Improved Floodplain Delineation Method Using ... - Semantic Scholar
Computer-Aided Civil and Infrastructure Engineering 28 (2013) 68–79
Improved Floodplain Delineation Method Using High-Density LiDAR Data Sagar S. Deshpande∗ Leonard Jackson Associates, Vienna, VA, USA
Abstract: With the improvements in sensor technologies over the past decade, there has been a significant decrease in the cost of acquisition and increase in the density and accuracy of Light Detection and Ranging (LiDAR) data. Due to its advantages over traditional surveying techniques, LiDAR data are widely preferred for floodplain delineation. But, processing dense LiDAR data is time-consuming and memory intense. Therefore, it is divided into manageable areas/tiles or simplified to raster DEM (Digital Elevation Model) format for feature extraction process such as floodplain delineation. This results in increase in processing time and decrease in accuracy due to loss of true elevation. Furthermore, as floodplain boundaries are unknown prior to delineation, processing time also increases as LiDAR data over larger extent is processed. Hence, there is a need of improved, automated method that will process only the LiDAR data that contribute to the floodplain. This article, describes a time-efficient floodplain delineation method that divides the LiDAR data into regular tiles and processes only the tiles that contribute to floodplain. This method is experimented using LiDAR data saved in ArcGIS “Terrain” format at 0.0, 0.1, and 0.3 m pyramid levels. These data are then preprocessed to obtain elevation information which is used to filter and process only LiDAR data tiles that truly contribute to the floodplain boundary; thus, reducing processing time. Results from two pilot hydraulic models showed that this method saved 12–34% of processing time compared to the conventional method.
1 INTRODUCTION Floods are one of the most common, serious, and costly natural disasters. During past decades, the number of flood disasters occurring worldwide has grown significantly. Climate change combined with growing urban ∗ To
whom correspondence should [email protected].
be
addressed.
C 2012 Computer-Aided Civil and Infrastructure Engineering. DOI: 10.1111/j.1467-8667.2012.00774.x
E-mail:
areas, have increased the frequency and the severity of flood events (Sole et al., 2008). Due to increased threats of flooding, both inland and coastal, accurate floodplain delineation is essential to make decisions regarding construction, insurance, and other regulatory practices (Noman et al., 2003). In the United States, the Federal Emergency Management Agency (FEMA) manages the National Flood Insurance Program (NFIP) who produces Flood Insurance Rate Maps (FIRMs). The FIRMs show areas of high flood risk by mapping the 1% annual chance flood. This information is used in town planning, in purchasing homeowners insurance, by home buyers, and by existing home owners who need to know how to appropriately protect their property from flooding. For all of these users, accurate floodplain mapping is essential. Light Detection and Ranging (LiDAR) data are exceedingly accurate; therefore, it is being used for floodplain delineation. In this article, a time-efficient method is presented for floodplain delineation using high density LiDAR data. This method involves: preprocessing of LiDAR data to extract elevation information for each tile and use of this information to filter and process the tiles that truly contribute to the floodplain. Analysis has been performed to study the time saved using Terrain data at 0, 0.1, and 0.3 m pyramid levels. These levels were selected so that the accuracy of the topographic data is within FEMA specifications. The main objectives are to make the floodplain delineation process computationally efficient by avoiding processing of unnecessary data and to use the best resolution of LiDAR data to produce accurate floodplain delineation. 2 BACKGROUND INFORMATION AND LITERATURE REVIEW In FEMA’s NFIP program, countywide FIRMs are prepared that show flood hazard zones for riverine and
Floodplain delineation method using LiDAR data
coastal flooding sources. These zones are identified by performing hydrologic and hydraulic analysis, or coastal analysis, using atmospheric, topographic, and geospatial information. Topography is one of the prime components which influence the mapping accuracy. FEMA requires that the topographic data accuracy requirements, regarding map scale and contour intervals, be equivalent to the National Map Accuracy Standard (NMAS) (FEMA, 2003a). FEMA further defines two general categories for vertical accuracy of topographic data as: 1. Two-Foot Equivalent—Data that have an accuracy of ±36.5 cm (±1.2 ft) at the 95% confidence interval [i.e.: 95% of data points have accuracy with respect to the true ground elevation equal to ±36.5 cm (1.2 ft) or smaller]. 2. Four-Foot Equivalent—Data that have an accuracy of ±73.2 cm (±2.4 ft) at the 95% confidence interval. The horizontal accuracy requirement of topographic data is a function of the intended map panel scale. Due to its high vertical accuracy, laser sensor technology in the form of LiDAR has quickly become the prime source of topographic information for hydrologic/hydraulic analysis and floodplain delineation. Due to its advantages over traditional measurement equipment, several researchers have studied use of terrestrial as well as airborne laser technology in structural monitoring (Lee and Park, 2011; Park et al., 2007; Siringoringo and Fujino, 2009; and Zalama et al., 2011), transportation engineering (Cai and Rasdorf, 2008), and other fields of civil engineering. Although its horizontal accuracy is less than vertical accuracy (Cai and Rasdorf, 2008; Liu et al., 2009), the combined accuracy of LiDAR data exceeds that of the traditional topographic data used by floodplain managers (Maune, 2001), as well as FEMA’s quality standards. With the improvements in sensor technology over the past decade, there has been a significant increase in the resolution and a decrease in the cost of acquisition of LiDAR data (Chen, 2007). LiDAR data can also measure ground elevation in vegetated areas, therefore increasing the detail when compared to an aerial survey (Sasaki et al., 2008). This improved detail provides more accurate topography along a river or stream in vegetated areas. The basic principle of a LiDAR survey is based on laser distance measurement using a scanning mirror mechanism. At a sampling rate of 30 kHz or higher, a LiDAR system can produce spatial elevation data with an average horizontal spacing of 1 m or less (Liu et al., 2009); thus providing very dense elevation data. Use of dense LiDAR data is time-consuming and memory intense. Hence, it is processed by several meth-
69
ods to reduce the frequency of elevation points such as manual editing, spatial and statistical filtering, and multiple return analysis techniques (Bryant et al., 2002). Rath and Pasche (2004) presented a fast and efficient method to reduce huge LiDAR data sets based on slope classification for hydrodynamic simulation. Omer et al. (2003) used original data and seven filtered data sets and found that filtering to 4◦ can be performed without compromising cross-sectional geometry, hydraulic model results, or floodplain delineation results. Some research also studied tiling of the data files and irregular spatial pattern of the LiDAR data. Tiling is a process in which data are reorganized and stored in contiguous regular “tiles” (Chen, 2007). Hug et al. (2004) and Chen (2007) have proposed tools to efficiently organize and process LiDAR data using tiles. Conventionally, LiDAR data have been converted over a smaller extent to topographic surfaces, such as a raster Digital Elevation Model (DEM; a grid of squares with elevation information), Triangulated Irregular Network (TIN), or contours, to facilitate processing. The raster DEMs of different resolutions can be created by interpolating point elevation information from LiDAR data. Colby and Dobson (2010) compared flood modeling results using digital terrain model (DTM) of different resolutions and found that higher resolution terrain is needed to better represent floodplain in low-relief areas whereas higher resolution data may be useful in high-relief areas due to steep slopes. Several other studies have also investigated the effects of topographic data resolution and mapping (Garbrecht and Martz, 1994; Zhang and Montgomery, 1994; Molnar and Julien, 2000; Moglen and Hartman, 2001; Colby and Dobson, 2010). Interpolation of original random points to create DEM results in loss of true elevation values. There is a need to develop improved automated processing techniques that preserve the original random point data (Bryant et al., 2002). Unlike a raster DEM, a TIN maintains the exact features of the LiDAR data accurately. TIN is preferable to a DEM when it is critical to preserve the precise location of narrow or small surface features such as levees or narrow stream channels (Maune, 2001). The mapping accuracy of the TIN surface created by using LIDAR data is much more accurate than alternative mapping accuracies on an interpolated 2 ft contour surface, a 10 ft contour surface, or an NED 1 arc second DEM surface (Cohen, 2007). Gesch (2009) also revealed that the high vertical accuracy and spatial resolution of LiDAR data improve identification and delineation of vulnerable lands. Although increasingly available, efficient processing and extraction of useful information using LiDAR data remains a big challenge in several fields (Chen, 2007).
