can. j. remote sensing, vol. 30, no. 1, pp. 64–76, 2004 ... · 2002; parkes et al., 2002;...

Using topographic lidar to map flood risk fromstorm-surge events for Charlottetown,

Prince Edward Island, CanadaTim L. Webster, Donald L. Forbes, Steve Dickie, and Roger Shreenan

Abstract. As part of a recent project to determine coastal impacts of climate change and sea-level rise on Prince EdwardIsland (PEI), airborne scanning laser altimetry (lidar) was employed to acquire high-resolution digital elevation models(DEMs) and other landscape information. The study area included both the Charlottetown urban area and an extensiveportion of the rural North Shore of PEI. Problems with the lidar data included data gaps and incorrect classification of“ground” and “non-ground” laser hits along the waterfront. Accurate representation of wharves and other waterfront featuresin the DEM was achieved by combining “ground” and “non-ground” data. The importance of calibration and validation inlidar data acquisition and interpretation was demonstrated by three independent validation exercises that uncovered andadjusted for a vertical offset attributed to calibration problems. The ground DEM was adjusted to hydrographic chart datumand used to model flood extent at three storm-surge water levels, one observed in the record storm of 21 January 2000 andtwo higher levels representing flood scenarios under rising sea level. Flood modelling was executed in a geographicinformation system (GIS) on the gridded ground DEM. The resulting binary grids were vectorized along the flooding limit.Low-lying areas isolated from free exchange with the harbour were excluded from the flood area. Vectors depicting thestorm-surge water lines for the three flood scenarios were implemented on the geographic information system (GIS) in thecity planning department and overlain on property boundary and assessment layers. This study demonstrated that validatedDEMs derived from airborne lidar data are efficient and adequate tools for mapping flood risk hazard zones in coastalcommunities.

Résumé. Des données altimétriques d’un système laser à balayage aéroporté (lidar) ont été utilisées pour acquérir desmodèles numériques d’altitude (MNA) à haute résolution et d’autres informations sur le paysage dans le cadre d’un projetrécent pour déterminer les impacts du changement climatique et du rehaussement du niveau de la mer sur les côtes, de l’Île-du-Prince-Édouard (IPE). La zone d’étude comprenait la zone urbaine de Charlottetown et une portion importante de lazone rurale de la côte nord de l’IPE. Des problèmes associés aux données lidar incluaient des trous dans les données et desclassifications incorrectes de retours laser « au sol » ou « non au sol » le long du front de mer. Des représentations précisesde quais et d’autres traits du rivage dans le MNA ont été réalisées en combinant des données « au sol » et « non au sol ».L’importance de l’étalonnage et de la validation dans l’acquisition et l’interprétation des données lidar a été démontrée aucours de trois exercices indépendants de validation qui ont permis de découvrir et d’ajuster un décalage vertical attribuable àdes problèmes d’étalonnage. Le MNA au sol a été ajusté par rapport aux données de la carte marine et utilisé pour modéliserl’ampleur des inondations à trois niveaux d’eau d’ondes de tempête, dont un observé durant la tempête record du 21 janvier2000 et deux niveaux plus élevés représentant des scénarios d’inondation suite à la montée du niveau de la mer. Lamodélisation des inondations a été réalisée dans le cadre d’un système d’information géographique (SIG) sur un MNAréférencé au sol. Les grilles binaires résultantes ont été vectorisées le long de la limite d’inondation. Les zones basses àl’abri des échanges d’eau libre avec le port ont été exclues de la zone d’inondation. Les vecteurs représentant les limitesd’eau d’ondes de tempête pour les trois scénarios d’inondation ont été intégrés dans le SIG du service de planification de laville et superposés sur les limites des propriétés et les couches illustrant les données d’évaluation. Cette étude a démontréque les MNA validés dérivés des données lidar aéroporté constituent des outils efficaces et appropriés pour la cartographiedes risques d’inondation dans les communautés côtières.[Traduit par la Rédaction]

76Introduction

Background

Storm-surge flood risk mapping was one major objective of arecent Climate Change Action Fund (CCAF) project toevaluate coastal impacts of climate change and sea-level rise onPrince Edward Island (McCulloch et al., 2002). Two airborneimaging systems were employed in this project, and theresulting datasets provided an essential foundation for flood

64 © 2004 CASI

Can. J. Remote Sensing, Vol. 30, No. 1, pp. 64–76, 2004

Received 16 April 2002. Accepted 8 April 2003.

T.L. Webster1 and S. Dickie. Applied Geomatics ResearchGroup, Centre of Geographic Sciences, Nova Scotia CommunityCollege, Middleton, NS B0S 1M0, Canada.

D.L. Forbes. Geological Survey of Canada, Bedford Institute ofOceanography, Dartmouth, NS B2Y 4A2, Canada.

R. Shreenan. Terra Remote Sensing Inc., Sidney, BC V8L 5Y3,Canada.

1Corresponding author (e-mail: [email protected]).

risk mapping in the urban centre of Charlottetown and arepresentative rural area in the vicinity of North Rustico(Webster et al., 2002). It was recognized at the outset that ahigh-resolution representation of the coastal topography wouldbe essential for predicting areas at risk of storm-surge flooding(Webster et al., 2001). A multidisciplinary scientific team wasinvolved in this project and contributed related analyses of sea-level change, storm-surge climatology, wave and sea-iceclimatology, statistics of flood probability, coastal erosion,socioeconomic impacts, and adaptation options (Chagnon,2002; Forbes and Manson, 2002; Forbes et al., 2002; Manson etal., 2002; Milloy and MacDonald, 2002; Parkes and Ketch,2002; Parkes et al., 2002; Thompson et al., 2002). The imagerycollected for this project has wider applicability, beyondclimate-change impacts assessment, to geological andecological research, urban and regional planning, coastalmanagement, agriculture, forestry, and other fields.

