area-based quality control of airborne laser scanning 3d ...scanning 3d models for different land...

Full Terms & Conditions of access and use can be found athttp://www.tandfonline.com/action/journalInformation?journalCode=tres20

Download by: [Bulent Ecevit University] Date: 24 November 2015, At: 06:28

International Journal of Remote Sensing

ISSN: 0143-1161 (Print) 1366-5901 (Online) Journal homepage: http://www.tandfonline.com/loi/tres20

Area-based quality control of airborne laserscanning 3D models for different land classesusing terrestrial laser scanning: sample survey inHouston, USA

Umut G. Sefercik, Craig Glennie, Abhinav Singhania & Darren Hauser

To cite this article: Umut G. Sefercik, Craig Glennie, Abhinav Singhania & Darren Hauser(2015) Area-based quality control of airborne laser scanning 3D models for different landclasses using terrestrial laser scanning: sample survey in Houston, USA, International Journalof Remote Sensing, 36:23, 5916-5934, DOI: 10.1080/01431161.2015.1110260

To link to this article: http://dx.doi.org/10.1080/01431161.2015.1110260

Published online: 23 Nov 2015.

Submit your article to this journal

View related articles

View Crossmark data

http://www.tandfonline.com/action/journalInformation?journalCode=tres20http://www.tandfonline.com/loi/tres20http://www.tandfonline.com/action/showCitFormats?doi=10.1080/01431161.2015.1110260http://dx.doi.org/10.1080/01431161.2015.1110260http://www.tandfonline.com/action/authorSubmission?journalCode=tres20&page=instructionshttp://www.tandfonline.com/action/authorSubmission?journalCode=tres20&page=instructionshttp://www.tandfonline.com/doi/mlt/10.1080/01431161.2015.1110260http://www.tandfonline.com/doi/mlt/10.1080/01431161.2015.1110260http://crossmark.crossref.org/dialog/?doi=10.1080/01431161.2015.1110260&domain=pdf&date_stamp=2015-11-23http://crossmark.crossref.org/dialog/?doi=10.1080/01431161.2015.1110260&domain=pdf&date_stamp=2015-11-23

Area-based quality control of airborne laser scanning 3D models fordifferent land classes using terrestrial laser scanning: sample survey

in Houston, USA

Umut G. Sefercika*, Craig Glennieb, Abhinav Singhaniab, and Darren Hauserb

aDepartment of Geomatics Engineering, Bulent Ecevit University, Zonguldak, Turkey; bNationalCenter for Airborne Laser Mapping, University of Houston, Houston, USA

(Received 25 January 2015; accepted 11 October 2015)

Airborne laser scanning (ALS) is a remote-sensing technique that provides scale-accurate 3D models consisting of dense point clouds with x, y planimetric coordinatesand altitude z. Using ALS, very high-resolution (VHR) digital surface models(DSMs) have been widely used for commercial and scientific applications since theearly 1990s. Although there is widespread usage, there has been little comprehensiveinvestigation of quality control for ALS DSMs in the literature, as most studies havebeen limited to assessing point-based vertical accuracy. This article is dedicated toinvestigating the quality of ALS DSMs for different land classes using statistical andvisual approaches based on absolute and relative vertical accuracy metrics. Ratherthan a limited number of ground control points (GCP), the model-to-model-basedapproach is applied and DSMs derived from terrestrial laser scanning (TLS) pointclouds that have around 5 mm absolute and 3 mm relative geolocation accuracy wereused as the reference data for comparison. The results demonstrate that in open, grass,and building land classes, the ALS DSMs reached both standard deviation (σ) andnormalized median absolute deviation (NMAD) of 3–5 cm after the elimination ofany systematic biases. This result sufficiently satisfies the vertical accuracy require-ments for 1/1000-scale topographic maps determined by National Digital ElevationProgram (NDEP) specifications. In tall vegetation, a higher number of discrepancieslarger than 0.5 m exist, reversing the relation between σ and NMAD. These vegeta-tion errors also do not appear to be normally distributed. As an additional investiga-tion, the performance of ALS DEMs under dense high-vegetation areas was assessed.These under-canopy ALS DEMs, created using only classified ground returns, offerboth σ and NMAD of 12–14 cm, a performance level that is difficult to achieveunder-canopy using photogrammetric techniques.

1. Introduction

Airborne laser scanning (ALS) offers highly accurate 3D topographic point clouds thathave traditionally not been provided by competing remote-sensing technologies. Becauseof the ability of AL to provide rapid, high-resolution, and accurate data, it is graduallybeing adopted as the primary technique for mapping local- to regional-scale coverage, anarea historically dominated by photogrammetry (Baltsavias 1999; Hill et al. 2000).Accordingly, ALS digital surface models (DSMs), digital terrain models (DTMs), anddigital elevation models (DEMs) have undergone significant study in the literature (Lohr1998; McIntosh, Krupnik, and Schenk 2000; Shan and Aparajithan 2005; Mandlburger,Briese, and Pfeifer 2007; Liu 2008; Baligh, Valadan Zoej, and Mohammadzadeh 2008).

*Corresponding author. Email: [email protected]

International Journal of Remote Sensing, 2015Vol. 36, No. 23, 5916–5934, http://dx.doi.org/10.1080/01431161.2015.1110260

