A dynamic grid update and scheduling method and system for watershed-scale ecological restoration

Abstract

This application provides a dynamic grid update and scheduling method and system for watershed-scale ecological restoration. The system includes a data fusion access module, a dynamic grid generation module, and a central collaborative controller. The data fusion access module receives multi-source heterogeneous data, the dynamic grid generation module constructs an initial watershed grid based on the multi-source heterogeneous data using a quadtree spatial partitioning algorithm, and the central collaborative controller periodically scans the entire watershed grid set, comprehensively determining the information value density and update urgency of each grid based on multi-dimensional indicators. By adopting the above technical solution, the autonomous evolution and precise adaptation of watershed spatial discretization units can be achieved, dynamically balancing the breadth of global monitoring and the accuracy of local characterization, and improving the response sensitivity to sudden ecological events.

Description

Technical Field

[0001] This application relates to the field of environmental engineering technology, and in particular to a dynamic grid update and scheduling method and system for watershed-scale ecological restoration. Background Technology

[0002] Watershed-scale ecological restoration projects are rapidly evolving towards a deep integration of comprehensive collaboration, dynamic response, and intelligent decision-making. These projects are characterized by their wide coverage, strong element coupling, and non-steady-state processes, urgently requiring a high-spatiotemporal-resolution multi-source environmental sensing system and an efficient information processing architecture to achieve precise characterization and dynamic control of key ecological elements such as hydrology, water quality, vegetation, and soil. Against this backdrop, a spatial grid-based digital management paradigm is gradually becoming the mainstream technological approach. By discretizing the watershed into regular or irregular computational units, it unifies the organization of heterogeneous data streams from remote sensing, ground sensing, and model simulation, supporting closed-loop decision-making for restoration measure deployment, effectiveness evaluation, and adaptive adjustments.

[0003] Among them, grid-based systems for watershed management typically employ static or quasi-static partitioning strategies. In the early stages of the project, fixed-resolution grids are pre-set based on topography, administrative divisions, or hydrological units, and a four-layer architecture including data access, storage, analysis, and visualization scheduling is constructed. Such systems have advantages in structural stability and computational reproducibility in macro-planning and periodic assessment, but they are difficult to adapt to the new demands of ecological restoration that are developing towards refinement and real-time monitoring.

[0004] In existing technologies, static grid systems, due to their uniform global resolution, cannot balance the breadth of global coverage with the accuracy of local event characterization under limited computing resources: high resolution leads to an exponential expansion of the data volume across the entire watershed, causing bottlenecks in storage, transmission, and computation timeliness; low resolution sacrifices the ability to capture critical processes such as sudden pollution, flooding, or vegetation mutations. More seriously, existing technologies generally lack a dynamic update mechanism deeply coupled with real-time data streams. The update frequency often relies on manually set cycles or simple threshold triggers, making it difficult to adaptively balance data value density, event urgency, and resource load status, resulting in resource misallocation of "lagging updates in high-value areas and redundant computation in low-activity areas." The fundamental problem lies in the fact that existing methods treat spatial discretization as a one-time pre-modeling step, failing to embed "data-driven" principles into the continuous evolution logic of the grid topology. Summary of the Invention

[0005] This application provides a dynamic grid update and scheduling method and system for watershed-scale ecological restoration, which solves the problem that static or quasi-static grid systems in the prior art cannot achieve both full coverage and local focus. It realizes the autonomous evolution and precise adaptation of watershed spatial discretization units, and dynamically balances the breadth of full-area monitoring with the accuracy of local characterization.

[0006] To achieve the above objectives, the technical solution of this application embodiment is as follows:

[0007] In a first aspect, embodiments of this application provide a dynamic grid update and scheduling system for watershed-scale ecological restoration. The system includes: a data fusion access module, a dynamic grid generation module, and a central collaborative controller; the data fusion access module is connected to the central collaborative controller through the dynamic grid generation module; the central collaborative controller is directly connected to the data fusion access module.

[0008] The data fusion and access module is used to receive and preprocess multi-source heterogeneous data from remote sensing satellites, UAV aerial photography platforms, ground sensor networks, hydrological and meteorological station networks, ecological monitoring buoys, and numerical simulation outputs in real time. The multi-source heterogeneous data is divided into structured monitoring data, semi-structured remote sensing image metadata, and unstructured text logs according to data type.

[0009] The dynamic mesh generation module is used to construct an initial watershed mesh based on the multi-source heterogeneous data using a quadtree spatial partitioning algorithm. The dynamic mesh generation module includes an event-driven local refinement mechanism, which is used to perform recursive subdivision operations on the target area until a preset maximum level depth limit is met or a local information entropy threshold is reached. During the recursive subdivision process, the newly generated sub-mesh inherits the basic attributes of the parent mesh and is supplemented with a timestamp, event type label, and priority weight field. At the same time, a boundary smoothing constraint strategy is adopted to ensure that the resolution difference between adjacent meshes does not exceed one level.

[0010] The grid status assessment module in the central collaborative controller is used to periodically scan the entire watershed grid set, comprehensively determine the information value density and update urgency of each grid based on multi-dimensional indicators; acquire the amount of newly added effective data flowing into each grid area in the past 24 hours, the data type diversity index, the number of anomaly detection alarms, the change rate of ecological sensitive factors, and the implementation records of external intervention measures from the data fusion access module, and determine the dynamic score value Si of each grid through a weighted linear combination model; if the dynamic score value Si of the target grid is higher than twice the global mean by two consecutive assessment cycles, the area where the target grid is located is determined to be a high-priority area and a local refinement and high-frequency update process is triggered; or, if the dynamic score value Si of the target grid is lower than twice the global mean by three consecutive assessment cycles, the area where the target grid is located is determined to be a medium-low priority area and a merging and rollback mechanism is triggered to aggregate multiple fine grids into a coarse grid.

[0011] In one possible implementation, the system further includes a distributed computing scheduler connected to the central collaborative controller.

[0012] The distributed computing scheduler is used to connect a resource pool composed of multiple heterogeneous computing nodes. The resource pool includes CPU-intensive general-purpose servers, GPU-accelerated computing modules, and edge computing terminals. The built-in resource profiling module continuously collects performance indicators, which include the idle memory capacity, available bandwidth, energy consumption level, and task queuing latency of each node.

