Urban inland inundation refined simulation method based on earth surface DEM correction and image recognition

Abstract

The invention discloses an urban inland inundation refined simulation method based on earth surface DEM correction and image recognition, and relates to the technical field of urban hydrology and disaster prevention and reduction. The method comprises the following steps: performing image recognition on satellite remote sensing image data by using an improved U-Net model, and extracting data of different land utilization types; fusing the surface elevation data of the drainage pipe network and the original DEM data, and generating DEM correction data through interpolation screening and ground feature hydrological correction; and finally, constructing an earth surface runoff model, an urban one-dimensional drainage model and a two-dimensional earth surface overflow model, and realizing one-dimensional-two-dimensional coupled urban waterlogging refined simulation. By improving the model, the small-scale ground feature recognition precision is improved, the limitation of a single data source is broken through, the characterization accuracy of the urban complex underlying surface is enhanced, the simulation result is accurate and reliable, and the method is suitable for the fields of urban inland inundation simulation, urban planning, flood control and drainage and the like.

Description

Technical Field

[0001] This invention relates to the field of urban hydrology and disaster prevention and mitigation technology, and in particular to a refined simulation method for urban flooding based on surface DEM correction and image recognition. Background Technology

[0002] Urban flooding models are a key tool for constructing urban flood control and drainage systems, and their simulation accuracy directly affects the scientific nature of early warning and decision-making. Simulation accuracy depends not only on the quality of the model itself, but also, and more importantly, on the accuracy of the input data, especially the topographic elevation and underlying surface (land use) data.

[0003] However, currently widely used data products are insufficient to meet the needs of refined modeling. On the one hand, mainstream DEM data (such as SRTM, ASTER GDEM, etc.) have limited accuracy, making it difficult to characterize urban micro-topographic structures and leading to deviations in water accumulation simulation. On the other hand, land use products (such as Globeland, MODIS, etc.) are limited by untimely updates, ambiguous category definitions, and weak identification of small-scale land features, making it difficult to truly reflect the complex hydrological characteristics of the urban underlying surface.

[0004] In recent years, deep learning technologies (such as U-Net) have provided new methods for land cover classification in remote sensing images. However, mainstream models still have limitations when applied to complex urban environments: insufficient attention to small-scale land cover leads to misclassification or omission; the large number of model parameters easily leads to overfitting and increased computational burden. In addition, in existing urban flooding modeling, DEM data is usually used directly or only simple interpolation is performed, failing to effectively integrate data such as pipe network elevation, and lacking topographic correction for key hydrological processes such as water obstruction by buildings and road diversion, resulting in insufficient ability of the model to represent the hydrological effects of urban "micro-topography".

[0005] Therefore, proposing a refined urban flooding simulation method based on surface DEM correction and image recognition can improve accuracy from the data source and effectively integrate multi-source information to characterize urban hydrological features. This new flooding simulation technology aims to improve the accuracy and reliability of the model and is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a refined simulation method for urban flooding based on surface DEM correction and image recognition, which solves the problems of insufficient DEM data accuracy, inaccurate land use classification, and low accuracy of flooding simulation in the prior art, and realizes refined and accurate simulation of urban flooding.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A refined simulation method for urban flooding based on surface DEM correction and image recognition includes the following steps: S1 Data Acquisition Steps: Acquire current remote sensing image data of the study area; S2 Land Type Classification Steps: Using an improved deep learning network model, image recognition is performed on the remote sensing image data acquired in S1 to obtain data on different land use types, including buildings and roads, and thus obtain data on the distribution of impervious surfaces in the study area. S3 DEM data reconstruction steps with integrated land cover features: After fusing the original digital elevation DEM data and the surface elevation data of the drainage network in the study area, DEM raster data is generated by interpolation method; Based on the building and road data extracted in step S2, hydrological feature correction is performed on the currently generated DEM raster data to obtain DEM corrected data based on multi-source data fusion. S4 Steps for constructing a refined urban flooding model: Based on the land use type data obtained in step S2, the study area is preliminarily divided into catchment areas and relevant hydrological parameters are determined; based on the DEM correction data obtained in step S3 based on multi-source data fusion, a two-dimensional surface runoff model is constructed; combined with preprocessed drainage network data, a one-dimensional drainage network model and a surface runoff generation and collection model are constructed; the two-dimensional surface runoff model, the one-dimensional drainage network model, and the surface runoff generation and collection model are coupled to construct a refined coupled urban flooding model; S5 Simulation Steps: Input rainfall data into the refined coupled urban flooding model obtained in S4 to simulate urban flooding.

