A Spatial Inference Method for Surface Water Depth Based on Urban Flood Monitoring Points

Abstract

This invention discloses a method for spatially inferring surface water depth based on urban flooding monitoring points. First, the semantic segmentation network SegNet is modified: the number of channels in its input layer and the number of input data channels in its first convolutional layer are both set to... n +1, setting the number of output data channels of its last convolutional layer to 1, removing its Softmax layer, and keeping the remaining layers unchanged. The modified SegNet is then used as a model for spatial inference of surface water depth. t - n Time's up t Spatial interpolation data of water depth at urban flooding monitoring points at any given time are used as model input to infer... t The invention provides the surface water depth at any given time. By effectively combining spatial interpolation methods and semantic segmentation networks, and taking into account the spatiotemporal correlation and complexity of urban flooding depth, this invention achieves high-precision, large-scale estimation of surface water depth based on urban flooding monitoring points. It has broad application prospects in urban flooding prediction and early warning, drainage and flood control, and other fields.

Description

Technical Field

[0001] This invention belongs to the field of electronic digital data processing technology and relates to urban flooding prediction. Specifically, it relates to a method for spatially inferring the depth of surface water accumulation based on flooding monitoring points, which can be applied to urban flooding prediction and early warning, drainage and flood control, and emergency management. Background Technology

[0002] Urban flooding monitoring helps relevant departments and the public to promptly grasp the depth of floodwater and take necessary countermeasures, thereby reducing or mitigating the harm caused by urban flooding. While water level sensor-based flooding monitoring points are a conventional method for obtaining urban floodwater depth, the high construction costs and maintenance difficulties of these points mean that the current number of urban flooding monitoring points is limited and their spatial distribution is sparse. This means that only a few fixed locations can be identified, which is insufficient to meet practical needs. Furthermore, compared to floodwater depth information from a few monitoring points, obtaining information on the surface water depth over a large urban area is clearly more practical.

[0003] Inferring surface water depth from sparsely distributed urban flooding monitoring data can be considered a spatial interpolation task. However, the spatiotemporal distribution of urban surface water depth exhibits both temporal and spatial correlations and high complexity. Existing spatial interpolation methods, such as inverse distance weighted interpolation and kriging interpolation, primarily consider distance factors or spatial autocorrelation, lacking consideration for the complexity and temporal correlation of surface water distribution. Therefore, they are difficult to apply to large-scale, high-precision estimation of surface water depth.

[0004] In summary, there is an urgent need to develop a new method for spatial estimation of surface water depth based on urban flooding monitoring points, in order to achieve high-precision and large-scale estimation of surface water depth, thereby further reducing or mitigating the harm of urban flooding. Summary of the Invention

[0005] To address the shortcomings of existing methods, this invention proposes a spatial estimation method for surface water depth based on urban flooding monitoring points. It also considers the spatiotemporal correlation and complexity of surface water depth distribution and constructs a spatial estimation model for surface water depth based on a modified semantic segmentation network. This model takes the spatial interpolation data of water depth at urban flooding monitoring points at a given time and historical times as input, and the estimated value of surface water depth at a given time as output, which greatly expands the spatial range of urban flooding perception.

[0006] A method for spatially estimating surface water depth based on urban flooding monitoring points, comprising the following steps:

[0007] Step 1: Obtain historical monitoring data and basic geographic data

[0008] Historical precipitation intensity time series data were obtained from various rain gauge stations in the city by the meteorological department, historical water depth time series data were obtained from various urban flooding monitoring points by the urban water affairs department, and urban elevation data and drainage pipeline density data were obtained from the surveying and mapping department.

[0009] Step 2: Calibrate the urban stormwater model

[0010] Using historical water depth time series data from urban flood monitoring points, the urban stormwater model is calibrated. The calibration objective is to minimize the overall difference between the simulated water depth output by the urban stormwater model at the urban flood monitoring points and the monitored water depth at the monitoring points.

[0011] Step 3: Generate raster data of simulated surface water depth.

[0012] Using historical precipitation intensity time series data from rain gauges to drive the urban stormwater model calibrated in step 2, vector data of simulated surface water depth under historical rainfall are obtained.

[0013] A series of design rainfall sequence data are generated for each rain gauge station. The design rainfall sequence data is used to drive the urban stormwater model calibrated in step 2 to obtain vector data of simulated surface water depth under design rainfall.