[0013] Furthermore, the system is configured to dynamically adjust the allocation strategy based on the performance indicators, current grid load, and cooling efficiency, scheduling update tasks for high-priority areas to the GPU-accelerated computing module, and performing remote sensing image interpretation and deep learning inference tasks through the GPU-accelerated computing module; scheduling update tasks for medium- and low-priority areas to the CPU-intensive general-purpose server, and running traditional hydrological models or statistical interpolation algorithms through the CPU-intensive general-purpose server; and assigning local sensor data aggregation and lightweight feature extraction tasks to edge computing terminals in remote areas; wherein the edge computing terminals in remote areas are edge computing terminals whose distance from the CPU-intensive general-purpose server is greater than a preset distance threshold.

[0014] In one possible implementation, the system further includes a visualization feedback module connected to the central collaborative controller;

[0015] The visualization feedback module provides dual access functionality via web and mobile devices, and uses WebGL rendering technology to achieve real-time 3D watershed scene display; it also displays the grid distribution status at any time, heat maps of ecological health indices in various regions, historical evolution animation playback, and restoration project layout suggestions, so that users can easily view them through the interactive interface;

[0016] The visualization feedback module integrates a manual intervention channel, which determines the corresponding key monitoring areas, temporarily increases the update frequency of specific grids, or freezes the automatic merging operation of some areas based on the control commands input by the user. The control commands are directly injected into the front end of the task queue of the central collaborative controller after identity authentication and permission verification.

[0017] In one possible implementation, the data fusion access module has a built-in standardized protocol parsing unit that supports HTTP / HTTPS, MQTT, CoAP and FTP transmission protocols, and maps the original observations to a consistent expression framework under WGS84 geographic coordinate system and UTC timestamp through a unified spatiotemporal reference alignment mechanism.

[0018] The multi-source heterogeneous data is preprocessed by a quality control filter to remove invalid, duplicate, or physically unreasonable data points, and is stored in a columnar database, an object storage cluster, and a full-text search engine, respectively. The quality control filter is used to filter out abnormal values ​​that are higher than the upper limit of soil moisture saturation or lower than the lower limit of dissolved oxygen survival in water.

[0019] In one possible implementation, the initial watershed grid consists of regular square cells ranging from 100 meters to 1000 meters in size. Each grid node contains a unique spatial identifier, central latitude and longitude coordinates, elevation attributes, and the code of its respective hydrological sub-watershed. The resolution of the initial watershed grid is determined based on the watershed area, topographic complexity, and historical ecological disturbance frequency of the target region.

[0020] The dynamic mesh generation module simultaneously activates the neighborhood influence propagation model when performing local refinement operations, and includes related meshes within a 500-meter radius of the target mesh into the update candidate set; the neighborhood influence propagation model constructs a spatial dependency graph based on the water flow direction matrix and vegetation connectivity index;

[0021] The dynamic mesh generation module uses a depth-first traversal strategy to perform recursive subdivision operations. In each recursive subdivision process, the parent mesh is divided into four sub-meshes until the preset maximum level depth limit is 8 or the local information entropy threshold condition is 0.85.

[0022] In one possible implementation, the grid state evaluation module is used to assign differentiated weights to data within different time windows when determining the dynamic score value Si, in conjunction with a time decay factor adjustment strategy; the weight of data within the last hour is set to 1.0, the weight of data within the last 1-6 hours is set to 0.8, and the weight of data within the last 6-24 hours is set to 0.5.

[0023] In one possible implementation, the central collaborative controller is used to coordinate the interaction logic between the dynamic grid generation module, the grid state evaluation module and the distributed computing scheduler, and runs on a primary and backup dual-machine architecture with redundancy and fault tolerance capabilities, and adopts a ZooKeeper-based distributed lock mechanism to ensure the atomicity of critical operations.

[0024] The central collaborative controller is equipped with a task queue manager, which receives update requests and generates a list of tasks to be executed by sorting them by priority. At the same time, it maintains a global grid topology map to record the life cycle status, hierarchical level, parent-child relationship chain, and most recent update completion time of each grid. When performing local refinement or merge rollback operations on the target grid, it synchronously updates the global grid topology map and pushes a resource configuration requirement list to the distributed computing scheduler.

[0025] In one possible implementation, the distributed computing scheduler is configured with an energy consumption-aware scheduling algorithm; the energy consumption-aware scheduling algorithm is used to comprehensively consider the current PUE value of each computing node and the expected execution time of the task during the task assignment phase, and determine the node with the highest unit energy consumption output ratio to execute the job; and to proactively postpone non-urgent tasks to the early morning of the next day during off-peak electricity price periods at night, and switch to performance priority mode during extreme weather warnings.

[0026] Secondly, embodiments of this application provide a dynamic grid update and scheduling method for watershed-scale ecological restoration, applied to a dynamic grid update and scheduling system for watershed-scale ecological restoration as described in the first aspect. The method includes:

[0027] The data fusion access module receives and preprocesses multi-source heterogeneous data in real time from remote sensing satellites, UAV aerial photography platforms, ground sensor networks, hydrological and meteorological station networks, ecological monitoring buoys, and numerical simulation outputs. The multi-source heterogeneous data is divided into structured monitoring data, semi-structured remote sensing image metadata, and unstructured text logs according to data type.

[0028] Based on the aforementioned multi-source heterogeneous data, an initial watershed grid is constructed using a quadtree spatial partitioning algorithm through a dynamic grid generation module. The dynamic grid generation module includes an event-driven local refinement mechanism, which is used to perform recursive subdivision operations on the target region until a preset maximum level depth limit is met or a local information entropy threshold is reached. During the recursive subdivision process, newly generated subgrids inherit the basic attributes of the parent grid and are supplemented with timestamp markers, event type labels, and priority weight fields. At the same time, a boundary smoothing constraint strategy is adopted to ensure that the resolution difference between adjacent grids does not exceed one level.

[0029] The grid status assessment module in the central collaborative controller periodically scans the entire watershed grid set and comprehensively determines the information value density and update urgency of each grid based on multi-dimensional indicators. It acquires the amount of newly added effective data flowing into each grid area within the past 24 hours, the data type diversity index, the number of anomaly detection alarms, the change rate of ecological sensitive factors, and the implementation records of external intervention measures from the data fusion access module. A weighted linear combination model is used to determine the dynamic score Si for each grid. If the dynamic score Si of the target grid is higher than twice the global mean standard deviation for two consecutive assessment periods, the area where the target grid is located is determined to be a high-priority area, and a local refinement and high-frequency update process is triggered. Alternatively, if the dynamic score Si of the target grid is lower than twice the global mean standard deviation for three consecutive assessment periods, the area where the target grid is located is determined to be a medium-low priority area, and a merging and rollback mechanism is triggered to aggregate multiple fine grids into a coarse grid.

