Hill area small watershed runoff yield and collection analysis method and system based on model simulation

Abstract

The application discloses a hilly area small watershed runoff yield and concentration analysis method and system based on model simulation, relates to runoff yield and concentration data analysis and artificial intelligence technology, and comprises the following steps: acquiring satellite remote sensing images covering a target hilly area small watershed; determining a terrain key area and a water level sensitive area contained in the satellite remote sensing images within an influence range; collecting image data of each area and extracting image features; fusing the extracted image features with corresponding terrain area features; filling a ground surface of a region of interest based on the fused features and a pre-established terrain base model; acquiring current runoff yield and concentration correlation data, and predicting runoff yield and concentration based on the correlation data; and presenting the prediction result on the filled terrain base model by using a graph neural network. The application focuses on feature changes of a local area of a small watershed runoff yield and concentration, and improves the accuracy of analyzing nonlinear runoff yield and concentration characteristics of a small watershed in a valley terrain.

Description

Technical Field

[0001] This application relates to the fields of runoff generation and runoff data analysis and artificial intelligence technology, and in particular to a runoff generation and runoff analysis method and system for small watersheds in hilly areas based on model simulation. Background Technology

[0002] Small watershed flood generation and runoff characteristics are characterized by high flood peaks, large volumes, and steep rises and falls. Accurate understanding and prediction of the generation and runoff characteristics in valley topographic watersheds are crucial for flood control and disaster reduction, rational allocation of water resources, and ecological environmental protection.

[0003] Due to the complex topography and small catchment area, rainfall events in small valley watersheds often lead to rapid and intense surface runoff responses. Traditional linear hydrological models are difficult to accurately simulate the hydrological processes in these areas, especially in terms of nonlinear runoff generation and confluence characteristics.

[0004] Traditional methods often neglect heterogeneity in the spatiotemporal dimensions, leading to a further decline in the predictive power of the models. Existing techniques attempt to improve this situation by introducing more complex mathematical models, such as using the finite element method or finite difference method for numerical simulation. However, these methods are computationally expensive and have strict requirements for initial and boundary conditions, resulting in significant limitations in practical applications. Summary of the Invention

[0005] This application provides a model simulation-based method and system for analyzing runoff generation and confluence in small watersheds in hilly areas. This method focuses on the characteristic changes in local runoff generation and confluence areas of small watersheds, improving the accuracy of analyzing the nonlinear runoff generation and confluence characteristics of small watersheds in valley topography.

[0006] This application proposes a model simulation-based method for runoff generation and confluence analysis in small watersheds in hilly areas, including:

[0007] Acquire satellite remote sensing images covering the target hilly area's small watershed and identify the watershed's impact range;

[0008] Identify key topographic regions and water level-sensitive regions contained in satellite remote sensing images within the influence range, and designate the corresponding regions as regions of interest;

[0009] Collect image data for each region of interest and extract image features from the image data;

[0010] The extracted image features are fused with the corresponding terrain region features;

[0011] Based on the fused features and the pre-established terrain model, the surface of the region of interest is filled.

[0012] Obtain current production flow correlation data, and predict production flow based on the correlation data;

[0013] Using a graph neural network, the prediction results are presented on the filled terrain base model.

[0014] Optionally, identifying key topographic regions and water level-sensitive regions contained in the cropped satellite remote sensing image includes:

[0015] Establish the correlation region between the satellite remote sensing images within the influence range and the basic terrain model;

[0016] Using the Voronoi diagram algorithm, the associated region is divided into adaptive grids of similar size;

[0017] The adaptive grid is mapped onto the cropped satellite remote sensing image;

[0018] Based on the area variation relationship of the mapped adaptive grid, key terrain areas and water level sensitive areas are identified.

[0019] Optionally, based on the area variation relationship of the mapped adaptive grid, key terrain areas and water level sensitive areas are determined, including:

[0020] Compare the mapped area of ​​any adaptive grid with that of its neighboring grids one by one;

[0021] Several consecutive grids with similar area variation trends were identified to determine key terrain areas; and,

[0022] Key topographic regions and pixel discontinuities on both sides of the watershed in the cropped satellite remote sensing image are identified as water level sensitive areas.

