lidar data processing for detailed inundation simulation ... - Google Sites

0 downloads 145 Views 2MB Size Report
to the measured LiDAR data, simulated the inundation process, and compared the results, with and without a ..... We also
7th International Conference on Hydroinformatics HIC 2006, Nice, FRANCE

LIDAR DATA PROCESSING FOR DETAILED INUNDATION SIMULATION OF AN URBANIZED AREA RYOTA TSUBAKI Graduate School of Science and Technology, Kobe University, 1-1 Rokkodai-cho, Nada-ku, Kobe 657-8501, Japan ICHIRO FUJITA Department of Architecture and Civil Engineering, Kobe University, 1-1 Rokkodai-cho, Nada-ku, Kobe 657-8501, Japan HIROSHI TERAGUCHI Politecnique School, São Paulo University, Avenida Professor Luciano Gualberto, Travessa 3, 380 – Butantã – São Paulo, Brazil (CEP:05508-900) An automated unstructured grid generation system from raw LiDAR data for inundation analysis of urbanized and suburban areas is developed. With this system, it becomes possible to perform detailed inundation prediction easily and quickly after a LiDAR measurement is carried out, whose accuracy is much higher than the conventional GIS (Geographic Information System). This system is applicable to regions where there are no GIS data regarding detailed building arrangements. We applied the developed system to the measured LiDAR data, simulated the inundation process, and compared the results, with and without a vegetation model. INTRODUCTION The economic and human suffering caused by inundation induced by bank failure, overflow, concentrated heavy rain-fall and tsunami waves in urbanized areas are problems that occurs in the past and at present. When estimating the human damage due to flooding water, not only inundation depth and rate of water level rise but also local flow velocity will become of great influence (Figure 1). To estimate such local flow by using numerical simulation models, consideration of detailed topographical information is necessary, and not only detailed but also rational and efficient treatment of such information is essential. The purpose of this study is to establish a methodology to obtain an accurate, and detailed numerical model for inundation prediction. In order to achieve the aim mentioned above, the generation of an adequate unstructured grid with detailed topology, numerical accuracy, and stability for calculation is necessary. Topographical information considered in this paper is obtained by airborne laser altimetry (LiDAR), and a data processing procedure is built up to utilize the unique characteristics of LiDAR measurement.

Figure 1. Urban flooding [from Kobe Shimbun (Kobe times)]. RELATED WORKS Many researchers studied inundation by using numerical models including ground slope, surface state and so on. Xanthopoulos and Koutitas [1] developed a numerical inundation model based on a finite difference method in 1976. They simulated flood propagation on a Greek plane discretized with a 1 km mesh grid, and several roughnesses were considered to represent the ground nature. The grid spacing they used was rough in the present-day sense because there were limitation of not only computational availability but also geographical information they could use. However, now we have access to abundant computational performance and an enormous amount of geographical information. Mason et al. [2] proposed a model for a friction factor caused by vegetation such as trees and crops. With this method, vegetation is classified into three categories, short vegetation (height less than 1.2 m), tall vegetation (greater than 5 m) and intermediate vegetation. The friction of each category was estimated by different models. Cobby et al. [3] proposed a method for generate a mesh for a finite-element discretization. In this method, to represent the important topographic features such as hedges and trees accurately, an unstructured grid was used. River bathymetry was individually represented by a structured triangular mesh first, then the mesh was connected smoothly into the floodplain by changing grid size gradually using a distance matrix. Bates et al. [4] explored the optimum assimilation of high-resolution data into numerical models. To implement this, significant length scales were estimated, and significant points were determined, an unstructured grid was generated to represent topographic features well and finally sub-grid scale topographic information was incorporated into the model. However, detailed flow structure and modeling method in an urbanized area is as yet not well know. LIDAR DATA PROCESSING FOR DETAILED INUNDATION SIMULATION OF AN URBANIZED AREA Several steps of LiDAR data processing we developed in this study are depicted in Figure 2. The first step, we should implement while processing raw LiDAR data, is the extraction of digital terrain model (DTM; Figure 2d) from a directly observed digital

300m

Elevation (m)

45

a) Orthophoto. a) Actual state

5

b) DSM.

c) DTM Roughness

b) RAW data by LiDAR/DSM

d) Surface model in this study

Figure 3. Four expressions of the surface state. c) Surface classification into convex (white), ground (light gray) and concave (dark gray) region.

d) DTM interpolated from data belonging to the ground region.

e) Convexe region mask which suggests presence of building and vegetaion.

f) Vegetation mask obtained from color information analysis of aerial photograph.

