DRAINAGE BASIN AXIAL AND SHAPE PARAMETERS FROM

MOMENT MEASURES

E. D. ONGLEY
Queen’s University, Kingston

INAN EARLIER PUBLICATION^ the present author demonstrated the need for an ob-
jective working definition of the drainage basin axis. Measures used by Schumm2
and Maxwell? are demonstrably subjective and confusing when a basin departs
greatly from symmetry. Measures of basin shape and relief derived from axial
properties such as Schumm’s relief ratio and elongation ratio also suffer from
inherent subjective axial criteria.
Morisawa4 has shown that Schumm’s elongation ratio and Miller’s5 circularity
ratio are the two most successful shape measures in correlation procedures with
runoff data. However, Miller’s circularity ratio, although it is completely objective,
suffers from indeterminacy of perimeter length due to perimeter crenulations.
A measure of network symmetry suggested by Schumm and HadleyO calls for a
ratio of the lateral distances from the main channel to adjacent basin divides. This
measure, like other measures mentioned above, suffers from a lack of objectivity.
Ongley’ showed that the basin axis could be defined objectively by locating a basin
vectorial axis. The axis was collinear with the resultant vector on a vector diagram
made up of vectorial equivalents of all highest and second-highest Strahler segments
(Fig. 1 ~ ) It. was anticipated that the magnitude of the resultant vector should
relate to the shape of basin.
Abrahams7 has verified the author’s suspicions that the vectorial axes of real
basins frequently fall outside of the basin when great asymmetry exists (Fig. 1 ~ ) .
A further theoretical consideration renders the method less than satisfactory. This
arises from the fact that each Strahler segment of whatever order was given equal
weight in computing the resultant vector. Clearly, from hydrologic principles, a
weighting should be given to successively higher-order segments.
In attempting to find a simple solution to the problems inherent in the vectorial
axial method, an alternative procedure became apparent which is hydrologically
more satisfactory. In addition to the definition of an objective axial property, the
alternative procedure provides direct measures of basin shape and symmetry. This
is accomplished by reduction of the basin network to a simple frequency distribution
which may be described by standard statistical techniques.
Measures of drainage basin shape are potential indicators of basin runoff charac-
teristics. Inasmuch as channels develop in accordance with Schumm’s constant of
channel maintenance, basin shape is dependent upon the development and distribu-
tion of- network members. Measures of basin shape using areas or perimeters are
therefore indirect measures of terrain runoff characteristics. A direct measure con-
sists of evaluating the distribution of network members. The network pattern is
38
CANADIAN
GEOGRAPHER,
XIV, 1, 1970
 DRAINAGE BASIN PARAMETERS 39

FIGURE 1. A. Procedure for finding basin vectorial axis (after Ongley, 1968). B. Invalid basin
vectorial axis resulting from asymmetry of network.

reducible to a frequency distribution by joining all sources (pendant verticess) to

the mouth (Fig. 2). All lines are assumed coplanar and are vectorial equivalents
of flow paths from source to mouth. This procedure provides an inherent weighting
for higher segment orders. The azimuth and length of each vectorial equivalent are
read to the nearest degree and 0.01 inch respectively. Because the vectorial equiv-
alent of all sources is measured, the data represent a population rather than a
sample. The azimuth and frequency data may now be analysed as a grouped fre-
quency distribution, that is to say, grouped into one-degree classes. A simple com-
puter program (available from the writer), written in Fortran IV for the Queen’s
University IBM 360-50 computer, gives moment measures and values for the
weighted mean azimuth, standard deviation, skewness, and kurtosis. Geometrically
equivalent networks differing only a scale have identical moment measures and
derivative descriptive statistics. Scale, therefore, is of no consequence; length
values may be used directly and need not be transformed into percentage values.
Because confidence in moment measures decreases with decreasing population
size, Abrahams’? geometric axis is the most useful alternative where there are too
few sources to yield reliable moment measures, as for example in a first-order basin.
Orientation data (described by an angle or its supplement and therefore not a
vector) allow a certain freedom in choice of origin, that is, the point at which the
40 LE GB~GRAPHE CANADIEN

FIGURE
2. Frequency distribu-
tion of a network; azimuth and
lengths of vectorial equivalents
from sources to mouth.