Rationale and objectives

The original project design highlighted emergingtechnologies in airborne imaging and data management thatoffer significant advances in high-resolution digital elevationmodelling and land surface characterization. This technologywas employed to enable realistic assessment of potentialimpacts from climate change and sea-level rise in the coastalzone of Prince Edward Island (PEI). Airborne scanning laseraltimetry (lidar) was used to obtain detailed topographic data,primarily for flood risk mapping in Charlottetown and alongthe North Shore of PEI. Additional airborne imaging using thecompact airborne spectrographic imager (CASI) was carriedout over parts of Charlottetown and a narrow region includingshallow nearshore waters along the North Shore (Webster et al.,2002).

The objectives of this paper are to document critical issues inlidar data acquisition and processing, including the need forground validation, and to demonstrate the utility of high-resolution digital elevation models (DEMs) derived from lidardata for flood risk mapping.

Imaging and survey systemsLaser altimetry (lidar)

Light detection and ranging (lidar) technology has beenemployed for a number of years in atmospheric studies (e.g.,Post et al., 1996; Mayor and Eloranta, 2001) and as an airbornetechnique for shallow bathymetric charting (e.g., Guenther etal., 2000), although cost was initially an impediment towidespread acceptance for the latter purpose. The technologycan also be used to image the land and water surface (Hwang etal., 2000), as was done in the present study. Applications havebeen demonstrated in forestry (Maclean and Krabill, 1986),sea-ice studies (Wadhams et al., 1992), and glacier massbalance investigations (Krabill et al., 1995; 2000; Abdalati andKrabill, 1999). A general overview of airborne laser scanning

technology and principles is provided by Wehr and Lohr(1999).

Applications to coastal process studies in the United Stateshave been reported by Sallenger et al. (1999), Krabill et al.(1999), and Stockdon et al. (2002), among others. Preliminarytrials in Atlantic Canada were reported by O’Reilly (2000), andsubsequent experience was described by Webster et al. (2002).Most of the coast of the conterminous United States has nowbeen mapped using this technology (Brock et al., 2002).

As illustrated in Figure 1, lidar mapping involves an aircraftemitting laser pulses towards the ground and measuring thereturn time of the pulse. The laser scan is acquired by rapidrepetition of the laser pulse transmitter and cross-trackdeflection of the beam using an oscillating mirror to produce azigzag pattern of laser hits on exposed surfaces below theaircraft. Utilizing precise differential global positioning system(GPS) technology to determine the location of the aircraft(Krabill and Martin, 1987) and an inertial measurement unit(IMU) to measure the aircraft attitude (pitch, yaw, and roll), thelocation of individual laser returns can be determined. Thenominal accuracy of the system used in this study is ±30 cm inthe horizontal plane and ±30 cm in the vertical plane. Animportant step in ensuring accurate representation of thetopography involves comparing the processed lidar data withground validation points positioned to a higher accuracy thanthe lidar (U.S. Federal Geographic Data Committee, 1998).

Geodetic issues

The lidar system produces a series of point measurementswith associated heights above the ellipsoid. GPS navigationuses the WGS84 ellipsoid, a smooth mathematical surfacerepresenting the earth, as the reference datum. Elevations on

© 2004 CASI 65

Canadian Journal of Remote Sensing / Journal canadien de télédétection

Figure 1. Lidar schematic, showing scanning laser unit, zigzagscan pattern on the ground, and aircraft positioning and attitudemeasurement systems. Carrier-phase DGPS for aircraft positionand an inertial measurement unit (IMU) for recording pitch, yaw,and roll of the aircraft.

most land-based topographic maps are measured relative to ageodetic vertical datum. For Canada this is known as theCanadian Geodetic Vertical Datum of 1928 (CGVD28). Torelate the height measurements to sea level, an adjustment mustbe made for the local vertical separation between the ellipsoidand the geoid. The difference between the WGS84 ellipsoidand the CGVD28 geoid is obtained by using the HT1_01Emodel, since replaced by HTv2.0, with an accuracy of ±5 cmwith 95% confidence in southern Canada (Geodetic SurveyDivision, Natural Resources Canada, www.geod.nrcan.ca/index_e/products_e/software_e/gpsht_e.html).

To map flood limits in this project, it was also necessary torelate the lidar elevation data to hydrographic chart datum, thevertical datum used in tide-gauge records from which storm-surge flood levels are obtained. This datum is approximatelyequivalent to the lowest astronomical tide and varies from placeto place around the coast. Determining the relation betweenchart datum, the ellipsoid, and CGVD28 was undertaken as aseparate component of the study (King et al., 2002).

Lidar data acquisition and processingCoverage and logistics

Lidar data acquisition over PEI was undertaken as part of awider geomatics and remote sensing research initiative focusedon the coastal zone in the Maritime Provinces during thesummer of 2000. On the island, lidar data were acquired forwaterfront areas of the City of Charlottetown and a coastal stripalong the North Shore from Rustico to Greenwich (Figure 2).A Bell 206B helicopter was used with local GPS control in PEIto complete the survey on 1–2 August 2000. Down-lookingvideo was acquired simultaneously to assist in interpreting thelidar returns.

Instrumentation and calibration

The survey specifications called for a laser hit with aminimum spacing every 3 m on the ground (in most cases thedata density was considerably higher) and horizontal andvertical accuracy within 0.3 m. Data were collected undersuitable conditions for video acquisition (during daylight hourswith good visibility). The instrument was a diode-pumped I/RYAG laser pulsed at repetitions up to 10 kHz, scanning normalto the aircraft flight path to a maximum scan angle of 57°. Laserrange distances were determined to a resolution of 0.05 m, andindividual shots were positioned with a mirror and attitude datafrom the IMU. The aircraft was positioned using phasekinematic GPS, referenced to a geodetic ground monumentnorth of the Charlottetown Airport.