70
Deshpande
Fig. 1. Floodplain delineation procedure.
Danner et al. (2007) presented a four-stage method which involved conversion of terrain data to raster DEM for extraction of river network and watershed hierarchy. This method was able to process over 20 GB of raw data in fewer than 26 hours. Seventy-six percent of this time was spent in the initial DEM construction stage. Wallis et al. (2009) described a parallel algorithm to enhance hydrologic terrain preprocessing so that larger data sets can be efficiently computed. Several other studies have been reported which use LiDAR data for flood risk modeling, hydrologic analysis, and hydraulic analysis (Sole et al., 2008; Bryant et al., 2002; Sasaki et al., 2008; Miller and Shrestha, 2004; Tate et al., 2002; Colby and Dobson, 2010). The increase in processing time for LiDAR data is a result of thousands of ground elevation points which have been obtained in a small area. In general, floodplain delineation for tens of kilometers of river requires processing millions of LiDAR points. Moreover, floodplain delineation involves several iterations to obtain a hydraulically correct model and continuous floodplain boundary using preliminary floodplain. These iterations require even more time and are computationally intense. Due to these timeconsuming processes, it is important to create an efficient tool which can delineate floodplain for using LiDAR.
3 DEFINITION OF RIVERINE FLOODPLAIN DELINEATION Riverine floodplain delineation is a process of identifying floodplain boundaries for a river. This process can be performed either manually, using topographic maps, or also can be performed using digital data saved on computer. Floodplain delineation performed using computers is more accurate and time efficient than the manual method and errors due to misinterpretations are avoided. Typically, to perform floodplain delineation using computers, topographic data are converted to TIN (Figure 1a) or raster DEM. Then, flood water elevations, simulated by a hydraulic model or any other similar source are interpolated to create a water surface. Figure 1b shows a TIN water surface overlain on TIN topographic data. The line of intersection of these two surfaces defines the floodplain boundary. Figure 1c identifies the area below the water surface as floodplain.
4 DATA SET AND SOFTWARE DESCRIPTION To demonstrate the methodology, LiDAR data covering Oswego County, NY were used (Figure 2).
Floodplain delineation method using LiDAR data
71
Fig. 2. LiDAR data in ArcGIS Terrain format covering Oswego County, NY along with the locations of water surfaces of Models 1 and 2 (a) and area enlarged in (b).
These data were collected in June 2008 and consist of 843 million elevation points with an average spacing of 1.2 m (4 ft). These data require 77.3 and 30.4 GB of disk space in LAS format and ArcGIS Terrain format, respectively. The horizontal datum referenced was the North American Datum of 1983 (NAD83), and the vertical datum referenced was the North American Vertical Datum of 1988 (NAVD88). The vertical accuracy of this LiDAR data is 20 cm (0.66 ft) (LiDAR Quality Assurance (QA) Report, Oswego County, NY: Dated June 26, 2009) and thus, satisfies the FEMA standards for topographic data. Two riverine hydraulic models, shown in Figure 2 as Models 1 and 2, were used to study time efficiency of the new method. These models are approximately 27 and 34.8 km in length. It is assumed that the hydraulic simulation of the two models is complete and the water surfaces obtained from these models are used in the new floodplain delineation process. ArcGIS version 9.3 was used to store and process the LiDAR data in the ArcGIS-specific Terrain data format. A customized tool was developed in Visual Studio 2008 Express using the tools available in ArcGIS to implement the new procedure of floodplain delineation. The computer used for processing, analysis, and visualization of the data had 2 GB RAM and Intel Core 2 Duo ([email protected]) central processing unit (CPU).
5 FLOODPLAIN DELINEATION WORKFLOW USING HIGH DENSITY LIDAR Figure 3 shows the entire workflow of the new floodplain delineation method. This workflow is described in detail in following subsections. 5.1 Hydraulic model setup in GIS environment Importing and setting up a hydraulic model is a routine procedure; therefore, it is briefly described in this subsection. As described earlier, several hydraulic modeling softwares are available. Among these programs, Hydrologic Engineering Centers River Analysis System (HEC-RAS) hydraulic model developed by the U.S. Army Corps of Engineers (USACE) is the most common. It is designed to perform one-dimensional hydraulic calculations for a full network of natural and constructed channels (http://www.hec.usace.army.mil/ software/hec-ras/hecras-features.html). HEC-GeoRAS is another program, developed by the USACE, for an ArcGIS environment which can be used to transfer data from ArcGIS to HEC-RAS for modeling simulations. Once the hydraulic modeling is complete, the output from HEC-RAS, in the form of georeferenced crosssections with flood water elevations, can be imported into ArcGIS using HEC-GeoRAS for floodplain delineation. Flood water surfaces are created by interpolating the flood water elevations at each cross-section.
72
Deshpande
Fig. 3. Schematic diagram showing hydraulic model setup in GIS environment, preprocessing LiDAR data, and use of preprocessing data in the new floodplain delineation method.
Likewise, an existing historic hydraulic model is also a source of flood water elevations. Mostly, such model is available only in nondigital format as it was developed in 1980s or 1990s using Hydraulic Engineering Center’s HEC-2 program. The cross-sections used in this model are digitized in a GIS environment from a historic map. Flood water elevations from the historic hydraulic model are manually assigned to these crosssections. Virtual water surfaces are created by interpolating the flood water elevations at each digitized crosssection. This process is called redelineation (FEMA, 2003b). Yang et al. (2005) also presents another GISbased approach for delineating and displaying data generated by HEC-2 numerical model. In this study, Model 1 was imported using historic data and Model 2 was obtained by importing a HECRAS model. The flood water surfaces for both models (Figure 4) were then created as TINs by linear interpolation of the flood water elevations along the crosssections.