[0008] Optionally, the improved deep learning network model in S2 is an improved U-Net model. The improvements include: introducing a convolutional block attention mechanism into the network; replacing the traditional pooling layer with a convolution operation with a stride of 2 for downsampling; and replacing the ReLU activation function with the Leaky ReLU activation function.

[0009] Optionally, in the above method, S3 includes hydrological feature correction of the currently generated DEM raster data, including: Building area correction: The elevation values ​​corresponding to the polygonal areas of the building are uniformly increased by a first preset height value to simulate its water-blocking characteristics; Road area correction: Create buffer zones along both sides of the road, and uniformly increase the grid elevation values ​​within the buffer zones by a second preset height value to simulate the water-blocking and flow-guiding effect of the curb stones; Fusion processing: The elevation data of the corrected building area and road buffer area are fused with the base DEM data of the uncorrected area to generate DEM corrected data based on multi-source data fusion.

[0010] Optionally, in S3, the high-precision basic DEM raster data is generated using an interpolation method: sampling is performed from a composite elevation database formed by the fusion of multi-source elevation data, the accuracy of various interpolation methods is evaluated through cross-validation, and the interpolation method with the best evaluation index is selected to generate DEM raster data.

[0011] The above methods may optionally include the following evaluation metrics: mean error, root mean square error, and coefficient of determination. Various interpolation methods include: Kriging interpolation, polynomial interpolation, spline function interpolation, and inverse distance weighted interpolation.

[0012] Optionally, in the above method, the two-dimensional surface runoff model, the one-dimensional drainage network model, and the surface runoff generation and collection model are coupled in S4. Specifically, a one-dimensional-two-dimensional bidirectional dynamic coupling method is adopted, and the water volume between the surface and the network is dynamically calculated and exchanged based on the height difference between the water head of the drainage network nodes and the water head of the surface grid.

[0013] Optionally, S4 of the above method also includes: verifying the simulation results of the refined coupling model of urban flooding using historical flooding observation data, and optimizing and adjusting the model parameters based on the verification results.

[0014] As can be seen from the above technical solution, compared with the prior art, the present invention provides a refined simulation method for urban flooding based on surface DEM correction and image recognition, which has the following beneficial effects: (1) Improve the recognition accuracy of small-scale ground objects: The improved U-Net network effectively overcomes the limitations of traditional models in complex urban environments by introducing an attention mechanism, optimizing the downsampling method and activation function, and significantly improves the classification accuracy of small-scale ground objects and the training efficiency of the model. (2) Breaking through the limitations of a single data source: By using multi-source data fusion technology to correct the DEM and to perform hydrological corrections on key features, the accuracy of the complex underlying surface representation of the city is significantly improved from the data source, providing a more reliable topographic basis for urban flooding simulation; (3) This invention is universal and applicable to different types of urban areas and terrain conditions. It can be widely used in urban waterlogging simulation, urban planning and flood control and drainage, providing more scientific and accurate decision support for urban waterlogging prevention and control. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This invention discloses a flowchart of a refined simulation method for urban flooding based on surface DEM correction and image recognition; Figure 2 This is an intended representation of the terrain type and label information disclosed in this invention; Figure 3 This is a diagram of the improved U-net structure disclosed in this invention; Figure 4 This invention discloses a technical roadmap for the establishment and reconstruction of a DEM database. Figure 5 This is a schematic diagram of the water flow in the one- or two-dimensional bidirectional coupling model disclosed in this invention. Figure 6 This is a technical roadmap for constructing a refined urban flooding model as disclosed in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In this application, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0019] This invention proposes a DEM correction method based on multi-source data fusion. By combining surface elevation data of drainage pipe network and original digital elevation model (DEM) data, high-precision DEM raster data is generated, and hydrological corrections are made for key features such as buildings and roads. This significantly improves the ability to represent complex urban underlying surfaces and provides a key topographic foundation for the reliability of urban flooding models.

