Measuring urban waterlogging depths from video images based on reference objects

Camera surveillance systems can record urban waterlogging processes. Objects with regular shapes and fixed sizes captured by the camera can be utilized to calculate urban waterlogging depths based on geometric principles. In this study, we propose a machine learning‐based method to measure urban waterlogging depths using wheels and traffic buckets captured in video images as reference objects. This method is validated through laboratory experiments and observed data. The results demonstrate that: (1) the urban waterlogging depths calculated using urban reference objects show high consistency with the observed water level data; (2) in the laboratory scenario, the probability of error within 3 cm for measurements based on the hub, tire, and traffic bucket are 99.07%, 99.38%, and 81.55%, respectively; (3) in the real‐world scenario, the probability of error within 3 cm for measurements based on car hubs and pickup truck hubs are 97.30% and 95.14%, respectively. In conclusion, urban waterlogging depths can be accurately measured using reference objects with regular shapes. The proposed method can help obtain waterlogging data with higher temporal and spatial resolution at lower economic costs, which is of great significance for urban flood control.


| INTRODUCTION
In recent years, climate change and human activities have increased the likelihood of rainstorms and floods (Duan et al., 2022).The rapid urbanization has led to widespread and frequent urban waterlogging disasters.Direct economic losses caused by floods in China have been rising in recent decades (Duan et al., 2016).There are two primary reasons for this increase: one is the escalation of extreme precipitation events due to climate change, and the other is the pressure of urbanization caused by population growth and rapid economic development, which puts a greater number of individuals and properties at risk.In this context, water resources management presents significant challenges to the sustainable development of the socio-economic landscape, severely impacting urban operations, social management, and the safety of people's lives and property (Duan et al., 2021).Urban waterlogging depth serves as fundamental data for hydrological simulations, disaster early warning and prediction, flood control, and drainage.Furthermore, urban waterlogging depth is an essential indicator of urban waterlogging disasters and is widely employed to assess urban waterlogging losses (Huang et al., 2012).
The conventional approach to monitoring urban waterlogging depths utilizing water level monitoring devices demands substantial manpower and material resources, and these devices are susceptible to damage from floods.Consequently, developing a method for rapidly and accurately measuring urban waterlogging depth is of paramount importance.
In recent years, deep learning and image recognition technologies have been employed for urban waterlogging measurements (Hou, 2019).The method of monitoring urban waterlogging depth based on deep learning and image recognition eliminates the need for water level monitoring instruments and extensive manpower.It enables real-time water level data monitoring through video footage.There are three primary approaches: the first method involves training the entire video image as a sample to obtain the water level without labeling the reference object.The second method detects the water level gauge in the video image and calculates the water level value according to the gauge's characteristics.The third method identifies the reference object in the video image and computes the water level value using the geometric characteristic parameters of the reference object.
vanden Boomen et al. (2021) acquired the water level by training the video image of river monitoring using the convolutional neural network (CNN) Visual Geometry Group (VGG-16) without labeling the reference object in the video image.The errors of the recognition results in the three scenarios were 0.133, 0.243, and 0.211 feet, respectively.This method necessitates a separate training model for each monitoring scene and is prone to being influenced by external dynamic factors.
Numerous studies focus on calculating water levels by identifying water level gauges.Zhu et al. (2021) extracted water gauges from video images using the Seg-Net semantic segmentation model and obtained waterlogging depth based on an improved You Only Look Once (YOLO-v3) network with an accuracy rate exceeding 95%.Wang et al. (2020) proposed a real-time water level recognition method based on a deep learning algorithm that detects the water level gauge through YOLO-v3 and identifies the water gauge scale via ResNet to obtain the water level.The accuracy rate in actual monitoring of irrigation areas reached 95.3%, essentially meeting the practical monitoring requirements.Kim et al. (2014) developed a cloud-based image water level gauge system named "River Eye," which identified the water level gauge to obtain water levels and began operation in some regions of Korea, with an average measurement error rate of approximately 10%.Qu et al. (2020) designed an urban waterlogging monitoring system that obtained submerged depth data by locating and identifying the water level gauge, with the error of water level identification within 2 cm.Zhong (2020) employed image recognition technology to recognize the scale of water level gauges and automatically observed the water levels of small reservoirs.The recognition rate under different weather conditions ranged from 38.42% to 94.35%, and the accuracy met the reporting requirements.The method of monitoring inundation depth by identifying water gauges is easily affected by factors such as camera monitoring angle changes, water gauge pollution, and flood damage.
Several studies have explored calculating waterlogging depth using reference objects in video images.Jiang et al. (2019Jiang et al. ( , 2020) ) proposed a method to estimate urban waterlogging depth through reference objects in the video image and constructed a model based on the traffic bucket.In the measurement and testing stage of waterlogging depth, the root mean square error (RMSE) and mean absolute percentage error were 0.026 m and 19.968%, respectively.Chaudhary et al. (2019) constructed a mask region-based convolutional neural network (Mask R-CNN) model using person, car, bus, bicycle, and house as reference objects to estimate waterlogging depth.The average absolute error of the evaluation on the test dataset was consistently below 10 cm.Huang et al. (2020) developed an approach to measure urban waterlogging depth based on Mask R-CNN, which utilized the tire as the reference object and applied the Pythagorean theorem and height differences method to measure waterlogging depth.In the testing stage of the video dataset, the symmetrical mean absolute percentage error and RMSE were 38.24% and 0.186 m, respectively.However, these methods exhibit some limitations, such as not considering the visual error caused by the monitoring angle and lacking measured data of urban waterlogging depth.
In this study, we propose a method for measuring urban waterlogging depths based on the feature parameters of reference objects.Our method considers the visual error caused by the monitoring angle of reference objects in video images, expands the reference object, and optimizes the feature parameter extraction method.The accuracy of this approach in urban waterlogging depth measurement is validated using videos of urban waterlogging in both laboratory and real-world scenarios.The method comprises three main steps: first, the Mask R-CNN model is employed to detect reference objects in video images; second, OpenCV is used to obtain the feature parameters of the exposed parts of the reference objects; and third, the actual feature parameters of the reference objects are combined to calculate the urban waterlogging depth using geometric relationships.Wheels and traffic buckets were chosen as typical reference objects to demonstrate the method in detail, and an experimental platform was constructed to verify the accuracy of the measurement technique.Compared to previous research, the proposed method expands the reference objects to include hubs and refines the extraction method of reference object parameters.This approach enriches the available data sources and enhances the accuracy of waterlogging depth estimation.