5.2 LiDAR data preprocessing Generally, raw LiDAR data are converted to software specific formats which are capable of storing and re-
trieving a huge amount of data sets. This facilitates the use of tools which are available in the software to process the data. The subsections which follow describe the importing and preprocessing steps of LiDAR data into an ArcGIS environment. 5.2.1 Importing LiDAR data into an ArcGIS environment. ArcGIS’s Terrain datum is a relatively new format, which was introduced after the release of its 9.2 version. It is a TIN-based platform that can produce accurate surfaces quickly by efficiently managing pointbased data; such as LiDAR in a geodatabase. This data set is also capable of managing, editing, and producing accurate TINs and also allows creating TIN pyramid levels based on Z-tolerance. The lowest pyramid level (0.0 m) stores all of the elevation points. At a higher pyramid level, points are eliminated from the Terrain, based on the Z-tolerance value; thus, reducing the number of points. For example, data points are eliminated from the Terrain based on 0.1 m pyramid level (Z-tolerance) to produce surface that is within a 0.1 m vertical accuracy relative to the 0.0 m pyramid surface. To explore the benefits of this data set using this new method, the LiDAR data for Oswego County, NY were converted to Terrain format. Tools available in
Floodplain delineation method using LiDAR data
73
Fig. 4. Water surface and cross-section layout of Models 1 and 2.
ArcGIS 9.3 were used to convert the LiDAR data from Oswego County, NY to the Terrain format (Figure 2a). LiDAR elevation points classified as “bare earth” were used to produce the Terrain. Three pyramids: 0.0 m (base level), 0.1 m (0.5 ft), and 0.3 m (1 ft) were created to study the performance efficiency and change in floodplain area at different levels using the new method. These pyramids were selected so that the accuracy of the topographic data at each pyramid level would be within the two-foot equivalent vertical accuracy of ±36.5 cm (±1.2 ft) defined by FEMA. 5.2.2 Creation of tiles to divide the Terrain data into a regular grid pattern. The Terrain data have to be converted to TIN or raster DEM format to perform any analysis. In this new method, Terrain data are converted to TIN because it is more accurate than raster DEM. Converting the entire Terrain data, which contain millions of points, to a TIN data set will required huge memory and would slow down the entire process. The realistic limit for a TIN format in ArcGIS is about 10 million nodes (Atkinson, 2010). Furthermore, converting only the area of Terrain that overlaps the water surface of Models 1 and 2 is also memory intense and exceeds the node limits of a TIN. Therefore, it is required to convert a manageable portion of the entire Terrain to perform any analysis. A regular tile pattern covering the entire extent of the Terrain data is then created which divides the LiDAR data. Based on the average spacing of the LiDAR data, tiles of size 410 m × 410 m are created. Figure 2b shows an enlarged area of the Terrain data overlain by the tiles.
5.2.3 Preprocessing Terrain data. The purpose of preprocessing the Terrain data is to obtain the elevation information within each tile. Terrain data are clipped to the extents of each overlapping tile as a TIN surface. A tool named “Terrain to TIN,” available in ArcGIS, is used to obtain a TIN surface for the extent of each tile. Minimum and maximum elevation values are obtained from each clipped TIN data and are then saved as attributes of each tile. This step cannot be performed manually because thousands of tiles cover the entire LiDAR data set. Hence, a tool is developed to implement this step. After extracting the elevation information, the clipped TINs are deleted, because they are duplicates of the Terrain and also require larger disk space. The elevation parameters obtained by preprocessing are then used to filter only those grids which truly contribute to the floodplain. This process is described in the next subsection.
5.3 New floodplain delineation method As described earlier, a floodplain boundary is a line where the dry ground intersects the water surface. Figure 5 shows a schematic representation of three possible scenarios that can occur between a triangle of ground surface TIN and floodwater surface in the floodplain delineation process. Figure 5a represents the first scenario where the entire water surface is above the ground surface and Figure 5b represents the second scenario where the water surface is entirely below the ground surface. It
74
Deshpande
Fig. 5. The three scenarios that can occur between ground surface TIN and water surface.
should be noted that these scenarios represent all triangles of ground surface TIN that are entirely inside or outside the boundaries of the floodplain. In the first scenario, the resultant floodplain is represented by the water surface area, as the ground data are below water whereas in the second scenario, the area covered by the water surface will not contribute to the floodplain boundary. Thus, such triangles are not processed to identify floodplain boundaries as they are entirely below or above the water surface. Figure 5c represents the third scenario in which the water surface intersects the ground surface TIN. Such triangles are processed to identify the line of intersection, which will delineate the floodplain boundary. This situation would occur when the elevation of water surface is between the minimum and maximum elevations of the overlapping topographic data. Considering the three different scenarios discussed earlier, the computational time can be reduced only if the topographic data and the water surface data, that satisfy the third scenario, are processed. These triangles can be identified if the minimum and maximum value of the topographic data is available by preprocessing. However, it is not time-efficient to preprocess and check millions of triangles in a Terrain data set which is based on the above three scenarios. Therefore, the above three scenarios are generalized to filter the tiles that contribute to the floodplain. Figure 6a shows the tiles that overlap the water surface of Model 1 and its vicinity. To reduce the number of tiles, only the tiles that spatially overlap the water surface are selected (Figure 6b). Processing each overlapping tile to obtain the floodplain boundary is a timeconsuming task because it involves clipping the Terrain
data to TIN format, which covers thousands of points and then performing floodplain delineation by intersecting the TIN format with the virtual water surface. Thus, the tiles that truly contribute to the floodplain boundary are filtered by implementing all three of the criterion described in the previous subsection. A step-by-step approach, as described below, is implemented for each tile that overlaps the water surface: 1. Clip the flood water surface TIN to the extents of the overlapping tile. (Note: It is faster to clip the flood water surface TIN as it has few elevation points within a tile.) 2. Extract the minimum (WSmin ) and maximum (WSmax ) elevation value for the clipped water surface TIN. 3. Retrieve the minimum (TOPOmin ) and maximum (TOPOmax ) elevation values saved as attributes to the tile during the preprocessing stage. These values represent the maximum and minimum elevation for the topographic data within the tile. 4. Perform the following checks to identify if the tile would contribute to the floodplain boundary: i. Case 1: (WSmax ) < (TOPOmin ) = True. In this case, the ground surface within this tile is entirely free from flooding and doesn’t contribute to the floodplain. Therefore, LiDAR data are not processed to perform floodplain delineation. ii. Case 2: (WSmin ) > (TOPOmax ) = True. In this case, the ground surface within the tile is entirely covered with water and, therefore, this entire tile contributes to the floodplain. Because it
Floodplain delineation method using LiDAR data
75
Fig. 6. Water surface of Model 1 and the overlapping tiles.