© 2015 Taylor & Francis

Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

http://www.tandfonline.com

While DSMs are the 3D cartographic representation of the earth’s surface including all-terrain and non-terrain formations, DTMs and DEMs contain only the bare ground; themain difference between DTMs and DEMs is the additional information derived frombreak-lines and mass points (in DTMs). Using ALS DSMs, DTMs, and DEMs, a widevariety of mapping applications is realized such as city modelling, forestry, agriculture,hydrology, bathymetry, and disaster monitoring and management (Haala and Brenner1997; Guenther et al. 2000; Carson et al. 2004; Hollaus, Wagner, and Kraus 2005;Hyyppä et al. 2006; McCoy, Asner, and Graves 2011; Mandlburger et al. 2011; Trinderand Salah 2011). Although widely utilized, the quality of ALS 3D earth models has notbeen comprehensively examined in the literature to date. There are a significant numberof studies on ALS accuracy; however, in almost all of these studies the analysis wasbased on vertical comparisons with ground control points (GCPs) (e.g. Maas 2003;Csanyi and Toth 2007; Dahlqvist et al. 2011; Pourali et al. 2014). Considering the nearlycontinuous nature of a LiDAR point cloud, a limited number of GCPs is not sufficient forcomparison and the results may be misleading, especially for varied land-cover classifi-cations. Additionally, GCPs are commonly compiled using real-time kinematic (RTK)global positioning system (GPS) instruments, which can have up to 3 cm horizontal and5 cm vertical geolocation errors in most cases (Mekik and Arslanoglu 2009). With staticGPS surveying, higher-accuracy GCPs can be collected but the longer observation timesmake it almost impossible to achieve a sufficient number of them. In this article, wepropose a direct model-to-model-based approach to quantify the quality of ALS DSMs indifferent land-cover classes (Ressl, Kager, and Mandlburger 2008; Ressl, Mandlburger,and Pfeifer 2009). We use terrestrial laser scanning (TLS) measurements as control andto generate reference DSMs for comparison with the ALS. The comparisons wereundertaken for six study areas located on the University of Houston’s central campusand included four different land classes to investigate the quality of ALS DSMs forseparate terrain and non-terrain formations. The quality assessment was performed usingstatistical and visual approaches based on absolute vertical accuracies (AVA) and relativevertical accuracies (RVA).

In addition to studies in different land classes, we also investigated the accuracy ofALS DEMs under a vegetation canopy. In 3D earth modelling using remote-sensingtechniques, one of the most significant issues is correct surface determination underdense vegetation. In operator-based stereo-image evaluations such as aerial photogram-metry and space-borne optical imaging, it is impossible in most cases to see bare surfaceand collect data under dense high vegetation. As a consequence, these are the mostinaccurate surfaces in DEMs. In contrast to passive optical techniques, ALS is normallymore adept at collecting data under high vegetation due to the multiple-return capabilityof modern ALS systems (Glennie et al. 2013). As a result, we also assessed ALS DEMperformance under high vegetation.

2. Study areas and instruments

The investigations were performed on data sets acquired over the central campus of theUniversity of Houston (UH) in Houston, TX, USA. For the quality assessment of a DSMderived from remotely sensed data, study areas with different land classes are veryimportant. It is well established that most remote-sensing data collection systems haveproblems in accurately modelling vegetated areas and inclined/complex terrain and non-terrain formations. In this context, the performance of the DSM needs to be estimated for

International Journal of Remote Sensing 5917

Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

different land classes, including open areas, to comment on overall quality. From the ALSsurvey of the UH campus, six representative target areas were selected. Areas 1 and 2consisted of open, grass, and high-vegetation layers, while 3, 4, 5, and 6 containedbuildings. In the study areas, the orthometric heights varied between 10 and 35 m.Figure 1 shows the study areas overlaid on an orthophoto of the UH central campus.The digital photography was collected simultaneously with the ALS data, and the landclasses were obtained by manual vectorization of the imagery using NetCAD 5.2 software.

The ALS data of University of Houston campus were collected by the NationalCenter for Airborne Laser Mapping (NCALM) on 23 June 2012 using an Optech Geminiairborne laser terrain mapper (ALTM). The data were post-processed and calibrated(Glennie et al. 2013) by NCALM, and a final point cloud was generated. For ease ofuse, the full point cloud was divided into 27 independent tiles each covering an area of1 km2. TLS data was also acquired for all six study areas using a Riegl VZ-400instrument, although not coincident with the airborne collection. Detailed informationfrom both the ALS and TLS instruments and measurements is presented in Table 1. TheTLS point clouds were georeferences using external targets whose locations were pre-cisely determined with long-period (>1 hour) static surveys using dual-frequency GlobalNavigation Satellite System (GNSS) receivers. Under 100 m range, the data has around5 mm absolute geolocation accuracy and 3 mm relative accuracy (precision, reproduci-bility, repeatability) (Riegl 2015).

3. Methodology

In order to study the accuracy of ALS for our six chosen cases, we designed amethodology (see Figure 2) for the generation of DSMs for four classification types inour study area, and then added a separate methodology for DEM generation under tree

Figure 1. Study areas (left) and land classes for each area (right).

5918 U.G. Sefercik et al.

Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

Table

1.Specificationsof

theALSandTLSinstruments

used

inthisstudy(D

I2014;Riegl

2015).

Param

etre

ALS

TLS

Area1

Area2

Area3

Area4

Area5

Area6

Instrument

OptechGem

iniALT

MRiegl

VZ-400

Date

23/06/12

27/06/12

13/11/13

02/04/14

04/07/14

04/07/14

04/07/14

Day

oftheyear

175

179

317

9297

9797

Point

density(m

2)

45around

10,000

points

≤10

mhorizontal

distance

Pulse

rate

(kHz)

167

300

Wavelength(nm)

1064

1550

Scanfrequency(H

z)0–

70N/A

Beam

divergence

(mrad)

0.25

0.35

Field

ofview

(º)

−25

to+25

360horizontal/–40

to+60

vertical

Flightaltitude

(m)

1000

N/A

Num

berof

returns

4Unlim

ited


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

canopy. The methodology used the commercial software packages TerraScan forMicroStation (v14) and Surfer (v11). For quality assessment we utilized the 3D modelsoftware BLUH (Bundle Block Adjustment Leibniz University of Hannover). Theoriginal ALS data were divided to 27 independent tiles of area 1 × 1 km to facilitateprocessing. For our analysis the DSM generation process was started with the combina-tion of the two ALS tiles that included all study areas containing more than 90 millionpoints. The tiles were further segmented to rectangular blocks for each study area forrapid processing with fewer points.

