Académique Documents
Professionnel Documents
Culture Documents
ABSTRACT
Maps are made to scale. Scale represents the ratio of distance on the map to distance on a projected
coordinate system, i.e., Universal Transverse Mercator (UTM). This specification is, however, limited
by the fact that different lines produce different scales. This often has disadvantages for users who may
require an optimal scale; one scale representing the entire map. Scale variations clearly show that
measurements on maps are deteriorated by biases. It is, therefore, desirable to have a unique scale
independent of linear measurements to enhance the accuracy of further data processing. In this
contribution, an optimal scale based on the relationship between the areas on a map and a reference
ellipsoid is proposed to reduce the distortions of the projected coordinate system. The motivation behind
the area approach is the fact that as the number of lines approaches infinity on a map, a surface is built
which is accurately represented by an area as opposed to linear features. Using several map
projections, this paper demonstrates that linear-scale optimization is achieved through areal-scale.
Almost all of the commercial software measure the linear-scale based on one line. The linear scale
remains unchanged even if the projection is changed or map view moved. Therefore, this contribution
can pave the way for GIS industry to present a better indication of scale and more accurate data
processing results.
KEYWORDS: Scale, Areal-scale, Linear-scale, Map projection, Distortion, GIS
INTRODUCTION
Maps are representations of earth surface. This is generally made possible through the
use of scale, where a unit on a map represents several units on the ground. The
relationship between distances on a map and their corresponding values on the earth's
surface are normally used to define map scales [14]. In spatial sciences, scale is known
as the map abstraction level of spatial data, and is the main criteria for data
generalization [5]. Scales, therefore, play a vital role and should be as accurate as
possible to leave reliable information. An incorrect scale leads to misinterpretation of
data, which is undesirable to users. Apart from the actual use of representing positions
of the earth on maps, scales are also used in map projection selections [8], [12].
Currently, most operating spatial statistics software, e.g., WinBUGS [13] and
geoR [9] allow specification of only two-dimensional Euclidean coordinates (Map
projection coordinate systems). In ArcGIS for example, the distance between two cities
(e.g., Baghdad and Tabriz) could be measured with remarkable variation in various
map projections. For instance, the distance on WGS84-UTM-zone38N between the
two points is 553590.84m, while the same distance on the Asia-North-LambertConformal-Conic is 504337.13m, thereby, differing by about 50 km. Both Tabriz
(461724.43E, 38451.52N) and Baghdad (442352.21E, 33202.54N) are
located in one UTM zone and the used projections (LCC and UTM) are in the same
geodetic system (WGS84), hence, no influence of geodetic datum transformation in
this computation. The software computes the distances by using a projected coordinate
Contact: H Helali e-mail: hhelali@yahoo.com
2010 Survey Review Ltd
375
DOI 10.1179/003962610X12747001420627
system, the coordinate system and scale change by changing the map projection.
Obviously, this amount of variation increases the undesirable data processing errors.
In conventional topographic and thematic maps, scales specify the location accuracy
of details shown and the grayness or resolution of information. In contrast to
topographic maps, hardly any standards exist for thematic maps, which generally use
small-scale map projections. For digital representation of geo-data, many of the
analysis and transformation used in maps are directly related to the scale of maps. For
example, generalization of spatial data, measurements (e.g. areas) and automatic
selection of map projections, require prime parameters. It is in such digital
representation that the importance of accurately specifying scale comes into focus.
In the commercially available geo-information software such as ArcGIS, there is no
sensitivity to change in scale when moving the map (i.e., portrayal operations like pan)
or when changing map projections [4]. In these software, scale is recalculated just by
vertical movements of map (zoom in, zoom out, zoom all, and zoom to objects)
functions. In reality, however, scale changes when a line position, its direction or its
length on a digital map is changed. Another fact often ignored in most software is that
only one scale is adopted based on linear relationship between a distance on a map and
its equivalent on the earth. However, several lines (distances) exist on a map and each
of them would give a slightly different scale, depending on the distortions incurred
during map projections. This is clearly undesirable. It is in this regard that the present
contribution proposes an optimal scale based on the relationship between areas on
maps and their ground equivalent, here known as areal-scale.
The basic idea of such optimal scale is to get a unique scale on a map window (in
this paper map refers to view on a monitor). This unique scale is normally
estimated by mean linear-scale. The mean is computed through many lines which tend
towards infinity. This leads to the idea of using surface areas instead of lines to define
optimal scale.