Figure 2. Steps of LiDAR and color data processing. surface model (DSM; Figure 2b). Several approaches are developed for this purpose, and it provides acceptable accuracy in a sense. By using DTM and DSM, we can divide the whole area into three segments; convex, ground and concave areas. A sample of the convex region is shown in Figure 2e. In the next step, we sub-divide the convex region into a built region (has solid surface) and vegetated region. In order to perform a detailed inundation analysis to estimate local flow on an individual street, the vegetated zone and buildings should be represented using different models. Meanwhile, if both buildings and vegetation are represented as solid surface as in Figures 3b, inundation flow is divided not only by buildings but also by vegetation, and flooded water will accumulated due to vegetation before overflowing over the top of the vegetation, which will yield a non-realistic result. In reality, the flow may go through the vegetated area while receiving a drag force from the vegetation (Figure 3d). On the other hand, in an urbanized area, vegetated areas are often situated close to buildings. However, it is not easy to separate those by using only the height distribution obtained by LiDAR.

a) Actual state

c) DTM Roughness

b) RAW data by LiDAR/DSM

d) Surface model in this study

Figure 3. Four expressions of the surface state. The interior of submerged building may be filled by inundated water so modeling the building area without submerging boundary is not quite realistic. However, the instantaneous flow rate invading the inner area of buildings is not easily measured on site. Obviously, inundation into buildings damages the interior and people in the buildings, however invasion of flow into buildings may have a small effect on the inundation flow on the street because exterior of building separates the flow. In order to distinguish between vegetated and built areas, we first tried to use height information such as shape detection technique [5], and the difference of the first and the last pulse heights [6], but we concluded that it is difficult to achieve this while using data with 1 m grid spacing because: #1 the shape of a small house is difficult to distinguish from that of a tree, #2 trimmed shrubbery has no specific shape and there is only a small difference between first and last pulse height. To overcome the difficulty in distinguishing buildings and vegetation by using only the height information, we utilize the combined information of height and color to detect buildings and vegetation, instead of using only the height data, as in the multi-spectral approach of Rottensteiner et al. [7]. The question we have to ask here is how to identify the vegetation area by using color information. The normalized difference vegetation index (NDVI) is widely used for the purpose of estimating vegetation existence and condition in the remote sensing survey. However, we do not have the near infrared images, which are used to compute NDVI, so we use optical (usual color) images represented by digital camera obtained from an airplane while the LiDAR measurement is underway. The color property of the vegetated area may depend on season, kind of vegetation, weather, and time when image is recorded, so the sample region of vegetation is extracted manually (see Figure 4). Then characteristics within the sample region and the whole recorded area are compared and the vegetation area sparsely located within the whole area can be detected such as shown in Figure 2f. The vegetated area within the convex mask is removed and finally we obtain a building mask. So far, region segmentation has been conducted and we now generate an unstructured grid of ground via the following steps. First, the boundary of each building is picked up from the building mask and simplified while preserving its original shape. Second, the boundary grid is arranged using boundary information prepared in the first

Figure 4. Sample region for vegetation color information extracted from aerial photograph. Color of each pixel within this sample images is used to extract characteristic of the vegetation region. Pattern itself is not used.

m 0 0 3

N 100m a) Unstructured grid (white = buildng area, light gray = vegetated area; dark gray = ground).

b) Conceptual representation of grid shown in persective angle.

Figure 5. Generated grid. The area shown here is the same as Figure 3.

B

1500 m

A C

Elevation (m)

50

5 a) Bed evelation

b) Orthophoto

Figure 6. Subject area. step. Third, the inner region of the boundary grid is filled up by using the advancing front method. Forth, locations of nodes in the inner region are modified by using a combination of the Laplacian smoothing and the local optimization techniques. Finally, local ground height and roughness are attributed to each grid. A generated unstructured grid with height and roughness information is depicted in Figure 5.

Inflow

Water depth (m)

3

a) Water depth predicted w ith vegetation model.

b) Water depth predicted w ithout vegetation model.

0

Velocity magnitude (m/s)

1.5

c) Velocity magnitude predicted w ith vege- d) Velocity magnitude predicted w ithout tation model. vegetation model.

0

Figure 7. Calculated result at t=2000 s obtained using with vegetation model (a, b) and without vegetation treatment (modeling as wall boundary; Figure c, d). INUNDATION SIMULATION USING FIELD DATA So far we have outlined the way to obtain the calculation grid. In this section we would perform inundation simulation on an urbanized area by using actual LiDAR data to confirm detailed topographic effects on the flood flow. The computational domain discussed in this section is a rectangular region, 1500m on a side, as shown in Figure 6. General ground slope in this area is comparatively steep (about 30 m / 1500 m = 0.02). The numerical model is generated by using the procedure described above and we obtained unstructured grid consisting of 37000 nodes and 46000 triangles. In this study, we focused on the effect on detailed topography modeling, a simple inflow condition is set, a constant flow rate of 200 m3/s where depicted with arrows in Figure 7a. This condition is based on the assumption that flood caused by bank failure.