When generating a very high-resolution (VHR) DSM, DEM, or DTM, the computa-tion of height values for all pixels is not possible because of the very small grid spacing.This is especially true for ALS data because of the presence of gaps in the point cloudsthat depend on resolution (point density), geometry of flight strip acquisition based onscanning type (Z-shaped, parallel lines, or elliptical scan), scanner characteristics (scanangle, field of view, number of returns, etc.), and localized occlusions due to buildingsand vegetation. Due to the irregular nature of raw LiDAR returns, interpolation is anecessary step for the generation of raster DSMs and therefore the rectangular blocks

Figure 2. Generation and assessment methodology.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

were chosen to be slightly larger than the study areas to ensure no edge effects as a resultof interpolation. For filtering of any blunder or noise points from both the ALS and TLSpoint clouds, a four-stage process was used as follows: 1) isolated and low (i.e. below-ground) points were removed using automated filters in the TerraScan software; 2)vertical profiles were plotted and height intervals were determined that represented theground truth and maximum object heights; 3) DSM classes were created; and 4) pointclouds were transferred to the DSM class by absolute elevation, considering predeter-mined height intervals from vertical profiles. For DEM generation only the ground pointswere used, and therefore only the first of these steps was required.

One of the most significant advantages of ALS over other remote-sensing techniques isthe density of the point clouds, which enables the generation of VHR 3D earth models.Studies have shown that the optimal original grid spacing of 3D earth models derived fromremotely sensed images is given as 3 × ground sampling distance (GSD), which defines thedistance between pixel centres measured on the ground; otherwise interpolation causessubstantial loss of vertical accuracy due to an insufficient amount of data (e.g. Buyuksalihand Jacobsen 2007; Baltsavias et al. 2008; Sefercik and Alkan 2009; Jacobsen 2012). Weadopted this theory for the ALS point clouds and used a metric of 3 × mean point spacing,calculated on the basis of the number of collected points per square metre (point density).Depending on the scanning mechanism, a different point spacing might occur in the scan linecompared with the distance between scan lines; that is why this theory is acceptable in thecase of sufficiently regular point distribution. Using this figure of merit, the 5–10 cm averagepoint spacing of both the TLS and ALS data enabled the creation of 25 cm-gridded DSMsand DEMs. For VHR DSM generation with ALS data, a problem can occur with verticalobjects which is distinct from other remote-sensing techniques, especially for high-pointdensity surveys. For primarily vertical objects such as trees and walls, several points ofdifferent elevation are present in the same pixel of the DSM. In these cases, it is preferablewhen using the interpolation method to use the maximum point elevation within the pixelinstead of averaging all points within the pixel (Hollaus et al. 2010). Therefore, the assign-ment of a maximum elevation for each pixel was preferred for the interpolation of DSMsrather than kriging (Yang et al. 2004), which is more commonly used for other remote-sensing data types. For DEM generation, a simple nearest-neighbour interpolation methodwas used, due to the high point density. After surface generation, the models (both TLS andALS) for each of the study blocks were cropped to the vectorized limits of the study areas.

For the quality control of 3D earth models derived from remotely sensed data, bothpoint- and model-based methods have previously been presented. Point-basedapproaches validate the quality of a model compared with measured GCPs on the groundor on available high-resolution maps. However, point-based approaches often use alimited number of GCPs and, considering the VHR modelling capability of ALS, theresults may be misleading and not fully evaluate accuracy for different land classes andchanging terrain slope. Point-based methods also make it difficult to evaluate the effectsof the interpolation method used. Therefore, a more effective method for quality assess-ment is comparison to a reference model (Lin, Vesecky, and Zebker 1994). To use amodel as a reference requires the following conditions be met: 1) it must be generatedwith a more accurate technique; 2) it must cover the entire study area with no gaps ordistortions; and, 3) it must have original (resampling is not valid) grid spacing smallerthan or equal to the tested model. In this study, the generated TLS DSMs have a 25 cmoriginal grid spacing, higher accuracy, and no significant data gaps, satisfying all therequirements for a reliable model-based quality assessment of ALS DSMs and DEM.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

For the quality assessment, AVAs and RVAs were used as the basic metrics;however, for correct calculation of these values, some preprocessing steps need to beapplied. First, the vertical datum has to be fixed as either the geoid or ellipsoid, andgeoid undulations have to be calculated if required. Second, the coordinate systems ofthe test and reference models must be the same and any minor horizontal offsets whichoccur due to horizontal geolocation errors of both models based on GPS, IMUmeasurements, or national datum based errors must be compensated. In our study,these residual horizontal errors are likely a result of inaccuracy in the ALS data set,and should be lower than the overall expected horizontal error of the ALS survey. Inthe study, the common coordinate system and the datum were Universal TransverseMercator (UTM) zone 15 on the National American Datum 1983 (NAD83), and thegeoid undulations were calculated based on the National Geodetic Survey Geoid12Amodel (NGS 2014) to obtain orthometric heights. The horizontal offsets were elimi-nated by shifting based on an areal matching cross-correlation method (Baltsaviaset al. 2008; Alobeid, Jacobsen, and Heipke 2010). Table 2 shows the calculated geoidundulations for each study area and the estimated horizontal offsets. Note that theestimated horizontal offsets are well below the expected horizontal accuracy of theALS survey (see, for example, Glennie 2007; Goulden and Hopkinson 2010).

The vertical accuracy of a DSM can be described using many different criteria. Insome applications, it is described using root mean square error (RMSE) or the standarddeviation of height discrepancies (RMSE�Z or σ) between tested and reference models,calculated as in Equations (1) and (3). In this study, σ that estimates the vertical accuracywith a 68% probability level was used as the main evaluation criteria:

RMSE�Z ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPn

i¼1 �Zi2

n

s; (1)

where n is the number of pixels (comparison points). If the arithmetic mean of the heightdiscrepancies is given as

μ ¼Pn

i¼1�Zin

; (2)

then the standard deviation of height discrepancies is equal to

σ ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPn

i¼1 �Zi � μð Þ2n� 1

s; (3)

Table 2. Geoid undulations (N) and eliminated horizontal offsets.

