DEM Generation by Various Methods: A Comparative Study

M. Tech Seminar

By Ajay Kumar(133040061)

Overview 1. What is DEM2. Data structure in DEM3. Data source for DEM4. Spot heights5. Case study6. Stereo imagery7. Height measurement8. Stereoplotters9. Very high resolution satellite10.Radar 11.Lidar 12.Summery

What is DEM ?Digital elevation model (DEM) is a digital representation 3-dimensional information (X, Y,

Z) of the continuous topography of the bare earth surface in a particular reference

coordinate system.

DTM ( Include all terrain, geological,

climatic, geomorphology, climatology, meteorology, and

oceanology factors)

DSM ( Include terrain & terrain features like natural features & man made

features.

DEM ( Include only bare

terrain)

Continued.. Initially elevation models were physical models made of rubber, plastics, clay,

sand etc.

Roberts was the first to propose DEM and Millar and laflamme of MIT described

the development in detail.

They extracted the road profile from the stereo photographs and displayed it

digitally in computer to assist road design.

DEM development

Data acquisition

Data management & manipulation

Computation and modelling

Applications

Initially photogrammetrists & civil engineers are involved

Scientist from computation geometry & applied mathematics involved in algorithm

development

Scientist from computation technology are involved in data management & system

development

Now specialists from various geo-disciplines are involved in the applications

Data Structure for DEM

Grid structure

There are two main data structure in which DEM data can be stored:

TIN structure (Triangulated Irregular network)

http://www.ncgia.ucsb.edu/cctp/units/unit06/06_f.html

A. Only elevation (Z) at each

node of grid is recorded.

B. All undulations of terrain can’t

be covered in a cell size of

grid.

C. Very easy to retrieve, analyze

& manipulate data for

algorithms or interpolation on

data.

D. Redundancy in data.

E. Surface generated appear

more natural.

A. X, Y, Z at surface specific points of

terrain is recorded.

B. Represent more true surface.

C. Developing algorithm or applying

mathematical model for TIN data.

D. Only surface specific point is

recorded hence no redundancy in

data.

E. Does not appear natural due to

edge of triangle

Grid

TIN

Data Source for DEM Generation

DEM Data Collection Methods

Spot Heights

Stereo Images

Active Remote Sensing Methods

Radar

Lidar

Various methods for collecting DEM data can be grouped as:

Spot Heights

This include all methods in which X, Y, Z coordinate of a point can be found e.g.

digital levels, theodolite, total station, global positioning system (GPS) etc..

Data can collected in form of grid or TIN, better option is TIN as less no. of points

needs to be recorded & later TIN data can be converted into grid data for

analysis purpose.

These are good & cheaper tools to obtain the point data to create highly

accurate DEM for small areas.

Topographic map generally prepared by these methods is also a good and

cheaper source for DEM generation.

Case studyThis case study is about creating DEM from topographic map:

Study area is in east of Mexico City & highly undulated. Area is 53.8 ×36 km2 (855 lines,

1275 columns), map scale is 1:250,000 & contour interval is 100m.

The data is obtained by scanning and labeling contours.

Accuracy of DEM entirely depends on contour’s accuracy.

Methodology :

1. Connectivity: The relationships between the pixels

describing a contour are determined by their connectivity.

a. Four connected- contour is a curve formed by pixels linked only by their edges.

b. Eight connected- curve formed by straight segments of 1 to n pixels linked only by

their corners.

Four connected will give larger distances & eight connected give smaller distances.Image source - http://www.imageprocessingplace.com/ downloads_V3/root_downloads/tutorials/contour_tracing_Abeer_George_Ghuneim/8con.html

Four connected Eight connected

Continued ….2. Dilation:

Dilation is applied in order to expand the surface that describes lines in the raster mode.

New contours are obtained by an iterative dilation applied to the earlier contour lines that

increase their surfaces on both sides, until they become contiguous.