[0023] Optionally, fusing the extracted image features with the corresponding terrain region features includes:

[0024] The topographic humidity of the corresponding topographic region is obtained, and the corresponding hydrological and topographic parameters are extracted based on the topographic basic model as topographic region features;

[0025] The extracted image features and the terrain region features are used to calculate an association matrix through a physically constrained cross-attention mechanism, and feature fusion is performed based on the association matrix.

[0026] Optionally, the extracted image features and the terrain region features are used to calculate an association matrix through a physically constrained cross-attention mechanism, satisfying the following:

[0027]

[0028] in, Let Q represent the correlation matrix calculated by the cross-attention mechanism, where Q represents the terrain region features, and K and V represent the key and value of the image features, respectively. This represents the physical prior matrix generated from the relationship between the curvature and local permeability of the adaptive mesh.

[0029] Optionally, feature fusion based on the correlation matrix further includes:

[0030] Acquire current meteorological data and determine rainfall information for the corresponding terrain area based on the meteorological data;

[0031] Based on the rainfall information and the initial soil moisture pre-configured for the corresponding terrain area, a lightweight LSTM network is used to predict the dynamic fusion weights of each adaptive grid.

[0032] Feature fusion is performed based on the dynamic fusion weights and the calculated correlation matrix.

[0033] Optionally, based on the fused features and a pre-established terrain model, the surface of the region of interest is filled, including:

[0034] Calculate the eigenvalues ​​of the fused features;

[0035] An adaptive grid with calculated feature values ​​greater than a preset feature threshold is filled based on the fused features.

[0036] Optionally, using a graph neural network to present the prediction results on the filled terrain base model includes:

[0037] Using the prediction results of runoff generation and confluence from graph neural networks, and taking an adaptive grid as a node topology, the propagation of water flow in the watershed is simulated in the terrain model.

[0038] The newly added adaptive grid, which is affected by water flow in the simulation results, is dynamically adjusted.

[0039] Optionally, it also includes using subsequently acquired image data to fill and update the surface of the region of interest.

[0040] This application also proposes a model simulation-based runoff generation and confluence analysis system for small watersheds in hilly areas, characterized in that it includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the steps of the aforementioned model simulation-based runoff generation and confluence analysis method for small watersheds in hilly areas.

[0041] The method in this application establishes a terrain model and an adaptive grid division approach, focusing on the characteristic changes in local areas of runoff generation and confluence in small watersheds, thereby improving the accuracy of analyzing the nonlinear runoff generation and confluence characteristics of small watersheds in valley terrain.

[0042] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0043] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0044] Figure 1 This is a schematic diagram of the basic process of the runoff generation and confluence analysis method for small watersheds in hilly areas based on model simulation, as described in this embodiment of the application. Detailed Implementation

[0045] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0046] This application proposes a model simulation-based method for runoff generation and confluence analysis in small watersheds in hilly areas, such as... Figure 1 As shown, it includes the following steps:

[0047] In step S101, satellite remote sensing images covering the target hilly area watershed are acquired, and the influence range of the watershed is identified. In a specific example, a topographic model of the target hilly area watershed can be established in advance based on the topographic elevation data of the watershed. Then, the influence range of the watershed is identified, and then the influence area in part of the topographic model is extracted based on the identified influence range.

[0048] In step S102, key topographic regions and water level-sensitive regions contained in the satellite remote sensing images within the influence range are determined, and the corresponding regions are designated as regions of interest. In some embodiments, key topographic regions may be areas that are not affected by the watershed during the dry season or normal water season, but form a watershed during the wet season. Water level-sensitive regions may be, for example, areas that may experience landslides or flooding during heavy rains, resulting in significant economic losses.

[0049] In step S103, image data of each region of interest is acquired, and image features are extracted from the image data. In a specific example, acquiring satellite remote sensing images and acquiring image data of regions of interest can be performed asynchronously. The frequency of acquiring image data can be higher than the frequency of acquiring remote sensing images. Image data acquisition can be achieved using devices such as drones or fixed cameras. Image features are extracted from the acquired image data, for example, based on CNNs.