Reference model Tested model Area N (m) dx (cm) dy (cm)

TLS ALS 1 27.281 2.3 4.22 27.284 1.9 3.13 27.279 8.3 8.54 27.279 6.9 18.85 27.279 8.0 −2.56 27.279 2.4 9.3


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

which is the square root of the sample variance. As a consequence, the relation betweenσ�Z and RMSE�Z is given as

σ2 ¼ RMSE�Z2 � μ2; (4)

AVA, which is equal to the standard deviation of height discrepancies, is correlatedwith terrain inclination and is therefore normally given as a function of terrain slope. Therelationship between terrain inclination and AVA is given in Equation (5), where b is themultiplication factor for terrain slope and α is the terrain slope. For the AVA analyses, atwo-step operation was used to compute the final results. In the first step, any systematicbias between the test and reference DSMs was determined by calculating the verticalshift and scale differences (Huising and Gomes Pereira 1998; Crombaghs, Bruegelmann,and De Min 2000; Wang et al. 2008). In the second step, the AVAs were calculated withthe systematic bias eliminated:

AVA ¼ σ þ b tan αð Þ: (5)

In addition to σ, the normalized median absolute deviation (NMAD) is used as asecond absolute accuracy metric. NMAD is the derivative of median absolute deviation(MAD, which is a robust measure of the variability of a univariate sample of quantitativedata. MAD and NMAD are calculated by Equations (6) and (7), where ~xj is the median ofthe univariate data set of height discrepancies (�Z1, �Z2, �Z3, . . ..., �Zn) and ~xi is themedian of height discrepancies from ~xj.

MAD ¼ ~xi �Zi�~xj �Zj� �� ; (6)

NMAD ¼ 1:4815� MADð Þ: (7)

In the case of normally distributed height discrepancies between the tested and referencemodel, NMAD has a 68% probability level (identical to σ) with the multiplication of MADby a factor of 1.48. A NMAD bigger than σ�Z is an indicator of an abnormal distribution ofheight discrepancies. Although a robust estimator, NMAD is not as sensitive in regard tostandard deviation for the determination of minor outliers in a large data set (Hellerstein2008). The vertical accuracy can also be described as linear errors at 90 and 95% probabilitylevels (LE90 and LE95) by σ � 1:65 and σ � 1:96 in the frequency distribution of heightdiscrepancies, respectively, under the National Digital Elevation Program (NDEP) instruc-tions for the United States (NDEP, 2004) however, they are preferred for heterogeneous datasets which have larger height discrepancies (Jacobsen 2013), which is why they are notplaced in AVATables 3 and 5. The shortest distance between two 3Dmodels is not actually inthe vertical direction but is given by the Euclidian distance (ΔZ × cos αð Þ), considering theterrain slope. Therefore RMSE�Z in Euclidian distance is also presented as an accuracyfigure. However, for normally distributed height discrepancies the difference betweenRMSE�Z and RMSE�Z Euclid is negligible.

Quality assessment of the ALS DSMs was performed separately for six study areas,which included different land classes (open, grass, building, and high vegetation). For theAVA calculations, maximum thresholds were determined for height discrepanciesbetween the test and reference models to avoid misleading results because of blunders.The thresholds were selected considering the characteristics of the evaluated land class.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

Table

3.AVAswithandwithout

system

atic

bias.

Ref.DSM

Tested

DSM

Area

Num

ber

ofpoints

Landclass

AVAwithout

system

atic

bias

(m)

Excluded

points(%

)σ

RMSE�Z

RMSE�ZEuclid

MAD

NMAD

TLS25

cmALS25

cm1

134197

whole

area

(100%)

0.06

0.08

0.08

0.04

0.06

0.17

open

(32.70%)

0.05

+0.09tan(α)

0.07

0.07

0.02

0.03

0.29

grass(48.48%)

0.03

+0.02tan(α)

0.04

0.04

0.02

0.03

0.03

high

veg.

(18.81%)

0.48

0.48

0.44

0.41

0.61

0.17

2208711

whole

area

(100%)

0.07

0.13

0.12

0.05

0.08

0.36

open

(8.83%

)0.05

+0.06tan(α)

0.09

0.09

0.03

0.05

0.57

grass(39.38%)

0.05

+0.01tan(α)

0.07

0.07

0.03

0.05

0.07

high

veg.

(51.79%)

0.29

+0.10tan(α)

0.48

0.42

0.35

0.52

0.24

3142365

built-up

(100%)

0.05

+0.20tan(α)

0.09

0.08

0.03

0.04

2.48

419851

built-up

(100%)

0.04

+0.06tan(α)

0.04

0.04

0.02

0.03

0.03

522298

built-up

(100%)

0.05

+0.02tan(α)

0.05

0.05

0.03

0.05

0.00

69570

built-up

(100%)

0.03

+0.1tan(α)

0.04

0.04

0.02

0.03

0.18


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

Therefore, for example, a higher threshold was employed for high-vegetation areasbecause these may include greater height discrepancies, especially considering the effectof the temporal spacing between the ALS and TLS collection dates, and the differentviewing geometry of the ALS and TLS scans; TLS images the vegetation from below,whereas ALS images the vegetation top. For each class, all blunder pixels exceeding thethresholds were visualized to show the excluded points for the AVA analyses.

At the end of AVA analyses, colour-coded DEM of difference (DoD) maps (Wheatonet al. 2010) were created for deviation visualization using LISA 4.7 software (Linder2005) with Equation (8):

DoD ¼ DSMð Þtested � DSMð Þreference: (8)

The second main metric for quality assessment is the RVA, which indicates theinterior consistency of a DSM. RVA is calculated separately for distance groups: distanceof the current point to a reference point for both of which we have height differences dZ(difference of Z (reference) – Z (compared point in the distance group)). Namely, theheight discrepancies of neighbouring points are compared. If the height differences areindependent (correlation = 0), RVA is identical to σ�Z . If the correlation coefficient is1.0, RVA would be 0.0. In reality RVA is between 0.0 and σ�Z – in extreme case ofnegative correlation RVA may also be larger as σ. With RVA, the dependency (correlationof dZ) is checked, which means that local random and random errors can be separated.Previously published studies have shown that the RVAs of active remote-sensing systemssuch as radar and lidar are superior to their AVAs (Jacobsen 2003, 2012; ASPRS 2004).