circular distribution is divided to produce a linear function. This choice is made pos-
sible because a measure of orientation, for example 90", can also be arrayed as its
supplement, 270". Any change of origin results in different moment measure values
for the distribution as a whole (for a discussion see Jizba". Because of the directed
property of vectors, no portion of the data can be rearranged by the use of supple-
mentary angles; consequently moment measures are immutable.
It must be made clear that the azimuth and length data, though in fact part of a
circular distribution, are treated as a linear function. For this reason an enumeration
of azimuths must be continuous and cannot pass through the 360" mark. The
nature of the descriptive statistics of circular distributions mitigates against the use
of circular normal procedures. Whereas the arithmetic mean of the linear distribu-
tion is approximated by the azimuth of the resultant vector (vector mean), there is
no comparable measure of dispersion of the distribution amenable to standard
statistical procedures.1° Agterberg and Briggs'l found that a normal approximation
can be applied when 94 per cent of the circular distribution falls within i57" of
the vector mean. Furthermore, these two authors found no significant difference
between linear normal and circular normal procedures when the standard deviation
was less than 3 1O .
From a sample of 68 networks ranging from third to fifth order, taken from the
American Southwest and from the Western Division of the Australian state of New
South Wales, it has been found that there exists a probability of 0.088 that a net-
work will exceed the limits of the normal approximation for circular distributions.
Beyond these limits, normal descriptive statistics are useful parameter indicators,
providing that no statistical tests or inferences are drawn from these measures.
Basin moment axis. A line through the basin mouth with weighted mean azi-
muth, (x) as shown in Figure 3, is called the basin moment axis.
Basin shape. The property of shape involves the concepts of both elongation and
 DRAINAGE BASIN PARAMETERS 41

FIGURE 3. Basin moment axis and t one standard deviation ( 2 1 s ) for nine published
networks (networks from Melton, 1957).
42 THE CANADIAN GEOGRAPHER

degree of symmetry. Symmetry is directly measured by the skewness statistic; skew-

ness in any symmetrical distribution is zero12 and increases with growing asym-
metry. Following the evaluation of the skewness statistic a test may be made for
departure from normality. Elongation is a function of both standard deviation and
kurtosis. Kurtosis, however, is best used to compare the shapes of data distributions
when these approach a Gaussian curve. The application of kurtosis measurements
is indicative but not conclusive evidence for peakedness (ref. 12, p. 173).
The standard deviation measures the dispersion of the network. A long narrow
basin will have less dispersion than a wide basin of only moderate length. Values
for the standard deviation may be compared between basins, for the data have been
rendered dimensionless by moment measure analysis. The standard deviation may
be used with any distribution, though the frequency of observations between the
mean ( X )and any stated standard deviation (a)will be somewhat less than for the
normal curve.13 Figure 3 illustrates results for nine published networks for which

2
0
.90

r 2-0.50 P>.IO
S y , x = 0.1154

.80

1.
:. .70
-
0
0
3
.-L
u
-
c
g -60
m

-50

.40
.I

.30 , I I
10 20 30

FIGURE 4. Regression of basin circularity against network stand-

ard deviation. Numbers refer to basins in Table I and Figure 3.
 DRAINAGE BASIN PARAMETERS 43
morphometric data are readily available (Me1tonl4). Descriptive statistics and
Miller’s circularity ratio are listed in Table I for the nine networks of Figure 3.
A rather anomalous result is obtained when standard deviations are plotted
against circularity ( C ) values (Fig. 4) for the nine basins listed in Table I. Miller’s
circularity ratio assumes values between zero and unity, the latter when the basin
is perfectly circular. One might expect a positive correlation between circularity
and standard deviation (cr), that is, as C increases so too should the spread (cr)
of network vectorial equivalents. Figure 4 indicates a weak negative correlation
( r = -0.50, P > 0.10) for these nine basins. A regression of circularity on stan-
dard deviation for the same data minus the two end members is instructive in terms
of explaining the lack of interdependence of the two variables. With elimination
of the two extreme end members, the correlation coefficientis reduced to -0.3 1 ( P >
0.5). The change in value of the correlation coefficient is a result of an apparent
anomalous relationship between circularity and standard deviation for Whipple and
Cerro Pavo basins. The lower end member of the data (Cerro Pavo Basin) has a
large standard deviation but low circularity. This fact is readily explained by the
high positively skewed (Table I ) nature of the basin. The upper end member
(Whipple Basin) appears highly circular, yet has a low standard deviation. This
circumstance is attributable to the concentration of channels in the centre of the
basin and the relative paucity of channels near the periphery. It would appear that
terrain characteristics leading to overland flow and channel development change
significantly from basin centre to periphery. This fact, though not evident in the in-
direct ratios of circularity and elongation, is taken into account in the derivation of
the standard deviation of the network.
The proposed method for standardization of network parameters offers a con-
venient and objective means of describing basin axial and shape properties quan-
titatively. It remains to be seen whether these indices are significantly correlated
with runoff characteristics.