3. Boundary extraction: The extraction of a limit between two zones can be obtained as a new contour whose

altitude is the value intermediate between the two altitude values of the given surfaces.

Result: The resulting DEM generation by dilation (855 lines, 1275 columns) is generated in 13

min and 26 iterations by using a PC Pentium W/MMX 233 MHz, and in 10 min by the BDM

method.

Borgefors distance measurements (BDM) is used in GRASS & ILWIS GIS software.

The computational time of the DEM generation by dilation can be considerably reduced to

less than 10 min by stopping the iteration procedure when there are no new significant

contours generated.

DEM from Stereo ImagesWhen two images are captured from different locations (for same area) then in the

overlapped area can be seen in 3D and X, Y, Z for any point can be measured.

Stereo images can be aerial, satellite or radar images.

Images taken from aerial platform have good resolution but less coverage area.

Satellite stereo image can be acquired either along path or across path of satellite orbit.

Across path stereo images are obtained after revisit period and hence environmental

conditions may not be same.

VHR satellite generally capture in pan-band and multispectral band of visible region.

Better spatial resolution may reduce the error.

History of Stereoscopy

Jacopo Chimenti (c 1551 - 1640), an

artist from Empoli, made two

sketches of a young man holding a

compass and a plumb line. When

these were seen, mounted next to

one another, by Alexander Crum

Brown in 1859, he combined them

by over convergence and described

the stereoscopic depth he saw.

Digital elevation model can be

generated if we can extract X, Y, Z

for each point from a image. Aime Laussedat (1819-1907) is regarded by many as the “father of photogrammetry.” due his pioneering work in photogrammetry.

Height measurements 1. Relief measurements

h =

d – relief displacement r – radial distance

2. Parallax measurement

h = H -

Where,

B - Airbase,

f - focal length,

H - height above datum,

ha - height of terrain above datum

Triangulation To determine the horizontal location of points aerial triangulation is used:

1. Aerial triangulation: Location of point on image is determined by the principle of intersection and resection. Exterior orientation elements are determined with help of GCP’s with known horizontal,

vertical locations lies in overlapping region. Tie points are measured at the top and bottom edges of each image where strips overlap

and thus many of these will appear on 4 to 6 images.

2. Aerial triangulation by digital correlation:Sophisticated Aerial triangulation software packages available that can simultaneously determine the ground co-ordinates of all measured points and exterior orientation elements using the method of least squares adjustment with the help of sensor's geometric model:

Feature extraction in every digital image Pairwise feature matching and computing the correlation coefficient. Localization and gross errors elimination by means of an affine

transform(preserve points plane and parallelism of lines)

University of Stuttgart (1995) the first develop Silicon Graphics Indigo2 software

package:

Stereoplotters

Stereoplotters are the instruments used to extract the 3-D topography

from stereo images. types of stereoplotters available:

Optical

mechanical

Optical-mechanical

Analytical

Digital photogrammetric system

Automated stereoplotters

Stereoplotters basics In Stereoplotters 3 orientations helps to create the similar model of

ray as it was the time of photo taken:

1. Interior - To fulfill the colinearity condition of the rays.

2. Relative – To fulfill the coplanarity condition, it involve

3 translational (X, Y, Z) & 3 rotational (ω , , k) movements.

3. Absolute - In this orientation scaling and levelling is

ensured.

f

Zhilin 2004

Direct Linear Transformation model

x = −f

y =

XS, YS, ZS- set of ground coordinates of projection

center S, XA, YA, ZA -set of ground coordinates of point

in geodetic coordinate system.

ai , bi , and ci (i = 1, 2, 3) are the functions of the three angular

orientation elements (i.e., φ, ω, κ) .

a1 = cos φ cos κ + sin φ sin ω sin κb1 = cos φ sin κ + sin φ sin ω cos κc1 = sin φ cos ωa2 = −cos ω sin κb2 = cos ω cos κc2 = sin ωa3 = sin φ cos κ + cos φ sin ω sin κb3 = sin φ sin κ − cos φ sin ω cos κc3 = cos φ cos ω

Similarly other models also there like Rigorous Sensor Model (RSM), Rational Functional Model, Self Calibration Direct Linear Transformation (SDLT).