is entirely covered with water, there is no floodplain boundary within this tile and its LiDAR data are not processed to perform floodplain delineation. iii. Case 3: In this case, it can be observed that portions of the ground data intersect the water surface. The Terrain data within this tile are processed to identify the areas within the tile that are above and the areas that are below the water surface. A tool named “TIN difference” available in ArcGIS is used to identify these areas and perform floodplain delineation. By examining these three scenarios, and performing the appropriate checks, the tiles which directly contribute to the floodplain boundary can be identified. The time required to clip the Terrain data and to perform floodplain delineation is eliminated for the tiles that do not contribute to the floodplain boundary which significantly reduces processing time. To determine the time saved by implementing this method, floodplain delineation was also performed for Models 1 and 2 at three pyramid levels using the conventional methodology. The three scenarios used in the new method were not implemented and, as a result, all the tiles that overlapped the water surface were processed in the conventional method to obtain the floodplain boundary. The processing time required for floodplain delineation and floodplain area, at different pyramid
Table 1 Time required to preprocess the Terrain data Pyramid level (m) Preprocessing results Time required (hours) Percent of time required compared to base-level pyramid Average number of points within each tile Percent of average point density compared to base-level pyramid
0
0.1
0.3
40.4 100%
34.6 86%
17.1 42%
73,793
36,500
4,384
100%
49.50%
5.90%
levels for Models 1 and 2, are presented in the next section.
6 RESULTS AND DISCUSSIONS As described previously, the LiDAR data are converted to Terrain data at three pyramid levels: 0.0, 0.1, and 0.3 m and preprocessing is done based on the overlapping 24,252 tiles that cover the entire Oswego County, NY. Table 1 summarizes the time required to preprocess the Terrain data. The results show that higher preprocessing time was required for the lower level pyramids. It took around
76
Deshpande
Fig. 7. Tiles identified based on the three cases.
40 hours to preprocess the Terrain at 0.0 m pyramid; whereas only 35 and 17 hours were required to process the Terrain at 0.1 and 0.3 m pyramids, respectively. The average density of data points at 0.1 and 0.3 m pyramids was 50 and 6%, respectively, of the point density at 0.0 m pyramid. The percentage decrease in the average number of points is significant from 0.0 to 0.1 m and to 0.3 m pyramids, compared to percentage decrease in the processing time. Thus, the reduction in the time is primarily due to the decrease in the average point density within each tile, and is also affected by the time required to clip the data and save the attributes. In Figure 7, tiles classified using the three scenarios that overlap the water surface of Models 1 and 2 are shown. It can be seen that using the new method, only the nonshaded tiles are processed. Figure 8 shows the floodplain obtained for the two models using the 0.0 m level pyramid. Floodplains delineated using 0.1 and 0.3 m pyramid Terrain were not identical to those delineated using 0.0 m pyramid and showed variations along the floodplain boundaries. Floodplain delineation was also performed by using the conventional method, which did not use the preprocessed information from the three pyramid levels. Identical floodplains were obtained by the new method and the conventional method at each pyramid level. Results involving the floodplain areas at three pyramid levels,
processing time, and other factors for Models 1 and 2 are tabulated in Tables 2 and 3. The following points are based on the results listed in Tables 2 and 3: 1. All tiles that overlap the water surface were processed in conventional floodplain delineation method; whereas only the short-listed tiles were processed using the new method. The new method was able to reduce the number of processed tiles by 43 and 17% for Models 1 and 2, respectively. 2. Large numbers of Case 2 tiles were observed for both models. Overall floodplain width of Model 2 was narrower than Model 1. Therefore, a large number of Case 1 tiles were observed for Model 1. 3. Using this method, 29, 32, and 34% of time were saved for Model 1 and 12, 13, and 17% of time were saved for Model 2 using the three levels of pyramid Terrain, respectively. This shows that the time saved increases as the density of data points decreases. 4. Processing time of 122 and 45 minutes was required for Models 1 and 2, respectively, by conventional method using Terrain at 0.0 m pyramid. A time saving of 82 and 76% was observed for the two models, respectively, using the new method at 0.3 m pyramid.
Floodplain delineation method using LiDAR data
77
Fig. 8. Floodplain boundary at pyramid level 0.0 m.
5. The change in area of the floodplain obtained by using 0.1 m pyramid Terrain was less than 1%. Using the 0.3 m pyramid Terrain, the change was less than 3%. Close visual inspection showed that the floodplain width slightly increased in some areas whereas it decreased in other areas, therefore compensating for the change in area as a result of using the higher pyramid Terrains. 6. Spot check along the entire floodplain boundary obtained using 0.1 m pyramid Terrain showed less than 5 ft variations to that obtained using 0.0 m pyramid Terrain. But, discrepancy more than 10 ft was observed at several locations between floodplain boundary obtained using 0.3 m pyramid Terrain. Particularly, in low-relief terrain, floodplain delineated using the 0.0 m pyramid Terrain represented minor topographic variations accurately compared to floodplain from other two Terrains. When visually compared to the orthoimagery, the floodplain boundaries obtained at 0.0 and 0.1 m pyramids were more appropriate than that obtained at 0.3 m pyramid. It was observed that the following factors affected the number of skipped tiles: 1. Floodplain width: The count of tiles that were entirely within the floodplain boundary, depended on the width of the floodplain. There were a higher
number of these types of tiles in Model 1 compared to Model 2. 2. Cross-section extent: The number of tiles entirely outside the floodplain depends on the extent of the hydraulic model cross-sections. The cross-sections were extended to enclose the floodplain boundary between them because the expected floodplain boundary between them was not known. Therefore, for very narrow floodplains, the number of such tiles could increase significantly.
7 CONCLUSIONS Over the past decade, density of LiDAR data has increased due to development in sensor technology. This calls for development of task-specific and time-efficient methods for feature extraction. Over the past several years, methods to filter LiDAR data point density have been researched. This article presents a time-efficient floodplain delineation method using high-density LiDAR data. This method preprocesses LiDAR data divided into regular tiles to obtain elevation information. This elevation information is then used to filter and process only the tiles that contribute to the floodplain, thus reducing the processing time. Furthermore, this method can be applied to filtered as well as raw data to save additional processing time. Results based on two pilot studies showed that this method can improve time
78
Deshpande
Table 2 Comparison of parameters obtained for floodplain delineation of Model 1 using conventional method and the new method
Table 3 Comparison of parameters obtained for floodplain delineation of Model 2 using conventional method and the new method
Pyramid level (m) Processing results Tiles processed by conventional method Tiles processed by new method Percent reduction in number of tiles Time required by conventional method (minutes) Time required by new method (minutes) Percent of processing time saved by new method compared with conventional method Percent of processing time saved by the new method compared with 0.0 m pyramid Floodplain area (sq. meters) Percent change in floodplain area
Pyramid level (m)
0
0.1
0.3
Processing results
454
454
454
259
259
259
43%
43%
43%
122
84
33
86
57
22
29%
32%
34%
Tiles processed by conventional method Tiles processed by new method Percent reduction in number of tiles Time required by conventional method (minutes) Time required by new method (minutes) Percent of processing time saved by new method compared with conventional method Percent of processing time saved by the new method compared with 0.0 m pyramid Floodplain area (sq. meters) Percent change in floodplain area
NA
34%
19,696,854.24 19,696,854.24 NA
Improved Floodplain Delineation Method Using High-Density LiDAR Data Sagar S. Deshpande∗ Leonard Jackson Associates, Vienna, VA, USA
Abstract: With the improvements in sensor technologies over the past decade, there has been a significant decrease in the cost of acquisition and increase in the density and accuracy of Light Detection and Ranging (LiDAR) data. Due to its advantages over traditional surveying techniques, LiDAR data are widely preferred for floodplain delineation. But, processing dense LiDAR data is time-consuming and memory intense. Therefore, it is divided into manageable areas/tiles or simplified to raster DEM (Digital Elevation Model) format for feature extraction process such as floodplain delineation. This results in increase in processing time and decrease in accuracy due to loss of true elevation. Furthermore, as floodplain boundaries are unknown prior to delineation, processing time also increases as LiDAR data over larger extent is processed. Hence, there is a need of improved, automated method that will process only the LiDAR data that contribute to the floodplain. This article, describes a time-efficient floodplain delineation method that divides the LiDAR data into regular tiles and processes only the tiles that contribute to floodplain. This method is experimented using LiDAR data saved in ArcGIS “Terrain” format at 0.0, 0.1, and 0.3 m pyramid levels. These data are then preprocessed to obtain elevation information which is used to filter and process only LiDAR data tiles that truly contribute to the floodplain boundary; thus, reducing processing time. Results from two pilot hydraulic models showed that this method saved 12–34% of processing time compared to the conventional method.