A crucial step in the lidar survey involved daily calibrationpasses over the GPS base station. Although calibrationparameters were adequately defined from this information, theintroduction of elevated anthropogenic features would haveallowed further refinement of the calibration. Objects withlong, straight, elevated edges such as bridges and buildings areparticularly favoured because they provide a check for bothhorizontal and vertical alignment. This process has a directimpact on the overall accuracy of the survey, and a morerigorous calibration protocol has since been adopted.

Another aspect of lidar calibration is the determination ofany errors in the raw laser ranges. For a proper determination,multiple passes at varying altitudes are required over knownpoints. In this project a single range calibration was carried outat the proposed flying altitude. Unforeseen power reduction ofthe laser system forced a reduction of flying altitude, whichresulted in a vertical offset of approximately 0.9 m. Thecalibration passes were made at an altitude of approximately900 m, 300 m higher than the actual survey altitude. Thus theelevations matched at the calibration sites but contained a rangebias over the survey area. Comprehensive tests following thesurvey showed the need for laser range bias and range scalefactor corrections. The value of the elevation offset wasdetermined from ground validation surveys, as discussed laterin the paper, and several lines of the lidar data in the survey areahave since been reprocessed to confirm these parameters.

Reliable point classification and feature extraction(described in the following section) depend on an adequatequality and quantity of points in the dataset. Valid groundelevations cannot be determined if most or all laser returns arefrom the tree canopy or if there are significant gaps in the data.Limitations of the instrumentation used in this study resulted ina scarcity of ground returns from beneath tree canopies andfrom low-reflective targets in the near infrared, such as asphaltsurfaces on roads and roofs (Figure 3). We have since foundthat the introduction of a collimator attached to the end of thelaser head decreases the divergence of the laser beam to betterthan 0.3 mrad. This enables improved capture of weakerreflective signals from asphalt and better penetration throughtrees.

66 © 2004 CASI

Vol. 30, No. 1, February/février 2004

Figure 2. Geographic extent of lidar coverage in Prince EdwardIsland, showing location of the Charlottetown (a) and North Shore(b) study areas.

Gaps in the lidar coverage may result from power loss,vibration-induced data loss, or navigation errors resulting fromgaps between flight lines. Field quality control is critical toidentify and correct such gaps. In the present study someindividual laser scans or groups of scans were lost because ofaircraft vibration, and the laser data storage device has sincebeen reconfigured and repackaged to solve this problem.

Data classification and management

An important step in any successful lidar survey is theaccurate extraction of features from the lidar point cloud. Theability to perform this task accurately and reliably has a directbearing on the overall accuracy of the survey. The proprietarysoftware used to process the data presented in this paper isextremely effective for single-line corridor mapping but lacksalgorithms for processing multiple lines of lidar, and also lacksvisualization tools to facilitate the detection of outliers. As aresult, there was some misclassification of “ground” hits data inthis study as “non-ground” (misidentification as tree canopy orstructures). In particular, there were problems in defining theleading edges of wharves along the waterfront, which wereresolved by manual inspection and reclassification of data inthe “non-ground” file. Similarly, some data in the “ground” hitsfile were returns from water and had to be edited out using athreshold filter. The lack of returns from asphalt discussedpreviously caused some difficulty in data validation from roadsurfaces. Nevertheless, once these issues had been recognized,it was possible to generate a valid and detailed ground DEM.As a result of this study, new software has been acquired andimplemented to improve the classification and provide moreprocessing options.

The lidar data were compiled on CD-ROM in 1 km × 1 kmtiles (e.g., Figure 4), each having three associated files. Onefile contained all the laser hits prior to noise removal. The othertwo files contained cleaned data, separated into a “bald earth”file (points identified as “ground” hits) and a “non-ground” file(points identified as vegetation or buildings, but in factincluding also cliff top, dune crest, wharf edge, and otherfeatures as described previously).

The horizontal position for each point was given in UniversalTransverse Mercator (UTM) Zone 20 coordinates, based on theWGS84 ellipsoid. The ASCII data files contained five fields:(i) UTM easting (m), (ii) UTM northing (m), (iii) height abovethe WGS84 ellipsoid (m), (iv) orthometric height aboveCGVD28 (m), and (v) GPS time (s) from midnight at the startof each day. The data were imported into a GIS for furtheranalysis and processing.

Data validation

Lidar elevation data were compared with several sets ofground-control data to determine their accuracy and precision(Figure 3). Plots of raw ground hits in Charlottetown (e.g.,Figure 4) show dense coverage on land, but with evident gapsoutlining streets and some building roofs, illustrating thesurprising lack of returns from asphalt. The raw point plot(Figure 4) also clearly shows a pattern of laser returns from thewater surface, with varying density of points from line to line.

Table 1 shows summary statistics for comparison betweenwater level measured at the tide gauge in CharlottetownHarbour at the time of data acquisition and observed ellipsoidaland geoidal laser ground-hit elevations on the water surface.The lidar data are taken from samples (ranging in size from

© 2004 CASI 67


Figure 3. Ground lidar points and shoreline for Charlottetown.

1013 to 3634 points) in seven discrete areas off theCharlottetown waterfront in two arms of the harbour, fromLewis Point in the northwest to Parkdale in the east (Figure 3).The differences between water levels at the tide gauge and

observed mean lidar elevation on the water surface range from–0.64 to –1.04 m. The standard deviations range from 0.14 to0.31 m and are therefore close to the specified vertical accuracyof the lidar system. The wind speed was low, and waveroughness on the water surface can therefore be neglected. Therange in vertical offsets is thought to be partly attributable tovariation in water levels associated with tidal dynamics in theharbour. The mean offset from the eight sample sets was –0.85 m.