[0030] In one possible implementation, the method further includes:

[0031] Based on the update request, the central collaborative controller generates a task list according to priority, synchronously updates the global grid topology map, and pushes resource configuration requirements to the distributed computing scheduler.

[0032] The distributed computing scheduler continuously collects performance metrics, including the free memory capacity, available bandwidth, energy consumption level, and task queuing latency of each node.

[0033] Based on the performance indicators, current grid load, and cooling efficiency, the distributed computing scheduler dynamically adjusts the allocation strategy. Update tasks in high-priority areas are scheduled to GPU-accelerated computing modules, which then perform remote sensing image interpretation and deep learning inference tasks. Update tasks in medium- and low-priority areas are scheduled to CPU-intensive general-purpose servers, which then run traditional hydrological models or statistical interpolation algorithms. Edge computing terminals in remote areas are responsible for local sensor data aggregation and lightweight feature extraction. These remote edge computing terminals are those located at a distance greater than a preset distance threshold from the CPU-intensive general-purpose servers.

[0034] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0035] (1) By constructing a data-driven dynamic grid generation mechanism and a multi-dimensional grid state evaluation system, the autonomous evolution and precise adaptation of the spatial discretization unit of the watershed are realized, and the breadth of global monitoring and the accuracy of local characterization are dynamically balanced under the constraints of limited computing resources.

[0036] (2) By introducing a time decay factor adjustment strategy and a neighborhood influence propagation model, the response sensitivity to sudden ecological events and the ability to capture the upstream and downstream linkage effects of ecological processes are significantly improved.

[0037] (3) Adopting a central collaborative control architecture and heterogeneous resource scheduling strategy, the computing resources are adaptively allocated based on task priority and node performance indicators, effectively avoiding resource mismatch and reducing overall operating costs;

[0038] (4) An edge intelligent pre-screening unit is integrated into the data fusion access module. Anomaly detection and feature compression are achieved before data upload through a lightweight convolutional neural network model, which greatly reduces the load on the backbone network. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments of this application or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A block diagram of a dynamic grid update and scheduling system for watershed-scale ecological restoration provided in this application embodiment;

[0041] Figure 2 A block diagram of another watershed-scale ecological restoration dynamic grid update and scheduling system provided in this application embodiment;

[0042] Figure 3 A flowchart of a distributed computing scheduler provided in this application embodiment;

[0043] Figure 4 A flowchart of a visual feedback module provided in this application embodiment;

[0044] Figure 5 A flowchart illustrating the workflow of a data fusion access module provided in this application embodiment;

[0045] Figure 6 A flowchart illustrating the workflow of a dynamic mesh generation module provided in this application embodiment;

[0046] Figure 7 A flowchart of a grid state assessment module provided in this application embodiment;

[0047] Figure 8 A flowchart of a central collaborative controller provided in this application embodiment;

[0048] Figure 9 A flowchart of a dynamic grid update and scheduling method for watershed-scale ecological restoration provided in this application embodiment;

[0049] Figure 10 A flowchart of another dynamic grid update and scheduling method for watershed-scale ecological restoration provided in this application embodiment. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0051] In the description of the embodiments of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the embodiments of this application and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. The terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this application according to the specific circumstances.

[0052] As the background technology content shows, how to construct an intelligent mechanism that can autonomously identify key areas, dynamically reconstruct grid resolution, and collaboratively optimize computing resource scheduling based on the spatiotemporal characteristics of multi-source big data and the evolution law of ecological processes has become a core technical problem that urgently needs to be solved in this field. Therefore, this application provides a dynamic grid update and scheduling method and system for watershed-scale ecological restoration, which is used to realize the autonomous evolution and accurate adaptation of watershed spatial discretization units, dynamically balance the breadth of global monitoring and the accuracy of local characterization under the constraint of limited computing resources, improve the response sensitivity to sudden ecological events, and effectively avoid resource misallocation and reduce overall operating costs.

[0053] Figure 1 This is a block diagram of a dynamic grid update and scheduling system for watershed-scale ecological restoration provided in an embodiment of this application. Figure 1 As shown, the system may include: a data fusion access module 110, a dynamic grid generation module 120, and a central collaborative controller 130; the data fusion access module 110 is connected to the central collaborative controller 130 through the dynamic grid generation module 120; the central collaborative controller 130 is directly connected to the data fusion access module 110.

[0054] The data fusion access module 110 is used to receive and preprocess multi-source heterogeneous data from remote sensing satellites, UAV aerial photography platforms, ground sensor networks, hydrological and meteorological station networks, ecological monitoring buoys and numerical simulation outputs in real time. The multi-source heterogeneous data is divided into structured monitoring data, semi-structured remote sensing image metadata and unstructured text logs according to data type.

[0055] The dynamic mesh generation module 120 is used to construct an initial watershed mesh based on the multi-source heterogeneous data using a quadtree spatial partitioning algorithm. The dynamic mesh generation module includes an event-driven local refinement mechanism, which is used to perform recursive subdivision operations on the target area until the preset maximum level depth limit is met or the local information entropy threshold condition is reached. During the recursive subdivision process, the newly generated sub-mesh inherits the basic attributes of the parent mesh and is supplemented with timestamp markers, event type labels and priority weight fields. At the same time, a boundary smoothing constraint strategy is adopted to ensure that the resolution difference between adjacent meshes does not exceed one level.

[0056] The grid status assessment module 131 in the central collaborative controller 130 is used to periodically scan the entire watershed grid set, comprehensively determine the information value density and update urgency of each grid based on multi-dimensional indicators; acquire the amount of newly added effective data, data type diversity index, number of anomaly detection alarms, change rate of ecological sensitive factors, and implementation records of external intervention measures flowing into each grid area in the past 24 hours from the data fusion access module, and determine the dynamic score value Si of each grid through a weighted linear combination model; if the dynamic score value Si of the target grid is higher than twice the global mean by two consecutive assessment cycles, the area where the target grid is located is determined to be a high-priority area and a local refinement and high-frequency update process is triggered; or, if the dynamic score value Si of the target grid is lower than twice the global mean by three consecutive assessment cycles, the area where the target grid is located is determined to be a medium-low priority area and a merging and rollback mechanism is triggered to aggregate multiple fine grids into a coarse grid.