1 INTRODUCTION Floods are one of the most common, serious, and costly natural disasters. During past decades, the number of flood disasters occurring worldwide has grown significantly. Climate change combined with growing urban ∗ To
whom correspondence should [email protected].
be
addressed.
C 2012 Computer-Aided Civil and Infrastructure Engineering. DOI: 10.1111/j.1467-8667.2012.00774.x
E-mail:
areas, have increased the frequency and the severity of flood events (Sole et al., 2008). Due to increased threats of flooding, both inland and coastal, accurate floodplain delineation is essential to make decisions regarding construction, insurance, and other regulatory practices (Noman et al., 2003). In the United States, the Federal Emergency Management Agency (FEMA) manages the National Flood Insurance Program (NFIP) who produces Flood Insurance Rate Maps (FIRMs). The FIRMs show areas of high flood risk by mapping the 1% annual chance flood. This information is used in town planning, in purchasing homeowners insurance, by home buyers, and by existing home owners who need to know how to appropriately protect their property from flooding. For all of these users, accurate floodplain mapping is essential. Light Detection and Ranging (LiDAR) data are exceedingly accurate; therefore, it is being used for floodplain delineation. In this article, a time-efficient method is presented for floodplain delineation using high density LiDAR data. This method involves: preprocessing of LiDAR data to extract elevation information for each tile and use of this information to filter and process the tiles that truly contribute to the floodplain. Analysis has been performed to study the time saved using Terrain data at 0, 0.1, and 0.3 m pyramid levels. These levels were selected so that the accuracy of the topographic data is within FEMA specifications. The main objectives are to make the floodplain delineation process computationally efficient by avoiding processing of unnecessary data and to use the best resolution of LiDAR data to produce accurate floodplain delineation. 2 BACKGROUND INFORMATION AND LITERATURE REVIEW In FEMA’s NFIP program, countywide FIRMs are prepared that show flood hazard zones for riverine and
Floodplain delineation method using LiDAR data
coastal flooding sources. These zones are identified by performing hydrologic and hydraulic analysis, or coastal analysis, using atmospheric, topographic, and geospatial information. Topography is one of the prime components which influence the mapping accuracy. FEMA requires that the topographic data accuracy requirements, regarding map scale and contour intervals, be equivalent to the National Map Accuracy Standard (NMAS) (FEMA, 2003a). FEMA further defines two general categories for vertical accuracy of topographic data as: 1. Two-Foot Equivalent—Data that have an accuracy of ±36.5 cm (±1.2 ft) at the 95% confidence interval [i.e.: 95% of data points have accuracy with respect to the true ground elevation equal to ±36.5 cm (1.2 ft) or smaller]. 2. Four-Foot Equivalent—Data that have an accuracy of ±73.2 cm (±2.4 ft) at the 95% confidence interval. The horizontal accuracy requirement of topographic data is a function of the intended map panel scale. Due to its high vertical accuracy, laser sensor technology in the form of LiDAR has quickly become the prime source of topographic information for hydrologic/hydraulic analysis and floodplain delineation. Due to its advantages over traditional measurement equipment, several researchers have studied use of terrestrial as well as airborne laser technology in structural monitoring (Lee and Park, 2011; Park et al., 2007; Siringoringo and Fujino, 2009; and Zalama et al., 2011), transportation engineering (Cai and Rasdorf, 2008), and other fields of civil engineering. Although its horizontal accuracy is less than vertical accuracy (Cai and Rasdorf, 2008; Liu et al., 2009), the combined accuracy of LiDAR data exceeds that of the traditional topographic data used by floodplain managers (Maune, 2001), as well as FEMA’s quality standards. With the improvements in sensor technology over the past decade, there has been a significant increase in the resolution and a decrease in the cost of acquisition of LiDAR data (Chen, 2007). LiDAR data can also measure ground elevation in vegetated areas, therefore increasing the detail when compared to an aerial survey (Sasaki et al., 2008). This improved detail provides more accurate topography along a river or stream in vegetated areas. The basic principle of a LiDAR survey is based on laser distance measurement using a scanning mirror mechanism. At a sampling rate of 30 kHz or higher, a LiDAR system can produce spatial elevation data with an average horizontal spacing of 1 m or less (Liu et al., 2009); thus providing very dense elevation data. Use of dense LiDAR data is time-consuming and memory intense. Hence, it is processed by several meth-
69
ods to reduce the frequency of elevation points such as manual editing, spatial and statistical filtering, and multiple return analysis techniques (Bryant et al., 2002). Rath and Pasche (2004) presented a fast and efficient method to reduce huge LiDAR data sets based on slope classification for hydrodynamic simulation. Omer et al. (2003) used original data and seven filtered data sets and found that filtering to 4◦ can be performed without compromising cross-sectional geometry, hydraulic model results, or floodplain delineation results. Some research also studied tiling of the data files and irregular spatial pattern of the LiDAR data. Tiling is a process in which data are reorganized and stored in contiguous regular “tiles” (Chen, 2007). Hug et al. (2004) and Chen (2007) have proposed tools to efficiently organize and process LiDAR data using tiles. Conventionally, LiDAR data have been converted over a smaller extent to topographic surfaces, such as a raster Digital Elevation Model (DEM; a grid of squares with elevation information), Triangulated Irregular Network (TIN), or contours, to facilitate processing. The raster DEMs of different resolutions can be created by interpolating point elevation information from LiDAR data. Colby and Dobson (2010) compared flood modeling results using digital terrain model (DTM) of different resolutions and found that higher resolution terrain is needed to better represent floodplain in low-relief areas whereas higher resolution data may be useful in high-relief areas due to steep slopes. Several other studies have also investigated the effects of topographic data resolution and mapping (Garbrecht and Martz, 1994; Zhang and Montgomery, 1994; Molnar and Julien, 2000; Moglen and Hartman, 2001; Colby and Dobson, 2010). Interpolation of original random points to create DEM results in loss of true elevation values. There is a need to develop improved automated processing techniques that preserve the original random point data (Bryant et al., 2002). Unlike a raster DEM, a TIN maintains the exact features of the LiDAR data accurately. TIN is preferable to a DEM when it is critical to preserve the precise location of narrow or small surface features such as levees or narrow stream channels (Maune, 2001). The mapping accuracy of the TIN surface created by using LIDAR data is much more accurate than alternative mapping accuracies on an interpolated 2 ft contour surface, a 10 ft contour surface, or an NED 1 arc second DEM surface (Cohen, 2007). Gesch (2009) also revealed that the high vertical accuracy and spatial resolution of LiDAR data improve identification and delineation of vulnerable lands. Although increasingly available, efficient processing and extraction of useful information using LiDAR data remains a big challenge in several fields (Chen, 2007).