A similar comparison was made between the CanadianHydrographic Service (CHS) benchmark CHTN 1-1963 on theCoast Guard wharf at the foot of Queen Street and lidar hitswithin a radius of 3 m. The ellipsoidal and geoidal (CGVD28)elevations of the benchmark were determined as part of thevertical datum control survey reported in King et al. (2002).The vertical coordinates of this control point are +4.34 m chartdatum (–14.90 m ellipsoidal, +2.65 m CGVD28). Initialcomparison with “ground” points in the lidar dataset gave anoffset of –2.2 m and created some concern. However, we soonrealized that the classification algorithm had identified thewharf top as “non-ground” owing to the abrupt rise from the

68 © 2004 CASI


Figure 4. Raw lidar points for a single tile, showing Hillsborough Bridge (with traffic) andadjacent Charlottetown waterfront, and several swaths of water-surface returns. Note the lack ofreturns over the road asphalt surface. Grid coordinates are UTM Zone 20 eastings and northingsbased on the WGS84 ellipsoid.

ζE (m) ζG (m) s (m) N ηCD (m) ηG (m) ∆ζ (m)

–17.48 0.15 0.22 2543 2.77 1.08 –0.93–17.47 0.17 0.24 1616 2.77 1.08 –0.91–17.19 0.44 0.14 2316 2.82 1.13 –0.69–17.18 0.44 0.31 1757 2.77 1.08 –0.64–17.61 0.01 0.28 2004 2.67 0.98 –0.98–17.32 0.29 0.28 3634 2.64 0.95 –0.66–18.25 –0.65 0.26 1013 1.97 0.28 –0.93–18.32 –0.72 0.26 1632 2.01 0.32 –1.04Mean –0.85

Table 1. Mean lidar ellipsoidal and geoidal (CGVD28) elevations(ζE and ζG with standard deviation s) for sets of laser returnsfrom the water surface in Charlottetown Harbour, with thenumber of points in each sample (N), equivalent tide-gauge waterlevel relative to chart datum (ηCD) and CGVD28 (ηG), and thecorresponding lidar offset (∆ζ) in metres.

adjacent water surface and had classified water-surface pointserroneously as “ground” points. Subsequent validation ofpoints on the wharf deck in the vicinity of the benchmark gavean offset of –0.92 m.

Both of the foregoing validation exercises indicate anequivalent offset within the resolution of the lidar system. Inother words, the initially determined lidar elevations were0.9 m lower (to the nearest decimetre) than their true values.Based on these results, an adjustment of +0.9 m was applied toall data points prior to incorporation in the DEM. Thisproduced realistic flooding levels for a simulation of the21 January 2000 storm surge, as confirmed from eyewitnessreports, whereas the original DEM without adjustment hadsuggested much more extensive flooding.

Planimetric base-map data were provided by the City ofCharlottetown in double stereographic (ATS77) projection inARC/INFO format. To integrate the map and lidar data, theplanimetric vectors were converted from double stereographicto UTM projection. Horizontal offsets between the lidar “all-hits” surface and the vector information ranged fromapproximately 0 to 3 m. The variation in the horizontal offset isprobably a result of the lidar data being gridded at a 2 mresolution, the variable density of the points for low reflectivetargets, the scale of the planimetric vector layers, and the mapprojection transformations.

To confirm that the vertical adjustment of +0.9 m to the lidarelevations was appropriate, a follow-up high-precision GPScampaign was carried out in Charlottetown during the summerof 2001. Fifteen control points were occupied throughout thelidar coverage area and processed using carrier-phaseinformation, providing accuracy on the order of centimetres.Grass-covered flat fields in city parks were selected for the GPS

campaign. This was done to ensure a significant number oflidar returns in these areas to provide a ground surface atsufficient detail, in view of the poor lidar reflection off asphaltsurfaces.

The GPS elevations were converted from ellipsoidal heightsto orthometric heights using the HT1_01E model. The GPSorthometric heights could then be compared with the lidar-derived DEM ground surface orthometric heights (after thevertical offset adjustment described previously). The GPSpoints were overlaid on the DEM surface in a GIS (Figure 5),and the orthometric height differences between the GPS andlidar surface were calculated (Table 2). The mean differencewas 0.04 m, confirming that the adjustment of +0.9 m wasappropriate (Figure 5). The lidar ground DEM and GPSmeasurements were highly correlated, with a coefficient ofdetermination r2 = 0.99 (Figure 6). The larger differences atindividual points are attributed to surface variability withinDEM grid cells.

Digital elevation model and flood riskmaps for Charlottetown

Surface construction

The multiple ground-hits lidar point files (40 tiles eachcovering 1 km2) were used to construct a topographic surfacerepresentation using the adjusted orthometric height values. Atriangular irregular network (TIN) was constructed from thepoints. The point distribution was dense and reasonablyhomogeneous, except over asphalt surfaces (Figure 4).Nevertheless, the point density was sufficient to constructappropriate triangles (near equilateral) for the topographic

© 2004 CASI 69


Figure 5. Colour shaded-relief image of the bald earth DEM from the ground lidar returns. Theblack triangles represent the locations of carrier-phase DGPS points used to validate the DEM.

model. The TIN was then used to construct a gridded surfacewith a cell size of 2 m using quintic interpolation (fifth-orderpolynomial). This method ensures that the grid surface passesthrough all the points and the resultant surface is smooth andcontinuous. Another digital surface model (DSM), representingall hits on the ground, buildings, and vegetation, was generatedfor Charlottetown and a separate one for the North Shore studyarea by gridding the lidar data in the public-domain image-processing system GRASS (Forbes and Manson, 2002;Webster et al., 2002).

Surface refinement

The lidar survey covered the Charlottetown waterfront,including parts of the north bank of the Hillsborough (East)River above the bridge and both banks of the Yorke (North)River to a short distance above the highway bridge (Figure 3).As a result, the TIN was constructed across water bodies and ablank zone within the “V” of the survey area, producing aninterpolated surface that was unreliable in areas of sparse andmissing data. The TIN was gridded and the resultant grid wasclipped to the shoreline and data extents, removing theerroneous surface values over water and in areas outside thesurvey limits.