The RVAs of the DSMs were calculated using Equation (9), where D represents thedistance groups and Dl and Du are the lower and upper distance limits, respectively. Thedistance groups change from 1st to 10th pixel (0.25–2.5 m considering the 0.25 m pixelsize of reference data) for this study because, for each pixel, the neighbouring 10 pixelswere used to calculate the RVAs. The term nx is used for the number of point combina-tions in the distance group – n is used for the total number of differences betweenreference and actual values. The factor 2 for multiplication with nx is used for normingRVA to the size of σ – if the height differences of the points in the distance group areindependent, corresponding to error propagation, the RMS would be the square root of2.0 larger as σ – this is respected by 2 × nx:

RVA ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiP ðdZi � dZjÞ2

2� nxð Þ

s;Dl < D < Du: (9)

4. Results and discussion

4.1. Results of ALS DSMs

Figure 3 shows the generated ALS DSMs (colour-coded) for the six study areas. Aspreviously mentioned, the land classes in study areas were determined by manualvectorization of 42 orthophotos with a 4 cm GSD in UTM 15º NAD83 datum thatwere acquired concurrently with the ALS data collection. The orthophotos weremosaicked in ENVI 5.0. When comparing ALS and TLS DSMs, the raster boundaryimages derived from manual vectorization (see Figure 1) were used to define each of theindependent land classes. Any height discrepancies including systematic bias higher than


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

1 m for open, grass, and building land classes and 2 m for high vegetation were classifiedas outliers and removed. We also set a maximum accepted tangent of slope of 1.0 (45º)for inclusion of points in the statistics. Table 3 shows the AVA statistics for ALS DSMs,including the number of compared points (pixels) and the percentage of excluded points.The graphs in Figure 4 illustrate the distribution of height discrepancies between the testand reference DSMs.

Figure 3. Generated ALS DSMs (colour-coded) with height scale.

Figure 4. Distribution of height differences between test and reference DSMs.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

Table 3 shows that the AVAs based on σ and NMAD for open, grass, and buildingland classes are very similar and range between 3 and 5 cm. The distributions of theheight differences (see Figure 4) are normal and symmetric for these land classes.However, the results in both Table 3 and Figure 5 show significant problems in regardto high vegetation. For both high-vegetation areas, NMAD is larger than σ; this indicatesan abnormal distribution of height discrepancies. Although high-vegetation height differ-ences were not normally distributed, we did not observe bimodal distributions. The peaksof both distributions show a mean offset of about 15 cm, but due to the significant spreadof the height differences (±1.45 m) the σ�Z and NMAD are between 29–48 cm and52–61 cm, respectively.

As can be seen in Table 3, a small percentage of excluded points existed for each ofthe different land classes. To visualize the excluded points, Figure 5 highlights these in

Figure 5. Excluded points (white) in the AVA analyses (a = area1 – open; b = area1 – grass; c =area1 – high vegetation; d = area2 – open; e = area2 – grass; f = area2 – high vegetation; g = area3– building; h = area4 – building; I = area5 – building; j = area6 – building).


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

white. From an analysis of Figure 5, the majority of excluded points in open areas aredue to cars and pedestrians in the ALS data. Similarly the building classes were selectedfrom open-roof parking garages on the university campus (to enable TLS measurements),and therefore here as well the excluded points are mostly the result of parked cars. In thegrassy areas, the excluded points are due to campus artwork (sculptures) that were notseparated out by manual vectorization. In high-vegetation areas the main cause ofexcluded points is interpolation errors at the edges of the vegetation resulting fromsignificant elevation differences between neighbouring pixels. This is a typical problemalso common in the generation of DSMs derived from photogrammetric and space-borneimaging data.

Figure 6 shows colour-coded DoD maps of the study areas based on σ�Z from theAVA. The height differences were scaled as lower than −15 cm (15 cm) for the whole areas. When examining thehistograms in Figure 6, ignoring 15 cm points which are mostly located inthe high-vegetation layers in areas 1 and 2, the coherence between the test and referenceDSMs is very good and the height differences are mostly concentrated around zero.

NDEP defines AVAs into three different accuracy types: fundamental vertical accu-racy (FVA), supplemental vertical accuracy (SVA), and consolidated vertical accuracy

Figure 6. Colour-coded DoD maps of the ALS DSMs (white = no data).


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

(CVA). FVA is calculated based on the results in open areas that is generally the mostreliable land class because it does not change temporally. The FVA of a DSM in openareas is also identical for both the DEM and DTM. SVA represents the AVA of otherindividual land classes and CVA is the arithmetic mean of all classifications included inthe study area. NDEP guidelines state that FVA has to be reported for quality validationof any height model, but SVA and CVA are optional. Table 4 shows the calculated FVAs,SVAs, and CVAs, neglecting the effect of slope, based on σ and NMAD. In NDEPstandards (NDEP 2004), the required AVAs for mapping at different scales have beendescribed; for the generation of 1/1000-scale topographic maps including a 1 m contourinterval, 0.3–0.4 m vertical accuracy is required. Given this requirement, the AVAs of theALS DSMs are more than satisfactory for the generation of 1/1000-scale topographicmaps.

The graphs in Figure 7 show the RVAs of the test ALS DSMs. In open, grass, andbuilding areas, after a range of three to four pixels, the neighbouring pixels are veryhomogeneous and thus the RVAs are stable. These results support the coherence of σ�Z

Table 4. AVAs based on NDEP specifications.

ReferenceDSM

TestDSM

FVA(cm)

SVA (cm)

CVA(cm)

Accuracyfiguregrass building

highvegetation

TLS ALS 5.00 4.00 4.25 38.50 12.94 σ4.00 4.00 3.75 56.50 17.06 NMAD

Figure 7. RVAs of ALS DSMs.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

and NMAD in the AVA analyses. In high-vegetation areas, the RVAs show a similar trendand indicate about 35 cm error for every first pixel. This shows that there is an unsolvedsystematic bias for high vegetation. This bias is most likely caused by the differing viewgeometry between the TLS and ALS observations. ALS looks downward on the vegeta-tion, and therefore the top of the canopy represents the first return surface in the ALSdata. TLS, captured from the ground, must pass through the vegetation understorey toreach the top of canopy. Therefore, it is very likely that the top of canopy is not as wellrepresented in the TLS data set, despite its much higher resolution. This seems to beconfirmed by the analysis results, as the TLS DSMs for high vegetation are system-atically lower than that of the ALS data set.

4.2. Results of ALS DEMs under high vegetation

Table 5 indicates the performance of ALS DEMs underneath high vegetation. The AVAsare similar for areas 1 and 2 (12–14 cm), with no excluded points. This means thatground points which are generally achieved with last returns of the signal correctlyrepresent the bare earth surface. Although the AVA is two- to threefold worse than theAVA in open areas (Table 3), it is still more than satisfactory for 1/1000-scale topo-graphic mapping. In Figure 8, the height difference modes are 0–5 cm while the RVAsshow similar trends and are also markedly better than the high-vegetation DSMs for thesame area. It should be noted that the areas under the high vegetation in the study areaconsisted mostly of manicured grass with little or no low-lying bushes or shrubs.