70
Deshpande
Fig. 1. Floodplain delineation procedure.
Danner et al. (2007) presented a four-stage method which involved conversion of terrain data to raster DEM for extraction of river network and watershed hierarchy. This method was able to process over 20 GB of raw data in fewer than 26 hours. Seventy-six percent of this time was spent in the initial DEM construction stage. Wallis et al. (2009) described a parallel algorithm to enhance hydrologic terrain preprocessing so that larger data sets can be efficiently computed. Several other studies have been reported which use LiDAR data for flood risk modeling, hydrologic analysis, and hydraulic analysis (Sole et al., 2008; Bryant et al., 2002; Sasaki et al., 2008; Miller and Shrestha, 2004; Tate et al., 2002; Colby and Dobson, 2010). The increase in processing time for LiDAR data is a result of thousands of ground elevation points which have been obtained in a small area. In general, floodplain delineation for tens of kilometers of river requires processing millions of LiDAR points. Moreover, floodplain delineation involves several iterations to obtain a hydraulically correct model and continuous floodplain boundary using preliminary floodplain. These iterations require even more time and are computationally intense. Due to these timeconsuming processes, it is important to create an efficient tool which can delineate floodplain for using LiDAR.
3 DEFINITION OF RIVERINE FLOODPLAIN DELINEATION Riverine floodplain delineation is a process of identifying floodplain boundaries for a river. This process can be performed either manually, using topographic maps, or also can be performed using digital data saved on computer. Floodplain delineation performed using computers is more accurate and time efficient than the manual method and errors due to misinterpretations are avoided. Typically, to perform floodplain delineation using computers, topographic data are converted to TIN (Figure 1a) or raster DEM. Then, flood water elevations, simulated by a hydraulic model or any other similar source are interpolated to create a water surface. Figure 1b shows a TIN water surface overlain on TIN topographic data. The line of intersection of these two surfaces defines the floodplain boundary. Figure 1c identifies the area below the water surface as floodplain.
4 DATA SET AND SOFTWARE DESCRIPTION To demonstrate the methodology, LiDAR data covering Oswego County, NY were used (Figure 2).
Floodplain delineation method using LiDAR data
71
Fig. 2. LiDAR data in ArcGIS Terrain format covering Oswego County, NY along with the locations of water surfaces of Models 1 and 2 (a) and area enlarged in (b).
These data were collected in June 2008 and consist of 843 million elevation points with an average spacing of 1.2 m (4 ft). These data require 77.3 and 30.4 GB of disk space in LAS format and ArcGIS Terrain format, respectively. The horizontal datum referenced was the North American Datum of 1983 (NAD83), and the vertical datum referenced was the North American Vertical Datum of 1988 (NAVD88). The vertical accuracy of this LiDAR data is 20 cm (0.66 ft) (LiDAR Quality Assurance (QA) Report, Oswego County, NY: Dated June 26, 2009) and thus, satisfies the FEMA standards for topographic data. Two riverine hydraulic models, shown in Figure 2 as Models 1 and 2, were used to study time efficiency of the new method. These models are approximately 27 and 34.8 km in length. It is assumed that the hydraulic simulation of the two models is complete and the water surfaces obtained from these models are used in the new floodplain delineation process. ArcGIS version 9.3 was used to store and process the LiDAR data in the ArcGIS-specific Terrain data format. A customized tool was developed in Visual Studio 2008 Express using the tools available in ArcGIS to implement the new procedure of floodplain delineation. The computer used for processing, analysis, and visualization of the data had 2 GB RAM and Intel Core 2 Duo ([email protected]) central processing unit (CPU).
5 FLOODPLAIN DELINEATION WORKFLOW USING HIGH DENSITY LIDAR Figure 3 shows the entire workflow of the new floodplain delineation method. This workflow is described in detail in following subsections. 5.1 Hydraulic model setup in GIS environment Importing and setting up a hydraulic model is a routine procedure; therefore, it is briefly described in this subsection. As described earlier, several hydraulic modeling softwares are available. Among these programs, Hydrologic Engineering Centers River Analysis System (HEC-RAS) hydraulic model developed by the U.S. Army Corps of Engineers (USACE) is the most common. It is designed to perform one-dimensional hydraulic calculations for a full network of natural and constructed channels (http://www.hec.usace.army.mil/ software/hec-ras/hecras-features.html). HEC-GeoRAS is another program, developed by the USACE, for an ArcGIS environment which can be used to transfer data from ArcGIS to HEC-RAS for modeling simulations. Once the hydraulic modeling is complete, the output from HEC-RAS, in the form of georeferenced crosssections with flood water elevations, can be imported into ArcGIS using HEC-GeoRAS for floodplain delineation. Flood water surfaces are created by interpolating the flood water elevations at each cross-section.
72
Deshpande
Fig. 3. Schematic diagram showing hydraulic model setup in GIS environment, preprocessing LiDAR data, and use of preprocessing data in the new floodplain delineation method.
Likewise, an existing historic hydraulic model is also a source of flood water elevations. Mostly, such model is available only in nondigital format as it was developed in 1980s or 1990s using Hydraulic Engineering Center’s HEC-2 program. The cross-sections used in this model are digitized in a GIS environment from a historic map. Flood water elevations from the historic hydraulic model are manually assigned to these crosssections. Virtual water surfaces are created by interpolating the flood water elevations at each digitized crosssection. This process is called redelineation (FEMA, 2003b). Yang et al. (2005) also presents another GISbased approach for delineating and displaying data generated by HEC-2 numerical model. In this study, Model 1 was imported using historic data and Model 2 was obtained by importing a HECRAS model. The flood water surfaces for both models (Figure 4) were then created as TINs by linear interpolation of the flood water elevations along the crosssections.