[0020] This invention makes a series of improvements to the U-Net network model, including constructing a feature extraction network based on encoder-decoder and introducing a convolutional block attention mechanism (CBAM) to enhance the model's ability to capture terrain features in spatial and channel dimensions; using convolution operations with a stride of 2 to replace traditional pooling layers for downsampling, reducing the number of network parameters; and using the Leaky ReLU activation function to replace ReLU, reducing the impact of gradient vanishing and thus improving classification accuracy.

[0021] This invention constructs a one-dimensional-two-dimensional coupled refined urban flooding model. By combining DEM correction data reconstructed from multi-source data, a two-dimensional surface runoff model is established and coupled with a surface runoff generation and runoff model and a one-dimensional urban drainage model, thereby realizing a refined simulation of urban flooding and providing more scientific and accurate decision support for urban flooding prevention and control.

[0022] The specific technical solution is as follows: See Figure 1 As shown, this invention discloses a refined simulation method for urban flooding based on surface DEM correction and image recognition, comprising the following steps: S1 Data Acquisition Steps: Acquire current remote sensing image data of the study area; S2 Land Type Classification Steps: Using an improved deep learning network model, image recognition is performed on the remote sensing image data acquired in S1 to obtain data on different land use types, including buildings and roads, and thus obtain data on the distribution of impervious surfaces in the study area. S3 DEM data reconstruction steps with integrated land cover features: After fusing the original digital elevation DEM data and the surface elevation data of the drainage network in the study area, DEM raster data is generated by interpolation method; Based on the building and road data extracted in step S2, hydrological feature correction is performed on the currently generated DEM raster data to obtain DEM corrected data based on multi-source data fusion. S4 Steps for constructing a refined urban flooding model: Based on the land use type data obtained in step S2, the study area is preliminarily divided into catchment areas and relevant hydrological parameters are determined; based on the DEM correction data obtained in step S3 based on multi-source data fusion, a two-dimensional surface runoff model is constructed; combined with preprocessed drainage network data, a one-dimensional drainage network model and a surface runoff generation and collection model are constructed; the two-dimensional surface runoff model, the one-dimensional drainage network model, and the surface runoff generation and collection model are coupled to construct a refined coupled urban flooding model; S5 Simulation Steps: Input rainfall data into the refined coupled urban flooding model obtained in S4 to simulate urban flooding.

[0023] Furthermore, land use type is one of the core parameters of urban flooding models, and its classification accuracy directly affects the simulation accuracy of key hydrological processes such as surface permeability, runoff coefficient, and flood evolution path. Multispectral images covering the study area, acquired by the Landsat 8 satellite, were used as the data source in S1. These images have a resolution of 30 meters, contain 11 bands, and have a cloud cover of 4.54%. Landsat 8 carries a land imager and a thermal infrared sensor. The land imager includes nine bands with a spatial resolution of 30 meters, including one 15-meter panchromatic band, with an imaging width of 185 × 185 km. The thermal infrared sensor includes two independent thermal infrared bands with a resolution of 100 meters. To obtain more effective image information, ENVI software was used to perform preprocessing operations on the images, including geolocation, radiometric calibration, atmospheric correction, mosaicking, and cropping.

[0024] Besides natural vegetation, the study area is mainly composed of non-vegetated areas formed by the combined effects of human activities and the natural environment, including land features such as buildings, water bodies, roads, and farmland. Based on this, this invention classifies the topographic types of the study area into six categories. Figure 2 The map types and their corresponding labels are listed. The images of the study area, cropped from SHP files, are irregularly shaped. To further adapt to the model input requirements, all images were uniformly cropped to 256×256 pixels, and images with missing information or incomplete coverage were removed. Subsequently, LabelMe software was used to manually label each feature area, assigning it corresponding category labels. Due to the geographical characteristics of the study area, features such as water bodies and roads are less representative in spatial distribution, resulting in a relatively small number of samples, while the samples for cultivated land and buildings are more concentrated. To alleviate the class imbalance problem and expand the effective data scale, this study implemented various data augmentation operations on the original dataset, including 90° rotation, mirror flipping, and image contrast and color enhancement. After augmentation, the total number of images in the dataset reached 4370, and they were randomly divided into training, validation, and test sets in a 6:2:2 ratio. At this point, the dataset construction work required for the remote sensing image feature classification research was completed.