[0057] Based on the above technical solution, by constructing a data-driven dynamic grid generation mechanism and a multi-dimensional grid state evaluation system, the autonomous evolution and precise adaptation of watershed spatial discretization units are realized, and the breadth of global monitoring and the accuracy of local characterization are dynamically balanced under the constraints of limited computing resources.

[0058] Figure 2 A block diagram of another dynamic grid update and scheduling system for watershed-scale ecological restoration provided in this application embodiment. (Refer to...) Figure 2The system may further include: a distributed computing scheduler 140 connected to the central collaborative controller 130; the distributed computing scheduler 140 connects to a resource pool composed of multiple heterogeneous computing nodes, the resource pool including CPU-intensive general-purpose servers, GPU-accelerated computing modules, and edge computing terminals, and has a built-in resource profiling module that continuously collects performance indicators; these performance indicators include the free memory capacity, available bandwidth, energy consumption level, and task queuing latency of each node; and, based on these performance indicators, the system dynamically adjusts the allocation strategy according to the current power grid load and cooling efficiency, scheduling update tasks in high-priority areas to the GPU-accelerated computing module, and executing remote sensing image interpretation and deep learning inference tasks through the GPU-accelerated computing module; scheduling update tasks in medium- and low-priority areas to the CPU-intensive general-purpose servers, and running traditional hydrological models or statistical interpolation algorithms through the CPU-intensive general-purpose servers; and assigning local sensor data aggregation and lightweight feature extraction tasks to edge computing terminals in remote areas; the edge computing terminals in remote areas are those whose distance from the CPU-intensive general-purpose servers is greater than a preset distance threshold. For example, the range of the preset distance threshold can be between 100 kilometers and 200 kilometers, such as 100 kilometers, 150 kilometers or 200 kilometers, etc., without limitation here; the distance between the aforementioned edge computing terminal and the CPU-intensive general-purpose server can be the straight-line distance between the two.

[0059] In some embodiments, the resource pool of the distributed computing scheduler 140 includes 16 CPU-intensive general-purpose servers, 8 GPU-accelerated computing modules, and 8 edge computing terminals. The scheduler continuously collects performance metrics of each node at 10-second intervals. The GPU-accelerated computing module can be a high-performance terminal equipped with an NVIDIA A100 Tensor Core GPU.

[0060] For example, refer to Figure 3 The flowchart of the distributed computing scheduler is shown. The distributed computing scheduler 140 may include a resource profiling module and heterogeneous computing nodes, which are used to continuously collect performance indicators and dynamically adjust the allocation strategy of computing nodes.

[0061] In one possible implementation, refer to Figure 2The system may further include: a visualization feedback module 150, which is connected to the central collaborative controller 130; the visualization feedback module 150 provides dual-channel access via web and mobile terminals, and uses WebGL rendering technology to achieve real-time 3D watershed scene display; it also displays the grid distribution status at any time, heat maps of ecological health indices in various regions, historical evolution animation playback, and restoration project layout suggestions, so that users can view them through the interactive interface; the visualization feedback module 150 integrates a manual intervention channel, which determines the corresponding key monitoring areas, temporarily increases the update frequency of specific grids, or freezes the automatic merging operation of some areas based on the control commands input by the user; the control commands are directly injected into the task queue front end of the central collaborative controller after identity authentication and permission verification.

[0062] In some embodiments, the user's control instructions can be verified through two-factor authentication or RBAC permission model. These control instructions can be used to specify key monitoring areas, temporarily increase the update frequency of specific grids to once every 5 minutes, or freeze the automatic merging operation of some areas.

[0063] For example, refer to Figure 4 The workflow diagram of the visualization feedback module 150 includes web and mobile access, real-time 3D watershed scene display, grid distribution status, ecological health index heat map, historical evolution animation and restoration project layout suggestions, manual intervention channel, RBAC permission model verification, and task queue front-end injection, used to realize scene display and manual intervention.

[0064] In one possible implementation, the data fusion access module 110 has a built-in standardized protocol parsing unit that supports HTTP / HTTPS, MQTT, CoAP, and FTP transmission protocols. It maps the original observations to a consistent expression framework under the WGS84 geographic coordinate system and UTC timestamps through a unified spatiotemporal benchmark alignment mechanism. The multi-source heterogeneous data is preprocessed by a quality control filter to remove invalid, duplicate, or physically unreasonable data points, and is stored in a columnar database, an object storage cluster, and a full-text search engine, respectively. The quality control filter is used to filter out abnormal values ​​that are higher than the upper limit of soil moisture saturation or lower than the lower limit of dissolved oxygen survival in water.

[0065] For example, you can refer to Figure 5 The flowchart of the data fusion access module is as follows. The data fusion access module 110 includes a standardized protocol parsing unit, a spatiotemporal reference alignment mechanism, a quality control filter, and classification storage, which are used to realize the reception, preprocessing, and classification storage of data.

[0066] In one possible implementation, the initial watershed grid consists of regular square cells ranging from 100 to 1000 meters in size. Each grid node contains a unique spatial identifier, central latitude and longitude coordinates, elevation attributes, and the code of its corresponding hydrological sub-watershed. The resolution of the initial watershed grid is determined based on the watershed area, topographic complexity, and historical ecological disturbance frequency of the target region. The dynamic grid generation module simultaneously activates a neighborhood influence propagation model during local refinement operations, incorporating associated grids within a 500-meter radius of the target grid into the update candidate set. This neighborhood influence propagation model constructs a spatial dependency graph based on the flow direction matrix and vegetation connectivity index. The dynamic grid generation module employs a depth-first traversal strategy to perform recursive subdivision operations, dividing the parent grid into four equal sub-grids in each recursive subdivision process, until the preset maximum depth limit is 8 levels or the local information entropy threshold condition is 0.85.

[0067] For example, you can refer to Figure 6 The flowchart illustrates the workflow of the dynamic mesh generation module 120, which may include an event-driven local refinement mechanism and a boundary smoothing constraint strategy to achieve recursive mesh subdivision and resolution adjustment.

[0068] In one possible implementation, the grid state assessment module 131 is used to combine a time decay factor adjustment strategy when determining the dynamic score value Si to assign differentiated weights to data in different time windows; the weight of data within the last hour is set to 1.0, the weight of data within the last 1-6 hours is set to 0.8, and the weight of data within the last 6-24 hours is set to 0.5.