[0014] Convert vector data of simulated surface water depth under historical and designed rainfall into raster data with a specified spatial resolution.

[0015] Step 4: Obtain simulated water depth data from urban flooding monitoring points.

[0016] From the raster data of simulated surface water accumulation under designed rainfall, the simulated surface water depth data of each waterlogging monitoring point in the raster cell is obtained as the simulated water depth value of the corresponding waterlogging monitoring point.

[0017] Step 5: Spatial interpolation of water depth at urban flooding monitoring points.

[0018] The spatial resolution of urban elevation data and drainage pipe density data is converted to a specified spatial resolution. The urban elevation and drainage pipe density after resolution conversion are used as auxiliary variables. Spatial interpolation algorithms are used to spatially interpolate the simulated water depth values ​​of urban flooding monitoring points according to the specified spatial resolution, resulting in spatially interpolated raster data of the simulated water depth values ​​of urban flooding monitoring points. Then, spatial interpolation is performed on the historical water depth time series data of urban flooding monitoring points according to the specified spatial resolution, resulting in spatially interpolated raster data of the monitored water depth values ​​of urban flooding monitoring points.

[0019] Step 6: Construct a spatial estimation model for surface water depth

[0020] Modify SegNet: Set the number of channels in the SegNet input layer to... n +1, setting the number of input data channels of the first convolutional layer of SegNet to... n +1; Set the number of output data channels of the last convolutional layer of SegNet to 1, and remove the Softmax layer of SegNet; the remaining layers of SegNet remain unchanged.

[0021] The modified SegNet is used as the spatial inference model for surface water depth. The input of the spatial inference model for surface water depth is... t - n ,…, t -1, t Spatial interpolated raster data of water depth at urban flooding monitoring points at any given time, output as t Raster data of estimated surface water depth at any given time.

[0022] Step 7: Train the spatial prediction model for surface water depth

[0023] To enrich the training set data and improve the model's generalization ability, a rainfall scenario was designed: t - n ,…, t -1, t The spatial interpolation raster data of the simulated water depth at urban flooding monitoring points under constant rainfall is used as the input feature value. t The simulated surface water depth under designed rainfall is used as the output label value to generate a training set for spatial inference of surface water depth under designed rainfall.

[0024] Regarding historical rainfall scenarios: t - n ,…, t -1, t Spatial interpolated raster data of water depth monitoring values ​​at urban flooding monitoring points under historical rainfall at specific times are used as input feature values. t The simulated surface water depth under historical rainfall is used as the output label value to generate a training set for spatial inference of surface water depth under historical rainfall.

[0025] The spatial inference model for surface water depth described in step 6 is trained using the two training sets mentioned above.

[0026] Step 8: Apply the spatial estimation model for surface water depth

[0027] Get the given time and the time before the given time nThe water depth monitoring values ​​at each waterlogging monitoring point are obtained, and spatial interpolation is performed using a spatial interpolation method. The resulting spatially interpolated raster data is input into the surface water depth spatial inference model trained in step 7. The model is then used to output the surface water depth inference raster data at a given time, thus realizing the spatial inference of surface water depth.

[0028] The present invention has the following beneficial effects:

[0029] 1. Using the modified semantic segmentation model to estimate the depth of surface water can fully leverage the advantages of the segmentation model in feature extraction and spatial information preservation, thereby effectively capturing the complexity of the distribution of surface water depth and ensuring the accuracy of the estimation of surface water depth.

[0030] 2. This invention effectively combines spatial interpolation methods and semantic segmentation models, taking into account the spatiotemporal correlation and complexity of urban flooding depth. This allows for the full utilization of data from sparsely distributed urban flooding monitoring points to infer the surface water depth data over a large area of ​​the city, greatly expanding the spatial range of urban flooding perception.

[0031] 3. This invention enables large-scale and high-precision estimation of surface water depth, which helps relevant departments and the public to fully and timely grasp the urban flooding situation, providing reference and guidance for taking reasonable countermeasures, thereby reducing and mitigating the losses caused by flooding. Attached Figure Description

[0032] Figure 1 This is a flowchart of a method for estimating surface water depth based on urban flooding monitoring points;

[0033] Figure 2 This is a schematic diagram of a spatial inference model for surface water depth based on the modified SegNet. Detailed Implementation

[0034] The present invention will be further explained below with reference to the accompanying drawings;

[0035] like Figure 1 As shown, a method for estimating surface water depth based on urban flooding monitoring points includes the following steps:

[0036] Step 1: Obtain historical monitoring data and basic geographic data

[0037] Historical precipitation intensity time series data were obtained from various rain gauge stations in the city by the meteorological department, historical water depth time series data were obtained from various urban flooding monitoring points by the urban water affairs department, and urban elevation raster data and drainage pipe density raster data were obtained from the surveying and mapping department.