Table 5. AVAs of ALS DEMs under high vegetation.

Ref.DSM

TestDSM Area

Numberof points Land class

AVA (m)

Excludedpoints (%)σ RMSE�Z

RMSE�ZEuclid MAD NMAD

TLS25 cm

ALS25 cm

1 134197 high vegetation(18.81%)

0.12 0.13 0.12 0.09 0.13 0.00

2 208711 high vegetation(51.79%)

0.14 0.15 0.15 0.10 0.14 0.00

Figure 8. Distribution of height differences between tested ALS DEMs and reference DEMs(left) and RVAs (right).


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

Therefore, the decrease in accuracy is directly attributable to the presence of highvegetation only, and not to the presence of understorey obscuring the ground.

5. Conclusions

In this article, the quality of VHR ALS 3D earth models was evaluated through the use ofcomprehensive statistical and visual approaches based on comparison with TLS-basedreference models. This is significant because the approach provides a more realisticassessment of the overall accuracy of the ALS data over a variety of land classescompared with the traditional ALS accuracy analysis using only GCPs. The area-basedapproach presented allows the analysis of a large number of spatially distributed obser-vations. Due to the direct model-to-model comparison, each pixel in the ALS 3D data setwas evaluated as a part of the analyses to determine the quality in different land classesseparately: open, grass, building, and high-vegetation layers.

The AVA of VHR ALS DSMs in open, grass, and building areas is very similar andin the range 3–5 cm based both on σ and NMAD. The height discrepancies between thetest ALS and reference TLS DSMs were also normally distributed for these classes, andthe ALS DSMs easily satisfy the vertical accuracy requirement for 1/1000-scale topo-graphic maps. However, in high-vegetation areas, a greater number of differences higherthan 0.5 m exist than would be expected for normally distributed errors. The interiorhomogeneity of ALS DSMs was evaluated by RVA analysis. The results show thatneighbouring pixels are homogeneous for all land classes, with the exception of high-vegetation areas where about 35 cm systematic error is observed. This larger discrepancyoccurs due to the different viewing geometry of the ALS and TLS data sets. The TLS iscaptured from the ground, and needs to pass through significant vegetation understoreyand therefore may not represent the top of canopy well. The ALS DEMs also showedremarkable accuracy under high-vegetation areas, albeit 2–3-fold worse than that foropen areas. Both AVAs and RVAs are better than the vegetation DSMs and the trends ofRVAs are very similar for two DEMs. Overall, the vertical accuracy under canopy stillsatisfies the requirement for 1/1000-scale topographic maps. Future work will examinethe deviations between the ALS and TLS in high-vegetation areas to determine whetherthe amount of understorey is correlated to the magnitude of the height differencesbetween the ALS and TLS DEMs.

AcknowledgementsThe Scientific and Technological Research Council of Turkey (TUBITAK) is thanked for support-ing this study within the scope of a 2219 post-doctoral scholarship to the first author. Many thanksto the University of Houston, National Center for Airborne Laser Mapping (NCALM) for ALSdata, TerraScan, Surfer, and ENVI software. Drs Karsten Jacobsen and Wilfred Linder are alsothanked for their support in regard to the BLUH and LISA software. Finally, many thanks forcontributions by Bulent Ecevit University.

Disclosure statementNo potential conflict of interest was reported by the authors.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

ReferencesAlobeid, A., K. Jacobsen, and C. Heipke. 2010. “Comparison of Matching Algorithms for DSM

Generation in Urban Areas from IKONOS Imagery.” Photogrammetric Engineering & RemoteSensing 76 (9): 1041–1050. doi:10.14358/PERS.76.9.1041.

ASPRS (American Society of Photogrammetry and Remote Sensing). 2004. “ASPRS GuidelinesVertical Accuracy Reporting for Lidar Data.” http://www.asprs.org/a/society/committees/lidar/Downloads/Vertical_Accuracy_Reporting_for_Lidar_Data.pdf

Baligh, A., M. J. Valadan Zoej, and A. Mohammadzadeh. 2008. “Bare Earth Extraction fromAirborne Lidar Data Using Different Filtering Methods.” The International Archives of thePhotogrammetry, Remote Sensing and Spatial Information Sciences XXXVII (B3b): 237–240.

Baltsavias, E. 1999. “A Comparison between Photogrammetry and Laser Scanning.” ISPRSJournal of Photogrammetry and Remote Sensing 54 (2–3): 83–94. doi:10.1016/S0924-2716(99)00014-3.

Baltsavias, E., A. Gruen, H. Eisenbeiss, L. Zhang, and T. Waser. 2008. “High-Quality ImageMatching and Automated Generation of 3D Tree Models.” International Journal of RemoteSensing 29 (5): 1243–1259. doi:10.1080/01431160701736513.

Buyuksalih, G., and K. Jacobsen. 2007. “Comparison of DEM Generation by Very High ResolutionOptical Satellites.” In New Developments and Challenges in Remote Sensing, edited by Z.Bochenek, 627–637. Rotterdam: Millpress.

Carson, W. W., H. E. Andersen, S. E. Reutebuch, and R. J. McGaughey. 2004. “Lidar Applicationin Forestry - An Overview.” Paper Presented at the Annual Conference of American Society ofPhotogrammetry and Remote Sensing, Denver, Colorado, USA, May 23-28, p. 9

Crombaghs, M. J. E., R. Bruegelmann, and E. J. De Min. 2000. “On the Adjustment ofOverlapping Strips of Laser Altimeter Height Data.” International Archives ofPhotogrammetry and Remote Sensing 33 (B3/1): 230–237.

Csanyi, N., and C. K. Toth. 2007. “Improvement of Lidar Data Accuracy Using Lidar-SpecificGround Targets.” Photogrammetric Engineering & Remote Sensing 73 (4): 385–396.doi:10.14358/PERS.73.4.385.