A colour shaded-relief image was produced from the clippedground grid. Visual examination of this image in combinationwith the “ground” lidar point data revealed that many of thelaser returns along the waterfront, specifically on wharves,docks, and seawalls, were missing. As noted earlier, becausethe automated cleaning routines examined the height values forareas of rapid change in elevation, points identified as havinganomalously high elevations were categorized as “non-ground”points representing buildings or vegetation. In the Charlottetownstudy area, most of the laser hits along the edge of the

waterfront were thus removed from the ground dataset. Theresulting ground surface showed the wharves as having slopingsides, whereas in reality they have steep vertical faces. Becausethe gridded surface was intended for use in modelling floodextent, it was critical to obtain an accurate representation ofwaterfront features. To solve this problem, the combined filesof “ground” and “non-ground” points were examined and datafrom the latter representing hits on waterfront structures wereextracted and used to construct a new TIN and gridded surface.This new waterfront grid was merged with the existing groundgrid, producing a more realistic surface representation.

In addition to the ground DEM for use in flood simulation, aDSM was constructed from a combination of all the lidarreturns to represent the ground, trees, and buildings. This DSMwas used for visualizing the flood extents determined from theground DEM.

Flood simulation modellingFlood levels

Water levels resulting from storm-surge events are definedwith respect to height above local hydrographic chart datum(CD). Three water levels were selected (McCulloch et al.,2002) for use in the flood simulations (Figure 7): (i) 4.23 mCD, (ii) 4.70 m CD, and (iii) 4.93 m CD.

The first flood level was observed during the 21 January2000 storm-surge event in which significant coastal floodingcovered parts of the Charlottetown waterfront and downtowncore (Parkes and Ketch, 2002). The third level corresponds toan equivalent event after 100 years of anticipated relative sea-level rise to a level 0.7 m above present mean water level(Parkes et al., 2002). The second level represents anintermediate event with a higher probability. These flood levelswere converted to heights above geodetic datum because thelidar surface was created using orthometric heights (CGVD28).In Charlottetown, chart datum is 1.685 m below CGVD28 (G.

70 © 2004 CASI


GPS (m) Lidar (m) ∆ζ (m)

4.46 5.36 –0.904.60 4.02 0.574.22 4.10 0.13

10.42 9.77 0.655.26 4.27 0.989.07 8.84 0.222.34 2.92 –0.59

14.99 15.54 –0.5519.30 19.81 –0.52

6.20 6.19 0.0018.64 18.93 –0.2919.48 19.86 –0.37

6.90 7.03 –0.132.09 2.44 –0.352.78 2.28 0.50

Mean 0.04Standard deviation 0.54

Table 2. Carrier-phase GPS elevationscompared with lidar ground surfaceorthometric heights.

Figure 6. Scatterplot of GPS and lidar orthometric ground heights.The coefficient of determination between the lidar elevations andthe validation points is r2 = 0.99.

King, Canadian Hydrographic Service, personal communication,2001).

Flooding the DEM

The flood modelling was done on the lidar ground surfaceDEM. The flood limit for each water level was captured as avector boundary. These vectors were used to select onlypolygons that were contiguous with and open to flooding fromthe harbour (Figures 7 and 8). In some cases, culverts allowwater to flow past barriers, such as causeways, which otherwiseprotect low-lying land behind. To determine if these areas couldbe flooded, we consulted engineers with the City ofCharlottetown. Most of the culverts have one-way valves,allowing water to flow into the harbour but not back. Therefore,areas upstream of these culverts were not included in the floodextents.

The three flood vector extents were projected from UTM tothe PEI double stereographic (ATS77) projection and deliveredto the City of Charlottetown, where the information is beingused for planning and development of adaptation strategies, incollaboration with the Federation of Canadian Municipalities.The flood extents were also used for analysis of socioeconomicimpacts (Milloy and MacDonald, 2002).

Flood visualization

In addition to the extent of inundation for each flood level,another issue is the depth of water at various locations within

the flooded area. To determine this, the flood extent raster layerwas assigned the water level orthometric height and the groundDEM was subtracted from this layer to develop a representationof flood water depth. This information is important forassessing the potential damage associated with a given waterlevel.

In addition to the three water levels defined previously,several other flood levels were also modelled for visualizationpurposes. A procedure was written to automate the floodmodelling process, raising the water in increments of 10 cmover the all-hits lidar DSM. This enabled us to produceanimations of the rising storm-surge flooding. Each floodincrement layer was processed to ensure that the floodingextent was connected to the harbour. Because the DSM wasused in this exercise and not the bald-earth DEM, the area ofinundation for a given flood level was limited by someobstructions. The flood-limit layers were merged with the DSMfor the visualization process. A series of plan-view layers wereused to construct an animation of the flooding sequence. Inaddition, a set of perspective views was also generated and usedto produce another animation. Examples of the perspectivescenes are shown in Figure 9, where Figure 9a shows waterlevel at geodetic datum (approximate mean sea level), andFigure 9b shows a flood level of 4.9 m CD (the third flood leveldefined previously). The extent of the flooding is somewhatless than that shown in Figures 7 and 8 because the DSM wasused instead of the bald-earth DEM.

© 2004 CASI 71


Figure 7. The three modelled flood-limit extents over the colour shaded-relief surface of all thelidar returns (i.e., includes buildings and trees along with the ground). The yellow line is 4.23 m,the white line 4.70 m, and the red line 4.93 m above chart datum for Charlottetown. The blackbox in the lower centre of the image shows the location of Figure 8.

Summary and discussionThis study demonstrates the high level of topographic detail

obtainable using airborne scanning laser altimetry and itspotential utility for coastal hazard mapping. As noted byWebster et al. (2002), Brock et al. (2002), and Stockdon et al.(2002), this technology now has the potential to become afundamental tool in coastal studies, comparable to establishedtechniques such as aerial photogrammetry and ground surveys.