[0069] For example, you can refer to Figure 7 The flowchart of the grid state assessment module 131 is shown. This grid state assessment module 131 may include a weighted linear combination model and a time decay factor adjustment strategy to achieve dynamic adjustment of the score value.

[0070] In one possible implementation, the central coordinating controller 130 coordinates the interaction logic between the dynamic grid generation module, the grid state assessment module, and the distributed computing scheduler. It operates on a dual-machine architecture with redundancy and fault tolerance, and employs a ZooKeeper-based distributed lock mechanism to ensure the atomicity of critical operations. The central coordinating controller 130 internally includes a task queue manager. This manager receives update requests, prioritizes them to generate a list of tasks to be executed, and maintains a global grid topology map recording the lifecycle state, hierarchy, parent-child relationship chain, and most recent update completion time of each grid. When performing local refinement or merge / rollback operations on a target grid, it synchronously updates the global grid topology map and pushes a resource configuration requirement list to the distributed computing scheduler.

[0071] For example, you can refer to Figure 8 The flowchart of the central coordinating controller 130 is shown. The central coordinating controller 130 may include a task queue manager and a global grid topology map for generating a list of tasks to be executed and maintaining the global grid topology map.

[0072] In one possible implementation, the distributed computing scheduler 140 is configured with an energy consumption-aware scheduling algorithm. This energy consumption-aware scheduling algorithm is used to comprehensively consider the current PUE value of each computing node and the expected execution time of the task during the task assignment phase, and to determine the node with the highest unit energy consumption output ratio to execute the job. It also proactively postpones non-urgent tasks to the early morning of the next day during off-peak electricity price periods at night, and switches to a performance-priority mode during extreme weather warnings.

[0073] The workflow of the technical solution of this application will be described below with reference to an embodiment. In the ecological restoration demonstration area of ​​a typical watershed in the middle reaches of the Yangtze River, this system receives and preprocesses monitoring information from multi-source heterogeneous data streams in real time through a data fusion access module.

[0074] The data fusion access module is deployed in the data center of the watershed management center, equipped with dual 10 Gigabit fiber optic network interfaces and redundant power supplies. It continuously receives multispectral images from remote sensing satellites, high-resolution orthophotos from UAV aerial photography platforms, soil moisture and water quality parameters transmitted from ground sensor networks, precipitation and evaporation data recorded by hydrological and meteorological stations, chlorophyll concentration indices in water bodies uploaded by ecological monitoring buoys, and runoff prediction results output by numerical simulation systems. The data fusion access module has a built-in standardized protocol parsing unit to identify and decode the input data streams. It supports HTTP protocol for batch downloading satellite data, HTTPS protocol to ensure the security of sensitive data transmission from meteorological stations, MQTT protocol for handling high-frequency pushes from ground sensors, CoAP protocol for low-power communication of edge devices, and FTP protocol for transferring large remote sensing image files. All raw observation data are processed through a unified spatiotemporal benchmark alignment mechanism, which uniformly converts spatial coordinates from different sources to the WGS84 geographic coordinate system and uniformly calibrates timestamps to UTC standard time, forming a spatiotemporally consistent data representation framework. The quality control filter performs multi-level verification on the input data, eliminating invalid data points caused by sensor malfunctions, identifying and deleting duplicate records generated during transmission, and automatically filtering out abnormal values ​​exceeding the upper limit of soil moisture saturation or falling below the lower limit of dissolved oxygen survival in water bodies by setting reasonable threshold ranges based on physical laws. After processing, the data is classified and stored according to structural characteristics: structured monitoring data, including sensor readings, meteorological elements, and other field-based information, is stored in a columnar database; semi-structured remote sensing image metadata, including tagged attributes such as shooting time and spatial range, is stored in an object storage cluster; and unstructured text logs, including equipment operation records and manual inspection reports, are stored in a full-text search engine.

[0075] The dynamic grid generation module constructs an initial watershed grid system based on a quadtree spatial partitioning algorithm. Considering the 5000 square kilometer area of ​​the demonstration watershed, its low to medium topographic complexity, and the frequency of historical ecological disturbances, the initial grid resolution is set to a 500-meter regular square cell. Each grid node contains a 32-bit unique spatial identifier, center latitude and longitude coordinates stored in double-precision floating-point numbers, elevation attributes extracted from the digital elevation model, and hydrological sub-watershed codes defined by the water resources department. This dynamic grid generation module introduces an event-driven local refinement mechanism. Upon receiving a local refinement command, it automatically performs recursive subdivision operations on the grid of the target area. The subdivision process employs a depth-first traversal strategy, recursively dividing the parent grid into four sub-grids each time, until the preset maximum depth limit of 8 levels is met or the local information entropy threshold condition of 0.85 is reached. The newly generated sub-grids inherit the basic attributes of the parent grid and are additionally marked with millisecond-level timestamps, event type labels, and priority weight fields represented by floating-point numbers in the 0-1 range. The boundary smoothing constraint strategy calculates the resolution ratio of adjacent grids to ensure that the level difference between any two adjacent grids does not exceed one level, avoiding abrupt grid changes.

[0076] This dynamic grid generation module simultaneously activates the neighborhood influence propagation model during local refinement operations, including associated grids within a 500-meter radius of the target grid in the update candidate set. This neighborhood influence propagation model constructs a spatial dependency graph based on the flow direction matrix extracted from a 30-meter resolution digital elevation model and the vegetation connectivity index retrieved from Landsat satellite imagery. In this spatial dependency graph, nodes represent grid cells, and edge weights are jointly determined by hydrological connectivity strength and ecological corridor integrity. Adjacent grids with weights exceeding 0.7 automatically trigger a collaborative update mechanism.

[0077] The central collaborative controller is equipped with a grid status assessment module that scans approximately 20,000 grid sets across the entire watershed every 30 minutes. This module extracts five core parameters from the data fusion and access module: the amount of newly added valid data flowing into each grid area within the past 24 hours, the statistical data type diversity index, the number of anomaly detection alarms, the rate of change of ecologically sensitive factors, and the implementation records of external intervention measures. A weighted linear combination model is used to calculate the dynamic score Si for each grid, with the weighting coefficients dynamically adjusted according to watershed management objectives. During the high-water season, hydrological parameters are emphasized, while during the low-water season, ecological indicators are emphasized. Specifically, a time decay factor adjustment strategy is introduced: data from the most recent hour is assigned a full weight of 1.0, data from the 1-6 hour interval is reduced to a weight of 0.8, data from the 6-24 hour interval is further reduced to a weight of 0.5, and data older than 24 hours is no longer included in the scoring. If the dynamic score Si of a certain grid is higher than twice the standard deviation of the global mean for two consecutive evaluation periods, it is marked as a high-priority region and the local refinement and high-frequency update process is triggered; if the dynamic score Si of the target grid is lower than twice the standard deviation of the global mean for three consecutive evaluation periods, the merging and rollback mechanism is initiated to aggregate multiple fine grids into a coarse grid.