[0050] In step S104, the extracted image features are fused with the corresponding terrain region features. In some embodiments, the image data of the region of interest can be roughly aligned with the influence range of the small watershed, and feature fusion can be performed. In specific examples, the rough alignment only needs to meet the accuracy of the subsequent adaptive grid.

[0051] In step S105, the surface of the region of interest is filled based on the fused features and the pre-established terrain model. In a specific example, the fused features can be encoded and decoded to fill the surface of the region of interest in the terrain model. The terrain model established in this example is based solely on surface elevation data and initially does not cover vegetation. By filling in the content, users can focus more on the area affected by runoff generation and confluence in subsequent displays.

[0052] In step S106, current runoff correlation data is obtained, and runoff is predicted based on the correlation data. In a specific example, models such as LSTM can be used to make predictions based on meteorological data, current water levels, etc.

[0053] In step S107, a graph neural network is used to present the predicted results on the filled terrain base model. In some embodiments, the presentation is dynamic based on the predicted results and the filled content, thereby facilitating the analysis and judgment of runoff generation and confluence in small watersheds in hilly areas.

[0054] The method in this application establishes a terrain model and an adaptive grid division approach, focusing on the characteristic changes in local areas of runoff generation and confluence in small watersheds, thereby improving the accuracy of analyzing the nonlinear runoff generation and confluence characteristics of small watersheds in valley terrain.

[0055] In some embodiments, determining the key topographic regions and water level-sensitive regions contained in the cropped satellite remote sensing image includes:

[0056] Establish the association region between the satellite remote sensing image within the influence range and the basic terrain model. In a specific example, multiple association regions can be established based on the obvious terrain boundaries in the terrain model and the corresponding boundaries on the satellite remote sensing image; for example, the boundaries can be mountain ranges or watershed areas.

[0057] Using the Voronoi diagram algorithm, the associated region is divided into adaptive grids of similar size. In specific examples, the specifications of the adaptive grids can be set according to the needs of the analysis, such as 100m*100m, 50m*50m, etc.

[0058] The adaptive grid is mapped onto the cropped satellite remote sensing image, also based on the aforementioned boundaries. By mapping the adaptive grid, the boundaries are adjusted according to the terrain undulations, meaning that the grid does not appear as a regular rectangle in gully areas.

[0059] Based on the area variation relationship of the mapped adaptive grid, key terrain areas and water level sensitive areas are identified.

[0060] In some embodiments, determining key terrain regions and water level-sensitive regions based on the area variation relationship of the mapped adaptive grid includes:

[0061] The mapped area of ​​any adaptive grid is compared one by one with that of neighboring grids. In this application example, the boundary range of the adaptive grid is adjusted according to the undulation of the terrain, and the planar visible area represented on the satellite remote sensing image is different. By comparing the mapped area with that of neighboring grids one by one, there is a significant area change, especially in areas with valleys and ravines.

[0062] Several consecutive grids with similar areas of variation are identified to determine key terrain regions. As mentioned above, key terrain regions are identified by identifying small grids with similar variations, for example, in a certain direction.

[0063] Key topographical regions and pixel discontinuities on both sides of the watershed in the cropped satellite remote sensing image are identified as water level sensitive areas. These pixel discontinuous areas may indicate landslides or other geological hazards. In some examples, water level sensitive areas can also be manually designated based on the economic losses from flooding, and these areas can be aggregated as the total water level sensitive area.

[0064] In some embodiments, fusing the extracted image features with the corresponding terrain region features includes:

[0065] The topographic humidity of the corresponding terrain region is obtained, and the corresponding hydro-topographic parameters are extracted based on the terrain base model as terrain region features. In a specific example, the topographic humidity determines the water absorption of the soil to a certain extent, and hydro-topographic parameters such as terrain slope affect the instantaneous flow of small watersheds in hilly areas. By introducing the terrain region features, areas sensitive to runoff generation and confluence can be further predicted from the covered features.

[0066] The extracted image features and the terrain region features are used to calculate an association matrix through a physically constrained cross-attention mechanism, and feature fusion is performed based on the association matrix.