Lidar data were used in this study to produce detailed, high-resolution, ground DEMs for flood simulation modelling, bothin Charlottetown (this paper) and in a representative area on thecentral North Shore of PEI (Figure 2) (Forbes and Manson,2002; Webster et al., 2002). Calibration and careful validationof the lidar results in relation to orthometric elevations tied tothe hydrographic chart datum enabled simulated flooding of theDEM at the observed water level of 21 January 2000 and twohigher levels related to storm-surge flooding on top of highersea levels anticipated over the coming century. The flood limitsobtained for the 21 January 2000 event in the digital modelwere compared with direct observations by Charlottetownresidents and city staff at the time of the storm to validate theflood simulation.

Problems observed in the lidar data included data gapsattributed to onboard data loss and lack of laser returns fromasphalt paving and roofs. Issues related to data gaps werediscussed previously. The lack of returns from asphalt (most

clearly demonstrated by the distinct delineation of roadvehicles on the Hillsborough Bridge and approaches inFigure 4) reflected gain and collimation problems in theequipment used for this study. These problems have since beenrectified. In the present study, the lack of returns from asphaltroof surfaces frustrated the identification and mapping of somemajor buildings (notably the Delta Prince Edward Hotel nearthe waterfront) but otherwise did not seriously jeopardize theflood mapping exercise.

Another problem uncovered in the initial validation workwas incorrect classification of “ground” and “non-ground”laser returns. As described previously, this occurred along theCharlottetown waterfront at the edges of wharves and onforedune ridges and cliffs along the North Shore. “Non-ground” hits were identified based on anomalously highelevations relative to nearby data, such that any hits on top ofabrupt elevation changes such as wharf edges, cliff tops, orcliffed dune crests were misclassified as “non-ground”. Thus,the initial DEM constructed for Charlottetown from the“ground” hits file did not correctly resolve the details ofwaterfront wharf structures. As described previously, carefulinspection of the “non-ground” points enabled the constructionof a refined waterfront grid that could be merged with theoriginal.

The ability to perform accurate and reliable differentiation ofground returns from buildings or vegetation and fromatmospheric aerosols has a direct bearing on the overall

72 © 2004 CASI


Figure 8. Close-up of the three flood extents for the waterfront and low-lying area ofCharlottetown. The flood vectors are the same as in Figure 7 (which shows the location of thisfigure).

accuracy of the survey. As a result of this study, new softwarehas been implemented to handle these tasks. The proprietarysoftware used in this study was extremely effective for single-line corridor mapping but lacked the algorithms for processingmultiple lines of lidar data. It also lacked visualization tools asrequired for adequate detection of outliers. In the revisedprocessing methodology, lidar data are processed in relation toneighboring hits rather than in the sequence of capture. Several

new visualization techniques have been implemented,including video and digital mosaics, to review the processeddata and detect anomalies.

As discussed previously, the lidar data had to be adjusted byadding 0.9 m to the orthometric elevation values, reflecting aconsistent low bias in the laser returns. This resulted primarilyfrom an unexpected requirement to fly the survey at a lowerelevation than planned, leading to slight miscalibration of the

© 2004 CASI 73


Figure 9. Perspective views of the DSM looking to the northwest over the Charlottetownwaterfront to the downtown core (foreground), with the Yorke River in the background:(a) represents the situation at mean water level (1.7 m chart datum or �0 m CGVD28);(b) represents a water level of 4.9 m above chart datum.

laser range data in the onboard system. Three independentvalidation exercises were undertaken in Charlottetown, usingground- and water-surface elevations to determine the offsetand confirm the validity of the linear adjustment. Furthervalidation of lidar data along the North Shore (Figure 2)confirmed the validity of the offset value there as well (Websteret al., 2002).

Two digital topographic models were developed: one, theDSM, was based on all “ground” and “non-ground” points; theother, the so-called bald-earth DEM, was based on “ground”points only, except where merged with the waterfront refinedgrid described previously. A shaded-relief image of the DSMprovides a view of Charlottetown with trees and buildings (e.g.,Figures 7–9), whereas the bald-earth DEM (e.g., Figure 5) isrequired to simulate flooding and is the model used in thevalidation exercises.

The bald-earth DEM was used to model three storm-surgewater levels (McCulloch et al., 2002). The flood modelling wascarried out on the gridded DEM with 2 m resolution.Depending on whether grid cells were flooded or not, an outputbinary grid was created to enable the definition of flood limitsby vectorizing the wet–dry boundary. The resulting vectorswere used to mark the limits of flooding on the municipal GISand to ensure that only areas with a free connection to theharbour were included in the flood zone for any given waterlevel. Detailed discussion with city engineers provided detailson culverts and stop valves to enable reliable identification ofprotected low-lying areas.

Overlay of the flood-limit vectors on the Charlottetown GISprovided an opportunity to assess the flood limits in relation tokey infrastructure assets, individual properties and businesses,and heritage structures. This provided the basis for asocioeconomic analysis of impacts (Milloy and MacDonald,2002) and provided a planning tool for city staff (D. Poole,Planning and Development Officer, City of Charlottetown,personal communication, 2001; O’Reilly et al., 2003). It alsoprovided graphic illustration of the potential for waterfrontflooding in Charlottetown from combinations of increasedstorm intensity and accelerated relative sea-level rise predictedunder climate warming (e.g., Figure 9). This resulted inimmediate adaptation of some building permit procedures and alonger term adaptation strategy in the city (cf. Forbes et al.,2002).

ConclusionsSpecific conclusions from this study include the following:

(1) Airborne topographic mapping using scanning laseraltimetry can provide high-resolution digital elevationmodels (DEMs) suitable for flood risk hazard mapping.

(2) High standards in aircraft attitude and position detectionand monitoring are required for accurate determination oflaser target position and elevation.

(3) Careful calibration of the laser system at the time of datacollection is necessary for unbiased topographic mappingand to ensure complete coverage of all surface types.