[0025] Furthermore, the improved deep learning network model in S2 is the improved U-Net model. The improvements include: introducing a convolutional block attention mechanism into the network; replacing the traditional pooling layer with a stride of 2 for downsampling; and replacing the ReLU activation function with the Leaky ReLU activation function.

[0026] Specifically, a series of improvements were made to the U-Net network model to enhance its performance in classification accuracy and training efficiency. An encoder-decoder-based feature extraction network was constructed, and a convolutional block attention mechanism (CBAM) was introduced to effectively enhance the model's ability to capture terrain features in both spatial and channel dimensions. Furthermore, to address the information loss and gradient vanishing problems that may result from traditional pooling layers, convolution operations with a stride of 2 were used to replace pooling layers for downsampling.

[0027] In the choice of activation function, Leaky ReLU was used instead of ReLU. This change helps reduce the impact of gradient vanishing during network training, thereby improving classification accuracy. These improvements not only optimize the network structure but also ensure effective gradient backpropagation while reducing the number of parameters.

[0028] The improved U-Net network model is shown below. Figure 3 Its enhanced structure clearly reflects the aforementioned improvements. Through these optimizations, the model exhibits higher accuracy and robustness when handling complex terrain features, significantly improving its performance in practical applications.

[0029] Convolutional and pooling layers are indispensable components of neural networks. Convolution is responsible for terrain extraction by multiplying the terrain map by the corresponding positions of the convolution kernel and summing the results. After convolution, the size of the terrain map remains unchanged, but the number of channels increases. On the other hand, pooling layers are usually located between convolutional layers, reducing the size of the terrain map while retaining the same number of channels. Convolution effectively compresses the terrain map by adjusting its size through techniques such as stride and padding, achieving size compression without the need for pooling. By using convolution instead of pooling, and by changing the number and size of the original convolution kernel structure, the number of network parameters and floating-point computations are reduced. Compared with U-Net, the improved network reduces the number of parameters by 7.26% and the floating-point computations by 18.02%.

[0030] To further enhance the focus on small-area terrain information, this study incorporates a convolutional block attention mechanism into the terrain extraction stage. An attention mechanism refers to the process of shifting attention from less important parts to the most important parts. In deep learning, the attention mechanism can be viewed as a dynamic selection process that assigns different weights to input data based on their importance. Attention terrain maps, calculated from a given terrain map in both channel and spatial dimensions, are then multiplied with the input terrain map to assign adaptive weights.

[0031]

[0032] in, This represents element-wise multiplication, where the values ​​of channel attention and spatial attention are multiplied along their respective dimensions. The resulting product is represented as... This represents the final output value.

[0033] The convolutional block attention mechanism consists of two parts: channel attention and spatial attention. Channel attention plays a crucial role in automatically learning terrain weights. It enhances the network's ability to extract useful information from specific channels while suppressing the extraction of less important channels. This helps reduce interference from irrelevant information and improves overall accuracy.

[0034]

[0035] Where σ represents the activation function, AvgPool represents the global average pooling operation, MaxPool represents the global maximum pooling operation, and W1 and W2 represent the MLP weights shared by both.

[0036] The Spatial Attention Module (SAM) complements the Channel Attention Mechanism. It compresses channels, performs global average pooling and max pooling operations, and learns a weight matrix from a topographic map of dimension H×W in a two-dimensional plane. The weight matrix is ​​then concatenated with the original topographic map. This process enhances attention to valuable terrain while attenuating or ignoring irrelevant terrain. The equation representing the Spatial Attention Mechanism is given in Equation (4):

[0037] Where σ represents the activation function, This indicates a convolution operation with a kernel size of 7×7.