Dahlqvist, S., P. Rönnholm, P. Salo, and M. Vermeer. 2011. “Evaluating the Correctness ofAirborne Laser Scanning Data Heights Using Vehicle-based RTK and VRS GPSObservations.” Remote Sensing 3 (12): 1902–1913. doi:10.3390/rs3091902.

DI (Direct Industry). 2014. “Catalogue of Optech Gemini ALTM.” http://pdf.directindustry.com/pdf/optech/altm-gemini/25132-53143.html

Glennie, C. 2007. “Rigorous 3D Error Analysis of Kinematic Scanning Lidar Systems.” Journal ofApplied Geodesy 1 (3): 147–157. doi:10.1515/jag.2007.017.

Glennie, C., W. E. Carter, R. L. Shrestha, and W. E. Dietrich. 2013. “Geodetic Imaging withAirborne LiDAR: The Earth’s Surface Revealed.” Reports on Progress in Physics 76 (8):086801. doi:10.1088/0034-4885/76/8/086801.

Goulden, T., and C. Hopkinson. 2010. “The Forward Propagation of Integrated System ComponentErrors within Airborne Lidar Data.” Photogrammetric Engineering & Remote Sensing 76 (5):589–601. doi:10.14358/PERS.76.5.589.

Guenther, G. C., A. G. Cunningham, P. E. LaRocque, and D. J. Reid. 2000. “Meeting the AccuracyChallenge in Airborne LIDAR Bathymetry.” Paper presented at EARSeL SIG Workshop onLIDAR Dresden/FRG, June 16 – 17.

Haala, N., and C. Brenner. 1997. “Generation of 3D City Models from Airborne Laser ScanningData.” Paper presented at 3rd EARSeL Workshop on LIDAR, Remote Sensing on Land andSea, Tallinn, Estonia, July 17-19, 105–112.

Hellerstein, J. M. 2008. Quantitative Data Cleaning for Large Databases. Technical ReportPresented at United Nations Economic Commission for Europe (UNECE), p. 42. http://db.cs.berkeley.edu/jmh/papers/cleaning-unece.pdf.

Hill, J. M., L. A. Graham, R. J. Henry, D. M. Cotter, A. Ping, and P. Young. 2000. “Wide-areaTopographic Mapping and Applications Using Airborne Light Detection and Ranging (Lidar)Technology.” Photogrammetric Engineering and Remote Sensing 66 (8): 908–914.

Hollaus, M., G. Mandlburger, N. Pfeifer, and W. Mücke. 2010. “Land Cover Dependent Derivation ofDigital Surface Models from Airborne Laser Scanning Data.” International Archives ofPhotogrammetry, Remote Sensing and the Spatial Information Sciences XXXVIII (3A): 221–226.

Hollaus, M., W. Wagner, and K. Kraus. 2005. “Airborne Laser Scanning and Usefulness forHydrological Models.” Advances in Geosciences 5: 57–63. doi:10.5194/adgeo-5-57-2005.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

http://dx.doi.org/10.14358/PERS.76.9.1041http://www.asprs.org/a/society/committees/lidar/Downloads/Vertical_Accuracy_Reporting_for_Lidar_Data.pdfhttp://www.asprs.org/a/society/committees/lidar/Downloads/Vertical_Accuracy_Reporting_for_Lidar_Data.pdfhttp://dx.doi.org/10.1016/S0924-2716(99)00014-3http://dx.doi.org/10.1016/S0924-2716(99)00014-3http://dx.doi.org/10.1080/01431160701736513http://dx.doi.org/10.14358/PERS.73.4.385http://dx.doi.org/10.3390/rs3091902http://pdf.directindustry.com/pdf/optech/altm-gemini/25132-53143.htmlhttp://pdf.directindustry.com/pdf/optech/altm-gemini/25132-53143.htmlhttp://dx.doi.org/10.1515/jag.2007.017http://dx.doi.org/10.1088/0034-4885/76/8/086801http://dx.doi.org/10.14358/PERS.76.5.589http://db.cs.berkeley.edu/jmh/papers/cleaning-unece.pdfhttp://db.cs.berkeley.edu/jmh/papers/cleaning-unece.pdfhttp://dx.doi.org/10.5194/adgeo-5-57-2005

Huising, E. J., and L. M. Gomes Pereira. 1998. “Errors and Accuracy Estimates of Laser DataAcquired by Various Laser Scanning Systems for Topographic Applications.” ISPRS Journal ofPhotogrammetry and Remote Sensing 53 (5): 245–261. doi:10.1016/S0924-2716(98)00013-6.

Hyyppä, J., X. Yu, H. Hyyppä, and M. Maltamo. 2006. “Methods of Airborne Laser Scanning forForest Information Extraction.” In International EARSeL Workshop 3D Remote Sensing inForestry Proceedings, Vienna, Austria, February 14–15, 63–78.

Jacobsen, K. 2003. “DEM Generation from Satellite Data.” Paper presented in EARSeLWorkshop,Ghent, Belgium, June 5–7, 273–276.

Jacobsen, K. 2012. “Characteristics of Nearly World Wide Available Digital Height Models.” 10thSeminar on Remote Sensing and GIS Applications in Forest Engineering, Curitiba, Brazil,October 15-18, p. 8.

Jacobsen, K. 2013. “Characteristics of Worldwide and Nearly Worldwide Height Models.” ISPRSWG IV/2 Workshop, Interexpo Geo-Siberia, Novosibirsk, Russia, April 16-18, 58–78.

Lin, Q., J. F. Vesecky, and H. A. Zebker. 1994. “Comparison of Elevation Derived from Insar Datawith DEM over Large Relief Terrain.” International Journal of Remote Sensing 15 (9): 1775–1790. doi:10.1080/01431169408954208.

Linder, W. 2005. “Program Description LISA, Module Basic.” http://www.fergi.uni-osnabrueck.de/module/dim/inhalt/1/practical/lisa.pdf

Liu, X. 2008. “Airborne LiDAR for DEM Generation: Some Critical Issues.” Progress in PhysicalGeography 32: 31–49. doi:10.1177/0309133308089496.

Lohr, U. 1998. “Digital Elevation Models by Laser Scanning.” The Photogrammetric Record 16(91): 105–109. doi:10.1111/0031-868X.00117.