[0078] The central coordinating controller, acting as the system's central hub, operates on a primary / standby dual-machine architecture and employs a ZooKeeper-based distributed lock mechanism to ensure the atomicity of critical operations. The internal task queue manager receives update requests from the grid state evaluation unit and generates a list of tasks to be executed, sorted by priority score in descending order. The global grid topology graph records the lifecycle state, hierarchy, parent-child relationship chain, and most recent update completion time of each grid using a graph data structure. When performing a local refinement operation on a grid, the central coordinating controller creates new child nodes and establishes parent-child relationships in the global grid topology graph, while simultaneously pushing a resource configuration requirement list to the distributed computing scheduler. When performing a merge / rollback operation on a grid, the central coordinating controller marks the parent node as active and deregisters the child nodes in the global grid graph, synchronously updating the resource requirement information.

[0079] The distributed computing scheduler connects a heterogeneous resource pool consisting of 32 computing nodes, including 16 CPU-intensive general-purpose servers, 8 GPU-accelerated computing modules, and 8 edge computing terminals. The scheduler has a built-in resource profiling module that continuously collects four performance metrics for each node at 10-second intervals: idle memory capacity, available bandwidth, energy consumption, and task queuing latency. Based on these performance metrics, current grid load, and cooling efficiency, the scheduler dynamically adjusts the allocation strategy. High-priority grid area update tasks are preferentially assigned to GPU-accelerated computing modules—high-performance nodes equipped with NVIDIA A100 Tensor Core GPUs—to perform remote sensing image interpretation and deep learning inference jobs, with each task allocated 4GB of GPU memory and 16 CPU cores. Medium- and low-priority areas are scheduled to CPU-intensive general-purpose server clusters to run SWAT hydrological models or Kriging statistical interpolation algorithms, allocated 2 CPU cores and 8GB of memory. Edge terminals in remote areas are only responsible for local sensor data aggregation and lightweight feature extraction tasks, with CPU utilization limited to no more than 50%.

[0080] The energy consumption-aware scheduling algorithm configured in this distributed computing scheduler takes into account the current PUE value of the computing node (1.2) and the expected execution time of the task during the task assignment phase. It prioritizes the node with the highest unit energy consumption output ratio to execute the job. During off-peak electricity price periods at night, it proactively postpones non-urgent tasks to the early morning of the next day. During extreme weather warnings, it switches to a performance-priority mode.

[0081] The visualization feedback module offers dual access via web and mobile devices, employing WebGL rendering technology to display real-time 3D watershed scenes. Users can view the grid distribution status at any given time through the interactive interface. The ecological health index of each region is presented as a red-yellow-green gradient heatmap, and historical evolution animations support playback over a time range of 1-30 days. Suggestions for restoration project deployment are automatically generated based on grid scoring results. This visualization feedback module also integrates a manual intervention channel. Authorized users can manually specify key monitoring areas after two-factor authentication, temporarily increase the update frequency of specific grids to once every 5 minutes, or freeze automatic merging operations for certain areas. User-submitted commands are directly injected into the task queue front end of the central collaborative controller after verification by the RBAC permission model, forming a closed loop of human-machine collaborative decision-making.

[0082] The workflow of the technical solution of this application will be described below with reference to Embodiment 2. In the alpine meadow ecological restoration area of ​​the upper reaches of the Yellow River Basin, this system adjusts its operating parameters to suit the special environment of high altitude and low temperature.

[0083] The data fusion and access module enhances the ability to identify anomalies in sensor data under low-temperature conditions, adjusting the reasonable range of soil temperature to -5℃ to 15℃ and adding the freezing index as a new quality control indicator. The dynamic grid generation module sets the initial grid resolution to 1000 meters, reducing the data collection density requirements in high-altitude and cold regions; the maximum layer depth limit is adjusted to 6 levels to avoid excessive subdivision leading to wasted computing resources. The grid status assessment module extends the scanning cycle to 1 hour to adapt to the slow ecological processes in high-altitude and cold regions; the dynamic scoring model adds two high-altitude and cold-region characteristic parameters—snow cover rate and permafrost depth—with corresponding adjustments to their weighting coefficients. The distributed computing scheduler prioritizes allocating computing tasks to computing nodes located in low-altitude areas, reducing the impact of the high-altitude and cold environment on hardware; the edge computing terminal adds a low-temperature protection module to ensure normal operation in -20℃ environments. The visualization feedback module specifically integrates a cold-region ecological thematic layer, highlighting the distribution of freeze-thaw erosion risk and the degradation trend of alpine meadows.

[0084] Figure 9 The flowchart of a dynamic grid update and scheduling method for watershed-scale ecological restoration provided in this application embodiment is applied to the aforementioned dynamic grid update and scheduling system for watershed-scale ecological restoration. The method may include the following steps.

[0085] S901, through the data fusion access module, receives and preprocesses multi-source heterogeneous data in real time from remote sensing satellites, UAV aerial photography platforms, ground sensor networks, hydrological and meteorological station networks, ecological monitoring buoys and numerical simulation outputs.

[0086] The multi-source heterogeneous data is divided into structured monitoring data, semi-structured remote sensing image metadata, and unstructured text logs based on data type.

[0087] S902. Based on this multi-source heterogeneous data, an initial watershed grid is constructed using a quadtree spatial partitioning algorithm through a dynamic grid generation module.

[0088] The dynamic mesh generation module includes an event-driven local refinement mechanism, which performs recursive subdivision operations on the target area until the preset maximum level depth limit is met or the local information entropy threshold is reached. During the recursive subdivision process, the newly generated sub-mesh inherits the basic attributes of the parent mesh and is supplemented with timestamp markers, event type labels and priority weight fields. At the same time, a boundary smoothing constraint strategy is adopted to ensure that the resolution difference between adjacent meshes does not exceed one level.

[0089] S903. Through the grid status assessment module in the central collaborative controller, the entire basin grid set is periodically scanned, and the information value density and update urgency of each grid are comprehensively determined based on multi-dimensional indicators.