[0038] To verify the effectiveness of the method, accuracy (A), intersection-over-union ratio (IoU), and Kappa coefficient (K) were selected (K coefficient is an indicator that measures classification accuracy and consistency). The calculation formulas for the three evaluation indicators are as follows: (5)-(7)

[0039] in, Representative for the first The number of pixels that are correctly classified into that class; d represents the first The number of pixels that do not belong to this class but are correctly classified into other classes; Representative for the first The number of pixels that do not belong to this class but are incorrectly classified as such; Representative for the first Class, the number of pixels that belong to this class but are incorrectly classified into other classes; N is the total number of classes (N=6).

[0040] The Kappa coefficient, calculated based on the confusion matrix, ranges from -1 to 1 and can be used to verify whether the predicted results match the actual results. The formula is as follows:

[0041] Where: N represents the total number of pixels; The diagonal elements of the confusion matrix are represented by r; r represents the number of categories. Represents the number of predicted samples for class i; This represents the number of real samples in class i.

[0042] Table 1 compares the evaluation metrics of the original U-net and the improved U-net model. The results show that the improved U-Net model has the highest extraction accuracy, with Acc, Iou, and Kappa coefficients improved by 2.6%, 6.47%, and 3.08% respectively compared to the original U-Net model.

[0043] Table 1 Model accuracy evaluation indicators

[0044] Furthermore, S3 performs hydrological feature corrections on the currently generated DEM raster data, including: Building area correction: The elevation values ​​corresponding to the polygonal areas of the building are uniformly increased by a first preset height value to simulate its water-blocking characteristics; Road area correction: Create buffer zones along both sides of the road, and uniformly increase the grid elevation values ​​within the buffer zones by a second preset height value to simulate the water-blocking and flow-guiding effect of the curb stones; Fusion processing: The elevation data of the corrected building area and road buffer area are fused with the base DEM data of the uncorrected area to generate DEM corrected data based on multi-source data fusion.

[0045] Furthermore, in S3, the high-precision basic DEM raster data is generated using interpolation methods as follows: sampling is performed from a composite elevation database formed by the fusion of multi-source elevation data, the accuracy of various interpolation methods is evaluated through cross-validation, and the interpolation method with the best evaluation index is selected to generate DEM raster data.

[0046] Furthermore, the evaluation indicators include: mean error, root mean square error, and coefficient of determination; Various interpolation methods include: Kriging interpolation, polynomial interpolation, spline function interpolation, and inverse distance weighted interpolation.

[0047] For details, see Figure 4 As shown, the content of S3 is: The accurate representation of urban surface features, especially micro-topographic structures, is crucial to the accuracy of urban flooding models. To overcome the limitations of previous studies that used single interpolation methods, this invention proposes a more scientific topographic DEM data reconstruction process that integrates land cover features. This process begins with the construction of a composite elevation database, combining the elevation values ​​of raster dataset transition points with the elevation attributes of pipeline network nodes to form a composite elevation database.

[0048] After data reconstruction, this sampling employed a simple random sampling method without replacement, using 70% of the samples as the training set to build different interpolation models, and the remaining 30% as an independent validation set. From a total population of 80,000 points, 30% of the points were selected as samples. This process ensured that before sampling, each point in the population had an equal probability of being included in the final sample. Furthermore, all possible sample combinations are selected with equal probability, ensuring that the samples are highly representative of the population. Based on the data validation set, the system evaluates the applicability of seven interpolation methods: ordinary kriging interpolation, simple kriging interpolation, generalized kriging interpolation, global polynomial interpolation, local polynomial interpolation, inverse distance weighting, and spline function interpolation. Through cross-validation, and using mean error (ME) and root mean square error (RMSE) as the core evaluation indicators, the optimal interpolation method is selected, thus ensuring the statistically optimal accuracy of the generated digital elevation model.

[0049] Average error:

[0050] Root mean square error:

[0051] in, It is an estimated value. is the measured value of sampling point i (i=1,...,n), and n is the number of values ​​used for estimation.

[0052] Coefficient of determination:

[0053] Among them, the coefficient of determination Used to measure the correlation between predicted and measured values. It is the average estimate. is the average measurement value, and n is the number of values ​​used for estimation.