| Obtain the urban reference object information from the video images
The process of measuring urban waterlogging depths using video images based on reference objects is illustrated in Figure 1.The critical aspect of this approach is to accurately extract the urban reference object information from the video images.
2.1.1 | Detection of reference objects using mask R-CNN model He et al. (2017) introduced a comprehensive object instance segmentation framework termed Mask R-CNN, applicable for object detection, object instance segmentation, and object keypoint detection.In this study, the Mask R-CNN model is employed to detect reference objects and acquire both the bounding box and object mask.The functionality of Mask R-CNN is demonstrated as follows (depicted in Figure 2): 1. Feed a preprocessed image into a pre-trained CNN to generate the corresponding feature map. 2. Assign a predetermined region of interest (ROI) for each point within the feature map to obtain multiple candidate ROIs.3. Input multiple candidate ROIs into a region proposal network to acquire object proposals and bounding box regression, subsequently filtering out some candidate ROIs. 4. Process the remaining ROIs via the RoIAlign layer to yield a fixed-size feature map. 5. Lastly, pass the remaining ROIs through the fully connected (FC) layer to predict the class label and bounding box (bbox), as well as to predict the object mask using the fully convolutional network layer.
F I G U R E 1 Process of urban waterlogging depth measurement based on reference objects.
The process of Mask R-CNN detecting reference objects.

| Obtaining the feature parameters of reference objects based on OpenCV
Various monitoring angles can lead to geometric distortion of reference objects.The bounding box of reference objects detected using the Mask R-CNN model is not suitable for acquiring feature parameters.The object mask associated with urban reference objects captures the object's feature parameters.In this study, OpenCV is employed to analyze the object mask of reference objects to obtain the feature parameters.

| Calculation of urban waterlogging depths
Certain urban reference objects possess regular shapes and fixed sizes, enabling the calculation of urban waterlogging depths.In this study, we select wheels and traffic buckets as reference objects to illustrate the method of measuring urban waterlogging depths from video images based on reference objects.

| Calculation of waterlogging depths based on wheel
The wheel, comprising a tire and hub, exhibits regular shapes and fixed sizes that can be determined by vehicle type.During urban waterlogging events, three scenarios of wheel inundation occur: (1) the urban waterlogging depth is less than the wheel radius, (2) the urban waterlogging depth is between the wheel radius and wheel diameter, and (3) the urban waterlogging depth is greater than the wheel diameter.In the third scenario, calculating the urban waterlogging depth becomes impossible; therefore, this paper presents detailed methods for determining urban waterlogging depths in the first two scenarios.