(4) Detailed ground validation of the resulting DEM isessential to ensure that the lidar data meet adequatespecifications in vertical and horizontal accuracy andprecision to enable reliable mapping.

(5) Further development and testing of feature-detectionalgorithms is required to ensure appropriate discriminationof ground surface laser hits, particularly in the vicinity ofabrupt elevation changes, such as along the fronts ofwharves and seawalls and on cliffs and coastal dunes.

(6) Adjustment of lidar-derived elevation data to thehydrographic chart datum (to which tidal elevations andstorm-surge water levels are referred) enables simulationof flood limits for historical and potential future floodlevels.

(7) Determination of the flood limits for past and futurewater levels in relation to the city property andassessment data in a geographic information system(GIS) was used to assess the socioeconomic impacts offlooding with and without climate change.

(8) Mapping of flood limits through the use of lidar-derivedDEMs provides an efficient method for defining floodrisk hazard zones as a basis for precautionary planning,climate-change adaptation, and emergency-measuresresponse.

In summary, this project has demonstrated the effectivenessand varied utility of airborne scanning laser altimetry for theanalysis of climate-change and storm-surge impacts in coastalregions. Although data acquisition, processing, and archivingissues require ongoing development and refinement, it is ourconviction that topographic lidar will become part of thestandard geomatics arsenal in future coastal studies.

AcknowledgmentsWe are happy to acknowledge the efforts of the data

acquisition contractors, including Rick Quinn and others (TerraRemote Sensing Inc.), Herb Ripley, Andrew Cameron, andLaura Roy (Hyperspectral Data International), the pilots, andother support staff. We thank Paul Fraser and Dan Deneau(Applied Geomatics Research Group, Centre of GeographicSciences, Nova Scotia Community College) for theCharlottetown GPS campaign. We acknowledge thecontributions of Mike Butler and Brent Rowley for helping inthe coordination of data acquisition during the summer of 2000.Gavin Manson (Geological Survey of Canada) provided criticalfield and office support. Glen King (CHS) assisted with verticalcontrol data, as did Charles O’Reilly (CHS), who was the primeinspiration behind the initiative to acquire laser altimetry forthis project. Don Poole (Planning and Development Officer,

74 © 2004 CASI


City of Charlottetown) provided invaluable assistance. Thisstudy was funded in large part by the Climate Change ActionFund (CCAF) of the Government of Canada, and we aregrateful to the entire CCAF project team for their ideas andsupport on the project. The manuscript benefited from criticalcomments by James Gibeaut and another anonymous journalreviewer and internal reviews by Bob Maher and GavinManson. This work is contribution 2003148 of the GeologicalSurvey of Canada.

ReferencesAbdalati, W., and Krabill, W.B. 1999. Calculation of ice velocities in the

Jakobshavn Isbrae area using airborne laser altimetry. Remote Sensing ofEnvironment, Vol. 67, pp. 194–204.

Brock, J.C., Wright, C.W., Sallenger, A.H., Krabill, W.B., and Swift, R.N.2002. Basis and methods of NASA airborne topographic mapper lidarsurveys for coastal studies. Journal of Coastal Research, Vol. 18, pp. 1–13.

Chagnon, R. 2002. Sea-ice climatology. In Coastal impacts of climate changeand sea-level rise on Prince Edward Island. Edited by D.L. Forbes andR.W. Shaw. CD-ROM. Geological Survey of Canada, Open File 4261,Supporting Document 5. 24 pp.

Forbes, D.L., and Manson, G.K. 2002. Coastal geology and shore-zoneprocesses. In Coastal impacts of climate change and sea-level rise onPrince Edward Island. Edited by D.L. Forbes and R.W. Shaw. [CD-ROM].Geological Survey of Canada, Open File 4261, Supporting Document 9.84 pp.

Forbes, D.L., Shaw, R.W., and Manson, G.K. 2002. Adaptation. In Coastalimpacts of climate change and sea-level rise on Prince Edward Island.Edited by D.L. Forbes and R.W. Shaw. [CD-ROM]. Geological Survey ofCanada, Open File 4261, Supporting Document 11. 18 pp.

Guenther, G.C., Brooks, M.W., and LaRocque, P.E. 2000. New capabilities ofthe SHOALS airborne lidar bathymeter. Remote Sensing of Environment,Vol. 73, pp. 247–255.

Hwang, P.A., Krabill, W.B., Wright, W., Swift, R.N., and Walsh, E.J. 2000.Airborne scanning lidar measurement of ocean waves. Remote Sensing ofEnvironment, Vol. 73, pp. 236–246.

King, G., O’Reilly, C., and Varma, H. 2002. High-precision three-dimensionalmapping of tidal datums in the southwest Gulf of St. Lawrence. In Coastalimpacts of climate change and sea-level rise on Prince Edward Island.Edited by D.L. Forbes and R.W. Shaw. [CD-ROM]. Geological Survey ofCanada, Open File 4261, Supporting Document 7. 16 pp.

Krabill, W.B., and Martin, C.F. 1987. Aircraft positioning using globalpositioning system carrier phase data. Navigation, Vol. 34, pp. 1–21.

Krabill, W.B., Thomas, R.H., Martin, C.F., Swift, R.N., and Frederick, E.B.1995. Accuracy of airborne laser altimetry over the Greenland ice sheet.International Journal of Remote Sensing, Vol. 16, pp. 1211–1222.

Krabill, W.B., Wright, C., Swift, R., Frederick, E., Manizade, S., Yungel, J.,Martin, C., Sonntag, J., Duffy, M., and Brock, J. 1999. Airborne lasermapping of Assateague National Seashore Beach. PhotogrammetricEngineering & Remote Sensing, Vol. 66, pp. 65–71.

Krabill, W., Abdalati, W., Frederick, E., Manizade, S., Martin, C., Sonntag, J.,Swift, R., Thomas, R., Wright, W., and Yungel, J. 2000. Greenland IceSheet: high-elevation balance and peripheral thinning. Science(Washington, D.C.), Vol. 289, pp. 428–430.