[0054] Table 2. Statistics of Cross-Validation Errors for Surface Elevation Interpolation

[0055] Table 2 shows that the root mean square error (RMSE) is ranked as follows: spline interpolation < ordinary kriging < universal kriging < local polynomial interpolation < inverse distance weighting < simple kriging < global polynomial interpolation; the mean absolute error (MAE) is ranked as follows: inverse distance weighting < spline interpolation < local polynomial interpolation < ordinary kriging < universal kriging < simple kriging < global polynomial interpolation. The coefficient of determination (COD) is ranked as follows: global polynomial interpolation < simple kriging < inverse distance weighting < universal kriging = ordinary kriging = local polynomial interpolation < spline interpolation. Overall, spline interpolation performs best, with the smallest ME (0.473) and R0. 2 The highest (0.856); among them, the inverse distance weight has the smallest RMSE, but in R 2 It is slightly inferior to spline functions. Global polynomials, on the other hand, are poorly fitted models based on global interpolation rather than exact interpolation, with R0... 2 The results of the three Kriging interpolation methods are similar. For highly developed urban centers, the ground elevation exhibits strong spatial heterogeneity due to human activities such as road leveling and underground space development. The semi-variogram model on which the Kriging method relies is difficult to effectively characterize the non-stationary spatial features of the surface caused by human intervention.

[0056] After obtaining the optimal interpolation surface, hydrological corrections are made to key urban features based on the imagery results to accurately reflect their true role in urban flooding.

[0057] Building treatment: To simulate its 100% impermeability and complete water resistance, the elevation of the building's polygonal area is raised by a sufficiently large value ΔHb (ΔHb = 10m in this study) to ensure that water flow cannot cross it in the model calculations. Its mathematical expression is as follows:

[0058] Road Treatment: Urban roads combined with curbs can be simplified into irregular channels serving as the main surface flood discharge channels. To avoid disrupting the elevation and topological relationship with the underlying drainage network, the blocking and guiding effects of the curbs on water flow are equivalently reconstructed at the model grid scale, resulting in an innovative generalization. The theoretical basis of this method is that the lifting effect of the curbs on water flow is equivalent to creating a micro-topographical protrusion on both sides of the road. Based on the minimum size (Δx) of the grid calculated using a two-dimensional surface runoff model, buffer zones with a width equal to the buffer zone radius (5m in this study) are created along both sides of the road, and the elevation of these buffer zones is uniformly raised by a height value ΔHc equivalent to the curb height (ΔHc = 0.15 m in this study). Its mathematical expression is as follows:

[0059] This treatment is equivalent to introducing a local terrain obstacle in fluid dynamics. Its water-blocking effect can be understood through a simplified Manning formula: when the water flow depth h encounters an obstacle, h > ΔHc in order for overflow to occur.

[0060] Finally, using the raster calculator on the ArcGIS platform, the con function is used to seamlessly integrate the elevation data of specially corrected building areas, road buffer zones, and other areas to generate a complete, coordinated final surface elevation model with enhanced hydrological features. Its mathematical expression is as follows.

[0061]

[0062] The systematic approach described in this invention, from data construction and method optimization to land feature reconstruction, significantly improves the ability to characterize complex urban underlying surfaces and provides a key topographic foundation for the reliability of urban flooding models.

[0063] Furthermore, in S4, the two-dimensional surface runoff model, the one-dimensional drainage network model, and the surface runoff generation and collection model are coupled. Specifically, a one-dimensional-two-dimensional bidirectional dynamic coupling method is adopted, and the water volume between the surface and the network is dynamically calculated and exchanged based on the height difference between the water head of the drainage network nodes and the water head of the surface grid.

[0064] Furthermore, S4 also includes: validating the simulation results of the refined coupled urban flooding model using historical flooding observation data, and optimizing and adjusting the model parameters based on the validation results. Specifically, the simulation results of the refined coupled model are validated using measured water depth and range data from historical flooding events in the study area; the model accuracy is evaluated by comparing the errors (such as absolute error and Nash efficiency coefficient NSE) between the simulated and measured values; and relevant parameters in the model (such as infiltration rate and Manning roughness) are calibrated and optimized based on the validation results to further improve the simulation accuracy and reliability of the model.

[0065] For details, see Figure 6 As shown, the specific content of S4 is as follows: This study uses IFMS software, which mainly includes modules such as a one-dimensional urban drainage network model and a two-dimensional surface water dynamics model. The urban network model is based on the urban flooding model SWMM, with improvements and extensions.