Urban waterlogging depth less than the wheel radius
In this scenario, the urban waterlogging depth is less than the wheel radius, as illustrated in Figure 3.By integrating the tire parameters observed in the video images with the actual tire radius, the urban waterlogging depth can be computed using Equation ( 1) where h w is the urban waterlogging depth; a t and b t are the tire width and the tire height of the exposed part in pixel units; R t is the actual tire radius.
Incorporating the hub parameters from the video images, the actual hub radius, and the actual tire thickness, the urban waterlogging depth can be determined using Equation ( 2) where a h and b h are the hub width and the hub height of the exposed part in pixel units; R h is the actual hub radius; h t is the actual tire thickness.
Urban waterlogging depth between wheel radius and wheel diameter.
The urban waterlogging depth lies between the wheel radius and wheel diameter, as depicted in Figure 4.
In accordance with the Pythagorean theorem, it is possible to determine the radius of a tire (or hub) in pixel units by utilizing Equation (3) in conjunction with the relevant parameters depicted in video images.
where r t h ð Þ is the tire (or hub) radius in pixel units; a t h ð Þ and b t h ð Þ are the tire (or hub) width and the tire (or hub) height of the exposed part in pixel units.
Equation (4) enables the calculation of the ratio coefficient between the tire (or hub) radius in pixel units and its actual radius.
where k t h ð Þ is the ratio coefficient between the tire (or hub) radius in pixel units and the actual tire (or hub) radius; R t h ð Þ is the actual tire (or hub) radius.
The urban waterlogging depth is less than the wheel radius.h w is the urban waterlogging depth; a t and b t are the tire width and the tire height of the exposed part in pixel units; r t is the tire radius in pixel units; a h and b h are the hub width and the hub height of the exposed part in pixel units; r h is the hub radius in pixel units; h t is the actual tire thickness.
Equation ( 5) allows for the computation of the urban waterlogging depth based on the relevant tire parameters.
Equation ( 6) provides a means of calculating the urban waterlogging depth based on the relevant hub parameters.

| Calculation of waterlogging depths based on traffic bucket
The city has numerous traffic buckets with regular shapes and standard sizes that can be utilized to compute the depth of urban waterlogging.Figure 5 depicts the state of a traffic bucket during urban waterlogging.
The sum of the traffic bucket height of the exposed portion and the depth of urban waterlogging equates to the height of the traffic bucket, and the width of the exposed portion remains constant.Nonetheless, due to the monitoring angle of the camera, visual errors occur in the recorded video images, as illustrated in Figure 6.
Using the position diagram of the camera and traffic bucket as illustrated in Figure 5, it is possible to compute the visual error angle by applying Equation (7).
where α is the visual error angle; h c is the actual camera height; l is the actual distance between camera and traffic bucket; H and W are the actual height and width of traffic bucket.Equation ( 8) can be used to determine the urban waterlogging depth based on the traffic bucket parameters.
where h w is the urban waterlogging depth; a and b are the width and height of traffic bucket in pixel units.

| The laboratory apparatus
To simulate urban waterlogging disasters and record video and water level data, a laboratory apparatus

F I G U R E 4
The urban waterlogging depth is between wheel radius and wheel diameter.h w is the urban waterlogging depth; a t and b t are the tire width and the tire height of the exposed part in pixel units; r t is the tire radius in pixel units; a h and b h are the hub width and the hub height of the exposed part in pixel units; r h is the hub radius in pixel units; h t is the actual tire thickness.
consisting of a rainfall simulator system, an electric fan system, an urban waterlogging simulation device, and a measurement system was developed.An overview of the laboratory apparatus is depicted in Figure 7a.

The rainfall simulator system
The rainfall simulator system has the capability to simulate rainfall with intensities ranging from 0 to 250 mm/h and drop diameters between 0.5 and 5.5 mm.It has a height of 12.5 m (Gao et al., 2021).