TABLE I

Mean Standard
azimuth deviation Skewness Kurtosis Circularity*
-
1 Cerro Pavo 228.7 25.24 +0.9179 4.4891 0.39
2 Whipple 303.1 16.21 -0.0115 2.3280 0.91
3 Peters Dome 279.7 13.58 -0.1166 2.5745 0.66
4 Finley and Adams
Canyon 1 286.9 17.08 +0.3866 2.8769 0.62
Finley and Adams
Canyon it 222.6 22.12 -0,0324 3.3489 0.50
Dory Hill 245.4 22.97 -0.4521 2.3532 0.60
Mesa Gulch 329.1 16.65 +0.0769 3.0177 0.67
Sycamore Canyon I 167.1 11.98 + O . 0031 1.7462 0.55
Sycamore Canyon II 116.8 22.41 - 0.0473 1.7701 0.60
*From Melton (1957).

The writer wishes to thank Dr H. French and Professor G. H. Dury who both
suggested improvements in the original manuscript. Some of the data result from
field work carried out while the writer was a Commonwealth Fellow (Australian)
under the auspices of the Commonwealth Department of Education and Science.
44 LE GEOGRAPHE CANADIEN

REFERENCES

1. ONGLEY,E. D., “Towards a Precise Definition of Drainage Basin Axis,” Australian Geogr.
Studies, 6 (1968), 84-8.
2. SCHUMM, S. A., Evolution of Drainage Systems and Slopes in Badlands at Perth Amboy,
New Jersey (Columbia Univ., Dept. of Geology Tech. Rept. No. 8, Office of Naval
Research Contract N6 ONR271, 1954).
3. MAXWELL,J. C., Quantitative Geomorphology of the Sun Dirnas Experimental Forest,
California (Columbia Univ., Dept. of Geology Tech. Rept. No. 19, Office of Naval
Research Contract N6 ONR 271-30, 1960).
4. MORISAWA, M. E., “Measurement of Drainage Basin Outline Form,” J . Geol. 66 (1958),
587-91.
5. MILLER,V. C., A Quantitative Geornorphic Study of Drainage Basin Clzaracteristics in the
Clinch Mountain Area, Virginia and Tennessee (Columbia Univ., Dept. of Geology,
Tech. Rept. No. 3, 1953).
6. SCHUMM, S. A. and R. F. HADLEY, Progress in the Application of Landform Analysis in
Studies of Semiarid Erosion (US. Geol. Survey Circ. 437, 1961).
7. ABRAHAMS, A. D., “Towards a Precise Definition of Drainage Basin Axis: Comment,”
Australian Geogr. Studies (in press).
8. SCHEIDEGGER, A. E., “Random Graph Patterns of Drainage Basins,” extract of Hydrological
Aspects o f the Utilization of Water (General Assembly o f Bern, Sept.-Oct. 1967),
415-25.
9. JIZBA,Z. V., “Mean and Standard Deviation of Certain Geologic Data - A Discussion,”
Anzer. J . Sci. 251 (1953), 899-906.
10. KRUMBEIN, W. C. and F. A. GRAYBILL, A n Introduction to Statistical Models in Geology
(New York, 1965), p. 131.
1 1 . AGTERBERG, F. P., and G. BRIGGS,“Statistical Analysis of Ripple Marks in Atokan and
Desmoinian Rocks in the Arkorna Basin of East-Central Oklahoma,” J . Sedirn. Petrol.
33 (1963), 393410.
12. MILLS,F. C., Staiistical Methods (New York, 1955), p. 171.
13. CROXTON, F. E., Elementary Statistics with Applications in Medicine and the Biological
Sciences (New York, 1959), p. 89.
14. MELTON,M. A., A n Analysis o f the Relations among Elements o f Climate, Surface Prop-
erties and Geomorphology (Columbia Univ., Dept. of Geology, Tech. Rept. No. 11,
Office of Naval Research Contract NR 389-042, 1957).