5.2 LiDAR data preprocessing Generally, raw LiDAR data are converted to software specific formats which are capable of storing and re-
trieving a huge amount of data sets. This facilitates the use of tools which are available in the software to process the data. The subsections which follow describe the importing and preprocessing steps of LiDAR data into an ArcGIS environment. 5.2.1 Importing LiDAR data into an ArcGIS environment. ArcGIS’s Terrain datum is a relatively new format, which was introduced after the release of its 9.2 version. It is a TIN-based platform that can produce accurate surfaces quickly by efficiently managing pointbased data; such as LiDAR in a geodatabase. This data set is also capable of managing, editing, and producing accurate TINs and also allows creating TIN pyramid levels based on Z-tolerance. The lowest pyramid level (0.0 m) stores all of the elevation points. At a higher pyramid level, points are eliminated from the Terrain, based on the Z-tolerance value; thus, reducing the number of points. For example, data points are eliminated from the Terrain based on 0.1 m pyramid level (Z-tolerance) to produce surface that is within a 0.1 m vertical accuracy relative to the 0.0 m pyramid surface. To explore the benefits of this data set using this new method, the LiDAR data for Oswego County, NY were converted to Terrain format. Tools available in
Floodplain delineation method using LiDAR data
73
Fig. 4. Water surface and cross-section layout of Models 1 and 2.
ArcGIS 9.3 were used to convert the LiDAR data from Oswego County, NY to the Terrain format (Figure 2a). LiDAR elevation points classified as “bare earth” were used to produce the Terrain. Three pyramids: 0.0 m (base level), 0.1 m (0.5 ft), and 0.3 m (1 ft) were created to study the performance efficiency and change in floodplain area at different levels using the new method. These pyramids were selected so that the accuracy of the topographic data at each pyramid level would be within the two-foot equivalent vertical accuracy of ±36.5 cm (±1.2 ft) defined by FEMA. 5.2.2 Creation of tiles to divide the Terrain data into a regular grid pattern. The Terrain data have to be converted to TIN or raster DEM format to perform any analysis. In this new method, Terrain data are converted to TIN because it is more accurate than raster DEM. Converting the entire Terrain data, which contain millions of points, to a TIN data set will required huge memory and would slow down the entire process. The realistic limit for a TIN format in ArcGIS is about 10 million nodes (Atkinson, 2010). Furthermore, converting only the area of Terrain that overlaps the water surface of Models 1 and 2 is also memory intense and exceeds the node limits of a TIN. Therefore, it is required to convert a manageable portion of the entire Terrain to perform any analysis. A regular tile pattern covering the entire extent of the Terrain data is then created which divides the LiDAR data. Based on the average spacing of the LiDAR data, tiles of size 410 m × 410 m are created. Figure 2b shows an enlarged area of the Terrain data overlain by the tiles.
5.2.3 Preprocessing Terrain data. The purpose of preprocessing the Terrain data is to obtain the elevation information within each tile. Terrain data are clipped to the extents of each overlapping tile as a TIN surface. A tool named “Terrain to TIN,” available in ArcGIS, is used to obtain a TIN surface for the extent of each tile. Minimum and maximum elevation values are obtained from each clipped TIN data and are then saved as attributes of each tile. This step cannot be performed manually because thousands of tiles cover the entire LiDAR data set. Hence, a tool is developed to implement this step. After extracting the elevation information, the clipped TINs are deleted, because they are duplicates of the Terrain and also require larger disk space. The elevation parameters obtained by preprocessing are then used to filter only those grids which truly contribute to the floodplain. This process is described in the next subsection.
5.3 New floodplain delineation method As described earlier, a floodplain boundary is a line where the dry ground intersects the water surface. Figure 5 shows a schematic representation of three possible scenarios that can occur between a triangle of ground surface TIN and floodwater surface in the floodplain delineation process. Figure 5a represents the first scenario where the entire water surface is above the ground surface and Figure 5b represents the second scenario where the water surface is entirely below the ground surface. It
74
Deshpande
Fig. 5. The three scenarios that can occur between ground surface TIN and water surface.
should be noted that these scenarios represent all triangles of ground surface TIN that are entirely inside or outside the boundaries of the floodplain. In the first scenario, the resultant floodplain is represented by the water surface area, as the ground data are below water whereas in the second scenario, the area covered by the water surface will not contribute to the floodplain boundary. Thus, such triangles are not processed to identify floodplain boundaries as they are entirely below or above the water surface. Figure 5c represents the third scenario in which the water surface intersects the ground surface TIN. Such triangles are processed to identify the line of intersection, which will delineate the floodplain boundary. This situation would occur when the elevation of water surface is between the minimum and maximum elevations of the overlapping topographic data. Considering the three different scenarios discussed earlier, the computational time can be reduced only if the topographic data and the water surface data, that satisfy the third scenario, are processed. These triangles can be identified if the minimum and maximum value of the topographic data is available by preprocessing. However, it is not time-efficient to preprocess and check millions of triangles in a Terrain data set which is based on the above three scenarios. Therefore, the above three scenarios are generalized to filter the tiles that contribute to the floodplain. Figure 6a shows the tiles that overlap the water surface of Model 1 and its vicinity. To reduce the number of tiles, only the tiles that spatially overlap the water surface are selected (Figure 6b). Processing each overlapping tile to obtain the floodplain boundary is a timeconsuming task because it involves clipping the Terrain
data to TIN format, which covers thousands of points and then performing floodplain delineation by intersecting the TIN format with the virtual water surface. Thus, the tiles that truly contribute to the floodplain boundary are filtered by implementing all three of the criterion described in the previous subsection. A step-by-step approach, as described below, is implemented for each tile that overlaps the water surface: 1. Clip the flood water surface TIN to the extents of the overlapping tile. (Note: It is faster to clip the flood water surface TIN as it has few elevation points within a tile.) 2. Extract the minimum (WSmin ) and maximum (WSmax ) elevation value for the clipped water surface TIN. 3. Retrieve the minimum (TOPOmin ) and maximum (TOPOmax ) elevation values saved as attributes to the tile during the preprocessing stage. These values represent the maximum and minimum elevation for the topographic data within the tile. 4. Perform the following checks to identify if the tile would contribute to the floodplain boundary: i. Case 1: (WSmax ) < (TOPOmin ) = True. In this case, the ground surface within this tile is entirely free from flooding and doesn’t contribute to the floodplain. Therefore, LiDAR data are not processed to perform floodplain delineation. ii. Case 2: (WSmin ) > (TOPOmax ) = True. In this case, the ground surface within the tile is entirely covered with water and, therefore, this entire tile contributes to the floodplain. Because it
Floodplain delineation method using LiDAR data
75
Fig. 6. Water surface of Model 1 and the overlapping tiles.
is entirely covered with water, there is no floodplain boundary within this tile and its LiDAR data are not processed to perform floodplain delineation. iii. Case 3: In this case, it can be observed that portions of the ground data intersect the water surface. The Terrain data within this tile are processed to identify the areas within the tile that are above and the areas that are below the water surface. A tool named “TIN difference” available in ArcGIS is used to identify these areas and perform floodplain delineation. By examining these three scenarios, and performing the appropriate checks, the tiles which directly contribute to the floodplain boundary can be identified. The time required to clip the Terrain data and to perform floodplain delineation is eliminated for the tiles that do not contribute to the floodplain boundary which significantly reduces processing time. To determine the time saved by implementing this method, floodplain delineation was also performed for Models 1 and 2 at three pyramid levels using the conventional methodology. The three scenarios used in the new method were not implemented and, as a result, all the tiles that overlapped the water surface were processed in the conventional method to obtain the floodplain boundary. The processing time required for floodplain delineation and floodplain area, at different pyramid
Table 1 Time required to preprocess the Terrain data Pyramid level (m) Preprocessing results Time required (hours) Percent of time required compared to base-level pyramid Average number of points within each tile Percent of average point density compared to base-level pyramid
0
0.1
0.3
40.4 100%
34.6 86%
17.1 42%
73,793
36,500
4,384
100%
49.50%
5.90%
levels for Models 1 and 2, are presented in the next section.