The urban waterlogging simulation device
The urban waterlogging simulation device was constructed and depicted in Figure 7b.The model comprises a platform for simulating urban waterlogging, a background wall for simulating urban scenes, and reference objects with a 1:1 scale.The platform's size is 3000 Â 2000 Â 500 mm 3 , enabling it to simulate urban waterlogging with depths ranging from 0 to 50 cm.The background wall's size is 5000 Â 1000 mm 2 , which simulates the city scene.In this study, the wheel and traffic bucket, both of 1:1 size, were selected as reference objects for detecting the depth of urban waterlogging.

The electric fan system
The electric fan system, illustrated in Figure 7e, is comprised of four industrial axial fans with a diameter of 1.0 m and a frequency converter.It can generate a wind field with a wind speed range of 0-7 m/s.

The measurement system
The measurement system comprises a video monitoring device and a water level measuring device, shown in Figure 7c, d, respectively.The video surveillance device consists of two 2-megapixel cameras with a focal length of 3.6 mm, a total pixel length of 1280 Â 720, and a frame rate of 25 fps.The water level measuring device is a pressure water level gauge, with an accuracy of 1 mm and the ability to adjust its sampling frequency to a minimum of 1 s.

| Experiment data
To obtain urban waterlogging video data and water level data, urban waterlogging simulation experiments were conducted in different scenarios.The experiment simulated 35 urban waterlogging processes and yielded 125 h of urban waterlogging video and 45,199 real-time water level monitoring data with a 5-s interval.
To train the Mask R-CNN model, every 10 s, a frame of urban waterlogging video with a size of 1920 Â 1080 pixels was extracted.From the video, 497 images were selected as training samples, and reference objects in each video image were manually labeled using a labeling tool named "labelme." To measure the depth of urban waterlogging, an urban waterlogging video was chosen, and the accuracy of the measurement was compared with the real-time water level monitoring data.The video duration was 54 min, and it generated 645 real-time monitoring water level data with water level changes ranging from 10.91 to 36.86 cm.

| Results of reference objects detection from video images
Figure 8 presents the detection results of reference objects derived from video images.In various scenarios (when the urban waterlogging depth is less than the wheel radius, and when the urban waterlogging depth lies between the wheel radius and wheel diameter), the Mask R-CNN model accurately detects and segments the reference objects.The tire size used was 145/70R12, with a wheel diameter of 490 mm and a hub diameter of 330 mm.The traffic bucket size was 400 Â 700 mm.

| Results of parameter acquisition of reference objects
Figures 9, 10 present the results of the wheel and traffic bucket parameters.To obtain the feature parameters, that is, the width and height of the exposed part of the urban reference object in the video image, OpenCV was used to detect the outer rectangle of the object mask of the urban reference object.

| The accuracy of measuring urban waterlogging depths in the laboratory scenario
To assess the performance of the urban waterlogging depth measurement method, a 54-minute urban waterlogging video was selected to measure the waterlogging depth based on the reference objects.The results were then compared with the real-time water level monitoring data, as shown in Table 1 and Figure 11.The water level changed from 10.91 to 36.86 cm during the experiment, with a total of 645 real-time monitoring water level data recorded.Four performance indexes were selected to evaluate the accuracy of the urban waterlogging depth measurement method: the probability of error within 2, 3, and 5 cm, and the average error value.
Table 1 presents the performance of the urban waterlogging depth measurement method using different reference objects.When the hub was used as the reference object, the probability of error within 2 cm was 94.73%, the probability of error within 3 cm was 99.07%, the probability of error within 5 cm was 100.00%, and the average error was 0.85 cm.When the tire was used as the reference object, the probability of error within 2 cm was 89.77%, the probability of error within 3 cm was 99.38%, the probability of error within 5 cm was 100.00%, and the average error was 0.85 cm.When the traffic bucket was used as the reference object, the probability of error within 2 cm was 64.03%, the probability of error within 3 cm was 81.55%, the probability of error within 5 cm was 96.12%, and the average error was 1.74 cm.
Figure 11 displays the distribution of observed water levels and measured urban waterlogging depths based on reference objects.The RMSE was used to evaluate the accuracy of waterlogging depths, with results of 1.1 cm for the hub, 1.1 cm for the tire, and 2.3 cm for the traffic bucket.
In conclusion, the highest accuracy for urban waterlogging depth measurement from video images was achieved using the hub as the reference object, possibly due to the clear boundary and significant color difference between the hub and the tire, resulting in more accurate measurement results.