Maclean, G.A., and Krabill, W.B. 1986. Gross-merchantable timber volumeestimation using an airborne lidar system. Canadian Journal of RemoteSensing, Vol. 12, pp. 7–18.

Manson, G.K., Forbes, D.L., and Parkes, G.S. 2002. Wave climatology. InCoastal impacts of climate change and sea-level rise on Prince EdwardIsland. Edited by D.L. Forbes and R.W. Shaw. [CD-ROM]. GeologicalSurvey of Canada, Open File 4261, Supporting Document 4. 31 pp.

Mayor, S.D., and Eloranta, E.W. 2001. Two-dimensional vector wind fieldsfrom volume imaging lidar data. Journal of Applied Meteorology, Vol. 40,pp.1331–1346.

McCulloch, M.M., Forbes, D.L., Shaw, R.W., and the CCAF A041 ScientificTeam. 2002. Coastal impacts of climate change and sea-level rise onPrince Edward Island. Edited by D.L. Forbes and R.W. Shaw. [CD-ROM].Geological Survey of Canada, Open File 4261, 62 pp. and 11 supportingdocuments.

Milloy, M., and MacDonald, K. 2002. Evaluating the socio-economic impactsof climate change and sea-level rise. In Coastal impacts of climate changeand sea-level rise on Prince Edward Island. Edited by D.L. Forbes andR.W. Shaw. [CD-ROM]. Geological Survey of Canada, Open File 4261,Supporting Document 10. 90 pp.

O’Reilly, C. 2000. Defining the coastal zone from a hydrographic perspective.Backscatter, Vol. 11, No. 2, pp. 20–24

O’Reilly, C.T., Forbes, D.L., and Parkes, G.S. 2003. Mitigation of coastalhazards: adaptation to rising sea levels, storm surges, and shoreline erosion.In Proceedings of the 1st CSCE Specialty Conference on Coastal, Estuaryand Offshore Engineering, 4–7 June 2003, Moncton, N.B. CanadianSociety for Civil Engineering, Montréal, Que. CSN-410, pp. 1–10.

Parkes, G.S., and Ketch, L.A. 2002. Storm-surge climatology. In Coastalimpacts of climate change and sea-level rise on Prince Edward Island.Edited by D.L. Forbes and R.W. Shaw. [CD-ROM]. Geological Survey ofCanada, Open File 4261, Supporting Document 2. 87 pp.

Parkes, G.S., Forbes, D.L., and Ketch, L.A. 2002. Sea-level rise. In Coastalimpacts of climate change and sea-level rise on Prince Edward Island.Edited by D.L. Forbes and R.W. Shaw. [CD-ROM]. Geological Survey ofCanada, Open File 4261, Supporting Document 1. 33 pp.

Post, M.J., Grund, C.J., Weickmann, A.M., Healy, K.R., and Willis, R.J. 1996.A comparison of the Mt. Pinatubo and El Chichon volcanic events: lidarobservations at 10.6 and 0.69 mm. Journal of Geophysical Research,Vol. 101, No. D2, pp. 3929–3940.

Sallenger, A.B., Jr., Krabill, W., Brock, J., Swift, R., Jansen, M., Manizade, S.,Richmond, B., Hampton, M., and Eslinger, D. 1999. Airborne laser studyquantifies El Niño-induced coastal change. Eos, Transactions of theAmerican Geophysical Union, Vol. 80, pp. 89, 92.

Stockdon, H.F., Sallenger, A.H., List, J.H., and Holman, R.A. 2002.Estimation of shoreline position and change using airborne topographiclidar data. Journal of Coastal Research, Vol. 18, No. 3, pp. 502–513.

Thompson, K., Ritchie, H., Bernier, N.B., Bobanovic, J., Desjardins, S.,Pellerin, P., Blanchard, W., Smith, B., and Parkes, G. 2002. Modellingstorm surges and flooding risk at Charlottetown. In Coastal impacts ofclimate change and sea-level rise on Prince Edward Island. Edited by D.L.Forbes and R.W. Shaw. [CD-ROM]. Geological Survey of Canada, OpenFile 4261, Supporting Document 6. 48 pp.

U.S. Federal Geographic Data Committee. 1998. Geospatial accuracystandards, parts 1 to 3. Documents FGDC-STD-007.1, FGDC-STD-007.2,FGDC-STD-007.3. Available from www.fgdc.gov/standards/status/textstatus.html [accessed March 2002].

© 2004 CASI 75


Wadhams, P., Tucker, W.B., III, Krabill, W.B., Swift, R.N., Comiso, J.C., andDavis, N.R. 1992. Relationship between sea ice freeboard and draft in theArctic Basin and implications for ice thickness monitoring. Journal ofGeophysical Research, Vol. 97, No. C12, pp. 20 325 – 20 334.

Webster, T.L., Dickie, S., O’Reilly, C., Forbes, D.L., Thompson, K., andParkes, G. 2001. Integration of diverse datasets and knowledge to producehigh resolution elevation flood risk maps for Charlottetown, Prince EdwardIsland, Canada. In Coast GIS2001, 4th International Symposium onComputer Mapping and GIS for Coastal Zone Management, Manuscripts.St. Mary’s University, Halifax, N.S. Vol. 2. 27 pp.

Webster, T.L., Forbes, D.L., Dickie, S., Colville, R., and Parkes, G.S. 2002.Airborne imaging, digital elevation models and flood maps. In Coastalimpacts of climate change and sea-level rise on Prince Edward Island.Edited by D.L. Forbes and R.W. Shaw. [CD-ROM]. Geological Survey ofCanada Open File 4261, Supporting Document 8. 36 pp.

Wehr, A., and Lohr, U. 1999. Airborne laser scanning — an introduction andoverview. ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 54,pp. 68–82.

76 © 2004 CASI