6 RESULTS AND DISCUSSIONS As described previously, the LiDAR data are converted to Terrain data at three pyramid levels: 0.0, 0.1, and 0.3 m and preprocessing is done based on the overlapping 24,252 tiles that cover the entire Oswego County, NY. Table 1 summarizes the time required to preprocess the Terrain data. The results show that higher preprocessing time was required for the lower level pyramids. It took around
76
Deshpande
Fig. 7. Tiles identified based on the three cases.
40 hours to preprocess the Terrain at 0.0 m pyramid; whereas only 35 and 17 hours were required to process the Terrain at 0.1 and 0.3 m pyramids, respectively. The average density of data points at 0.1 and 0.3 m pyramids was 50 and 6%, respectively, of the point density at 0.0 m pyramid. The percentage decrease in the average number of points is significant from 0.0 to 0.1 m and to 0.3 m pyramids, compared to percentage decrease in the processing time. Thus, the reduction in the time is primarily due to the decrease in the average point density within each tile, and is also affected by the time required to clip the data and save the attributes. In Figure 7, tiles classified using the three scenarios that overlap the water surface of Models 1 and 2 are shown. It can be seen that using the new method, only the nonshaded tiles are processed. Figure 8 shows the floodplain obtained for the two models using the 0.0 m level pyramid. Floodplains delineated using 0.1 and 0.3 m pyramid Terrain were not identical to those delineated using 0.0 m pyramid and showed variations along the floodplain boundaries. Floodplain delineation was also performed by using the conventional method, which did not use the preprocessed information from the three pyramid levels. Identical floodplains were obtained by the new method and the conventional method at each pyramid level. Results involving the floodplain areas at three pyramid levels,
processing time, and other factors for Models 1 and 2 are tabulated in Tables 2 and 3. The following points are based on the results listed in Tables 2 and 3: 1. All tiles that overlap the water surface were processed in conventional floodplain delineation method; whereas only the short-listed tiles were processed using the new method. The new method was able to reduce the number of processed tiles by 43 and 17% for Models 1 and 2, respectively. 2. Large numbers of Case 2 tiles were observed for both models. Overall floodplain width of Model 2 was narrower than Model 1. Therefore, a large number of Case 1 tiles were observed for Model 1. 3. Using this method, 29, 32, and 34% of time were saved for Model 1 and 12, 13, and 17% of time were saved for Model 2 using the three levels of pyramid Terrain, respectively. This shows that the time saved increases as the density of data points decreases. 4. Processing time of 122 and 45 minutes was required for Models 1 and 2, respectively, by conventional method using Terrain at 0.0 m pyramid. A time saving of 82 and 76% was observed for the two models, respectively, using the new method at 0.3 m pyramid.
Floodplain delineation method using LiDAR data
77
Fig. 8. Floodplain boundary at pyramid level 0.0 m.
5. The change in area of the floodplain obtained by using 0.1 m pyramid Terrain was less than 1%. Using the 0.3 m pyramid Terrain, the change was less than 3%. Close visual inspection showed that the floodplain width slightly increased in some areas whereas it decreased in other areas, therefore compensating for the change in area as a result of using the higher pyramid Terrains. 6. Spot check along the entire floodplain boundary obtained using 0.1 m pyramid Terrain showed less than 5 ft variations to that obtained using 0.0 m pyramid Terrain. But, discrepancy more than 10 ft was observed at several locations between floodplain boundary obtained using 0.3 m pyramid Terrain. Particularly, in low-relief terrain, floodplain delineated using the 0.0 m pyramid Terrain represented minor topographic variations accurately compared to floodplain from other two Terrains. When visually compared to the orthoimagery, the floodplain boundaries obtained at 0.0 and 0.1 m pyramids were more appropriate than that obtained at 0.3 m pyramid. It was observed that the following factors affected the number of skipped tiles: 1. Floodplain width: The count of tiles that were entirely within the floodplain boundary, depended on the width of the floodplain. There were a higher
number of these types of tiles in Model 1 compared to Model 2. 2. Cross-section extent: The number of tiles entirely outside the floodplain depends on the extent of the hydraulic model cross-sections. The cross-sections were extended to enclose the floodplain boundary between them because the expected floodplain boundary between them was not known. Therefore, for very narrow floodplains, the number of such tiles could increase significantly.
7 CONCLUSIONS Over the past decade, density of LiDAR data has increased due to development in sensor technology. This calls for development of task-specific and time-efficient methods for feature extraction. Over the past several years, methods to filter LiDAR data point density have been researched. This article presents a time-efficient floodplain delineation method using high-density LiDAR data. This method preprocesses LiDAR data divided into regular tiles to obtain elevation information. This elevation information is then used to filter and process only the tiles that contribute to the floodplain, thus reducing the processing time. Furthermore, this method can be applied to filtered as well as raw data to save additional processing time. Results based on two pilot studies showed that this method can improve time
78
Deshpande
Table 2 Comparison of parameters obtained for floodplain delineation of Model 1 using conventional method and the new method
Table 3 Comparison of parameters obtained for floodplain delineation of Model 2 using conventional method and the new method
Pyramid level (m) Processing results Tiles processed by conventional method Tiles processed by new method Percent reduction in number of tiles Time required by conventional method (minutes) Time required by new method (minutes) Percent of processing time saved by new method compared with conventional method Percent of processing time saved by the new method compared with 0.0 m pyramid Floodplain area (sq. meters) Percent change in floodplain area
Pyramid level (m)
0
0.1
0.3
Processing results
454
454
454
259
259
259
43%
43%
43%
122
84
33
86
57
22
29%
32%
34%
Tiles processed by conventional method Tiles processed by new method Percent reduction in number of tiles Time required by conventional method (minutes) Time required by new method (minutes) Percent of processing time saved by new method compared with conventional method Percent of processing time saved by the new method compared with 0.0 m pyramid Floodplain area (sq. meters) Percent change in floodplain area
NA
34%
19,696,854.24 19,696,854.24 NA