The contribution is organized as follows: Discussion of linear-scale and its
governing parameters are presented first. It is demonstrated how the spatial scale is
achieved based on lines. The concept of areal-scale is then presented. A detailed
description of implementing the contribution is given. The relationship between linearscales and areal-scales for different map projections are established and the empirical
results are assessed in this paper.
MEAN (OPTIMAL) LINEAR-SCALE
Maps are produced based on scale. In each case, the scale represents the ratio of a
distance on the map to the actual distance on the ground. However, the particular
distance (line) must be clearly understood [6]. In practice, by changing the position,
direction or length of a line, the scale value ( Sl ) changes. This is due to the variation of
scale-factor on different places of a projection. Scale Sl for the line l , on a map is
given by:
S
d
D
(1)
where dl is the length of any straight line on the map window and Dl is the equivalent
distance on a projected coordinate system. The projected coordinate system is a plan
coordinate and already has the effect of projection distortion. In this study, linear-scale
is chosen such that it is independent of map projection distortion. To maintain a
distortion free scale, the measurement of Dl on an ellipsoid (rather than the projected
376
Mi
,
.
2
cos ij
3e sin 2 i
1
2
2
4(1 e sin i )
(2)
where M is the radius of curvature of the meridian (see, e.g., Eq.(9), is the geodetic
latitude, e is the eccentricity of the ellipsoid and ij is the geodetic azimuth of point i
to j . Since the geodetic azimuth is obtained from the geodesic, Eq. (2) is computed
iteratively given some starting values of the azimuth.
To obtain an optimum linear-scale on the map, the number of lines should be
increased such that the probability of reaching an optimum scale is enhanced. The
optimum value is thus considered as the average of all scales S computed by all
possible lines in Eq. (3).
l
S=
S
1
(3)
where l is the number of participating lines. To increase the number of lines, one needs
to know projection distortions on the map. It should be pointed out that there are many
parameters influencing the amount of map distortions, which ultimately lead to
distortion of the measured linear-scale. These include: projection type, map position,
map window shape and line specification on the map.
Map Projection
Measuring distances on projected coordinate systems are affected by map
projections. It always distorts distances and can influence statistical estimation. The
polygonal methods treat 'distances' informally (e.g., actual roadway distance or rail
track distance), rather than purely geometric [2]. According to Kennedy [6], scale is
not maintained correctly by any projection throughout an entire map. Regarding
projection distortions, scale variation is not always the same in different directions. As
Figure 1 shows, various lines may have different lengths on the ellipsoid, caused by
the projection.
a
Projection
Map window
C
Projected view
Ellipsoid
map and globe (at map scale) are the same, regardless of whether it is a great or small
circle or a straight or a curved line. Such distances are said to be true. For example, in
Sinusoidal projection, the equator and all parallels have their true lengths. In other
equidistant projections, the equator and all meridians are true. Distances are always
distorted, with its extent varying by type, but typically conformal projections distort
distances much more than equal-area projections [2]. Regarding the diversity of
applications and different map projections, various directions of measurements are
needed to compute the optimum scale, thus, rendering the application of these
projections to very limited specified cases. Because of the mentioned necessity, there
is a need to present an optimum scale such that it is independent of distortion as much
as possible.
Map position and extension
Scale-factor of a map projection is not the same throughout the map; the scale of a
map is affected by its position. As the map position is placed around the standard lines
(e.g. parallels and meridians), map distortion is less and scale is more accurate than
when the map position is far from the standard lines.
Map Window Shape
Map window (framework) may have different shapes (e.g. rectangle, circle, etc.).
Each of them has different specifications, which make them suitable for particular
projections. To keep the vertical and horizontal symmetry, this study considers a
window, which is set to squares instead of rectangles.
Line specification on a map
In addition to different distortion propagation on position [3], there exist different
linear-scales for any change in position, length, and orientation of lines. The scales for
diagonal lines for example are different from mid-horizontal or vertical ones. In some
cases, when there is a true line to which linear-scale is referred, it turns out not to be
applicable to all map projections, since such lines do not exist in all maps. This is the
main reason why there are different linear-scales for a given map in commercial
software.
Optimal Linear-scale
An approach that can be used to obtain mean linear-scale (optimal scale) is
presented in this paper. This methodology will be used to develop the proposed arealscale. Line design is used in deciding which lines should be used in arithmetic mean
linear-scale ( S ) calculation. Since infinite number of lines exists, a statistical sampling
approach is used to choose lines to cover the entire map area. Therefore, the
influencing parameters (i.e., direction, position or length) are considered properly in
this process.