ID Satellite/Sensor

Country/Comp any

Date of lunch Resolution

(stereo image) Min/Max (m)

Swath Width Min/Max

(km) Stereo B/H

1 IKONOS 2 USA/GeoEye 24 Sep 1999

Pan (N) 0.8 Multi (N) 3.2

Pansharpened 0.8-1.0 11*11 Along -track 0.54-0.83

2 EROS -A1 Israel/ImageSat 5 Dec 2000

Pan (N) 1.9 14*14 Along –track Across-track

variable

3 QuickBird USA/Digital Globe 18 Oct 2001

Pan 0.61 Multi 2.4

16.5*16.5 Along -track

0-6 to 2.0 most collections

between 0.9 and 1.2.

4 Spot 5 France /Spot

image 4 May 2002

Pan 2.5-5 Multi 10

60*60 Along- track Across-track

Variable

6 CartoSat -1 India 5 May 2005

Pan 2.5 26*26 Along-track 0.62

7 ALOS(PRISM) Japan 24 jan 2006

Pan 2.5 35*35 Along-track triplet

of images 1

7 EROS –B1 Israel/ImageSat 25 Apr 2006

Pan 0.7 7*up to 21 Along –track Across-track

Variable

8 KOMPOSAT 2 Korea /KARI 28 July 2006

Pan 1 Multi 4 15 Across –track Variable

9 WorldView-1 USA/Digital Globe 18 Sep 2007

Pan (N)0.5 (20º off-N) 0.55

17.6*17.6 Along-track Variable

10 WorldView-2 USA/Digital Globe 8 Oct 2009

Pan 0.46 20º off-N) 0.52

48*110 Along -track Variable

11 GeoEye-1 USA/GeoEye 6 Sep 2008

Pan (N) 0.5 Multi(N) 2

Pansharpened

15.2 area 224*28

Along-track Variable

Very high resolution satellite also provide stereo-image [Deilami 2011]

DEM Product from Stereo satellite images

Product Positional accuracy

 

CE90 LE90 SCALEGeoStereo

(0.5m) 4m 6m 1:5000

GCPs 2m 3m 1:2,500 Geostereo

(1m) 15m 22m 1:20.000

GCPs 4m 6m 1:5000

Accuracy GeoEye products [Deilami et al. 2011]

Cartosat 1C/1D- 3.01m error on relative scale

with standard deviation of 3.83m in less than

100m elevation and -2.30m error with standard

deviation of 6.25m between 100-300 m elevations

Spot 5- Orthoimages led to absolute accuracy of

terrain heights in the order of 5 to 10 m height

variation, with standard deviations of about 2 to 4

m for single points and 4 to 7 m for the

interpolated DSM in comparison to the reference

[Deilami et al. 2011].

IKONOS 2

IDImage

sConvergence

Angle (di)δx δy δz

1 2 and 3 12.311° 0.511 1.506 2.949

2 1 and 2 15.210° 0.568 1.454 2.362

3 3 and 4 17.691° 0.768 1.151 2.321

4 1 and 3 27.521° 0.445 1.392 1.895

5 2 and 4 30.002° 0.554 0.980 1.502

6 1 and 4 45.212° 0.525 1.055 0.691

ID ImagesConvergence Angle (di)

δx δy δz

1 1 and 5 15.635° 0.6181.17

31.895

2 2 and 5 18.018° 0.4461.18

61.908

3 3 and 6 18.954° 0.6820.75

72.231

4 3 and 5 26.906° 0.7990.94

61.855

5 1 and 6 39.484° 0.6570.39

51.288

6 4 and 5 42.576° 0.6990.82

71.038

Rongxing et al. (2009)

Along track RMSE Across track RMSE

Radar

Unlike optical and infrared imaging sensors, imaging radar is able to take clear pictures

during day and night under all weather conditions.