[0038] Step 2: Calibrate the urban stormwater model

[0039] A PCSWMM model of urban stormwater was constructed. The model was calibrated using historical water depth time series data from waterlogging monitoring points. The calibration objective was to minimize the mean square error (MSE) between the simulated water depth output by the urban stormwater model at the waterlogging monitoring points and the monitored water depth at the monitoring points.

[0040] Step 3: Generate raster data of simulated surface water depth.

[0041] s3.1 Using historical precipitation intensity time series data obtained from rain gauges to drive the calibrated urban stormwater model, vector data of simulated surface water depth under historical rainfall are obtained.

[0042] s3.2 Generate a series of design rainfall sequence data for each rain gauge station, and use the design rainfall sequence data to drive the calibrated urban stormwater model to obtain vector data of simulated surface water depth under design rainfall.

[0043] s3.3. Using the RasterizeLayer() method in the GDAL library, the vector data of simulated surface water depth under historical rainfall and designed rainfall are converted into raster data with a spatial resolution of 1 meter.

[0044] Step 4: Obtain simulated water depth data from urban flooding monitoring points.

[0045] Assuming there is only one waterlogging monitoring point in each grid cell, the simulated surface water depth data of each waterlogging monitoring point is obtained from the simulated surface water depth raster data under the design rainfall, and used as the simulated water depth value of the corresponding waterlogging monitoring point.

[0046] Step 5: Spatial interpolation of water depth at urban flooding monitoring points.

[0047] s5.1 Use the gdalwarp tool in the GDAL library to convert the spatial resolution of the urban elevation raster data and drainage pipe density raster data to 1 meter.

[0048] s5.2 Set the spatial resolution of the co-kriging method in the PyKrige package to 1 meter, use urban elevation and drainage pipe density as auxiliary variables of the co-kriging method, and use the co-kriging method to spatially interpolate the simulated water depth values ​​of the urban flooding monitoring points to obtain spatial interpolated raster data of the simulated water depth values ​​of the urban flooding monitoring points.

[0049] s5.3. Set the spatial resolution of the co-kriging method to 1 meter, and use urban elevation and drainage pipe density as auxiliary variables for the co-kriging method. Use the co-kriging method to spatially interpolate the water depth monitoring values ​​at urban flooding monitoring points, obtaining spatially interpolated raster data of the water depth monitoring values ​​at these points.

[0050] Step 6: Construct a spatial estimation model for surface water depth

[0051] The SegNet image semantic segmentation model was chosen as the main framework for the spatial inference model of surface water depth. In order to apply the SegNet architecture to water depth spatial inference, the following modifications were made:

[0052] s6.1 Set the number of channels in the SegNet input layer to n +1, setting the number of channels in the input data of the first convolutional layer of SegNet to... n +1, n As hyperparameters, they are determined through a random search method.

[0053] s6.2 Set the number of channels in the output data of the last convolutional layer of SegNet to 1.

[0054] s6.3 Remove the Softmax layer of SegNet, and keep the other layers unchanged.

[0055] like Figure 2 As shown, the modified SegNet is used as a spatial inference model for surface water depth, using a given time and past data. n Spatial interpolation of water depth at waterlogging monitoring points at various times can be used to infer the surface water depth at a given time. This fully leverages SegNet's advantages in feature extraction and spatial information preservation, thereby ensuring the accuracy of surface water depth estimation.

[0056] Step 7: Train the spatial prediction model for surface water depth

[0057] s7.1, For the designed rainfall scenario, t - n ,…, t -1, t The spatial interpolation raster data of the simulated water depth at urban flooding monitoring points under constant rainfall is used as the input feature value. t The simulated surface water depth under designed rainfall is used as the output label to generate a training set for spatial inference of surface water depth under designed rainfall.

[0058] s7.2, Regarding historical rainfall scenarios, t - n ,…, t -1, t Spatial interpolated raster data of water depth monitoring values ​​at urban flooding monitoring points under historical rainfall at specific times are used as input feature values. tThe simulated surface water depth under historical rainfall is used as the output label value to generate a training set for spatial inference of surface water depth under historical rainfall.