Distortion patterns differ from one projection to another. It should, therefore, be
considered for a variety of the projections. In other words, one requires to know the
behaviour of S over l . To select lines as statistical observations to measure scale, and
then assign the average of linear-scale to the map scale, the issues of positions,
directions and lengths are considered in the following ways:
The position of the selected lines are symmetric
Lines are homogeneously distributed on the map
Lines cover various directions
Lines have different lengths
378
Number of lines is optimized, so that the computation times are minimal while the
scale shows the trend.
No single line meets all the conditions above for all map projections. In order to
distinguish the effect of increasing the number of lines, the effect of distortion on a
specified direction, position or length should be avoided. A grid template of points is
used to generate homogeneous distribution of lines in direction, position and length.
The number of possible unordered lines by using these grids is considered as a
combination of 2 points out of total grid points, n Ck in Eq. (4)as
n
n!
,
l = n C 2 = =
2 2!( n 2)!
(4)
where n is the total number of grid points (e.g., for a grid of 2 x 2, n = 4 ). Grids of 2 x
2, 3 x 3, 4 x 4, 5 x 5, 6 x 6 and 7 x 7 homogeneous points will have corresponding
values of combination l as 6, 36,120, 300, 630 and 1176 lines respectively. These
grids have been examined to observe the behaviour of the linear-scale. Such unordered
lines, influence the behaviour of mean linear-scale in Eq.(3). To avoid such influence
and to stabilize the linear-scale using minimum number of lines, a homogenous order
of sampling is considered. The sampling order is direction, position and length (i.e., a
vertical line is followed by a horizontal one; a line on the bottom of a view is chosen
after the top line; a short line is followed by a long one, etc). Figure 2 illustrates the
primary orders of lines for a grid of 5 x 5 points. The final coverage of this grid (300
lines) is also depicted in Figure 2.
...
b) Position
a) Direction
...
e) Direction
...
d) Position
c) Length
...
f) Position
...
Figure 2. Primary line order of homogenous sampling for a 5 x 5 point grid on a map
Sa =
a
A
(5)
Using two flat rectangles, areal-scale can be derived from Eq.(6). By splitting the
area to very small cells (Figure 3), rectangles on the earth can be considered to
approximate flat surfaces. For these infinitesimal flat rectangles, scale in all directions
is assumed similar. Linear-scale for each pair of rectangle is then given by:
di
Di
Si =
(6)
x
y
i = i
X i Yi
xi y i
=
X i Yi
ai
Ai
= S ai ,
where for a rectangular map i , d i is the distance, xi and y i are the dimensions of the
map and ai is its area. Di , X i , Yi and Ai are the corresponding values of other infinite
small rectangles on an ellipsoid. The independent mathematical proof of Eq. (6) can be
found in [11].
Regarding the effect of different map projections, the boundary lines ( AB , BC , CD
and DA in Figure 1) are neither geodesic nor graticules (same longitude and latitude).
In general, it will not be displayed as a rectangular area (most projections do not depict
parallels and meridians as perpendicular straight lines). In other words, the map
window will not correspond to the rectangular area seen on the display. For such a
complex outline of a map window on an ellipsoid, it is hard to compute the area in a
closed form. In addition, other methods like Crofton formula series used for computing
areas are tedious and time-consuming [7]. The challenging part of the areal-scale
method is to compute the map equivalent area on the ellipsoid.
To compute the ellipsoidal area, the shape has been divided into 100 x 100 small
cells (10,000) Ai . Figure 3 shows equivalent cells on the map and ellipsoid.
a i Ai
Map
Ellipsoidal
Figure 3. Dividing the ellipsoidal area to small rectangles equivalent to those on map
Each cell ( Ai ) is considered as two triangles (see, Figure 4). Each triangle is
approximated by a spherical triangle to develop a closed formula for areal
computation.
380
1
Ai1
3
p3
p2
The radius of curvature on the sphere is calculated for all grid points. For each
triangle the mean of radii in three vertices is used as the radius of the equivalent
ellipsoidal radius. The radius is used in Eq. (7), to calculate the area of each spherical
triangle [10]:
^
(7)
Ai1 = Ri12 (1 + 2 + 3 ),
where 1 , 2 and 3 are the space angles of spherical triangle and Ri1 is the equivalent
spherical radius. For each vertex, Eulers radius is the ellipsoid radius at the point and
is given by [10]:
(8)
R = MN ,
a (1 e 2 )
(1 e 2 sin 2 )
,
2
a
(1 e 2 sin 2 )
(9)
,
and is the geodetic latitude at point, and e is the eccentricity of the ellipsoid.