Active imaging radar (microwave region λ >1mm) is used to obtain the data for DEM,

there are two main methods for collecting data for DEM.

1. SAR (synthetic radar aperture)

2. InSAR (interferometric SAR)

1. SAR- SAR, synthesize the antenna length using Doppler shift

 of the received frequency from that of the transmitted

frequency due to flight motion.ωv = λ /w (angular fields in the cross flying direction

)

ωh = λ /L (angular fields in the flying direction )

WG ≈ (swath width) ∆x = (azimuth resolution flying direction) R = cτp/2 (slant range)

∆ y = ∆ R/(sin θi) = cτp/(2 sin θi)

Radar continued.. 2. Interferometric SAR:

Whereas SAR uses a single antenna, InSAR requires two antennas separated by

a baseline.

Signals from both antennas are recorded and processed to yield two complex

SAR images of the same scene.

Phases measured in each of the scenes are differenced on a pixel-by-pixel basis

to obtain additional geometrical information about the scene.

φ = φtopo + φscat + φdisp + φatm

R2 2 = R1 2 + B2 + 2 R1× B×sin(α−θ)Z = H – R1 cos θPhase change φ = δ

Radar satellite1. Shuttle Radar Topography Mission (SRTM)

It has been observed that C-band penetrate only quarter of canopy height. SRTM

project team have shown that the absolute vertical error of 5m with the most reliable

estimate.

an absolute horizontal circular accuracy of less than 20m and absolute and relative vertical

accuracy is less than 16m and 10m respectively [Tighe et al. 2009].

2. TerraSAR-X and TanDEM-X

TerraSAR-X and TanDEM-X are two radar Earth observation satellite systems by German

Aerospace Center (DLR) and EADS Astrium.

TerraSAR-X was launched on June 15, 2007 with its active phased array X-band SAR antenna

(wavelength 31 mm, frequency 9.6 GHz), circling Earth in a polar orbit at 514 km altitude.

It has 1 m spatial resolution, quick site access time of 2.5 days max. (2 days at 95%

probability) to any point on Earth, unique agility (rapid switches between imaging modes and

polarizations).

23,000 data sets from the first acquisition phase have been evaluated and visualized. About

25.9% of all DEMs are predicted to have a relative height error of better than 1.8 m.

Lidar

Laser scanning has triggered off a revolution in topographic terrain capturing, especially in the generation of

digital terrain models (DTM).

Laser scanner send pulse in green wavelength and record time taken by the pulse to reach backs or phase

difference is recorded to determine position of ground point.

Integration laser scanner with DGPS (Differential GPS) and IMU (accelerometers and

gyroscopes), can produce highly accurate & georeferenced DEM.

Due to small range, coverage area is less.

Applications:

Meteorology and atmospheric environment

Updating and Creating Flood Insurance Rate Maps

Coastal Change Mapping

Archaeology mapping

Forest and Tree Studies[Carter 2012]

Summary To generate DEM from contour map one approach called dilation algorithm is discussed.

This method has accuracy less than contour plot accuracy. It can be useful where high

accuracy of DEM is not required like environmental modelling.

Satellite imagery method is most popular due its high resolution images and its

characteristics like near real time data, temporal data, very good accuracy on horizontal

and vertical measurements etc. Stereoplotters are the instrument which are used to extract

the X, Y, Z of features for DEM. Fully automated stereoplotters are also available which

extract the required information very effectively and accurately.

Radar is good for accuracy on vertical scale measurement but poor in horizontal scale

measurement. Data processing is costly. Accuracy of DEM is less than 2m on relative

vertical scale in case of WorldDEM by TerraSAR.

Laser scanner can produce very high accuracy (1cm) DEM but accuracy decreases as

distance between scanner and target increases. Range up to which scanner can operate is

generally less than 500m. Laser scanner is generally terrestrial or mounted on aerial

platform.

Thank you