Maas, H. G. 2003. “Planimetric and Height Accuracy of Airborne Laserscanner Data: UserRequirements and System Performance.” Paper Published in Photogrammetric Week 2003,Edited by D. Fritsch, 117–125.

Mandlburger, G., C. Briese, and N. Pfeifer. 2007. “Progress in LiDAR Sensor Technology- Chanceand Challenge for DTM Generation and Data Administration.” Paper Presented at 51stPhotogrammetric Week, Stuttgart, Germany, September 3–7, 159–169.

Mandlburger, G., M. Pfennigbauer, M. Steinbacher, and N. Pfeifer. 2011. “Airborne HydrographicLidar Mapping -Potential of a New Technique for Capturing Shallow Water Bodies.” PaperPresented at 19th International Congress on Modelling and Simulation, Perth, Australia,December 12–16, 2416–2422.

McCoy, M. D., G. P. Asner, and M. W. Graves. 2011. “Airborne Lidar Survey of IrrigatedAgricultural Landscapes: An Application of the Slope Contrast Method.” Journal ofArchaeological Science 38 (9): 2141–2154. doi:10.1016/j.jas.2011.02.033.

McIntosh, K., A. Krupnik, and A. Schenk. 2000. “Improvement of Automatic DSM Generationover Urban Areas Using Airborne Laser Scanner Data.” International Archives ofPhotogrammetry and Remote Sensing XXXIII (B3): 563–570.

Mekik, C., and M. Arslanoglu. 2009. “Investigation on Accuracies of Real Time Kinematic GPSfor GIS Applications.” Remote Sensing 1 (1): 22–35. doi:10.3390/rs1010022.

NDEP (National Digital Elevation Program). 2004. “Guidelines for Digital Elevation Data (Version1.0).” http://www.ndep.gov/NDEP_Elevation_Guidelines_Ver1_10May2004.pdf

NGS (National Geodetic Survey). 2014. http://www.ngs.noaa.gov/GEOID/GEOID12A/Pourali, S., C. Arrowsmith, N. Chrisman, and A. Matkan. 2014. “Vertical Accuracy Assessment of

LiDAR Ground Points Using Minimum Distance Approach.” Proceedings of Research atLocate’14 in conjunction with Locate 2014 1142: 86–96.

Ressl, C., H. Kager, and G. Mandlburger. 2008. “Quality Checking of ALS Projects UsingStatistics of Strip Differences.” International Archives of Photogrammetry and RemoteSensing XXXVII (B3b): 253–260.

Ressl, C., G. Mandlburger, and N. Pfeifer. 2009. “Investigating Adjustment of Airborne LaserScanning Strips without Usage of GNSS/IMU Trajectory Data.” International Archives of thePhotogrammetry, Remote Sensing and Spatial Information Sciences XXXVIII (3/W8): 195–200.

Riegl. 2015. Datasheet Riegl VZ-400. http://www.riegl.com/uploads/tx_pxpriegldownloads/10_DataSheet_VZ-400_2014-09-19.pdf

Sefercik, U. G., and M. Alkan. 2009. “Advanced Analysis of Differences Between C and X BandsUsing SRTM Data for Mountainous Topography.” Journal of the Indian Society of RemoteSensing 37 (3): 335–349. doi:10.1007/s12524-009-0044-4.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

http://dx.doi.org/10.1016/S0924-2716(98)00013-6http://dx.doi.org/10.1080/01431169408954208http://www.fergi.uni-osnabrueck.de/module/dim/inhalt/1/practical/lisa.pdfhttp://www.fergi.uni-osnabrueck.de/module/dim/inhalt/1/practical/lisa.pdfhttp://dx.doi.org/10.1177/0309133308089496http://dx.doi.org/10.1111/0031-868X.00117http://dx.doi.org/10.1016/j.jas.2011.02.033http://dx.doi.org/10.3390/rs1010022http://www.ndep.gov/NDEP_Elevation_Guidelines_Ver1_10May2004.pdfhttp://www.ngs.noaa.gov/GEOID/GEOID12A/http://www.riegl.com/uploads/tx_pxpriegldownloads/10_DataSheet_VZ-400_2014-09-19.pdfhttp://www.riegl.com/uploads/tx_pxpriegldownloads/10_DataSheet_VZ-400_2014-09-19.pdfhttp://dx.doi.org/10.1007/s12524-009-0044-4

Shan, J., and S. Aparajithan. 2005. “Urban DEM Generation from Raw Lidar Data: A LabelingAlgorithm and Its Performance.” Photogrammetric Engineering & Remote Sensing 71 (2):217–226. doi:10.14358/PERS.71.2.217.

Trinder, J. C., and M. Salah. 2011. “Airborne Lidar as a Tool for Disaster Monitoring andManagement.” GeoInformation for Disaster Management, Antalya, Turkey, April 3-8, p. 6.

Wang, Z., R. Shu, W. Xu, H. Pu, and B. Yao. 2008. “Analysis and Recovery of Systematic Errorsin Airborne Laser System.” International Archives of the Photogrammetry, Remote Sensingand Spatial Information Sciences XXXVII (B1): 289–294.

Wheaton, J. M., J. Brasington, S. E. Darby, and D. A. Sear. 2010. “Accounting for Uncertainty inDEMs from Repeat Topographic Surveys: Improved Sediment Budgets.” Earth SurfaceProcesses and Landforms 35 (2): 136–156. doi:10.1002/esp.1886.

Yang, C. S., S. P. Kao, F. B. Lee, and P. S. Hung. 2004. “Twelve Different Interpolation Methods:A Case Study of Surfer 8.0.” XXth ISPRS Congress, Istanbul, Turkey, July 12-23, 778–785.


Dow

nloa

ded

by [

Bul

ent E

cevi

t Uni

vers

ity]

at 0

6:28

24

Nov

embe

r 20

15

http://dx.doi.org/10.14358/PERS.71.2.217http://dx.doi.org/10.1002/esp.1886

Abstract1. Introduction2. Study areas and instruments3. Methodology4. Results and discussion4.1. Results of ALS DSMs4.2. Results of ALS DEMs under high vegetation

5. ConclusionsAcknowledgementsDisclosure statementReferences

area-based quality control of airborne laser scanning 3d ...scanning 3d models for different land...

Documents