The space angles are calculated through [1]:
Cos ij = cos i cos j cos( i j ) + sin i sin j ,
(10)
where ( i , i ) and ( j , j ) are the spherical coordinate of points i and j . The area of Ai
is finally given by the sum of the areas of the two triangles. The areal-scale, S a in Eq
(5), can then be computed. Eq. (8) applies to the map projections based on the
ellipsoidal datum. For those map projections based on sphere, Eqs. (8) and (9) are
immaterial.
TEST SPECIFICATIONS
In order to test the proposed areal-scale, a state-of-the-art software environment of
ESRI ArcObjects (ArcGIS 9) is used. The software is further developed by Visual
Basics. The selection of ArcObjects is based on its variety of predefined functionality;
numerous well-known map projections, accessibility and the familiarity by the authors.
In developing the software to investigate the relationship between the mean linearscale and areal-scale, the following options are made:
controlling the number of lines and grid density
generating a linear-scale list for all lines
changing map projections
381
25,001
Scale
25,000
24,999
24,998
24,997
Mean Linear-Scale
Areal-Scale
24,996
24,995
1
51
101
151
201
251
301
351
401
451
501
551
601
No. of Lines
383
Scale
250,000
249,990
249,980
249,970
Mean Linear-Scale
249,960
Areal-Scale
249,950
1
51
101
151
201
251
301
351
401
451
501
551
601
No. of Lines
Series
7.5 minute
Max changes
on units of
scales (mean
linear and
areal) per 10
cm pan on map
1
UTM
UTM
1:49,735
1:99,471
2
4
1 degree by
2 degrees or
3 degrees
1:250,000
UTM
12
State maps
1:500,000
LCC
22
Equidistant
In Equidistant Conic projection, correct scale is true along the meridians and the
standard parallels. Scale is constant along any given parallel, but it changes from one
parallel to another. Equidistant projections are common for atlas at medium and smallscale maps of small countries (e.g., those used by the former Soviet Union) [6]. Figure
8 shows an Equidistant Conic projection at scale around 1:52M. The central meridian
is 60 W, the first and second standard parallels are 5 S and 42 S respectively, and
the latitude of the origin is 32 S. It is used for regional mapping of mid-latitude areas
with a predominantly eastwest extent [6].
384
Millions
51.5
Scale
51.0
Mean Linear-Scale
50.5
Areal-Scale
50.0
49.5
49.0
1
51
101
151
201
251
301
351
401
451
501
551
601
No. of Lines
Figure 8. Trend of mean linear-scale for Equidistance Conic projection at scale about 1:50000000
Estimating scale numbers are very sensitive to the number of lines used. The
computation gets stable if the number of lines gets greater than 50 as shown in Figure
8. The behaviour of mean linear-scale varies depending on the map window position.
The position of map window also influences the magnitude of areal-scale. Figure 9
shows this variation for Sinusoidal projection at scale around 1:100M. This projection
is pseudocylindrical in that areas are represented accurately. The scales along all
parallels and the central meridian of the projection are accurate. This projection is used
for continental maps of South America, Africa, and occasionally for the others, where
each land mass has its own central meridian. Further applications are discussed in [6].
Millions
112
Projection: Sinusoidal
view window position:
up left latitude: 63.4775
down left latitude: -8.0971
down left longitude: -8.4358
down right longitude: 63.5766
110
108
Scale
106
Mean Linear-Scale
104
Areal-Scale
102
100
98
96
1
51
101
151
201
251
301
351
401
451
501
551
601
No. of Lines
Figure 9. Trend of mean linear-scale for Sinusoidal projection at scale around 1:100M
the equator. The scale is also constant along any given latitude and symmetrical around
the equator.
Although the distortion of distances in this projection is more than that of equaldistance projection, the linear-scale stabilizes rapidly (i.e., it needs 100 lines in
comparison to 251 for equal-distance projection). This is due to the fact that in equalarea projection, distance distortion is less compared to the conformal and equaldistance projection. Position of the map window and scale also affect the stabilization.
However, the magnitude of these effects are not as those of equal-distance projections.
As Figs. 8, 9 and 10 show, the variation of the linear-scale is too high at smaller scales.
Moving on the map (at these scales) may result in changing the scale by up to 5%. This
magnitude of change in scale is not so crucial for such maps. Maps at such scales
(global maps) are not suitable for accurate measurement.