| The accuracy of measuring urban waterlogging depths in the real-world scenario
To validate the accuracy of the proposed method for monitoring urban waterlogging, a real-world video of waterlogging was selected and analyzed (see Figure 12).The video depicts the waterlogging scenario during Typhoon Rumbia on August 19, 2018, in Dongying City, Shandong Province, China.The video has a duration of 30 min, with a data acquisition interval of 10 s, resulting in 185 measurements of the waterlogging depth.The water level varied from 14.30 to 23.80 cm during the video recording.
The video image includes a pickup truck and a car, but the unclear tire boundary and absence of a traffic bucket make it challenging to use the tire and the traffic bucket for reference in monitoring waterlogging depth.Instead, only the hub is selected as a reference object.To ensure accuracy, the vehicle type is first identified, and the corresponding hub size is assigned based on established tire parameters for different types of vehicles.Past studies have explored the identification of vehicle models through video (Krause et al., 2013;Li et al., 2017;Yang et al., 2015), and this paper will not elaborate further.The urban waterlogging depth is measured based on the hub, and the results are compared with the observed waterlogging depth in Table 2 and Figure   Table 2 illustrates the performance of different types of vehicle hubs in measuring urban waterlogging depths under real-world conditions.For waterlogging depths measured using car hubs, the probability of error within 2 cm is 89.73%, within 3 cm is 97.30%, and within 5 cm is 100.00%.The average error result is 1.09 cm.For waterlogging depths measured using pickup truck hubs, the probability of error within 2 cm is 75.14%,within 3 cm is 95.14%, and within 5 cm is 100.00%.The average error result is 1.25 cm.
Figure 13 displays the distribution of observed water levels and measured urban waterlogging depths based on different vehicle hubs under real-world conditions.The accuracy of waterlogging depth measurements is evaluated using the RMSE.The RMSE of waterlogging depth measurement results using car hubs and pickup truck hubs are 1.3 and 1.5 cm, respectively.
In conclusion, the use of vehicle hubs for monitoring urban waterlogging depth has demonstrated high accuracy under real-world conditions.T A B L E 1 The performance of the urban waterlogging depth measurement method in the Laboratory Scenario.

The reference object
The probability of error within 2 cm (%) The probability of error within 3 cm (%) The probability of error within 5 cm (%)  This research focused on measuring urban waterlogging depths from video images using reference objects.The accuracy of the method was verified in both laboratory and real-world scenarios.This method has the advantages of being low-cost and having simple calculations.By using urban monitoring videos as the data source, this method does not require the installation of additional monitoring equipment.By identifying reference objects with fixed parameters and regular shapes in the video images and extracting their feature parameters, the urban waterlogging depth can be calculated based on the geometric relationship.
As urbanization continues to develop, a large number of cameras are being installed in urban areas, many of which can be used for measuring urban waterlogging depth using the method proposed in this paper.While water level monitoring equipment can obtain accurate data, they are limited in number and mostly installed in central urban areas.This method can be used as a supplement to water level monitoring equipment to obtain urban waterlogging data with higher temporal and spatial resolutions.This is significant for urban flood control and drainage.T A B L E 2 The performance of the urban waterlogging depth measurement method in the real-world scenario.

The reference object
The probability of error within 2 cm (%) The probability of error within 3 cm (%) The probability of error within 5 cm (%) The average error value (cm)

| Comparison with existing approaches
Recent research (Chaudhary et al., 2019;Huang et al., 2020;Jiang et al., 2019Jiang et al., , 2020) ) has explored methods for measuring urban waterlogging depths from video images using reference objects.These methods utilize objects such as traffic buckets, tires, and vehicles as reference objects and offer new ideas and methods for monitoring waterlogging depths in areas without water level monitoring equipment.However, these methods have some shortcomings, such as not considering the visual error caused by the monitoring angle and a lack of measured data on waterlogging depth in cities.
Building upon existing approaches, this paper introduces a new method for measuring urban waterlogging depths using the hub as a reference object.When extracting feature parameters from reference objects, the visual error caused by the monitoring angle is considered, and OpenCV is used to further improve the feature parameter extraction method.The accuracy of the method in measuring urban waterlogging depth is verified in laboratory and real-world scenarios.This study overcomes some of the shortcomings of previous methods, leading to more accurate measurement results of urban waterlogging depths.