[0067] In some embodiments, the extracted image features and the terrain region features are used to calculate an association matrix through a physically constrained cross-attention mechanism, satisfying the following:

[0068]

[0069] in, Let Q represent the correlation matrix calculated by the cross-attention mechanism, where Q represents the terrain region features, and K and V represent the key and value of the image features, respectively. This represents the physical prior matrix generated by the relationship between the edge curvature and local permeability of the adaptive mesh. Empirical formulas for terrain curvature and permeability are introduced as prior knowledge into the cross-attention layer, forcing the attention weights to conform to hydrological patterns.

[0070] Regarding the keys and values ​​K and V of image features, key (K) is a physically driven feature matching method used to determine which image regions are relevant to the hydrological response of the current adaptive grid by calculating the similarity with topographic region features Q. Key K is adjusted by physical priors (such as curvature-permeability relationships), for example, filtering high-permeability areas through curvature thresholding to retain only features that conform to Darcy's law.

[0071] Value (V): Retains and transmits original information, carrying raw image information (such as vegetation cover and surface water content) that has not been filtered by physical rules, ensuring no loss of detail. Even if some image features are downweighted due to physical constraints, their original data still participates in the fusion process through Value V, avoiding information loss. Further... The bond K in the forced high curvature region (steep slope) is only associated with physical laws, such as soil moisture content. This application avoids the black box model from deviating from the basic principles of hydrology through such design, and retains the original information to the maximum extent while following physical constraints.

[0072] In some embodiments, feature fusion based on the correlation matrix further includes:

[0073] Acquire current meteorological data and determine rainfall information for the corresponding terrain area based on the meteorological data;

[0074] Based on the rainfall information and the initial soil moisture pre-configured for the corresponding terrain area, a lightweight LSTM network is used to predict the dynamic fusion weights of each adaptive grid. This application further predicts the dynamic fusion weights of each adaptive grid based on the current rainfall information, thereby further improving the characterization effect of the fused features on the flow sensitivity of small watersheds in hilly areas.

[0075] Feature fusion is performed based on the dynamic fusion weights and the calculated correlation matrix.

[0076] In some embodiments, filling the surface of the region of interest based on the fused features and a pre-established terrain model includes:

[0077] Calculate the eigenvalues ​​of the fused features;

[0078] The adaptive grid with calculated feature values ​​greater than a preset feature threshold is filled according to the fused features. In this embodiment, the dynamic fusion weights allow for a greater focus on local areas. By filling the adaptive grid with feature values ​​greater than the preset feature threshold according to the fused features, key runoff generation and confluence areas in hilly areas can be identified, facilitating intuitive presentation for analysis.

[0079] In some embodiments, using a graph neural network to render the predicted results on the filled terrain base model includes:

[0080] Using the prediction results of runoff generation and confluence from graph neural networks, and taking an adaptive grid as a node topology, the propagation of water flow in the watershed is simulated in the terrain model.

[0081] By dynamically adjusting the newly added adaptive mesh that is affected by water flow in the simulation results, the simulation effect can be further improved.

[0082] In some embodiments, the method further includes using subsequently acquired image data to fill and update the surface of the region of interest, which reduces the amount of data processing compared to real-time updates of all data.

[0083] This application also proposes a model simulation-based runoff generation and confluence analysis system for small watersheds in hilly areas, characterized in that it includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the steps of the aforementioned model simulation-based runoff generation and confluence analysis method for small watersheds in hilly areas.

[0084] Furthermore, although exemplary embodiments have been described herein, their scope includes any and all embodiments based on this disclosure that have equivalent elements, modifications, omissions, combinations (e.g., schemes involving intersections of various embodiments), adaptations, or changes. They are not limited to the examples described in this specification or during the implementation of this application, and such examples are to be construed as non-exclusive.

[0085] The above description is intended to be illustrative and not restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. Other embodiments can be used by those skilled in the art when reading the above description.

[0086] The above embodiments are merely exemplary embodiments of this disclosure. Those skilled in the art can make various modifications or equivalent substitutions to this invention within the scope of the disclosure, and such modifications or equivalent substitutions should also be considered to fall within the protection scope of this invention.