Millions
190
180
170
Scale
160
150
140
Mean Linear-Scale
Areal-Scale
130
120
110
100
1
51
101
151
201
251
451
501
551
601
Figure 10. Trend of mean linear-scale for Quartic-Authalic projection at scale around 1:150M
In summary, in all cases considered above, (conformal, equal-area and equaldistance projections), the optimum value of the mean linear-scale coincides with the
areal-scale. This, therefore, implies that, instead of laboring to obtain the optimum
linear-scale; mean linear-scale, areal-scale is recommended.
CONCLUSIONS
This paper illustrates that computing scales based on measuring a single distance do
not produce the optimum value. Variation of linear-scale is the result of projection
distortion throughout the map. The empirical results indicate that when the number of
lines in estimating mean linear-scale increase, the magnitude of linear-scale
approaches areal-scale, which shows that the areal-scale offers an optimum equivalent
of the mean linear-scale on a map.
The paper, therefore, proposes an accurate method of computing areal-scale for the
entire earth and demonstrates the following advantages of areal scale over linear scale:
i. Areal-scale offers one optimal value of scale, which is desirable for any
measurement and computation on a map. Implementing an areal-scale in a GIS
and mapping software can, therefore, enhance the accuracy of data processing.
ii. Areal scale provides a unique value for a map window. The values of linearscale, depends on directions, lengths and positions of the selected distances.
386
The study has also demonstrated that by changing the map projection, the map scale
(mean liner-scale and areal-scale) changes. Changes in map scales are also noted to be
influenced by window motions (e.g., pan on a map). It should be pointed out that
calculating areas on ellipsoid for different maps is rather complicated and requires
further investigation.
ACKNOWLEDGEMENT
The first author wishes to acknowledge the financial support of University of Tabriz
Research Affairs for their funding. The second author acknowledges the support of
Curtin Research Fellowship.
References
1. Awange, J.L. and Grafarend E.W., 2005. Solving Algebraic Computational Problems
in Geodesy and Geoinformatics: The Answer to Modern Challenges, Springer Verlag,
p177.
2. Banerjee, S., 2005. On Geodetic Distance Computations in Spatial Modelling. Journal
of Biometrics Volume 61 Issue 2, pp. 617625.
3. Canters, F., 2002. Small-scale Map Projection Design. Taylor & Francis, London and
New York. p.44.
4. Helali, H., Alesheikh, A.A.and Voosoghi, B., 2005. Classification and Modeling Multi
Map Projection Parameters in Spatial Database, University of Tehran Journal of
Science, Vol 39, No. 5, pp 689-699.
5. Joao, E. M., 1998. Causes and Consequences of Map Generalisation, Taylor &
Francis. p.2.
6. Kennedy, M., 2000. Understanding Map Projections, Environmental Systems
Research Institute. ESRI Inc., ArcInfo 8,
http://web.clas.ufl.edu/users/jsloan/web/sloan_understanding_map_projections.pdf
7. Kubnov, L.Jancek, J., Guilak, F. and Opatrny, Z., 1999. Comparison of Several
Digital and Stereological Methods for Estimating Surface Area and Volume of Cells
Studied by Confocal Microscopy. Journal of Cytometry, Volume 36, Issue 2, pp. 85
95.
8. Maling, D.H., 1993. Coordinate Systems and Map Projections. Second Edition.
Oxford: Pergamon Press, 476p.
9. Ribeiro, P. J. and Diggle, P. J., 2003. The GeoR Package. Available at
http://www.est.ufpr.br/geoR.
10. Richardus, P. and Adler, R. K. 1972. Map Projections for Geodesists, Cartographers
and Geographers. North-Holland Pub. Co., Amsterdam. pp 99-100.
11. Rudin, W., 1976. Principles of Mathematical Analysis. 3rd edition, McGraw-Hill,
New York, Chapter 4, No. 14a.
12. Snyder, J.P., 1987. Map Projections: A Working Manual. USGS Professional Paper
1395. Washington, D.C. USGS.
13. Thomas, A., Best, N., Arnold, R., and Spiegelhalter, D., 2002. The GeoBUGS User
Manual. Available at http://www.mrc-bsu.cam.ac.uk/bugs.
14. USGS, U.S. Geological Survey, 1995. USGS Maps, U.S. Department of the Interior,
VA 20192, USA, Online Edition booklet at URL:
http://mac.usgs.gov/isb/pubs/booklets/usgsmaps/usgsmaps.html.
15. Vanicek, P. and Krakiwsky, E.W., 1986. Geodesy the Concepts (second edition),
Elsevier, Amsterdam. pp.40, 354.
387