| Problems and prospects
The accuracy of urban waterlogging depth measurements is significantly impacted by the quality of the video image.First, the definition of the video image affects the measurement results.If the image is unclear or the reference object color is too similar to the background, the detection and feature parameter extraction of the urban reference object may be inaccurate, resulting in incorrect waterlogging depth measurements.This can be resolved by using high-definition cameras and reference objects with a significant color difference.Second, fluctuations in the water surface due to wind and moving vehicles can also cause measurement errors.These uncertainties can be mitigated through multiple measurements at different times.Additionally, the lateral slope of lanes in urban areas limits the use of waterlogging depth at the wheel to represent the entire area's waterlogging depth.To address this issue, the solution is to convert the slope and obtain the water depth near the wheel.
Despite some inherent uncertainty in the measurement results of the method proposed in this article, it has significant practical implications for flood control in areas lacking waterlogging depth monitoring data.Moreover, future research will explore the accuracy of monitoring urban waterlogging depth using other reference objects under real-world conditions.

| CONCLUSIONS
Accurately measuring the urban waterlogging depth is essential for flood control and drainage management.In this study, a method is proposed to measure urban waterlogging depth by detecting and extracting characteristic parameters of regular-shaped urban reference objects in video images.The method calculates the waterlogging depth based on the geometric relationship and verifies its accuracy in both laboratory and realworld scenarios.The results demonstrate high consistency between the waterlogging depths calculated based on the urban reference object and the observed water level data.The laboratory experiments show a high probability of error within 3 cm of the measurement results based on the hub, tire, and traffic bucket, with RMSEs of 1.1, 1.1, and 2.3 cm, respectively.In the realworld scenario, the method shows a probability of error within 3 cm of the measurement results based on the car hub and the pickup truck hub of 97.30% and 95.14%, respectively, with RMSEs of 1.3 and 1.5 cm, respectively.Therefore, this study concludes that regular-shaped reference objects can accurately measure the urban waterlogging depth.

F
I G U R E 5 Schematic diagram of traffic bucket submerged by urban waterlogging.h w is the urban waterlogging depth; a and b are the width and height of traffic bucket in pixel units; h e is the visual error height.F I G U R E 6 Location diagram of camera and traffic bucket.α is the visual error angle; h c is the actual camera height; l is the actual distance between camera and traffic bucket; H and W are the actual height and width of traffic bucket; h e is the visual error height.

F
I G U R E 7 Laboratory apparatus.(a)-(e) show the overall experimental device, the urban waterlogging simulation device, the video monitoring, the water level measuring device, and the electric fan system, respectively.
13.The evaluation metrics used are the same as in Section 3.3.1.

F
I G U R E 8 Detection effect of reference objects in different scenarios.(a) The urban waterlogging depth is less than the wheel radius; (b) The urban waterlogging depth is between wheel radius and wheel diameter.

F
I G U R E 1 0 Results of traffic bucket parameters.(a) Results of traffic bucket inspection detection; (b) results of the parameters are obtained from the traffic bucket mask image.

F
I G U R E 9 Results of wheel parameters.(a) Results of wheel inspection detection; (b) results of the parameters are obtained from the tire mask image; (c) results of the parameters are obtained from the hub mask image.4 | DISCUSSION 4.1 | Practicability of the study Results of urban waterlogging depths measurement in the Laboratory Scenario.(a) Comparison between urban waterlogging depth measurement results and observed water level in the laboratory scenario.(b) Correlation between observed water level and urban waterlogging depth measurement results based on hub; (c) correlation between observed water level and urban waterlogging depth measurement results based on tire; (d) correlation between observed water level and urban waterlogging depth measurement results based on traffic bucket.F I G U R E 1 2 Video of urban waterlogging during Typhoon Rumbia in Dongying City, Shandong Province, China.

F
I G U R E 1 3 Results of urban waterlogging depths measurement in the real-world scenario.(a) Comparison between urban waterlogging depth measurement results and observed water level in the real-world scenario.(b) Correlation between observed water level and urban waterlogging depth measurement results based on car hub and pickup truck hub.