the hybrid assimilation engine includes: a machine learning proxy model prediction module for providing a fast background field, a physical constraint module for embedding physical conservation laws, and an ensemble Kalman filter algorithm module for performing state updates.

6. The real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grid as described in claim 5, characterized in that, In step 3, during the assimilation process using ensemble Kalman filtering, the state set of each dynamic hydrological response unit is used as the background field. The Kalman gain matrix is ​​calculated using real-time observation data, and the state set of each unit is analyzed and updated. The updated analysis field is further processed through physical constraint rules to ensure that the assimilation result conforms to the laws of mass conservation or energy conservation.

7. The real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grid as described in claim 1, characterized in that, In step 4, the event determination is specifically as follows: the relative rate of change of the state variable is calculated by a sliding window, and the rate of change is compared with a dynamic threshold that is adaptively adjusted according to the season and weather conditions; when the rate of change exceeds the dynamic threshold, or when the rate of water level rise exceeds the preset flood threshold or the rate of change of water quality parameters exceeds the preset pollution threshold, it is determined as an event that needs to trigger incremental synchronization.

8. The real-time assimilation method for water resource survey data based on the Internet of Things and kilometer grid as described in claim 1, characterized in that, In step 7, the water resources survey service products include real-time monitoring distribution maps, water resources statistical reports, hydrological forecasting and early warning products, and development and utilization analysis and evaluation reports; the products are released in the following ways: providing standardized geographic information services, visual display through web portals or mobile applications, generating and pushing standardized reports, or integrating with business systems through API interfaces.

Images