Claims

1. A model-based simulation-based method for runoff generation and confluence analysis in small watersheds in hilly areas, characterized in that, include: Acquire satellite remote sensing images covering the target hilly area's small watershed and identify the watershed's impact range; Identify key topographic regions and water level-sensitive regions contained in satellite remote sensing images within the influence range, and designate the corresponding regions as regions of interest; Collect image data for each region of interest and extract image features from the image data; The extracted image features are fused with the corresponding terrain region features; Based on the fused features and the pre-established terrain model, the surface of the region of interest is filled. Obtain current production flow correlation data, and predict production flow based on the correlation data; Using a graph neural network, the prediction results are presented on the filled terrain base model; The fusion of extracted image features with corresponding terrain region features includes: The topographic humidity of the corresponding topographic region is obtained, and the corresponding hydrological and topographic parameters are extracted based on the topographic basic model as topographic region features; The extracted image features and the terrain region features are used to calculate an association matrix through a physically constrained cross-attention mechanism, and feature fusion is performed based on the association matrix. The extracted image features and the terrain region features are used to calculate the correlation matrix through a physically constrained cross-attention mechanism, satisfying the following: in, Let Q represent the correlation matrix calculated by the cross-attention mechanism, where Q represents the terrain region features, and K and V represent the key and value of the image features, respectively. This represents the physical prior matrix generated from the relationship between the curvature and local permeability of the adaptive mesh.

2. The method for runoff generation and confluence analysis of small watersheds in hilly areas based on model simulation as described in claim 1, characterized in that, The key topographic regions and water level-sensitive regions included in the cropped satellite remote sensing image are as follows: Establish the correlation region between the satellite remote sensing images within the influence range and the basic terrain model; Using the Voronoi diagram algorithm, the associated region is divided into adaptive grids of similar size; The adaptive grid is mapped onto the cropped satellite remote sensing image; Based on the area variation relationship of the mapped adaptive grid, key terrain areas and water level sensitive areas are identified.

3. The method for runoff generation and confluence analysis of small watersheds in hilly areas based on model simulation as described in claim 2, characterized in that, Based on the area variation relationship of the mapped adaptive grid, the key terrain areas and water level sensitive areas are identified, including: Compare the mapped area of ​​any adaptive grid with that of its neighboring grids one by one; Several consecutive grids with similar area variation trends were identified to determine key terrain areas; and, Key topographic regions and pixel discontinuities on both sides of the watershed in the cropped satellite remote sensing image are identified as water level sensitive areas.

4. The method for runoff generation and confluence analysis of small watersheds in hilly areas based on model simulation as described in claim 1, characterized in that, Feature fusion based on the aforementioned correlation matrix also includes: Acquire current meteorological data and determine rainfall information for the corresponding terrain area based on the meteorological data; Based on the rainfall information and the initial soil moisture pre-configured for the corresponding terrain area, a lightweight LSTM network is used to predict the dynamic fusion weights of each adaptive grid. Feature fusion is performed based on the dynamic fusion weights and the calculated correlation matrix.

5. The method for runoff generation and confluence analysis of small watersheds in hilly areas based on model simulation as described in claim 1, characterized in that, Based on the fused features and the pre-established terrain model, the surface of the region of interest is filled, including: Calculate the eigenvalues ​​of the fused features; An adaptive grid with calculated feature values ​​greater than a preset feature threshold is filled based on the fused features.

6. The method for runoff generation and confluence analysis of small watersheds in hilly areas based on model simulation as described in claim 5, characterized in that, Using graph neural networks, the prediction results are presented on the filled terrain base model, including: Using the prediction results of runoff generation and confluence from graph neural networks, and taking an adaptive grid as a node topology, the propagation of water flow in the watershed is simulated in the terrain model. The newly added adaptive grid, which is affected by water flow in the simulation results, is dynamically adjusted.

7. The method for runoff generation and confluence analysis of small watersheds in hilly areas based on model simulation as described in claim 6, characterized in that, It also includes using subsequently acquired image data to fill and update the surface of the region of interest.

8. A model-based simulation-based runoff generation and confluence analysis system for small watersheds in hilly areas, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, implements the steps of the model simulation-based runoff generation and confluence analysis method for small watersheds in hilly areas as described in any one of claims 1 to 7.

Images