[0059] s7.3. Train the surface water depth spatial estimation model described in step 6 using the training sets for spatial estimation of surface water depth under designed rainfall and historical rainfall. The objective function for model training is MSE, and the optimization algorithm is AdamW.

[0060] Step 8: Apply the spatial estimation model for surface water depth

[0061] s8.1 Obtain the given time and the time before the given time n The water depth monitoring value at each waterlogging monitoring point at any given time;

[0062] s8.2 Using urban elevation and drainage pipe density as auxiliary variables, the co-kriging method is used to analyze the data at a given time and before that time. n Spatial interpolation was performed on the water depth monitoring values ​​at each waterlogging monitoring point at each time point.

[0063] s8.3. Input the spatial interpolation raster data obtained in s8.2 into the spatial inference model of surface water depth trained in step 7, and output the raster data of the inferred surface water depth at a given time to realize the spatial inference of surface water depth.

Claims

1. A method for spatially estimating surface water depth based on urban flooding monitoring points, characterized in that: The specific steps are as follows: Step 1: Obtain historical precipitation intensity time series data, historical water depth time series data of urban flooding monitoring points, as well as urban elevation data and drainage pipe density data; Step 2: Using urban elevation data and drainage pipe density data as auxiliary variables, spatial interpolation is performed on the water depth at the urban flooding monitoring points, and the resulting spatial interpolation raster data is used as the input feature value. The urban stormwater model is driven by historical precipitation intensity time series data. The simulated vector data of surface water depth is obtained and converted into raster data as output label values. The training set for spatial inference of surface water depth is constructed using input feature values ​​and output label values. Step 3: Set the number of channels in the SegNet input layer to... n +1. The number of input data channels for the first convolutional layer is set to... n +1. Set the number of output data channels of the last convolutional layer to 1, remove the Softmax layer, and keep the other layers unchanged. The modified SegNet is used as a spatial inference model for surface water depth, based on ( t - n )~ t Spatial interpolation raster data of water depth at urban flooding monitoring points at specific times, used to infer... t Raster data of surface water depth at any given time; n Hyperparameters are determined using random search. n value; Step 4: Train the spatial estimation model of surface water depth from Step 3 using the training set from Step 2; Given the time interval and the previous time interval... n Spatial interpolated raster data of water depth monitoring values ​​at waterlogging monitoring points at a given time are input into the trained model to obtain raster data of the estimated surface water depth at a given time, thus realizing spatial estimation of surface water depth.

2. The method for spatial estimation of surface water depth based on urban flooding monitoring points as described in claim 1, characterized in that: The urban stormwater model was constructed using PCSWMM.

3. The method for spatial estimation of surface water depth based on urban flooding monitoring points as described in claim 2, characterized in that: The urban stormwater model is calibrated using historical water depth time series data from urban flood monitoring points to minimize the overall difference between the simulated water depth values ​​output by the urban stormwater model at the urban flood monitoring points and the monitored water depth values ​​at the urban flood monitoring points.

4. The method for spatial estimation of surface water depth based on urban flooding monitoring points as described in claim 3, characterized in that: Mean squared error is used to measure the overall difference between simulated and monitored values ​​output by urban stormwater models.

5. The method for spatial estimation of surface water depth based on urban flooding monitoring points as described in claim 1, characterized in that: A series of design rainfall sequence data are generated to drive the urban stormwater model, obtaining vector data of simulated surface water depth under design rainfall, which is then converted into raster data and used as output label values. From the raster data of simulated surface water depth under design rainfall, the simulated surface water depth data of each urban flooding monitoring point is obtained, and spatial interpolation is performed on it to obtain spatially interpolated raster data of the simulated water depth of the urban flooding monitoring points, which is used as input feature values. Using the input feature values ​​and output label values, a training set for spatial inference of surface water depth under design rainfall is constructed. This training set supplements the training set obtained in step 2 and is used to train the spatial inference model of surface water depth constructed in step 3.

6. The method for spatial estimation of surface water depth based on urban flooding monitoring points as described in claim 1, characterized in that: The spatial interpolation method is the co-kriging method.

7. The method for spatial estimation of surface water depth based on urban flooding monitoring points as described in claim 1, characterized in that: The objective function for training the spatial inference model of surface water depth is MSE, and the optimization algorithm is AdamW.

8. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of any one of claims 1 to 7.

Images