Point A B C A B C

Water depth (m)

3 2.5 2 1.5 1 0.5 0 0

500

a) Water depth

Point A B C A B C

2.5 Velocity magnitude (m/s)

3.5

1000 Elapsed time (s)

1500

2000

2 1.5 1 0.5 0 0

500

1000 Elapsed time (s)

1500

2000

b) Velocity magnitude

Figure 8. Comparison of time series of local water depth and velocity. Black symbols indicate result calculated with vegetation model and white symbols without vegetation model. Locations of points A, B and C is shown in Figure 6a. To estimate flood flow, shallow water equations discretized by the finite volume method is used (Shige-eda et al. [8,9]). The flow parameter is located in each cell center and a flux difference scheme is introduced to achieve numerical stability. Bed friction is estimated by using Manning’s friction formula. Manning’s n is set constant 0.02 for the ground area, and 0.2 on the vegetation region. As demonstrated in Figure 7, the flood spreads through road networks, and some regions are predicted to have deep (over 3 m) water depth, and fast-flowing (about 1.5 m / s) inundation water on some streets. Comparing the local flow structure of the model with and without vegetation, there are differences in detail, however the extent of inundation itself is generally similar. To investigate the differences quantitatively, time series of water depth and velocity at three locations depicted in Figure 6a are compared in Figure 8. Point A is located in the playground surrounded by trees, and points B and C are in the center of streets. By comparison with model without vegetation, calculated water depth with vegetation model at 3 points are shallow, and the time submerge starts is late. A difference of velocity is more complex, though both maxima and variation of velocity of vegetation model are large compared with no vegetation modeling. CONCLUSION In this study, an unstructured grid generation system from raw LiDAR data for flood prediction in urbanized and suburban areas is developed. The approach distinguishes buildings from vegetation and treats them differently in the model, as wall boundaries for buildings and a rough bed for vegetation, respectively. The unstructured grid is generated considering the building shape and vegetation. We conducted hypothetical flood calculations and discussed the effect of modeling vegetation. The instantaneous distributions of water depth and velocity magnitude are compared and slight differences are observed. The time series of water depth and velocity at 3 locations are also compared. We found that the result without vegetation model compared with model regarding vegetation shows overestimate for water depth and underestimate for maximal velocity.

Out future plan is to measure actual inundation process caused by heavy rain fall in urban area in detail to estimate and improve detailed flood models for inundation of urbanized areas. ACKNOWLEDGEMENTS This research was conducted as the part of the research fellow of the JSPS and 21st COE program for Kobe University. We are grateful to Assistant Prof. Jeremy D. Bricker for many fruitful suggestion and advice. We also are grateful to Dr. Takeshi Kawatani for giving valuable advice. The authors wish to acknowledge Asian Air Survey Co., ltd. for processing LiDAR data. REFERENCES [1] Xanthopoulos T. and Koutitas C., “Numerical simulation of a two dimensional flood wave propagation due to dam failure”, Journal of Hydraulic Research, Vol. 14, No. 2, (1976), pp 321-339. [2] Mason D. C., Cobby D. M., Horritt M. S. and Bates P. D., “Floodplain friction parameterization in two-dimensional river flood models using vegetation heights derived from airborne scanning laser altimetry”, Hydrological processes, Vol. 17, (2003), pp 1711-1732. [3] Cobby D. M., Mason D. C., Horritt M. S. and Bates P. D., “Two-dimensional hydraulic flood modeling using a finite-element mesh decomposed according to vegetation and topographic features derived from airborne scanning laser altimetry”, Hydrological processes, Vol. 17, (2003), pp 1979-2000. [4] Bates P. D., Marks K. J. and Horritt M. S., “Optimal use of high-resolution topographic data in flood inundation models”, Hydrological processes, Vol. 17, (2003) , pp 537-577. [5] Vosselman G., “3D reconstruction of roads and trees for city modelling”, Proc. ISPRS working group III/3 workshop, Dresden, (2003). [6] Clode S. and Rottensteiner F., “Classification of trees and powerlines from medium resolution airborne laserscanner data in urban environments”, Proc. APRS Workshop on Digital Image Computing, Brisbane, (2005). [7] Rottensteiner F., Trinder J., Clode S. and Kubik K., “Building detection using LiDAR data and multi-spectral images”, Proc. VIIth Digital Image Computing: Techniques and Applications, Sydney, (2003), pp 673-682. [8] Shigeda M., Akiyama J., Ura M., Jha A. K. and Arita Y., “Numerical simulations of flood propagation in a flood plain with structures”, Journal of Hydroscience and Hydraulic Engineering, JSCE, Vol. 20, No. 2, (2002), pp 117-129. [9] Shige-eda, M., Akiyama, J., “Numerical and experimental study on 2D flood flows with and without structures”, Journal of Hydraulic Engineering, ASCE, Vol. 129, No. 10, (2003), pp 817-821.

Suggest Documents