[0066] IFMS software can execute one-dimensional and two-dimensional bidirectional coupled models. These coupled models can realize the function of water flow interaction, meaning that water in the pipe network can overflow to the surface, and surface water can also re-enter the pipe network (see...). Figure 5 Currently, the common practice for coupling surface flood models and underground pipe network models is to calculate the exchanged water volume and then substitute it into their respective models for calculation. The exchanged water volume is calculated using the following formula:

[0067] In the formula, For surface water head, The head of the drainage pipe network, M, is the flow coefficient. It represents the ground elevation.

[0068] After completing image recognition, land use type extraction, and DEM database establishment and data reconstruction, this invention further constructs a refined urban flooding model based on surface DEM correction and image recognition. First, the drainage network data is preprocessed and structurally generalized. By deleting pipes with diameters below a specific threshold and non-major branch pipes, the network is reasonably simplified, thereby improving the model's computational efficiency. Through these steps, the final determined number of pipes is 9500, with a total length of 253.68 km, the number of inspection wells is 9075, the number of drainage outlets is 236, and the simulated total area of ​​the study area is 38.68 km². 2 .

[0069] Based on road and water system information obtained from image recognition, combined with hydrological analysis, the study area was divided into catchment areas. The simulated area was artificially divided into 83 main catchment areas by merging the centerlines of water bodies and roads. Then, the Thiessen polygon method was used to divide the sub-catchment areas corresponding to the nodes, finally obtaining 9075 sub-catchment areas, with the smallest sub-catchment area being 20.78 m². 3 The maximum area is 290,757.07 m². 3 Parameters such as slope and impermeability were calculated using GIS spatial analysis tools, while the characteristic width was taken as the square root of the sub-catchment area. The infiltration model adopted the Horton model, in which the maximum / minimum infiltration rate and attenuation constant were initially determined based on empirical values ​​and the underlying surface.

[0070] Mesh 2D, a mesh generation software used in conjunction with IFMS, was employed with an average mesh side length of 25m. The meshes for the main streets were further refined with a densification size of 5m. Interpolation was performed on the meshes using the original topographic data and the DEM correction data fused from multi-source data. Rainfall data was also input, and simulation results from different data sources were calculated and compared.

[0071] A surface runoff generation and collection model, a one-dimensional urban drainage model, and a two-dimensional surface runoff model were constructed. The surface runoff generation and collection model simulates the generation and collection process of surface rainwater based on land use type data and DEM correction data. The one-dimensional urban drainage model uses preprocessed drainage network data to simulate the water flow transmission and drainage process within the drainage network. The two-dimensional surface runoff model combines DEM correction data reconstructed from multi-source data to simulate the diffusion and flow process of surface water. These three models are coupled (the coupling principle is shown in formula (14)) to realize the dynamic interaction between the surface and the pipe network, ensuring that water in the pipe network can overflow onto the surface and surface water can re-enter the pipe network, thereby more accurately simulating the entire process of urban flooding. This coupled model can effectively reflect the water flow exchange between the surface and the pipe network, improving the accuracy and reliability of flooding simulation.

[0072] The model was validated using historical flooding data and field observation data. The results (Table 3) show that the model constructed in this invention exhibits high accuracy at all waterlogging points, with the absolute error between the simulated maximum water depth and the measured value being less than 0.03 meters. Simultaneously, the Nash efficiency coefficient (NSE) at each waterlogging point is greater than 0.7, reaching a maximum of 0.90. This indicates that the model not only has high accuracy but also excellent stability and reliability across different geographical locations, accurately reproducing the actual process of urban flooding. This superior simulation performance directly benefits from the high-precision land use classification and DEM correction enhanced by hydrological features in the preceding steps. These provide accurate key input parameters for the model, ensuring the authenticity of the simulation results from the data source. The refined urban flooding model constructed in this invention can more accurately simulate the formation and development process of urban flooding, providing a scientific basis for urban flooding prevention and control.

[0073] Table 3. Error Statistics of Urban Flooding Simulation Results

[0074] Based on the verification results, the model parameters were optimized and adjusted to further improve the simulation accuracy of the model. Through this series of steps, the present invention constructs a refined urban flooding model that can more accurately simulate the formation and development process of urban flooding, providing a scientific basis for urban flooding prevention and control.