[0090] Specifically, the data fusion access module acquires the amount of newly added valid data flowing into each grid area within the past 24 hours, the data type diversity index, the number of anomaly detection alarms, the change rate of ecological sensitive factors, and the implementation records of external intervention measures. The dynamic score value Si of each grid is determined through a weighted linear combination model. If the dynamic score value Si of the target grid is higher than twice the standard deviation of the global mean for two consecutive evaluation periods, the area where the target grid is located is determined to be a high-priority area and a local refinement and high-frequency update process is triggered. Alternatively, if the dynamic score value Si of the target grid is lower than twice the standard deviation of the global mean for three consecutive evaluation periods, the area where the target grid is located is determined to be a medium-low priority area and a merging and rollback mechanism is triggered to aggregate multiple fine grids into a coarse grid.

[0091] Based on the above technical solution, by constructing a data-driven dynamic grid generation mechanism and a multi-dimensional grid state evaluation system, the autonomous evolution and precise adaptation of watershed spatial discretization units are realized, and the breadth of global monitoring and the accuracy of local characterization are dynamically balanced under the constraints of limited computing resources.

[0092] Figure 10 A flowchart illustrating another dynamic grid update and scheduling method for watershed-scale ecological restoration provided in this application embodiment. Figure 10 As shown, the method may also include the following steps.

[0093] S904. Based on the update request, the central collaborative controller generates a task list according to priority, synchronously updates the global grid topology map, and pushes resource configuration requirements to the distributed computing scheduler.

[0094] S905 continuously collects performance metrics through a distributed computing scheduler; these metrics include the idle memory capacity, available bandwidth, energy consumption level, and task queuing latency of each node.

[0095] S906. Combining this performance indicator with the current grid load and cooling efficiency, the allocation strategy is dynamically adjusted through this distributed computing scheduler.

[0096] Specifically, update tasks for high-priority areas are scheduled to GPU-accelerated computing modules, which then perform remote sensing image interpretation and deep learning inference. Update tasks for medium- and low-priority areas are scheduled to CPU-intensive general-purpose servers, which then run traditional hydrological models or statistical interpolation algorithms. Edge computing terminals in remote areas are responsible for local sensor data aggregation and lightweight feature extraction. These remote edge computing terminals are those located at a distance greater than a preset distance threshold from the CPU-intensive general-purpose servers.

[0097] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. Each embodiment focuses on describing the differences from other embodiments.

[0098] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of this application.

Claims

1. A dynamic grid update and scheduling system for watershed-scale ecological restoration, characterized in that, The system includes: a data fusion access module, a dynamic grid generation module, and a central collaborative controller; the data fusion access module is connected to the central collaborative controller through the dynamic grid generation module; the central collaborative controller is directly connected to the data fusion access module. The data fusion and access module is used to receive and preprocess multi-source heterogeneous data from remote sensing satellites, UAV aerial photography platforms, ground sensor networks, hydrological and meteorological station networks, ecological monitoring buoys, and numerical simulation outputs in real time. The multi-source heterogeneous data is divided into structured monitoring data, semi-structured remote sensing image metadata, and unstructured text logs according to data type. The dynamic mesh generation module is used to construct an initial watershed mesh based on the multi-source heterogeneous data using a quadtree spatial partitioning algorithm. The dynamic mesh generation module includes an event-driven local refinement mechanism, which is used to perform recursive subdivision operations on the target area until a preset maximum level depth limit is met or a local information entropy threshold is reached. During the recursive subdivision process, the newly generated sub-mesh inherits the basic attributes of the parent mesh and is supplemented with a timestamp, event type label, and priority weight field. At the same time, a boundary smoothing constraint strategy is adopted to ensure that the resolution difference between adjacent meshes does not exceed one level. The grid status assessment module in the central collaborative controller is used to periodically scan the entire watershed grid set, comprehensively determine the information value density and update urgency of each grid based on multi-dimensional indicators; acquire the amount of newly added effective data flowing into each grid area in the past 24 hours, the data type diversity index, the number of anomaly detection alarms, the change rate of ecological sensitive factors, and the implementation records of external intervention measures from the data fusion access module, and determine the dynamic score value Si of each grid through a weighted linear combination model; if the dynamic score value Si of the target grid is higher than twice the global mean by two consecutive assessment cycles, the area where the target grid is located is determined to be a high-priority area and a local refinement and high-frequency update process is triggered; or, if the dynamic score value Si of the target grid is lower than twice the global mean by three consecutive assessment cycles, the area where the target grid is located is determined to be a medium-low priority area and a merging and rollback mechanism is triggered to aggregate multiple fine grids into a coarse grid.

2. The system according to claim 1, characterized in that, The system also includes a distributed computing scheduler, which is connected to the central collaborative controller. The distributed computing scheduler is used to connect a resource pool composed of multiple heterogeneous computing nodes. The resource pool includes CPU-intensive general-purpose servers, GPU-accelerated computing modules, and edge computing terminals. The built-in resource profiling module continuously collects performance indicators, which include the idle memory capacity, available bandwidth, energy consumption level, and task queuing latency of each node. Furthermore, the system is configured to dynamically adjust the allocation strategy based on the performance indicators, current grid load, and cooling efficiency, scheduling update tasks for high-priority areas to the GPU-accelerated computing module, and performing remote sensing image interpretation and deep learning inference tasks through the GPU-accelerated computing module; scheduling update tasks for medium- and low-priority areas to the CPU-intensive general-purpose server, and running traditional hydrological models or statistical interpolation algorithms through the CPU-intensive general-purpose server; and assigning local sensor data aggregation and lightweight feature extraction tasks to edge computing terminals in remote areas; wherein the edge computing terminals in remote areas are edge computing terminals whose distance from the CPU-intensive general-purpose server is greater than a preset distance threshold.

3. The system according to claim 2, characterized in that, The system further includes a visual feedback module, which is connected to the central collaborative controller; The visualization feedback module provides dual access functionality via web and mobile devices, and uses WebGL rendering technology to achieve real-time 3D watershed scene display; it also displays the grid distribution status at any time, heat maps of ecological health indices in various regions, historical evolution animation playback, and restoration project layout suggestions, so that users can easily view them through the interactive interface; The visualization feedback module integrates a manual intervention channel, which determines the corresponding key monitoring areas, temporarily increases the update frequency of specific grids, or freezes the automatic merging operation of some areas based on the control commands input by the user. The control commands are directly injected into the front end of the task queue of the central collaborative controller after identity authentication and permission verification.

4. The system according to claim 3, characterized in that, The data fusion access module has a built-in standardized protocol parsing unit that supports HTTP / HTTPS, MQTT, CoAP and FTP transmission protocols, and maps the original observations to a consistent expression framework under WGS84 geographic coordinate system and UTC timestamp through a unified spatiotemporal reference alignment mechanism. The multi-source heterogeneous data is preprocessed by a quality control filter to remove invalid, duplicate, or physically unreasonable data points, and is stored in a columnar database, an object storage cluster, and a full-text search engine, respectively. The quality control filter is used to filter out abnormal values ​​that are higher than the upper limit of soil moisture saturation or lower than the lower limit of dissolved oxygen survival in water.

5. The system according to claim 3, characterized in that, The initial watershed grid consists of regular square units ranging from 100 meters to 1000 meters. Each grid node contains a unique spatial identifier, central latitude and longitude coordinates, elevation attributes, and the code of its respective hydrological sub-watershed. The resolution of the initial watershed grid is determined based on the watershed area, topographic complexity, and historical ecological disturbance frequency of the target region. The dynamic mesh generation module simultaneously activates the neighborhood influence propagation model when performing local refinement operations, and includes related meshes within a 500-meter radius of the target mesh into the update candidate set; the neighborhood influence propagation model constructs a spatial dependency graph based on the water flow direction matrix and vegetation connectivity index; The dynamic mesh generation module uses a depth-first traversal strategy to perform recursive subdivision operations. In each recursive subdivision process, the parent mesh is divided into four sub-meshes until the preset maximum level depth limit is 8 or the local information entropy threshold condition is 0.

85.

6. The system according to claim 3, characterized in that, The grid state evaluation module is used to assign differentiated weights to data within different time windows when determining the dynamic score value Si, in conjunction with a time decay factor adjustment strategy; the weight of data within the last hour is set to 1.0, the weight of data within the last 1-6 hours is set to 0.8, and the weight of data within the last 6-24 hours is set to 0.

5.

7. The system according to claim 3, characterized in that, The central collaborative controller is used to coordinate the interaction logic between the dynamic grid generation module, the grid status evaluation module and the distributed computing scheduler. It runs on a primary and backup dual-machine architecture with redundancy and fault tolerance, and adopts a ZooKeeper-based distributed lock mechanism to ensure the atomicity of critical operations. The central collaborative controller is equipped with a task queue manager, which receives update requests and generates a list of tasks to be executed by sorting them by priority. At the same time, it maintains a global grid topology map to record the life cycle status, hierarchical level, parent-child relationship chain, and most recent update completion time of each grid. When performing local refinement or merge rollback operations on the target grid, it synchronously updates the global grid topology map and pushes a resource configuration requirement list to the distributed computing scheduler.

8. The system according to claim 7, characterized in that, The distributed computing scheduler is configured with an energy consumption-aware scheduling algorithm; the energy consumption-aware scheduling algorithm is used to comprehensively consider the current PUE value of each computing node and the expected execution time of the task during the task assignment phase, and determine the node with the highest unit energy consumption output ratio to execute the job. Additionally, during off-peak electricity pricing periods at night, non-urgent tasks are proactively postponed until the early hours of the following day, and a performance-priority mode is switched during extreme weather warnings.

9. A dynamic grid update and scheduling method for watershed-scale ecological restoration, characterized in that, A dynamic grid update and scheduling system for watershed-scale ecological restoration, applicable to any one of claims 1 to 8, wherein the method comprises: The data fusion access module receives and preprocesses multi-source heterogeneous data in real time from remote sensing satellites, UAV aerial photography platforms, ground sensor networks, hydrological and meteorological station networks, ecological monitoring buoys, and numerical simulation outputs. The multi-source heterogeneous data is divided into structured monitoring data, semi-structured remote sensing image metadata, and unstructured text logs according to data type. Based on the aforementioned multi-source heterogeneous data, an initial watershed grid is constructed using a quadtree spatial partitioning algorithm through a dynamic grid generation module. The dynamic grid generation module includes an event-driven local refinement mechanism, which is used to perform recursive subdivision operations on the target region until a preset maximum level depth limit is met or a local information entropy threshold is reached. During the recursive subdivision process, newly generated subgrids inherit the basic attributes of the parent grid and are supplemented with timestamp markers, event type labels, and priority weight fields. At the same time, a boundary smoothing constraint strategy is adopted to ensure that the resolution difference between adjacent grids does not exceed one level. The grid status assessment module in the central collaborative controller periodically scans the entire watershed grid set and comprehensively determines the information value density and update urgency of each grid based on multi-dimensional indicators. It acquires the amount of newly added effective data flowing into each grid area within the past 24 hours, the data type diversity index, the number of anomaly detection alarms, the change rate of ecological sensitive factors, and the implementation records of external intervention measures from the data fusion access module. A weighted linear combination model is used to determine the dynamic score Si for each grid. If the dynamic score Si of the target grid is higher than twice the global mean standard deviation for two consecutive assessment periods, the area where the target grid is located is determined to be a high-priority area, and a local refinement and high-frequency update process is triggered. Alternatively, if the dynamic score Si of the target grid is lower than twice the global mean standard deviation for three consecutive assessment periods, the area where the target grid is located is determined to be a medium-low priority area, and a merging and rollback mechanism is triggered to aggregate multiple fine grids into a coarse grid.

10. The method according to claim 9, characterized in that, The method further includes: Based on the update request, the central collaborative controller generates a task list according to priority, synchronously updates the global grid topology map, and pushes resource configuration requirements to the distributed computing scheduler. The distributed computing scheduler continuously collects performance metrics, including the free memory capacity, available bandwidth, energy consumption level, and task queuing latency of each node. Combining the aforementioned performance indicators, current grid load, and cooling efficiency, the distributed computing scheduler dynamically adjusts the allocation strategy; update tasks in high-priority areas are scheduled to GPU-accelerated computing modules, which then perform remote sensing image interpretation and deep learning inference tasks; update tasks in medium- and low-priority areas are scheduled to CPU-intensive general-purpose servers, which then run traditional hydrological models or statistical interpolation algorithms; edge computing terminals in remote areas are responsible for local sensor data aggregation and lightweight feature extraction tasks; the edge computing terminals in remote areas are those located at a distance greater than a preset distance threshold from the CPU-intensive general-purpose servers.

Images