[0075] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements as follows: Figure 1 The steps of the method shown.

[0076] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A refined simulation method for urban flooding based on surface DEM correction and image recognition, characterized in that, Includes the following steps: S1 Data Acquisition Steps: Acquire current remote sensing image data of the study area; S2 Land Type Classification Steps: Using an improved deep learning network model, image recognition is performed on the remote sensing image data acquired in S1 to obtain data on different land use types, including buildings and roads, and thus obtain data on the distribution of impervious surfaces in the study area. S3 DEM data reconstruction steps with integrated land cover features: After fusing the original digital elevation DEM data and the surface elevation data of the drainage network in the study area, DEM raster data is generated by interpolation method; Based on the building and road data extracted in step S2, hydrological feature correction is performed on the currently generated DEM raster data to obtain DEM corrected data based on multi-source data fusion. S4 Steps for constructing a refined urban flooding model: Based on the land use type data obtained in step S2, the study area is preliminarily divided into catchment areas and relevant hydrological parameters are determined. Based on the DEM correction data obtained in step S3 based on multi-source data fusion, a two-dimensional surface runoff model is constructed; a one-dimensional drainage network model and a surface runoff generation and collection model are constructed by combining the preprocessed drainage network data; the two-dimensional surface runoff model, the one-dimensional drainage network model and the surface runoff generation and collection model are coupled to construct a refined coupled model for urban waterlogging. S5 Simulation Steps: Input rainfall data into the refined coupled urban flooding model obtained in S4 to simulate urban flooding.

2. The method for refined simulation of urban flooding based on surface DEM correction and image recognition according to claim 1, characterized in that, The improved deep learning network model in S2 is the improved U-Net model. The improvements include: introducing a convolutional block attention mechanism into the network; replacing the traditional pooling layer with a stride of 2 for downsampling; and replacing the ReLU activation function with the Leaky ReLU activation function.

3. The method for refined simulation of urban flooding based on surface DEM correction and image recognition according to claim 2, characterized in that, S3 includes the following hydrological feature corrections to the currently generated DEM raster data: Building area correction: The elevation values ​​corresponding to the polygonal areas of the building are uniformly increased by a first preset height value to simulate its water-blocking characteristics; Road area correction: Create buffer zones along both sides of the road, and uniformly increase the grid elevation values ​​within the buffer zones by a second preset height value to simulate the water-blocking and flow-guiding effect of the curb stones; Fusion processing: The elevation data of the corrected building area and road buffer area are fused with the base DEM data of the uncorrected area to generate DEM corrected data based on multi-source data fusion.

4. The method for refined simulation of urban flooding based on surface DEM correction and image recognition according to claim 3, characterized in that, In S3, the high-precision basic DEM raster data is generated by using interpolation methods. Specifically, samples are drawn from a composite elevation database formed by the fusion of multi-source elevation data. The accuracy of various interpolation methods is evaluated through cross-validation, and the interpolation method with the best evaluation index is selected to generate DEM raster data.

5. The method for refined simulation of urban flooding based on surface DEM correction and image recognition according to claim 4, characterized in that, Evaluation metrics include: mean error, root mean square error, and coefficient of determination; Various interpolation methods include: Kriging interpolation, polynomial interpolation, spline function interpolation, and inverse distance weighted interpolation.

6. The method for refined simulation of urban flooding based on surface DEM correction and image recognition according to claim 5, characterized in that, In S4, the two-dimensional surface runoff model, the one-dimensional drainage network model, and the surface runoff generation and collection model are coupled. Specifically, a one-dimensional-two-dimensional bidirectional dynamic coupling method is adopted. Based on the height difference between the water head of the drainage network nodes and the water head of the surface grid, the water volume between the surface and the network is dynamically calculated and exchanged.

7. The method for refined simulation of urban flooding based on surface DEM correction and image recognition according to claim 6, characterized in that, S4 also includes: verifying the simulation results of the refined coupling model of urban waterlogging using historical waterlogging observation data, and optimizing and adjusting the model parameters based on the verification results.

Images