See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/273507833

Effect of Rainfall Temporal Distribution on the Conversion Factor to Convert the

Fixed-Interval into True-Interval Rainfall

Article  in  Journal of Hydrologic Engineering · February 2015

DOI: 10.1061/(ASCE)HE.1943-5584.0001178

CITATIONS READS

13 691

3 authors, including:

Chulsang Yoo Changhyun Jun

Korea University Chung-Ang University
233 PUBLICATIONS   1,682 CITATIONS    43 PUBLICATIONS   220 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Changhyun Jun on 18 April 2016.

The user has requested enhancement of the downloaded file.

 Effect of Rainfall Temporal Distribution on the
Conversion Factor to Convert the Fixed-Interval
into True-Interval Rainfall
Chulsang Yoo 1; Changhyun Jun 2; and Changyeol Park 3
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

Abstract: In this study, the Weiss approach to derive the CF (the conversion factor to convert the fixed-interval annual maximum rainfall into
the true-interval one) was examined and revised to consider the rainfall temporal distribution. As examples, several rainfall temporal dis-
tribution models currently being used in the rainfall-runoff analysis, along with several simple distributions such as triangular or pentagonal,
were considered to derive the CF. The resulting CFs were then compared with the CFs estimated by analyzing the observed rainfall data, both
in Korea and in several other countries, such as the United States, the United Kingdom, Australia, and New Zealand. The findings from this
study can be summarized as follows. First, the effect of the temporal distribution of rainfall is very significant on the estimation of the CF. The
CF for the impulse rainfall was the smallest at 1.0, and that for the uniformly-distributed rainfall was the highest at 1.333. Second, the CFs
derived for the temporal distribution models considered in this study were higher than the empirical CFs used worldwide. Finally, it was found
that, among simple distributions and temporal distribution models analyzed in this study, the quadratic functional form and the Keifer and Chu
method provide the most similar CF value to the empirical CF values used in many countries mentioned above. DOI: 10.1061/(ASCE)HE
.1943-5584.0001178. © 2015 American Society of Civil Engineers.
Author keywords: Conversion factor; True-interval rainfall; Fixed-interval rainfall; Temporal distribution of rainfall; Weiss’ approach.

Introduction which is determined by the rainfall-runoff analysis and used for

the design of hydraulic structures, is proportional to the CF.
The probable rainfall for the design of hydraulic structures is There are many research studies related to this CF worldwide.
generally determined through frequency analysis of the annual For example, in the United States, a CF of 1.13 was suggested for
maximum hourly or daily rainfall data. However, the rainfall data the hourly and daily rainfall data (U.S. Weather Bureau 1956;
generally used in hydrological analysis are all fixed-interval ones Hershfield 1961; Miller et al. 1973; Frederick et al. 1977; Huff
collected following the given clock system. For example, the 1-h and Angel 1992). Recently, Young and McEnroe (2003) estimated
rainfall is the rainfall amount collected between T and T þ 1 the CFs, using the highly precise ALERT-type rain gauge data.
o’clock. If the hourly rainfall is only the data available, that is, They also showed that the derived CFs are all consistent with
if the observed data are provided in fixed-interval h, it becomes the previous studies mentioned above. In the United Kingdom, a
impossible to determine the probable 1-h rainfall in true intervals. CF of 1.167 was suggested for the daily rainfall data, which is
The true-interval annual maximum 1-h rainfall indicates the real a representative value, as the average of CFs estimated regionally
maximum rainfall amount in a year measured for an h from any ranging from 1.14 to 1.19 (Dwyer and Reed 1994). Furthermore,
starting time. The conversion factor (CF) is a concept to improve Fowler et al. (2005) presented CFs of 1.16, 1.11, 1.035, and 1.005
the limitation of rainfall observation that makes it difficult to obtain for durations of 1, 2, 5, and 10 days, using the daily rainfall data
the true-interval annual maximum rainfall data. The CF is defined from 1961 to 1990, measured at four rain gauge stations. For
as the ratio of the true-interval annual maximum rainfall to the Australia, the CFs for the daily rainfall data were estimated from
fixed-interval annual maximum rainfall (Institute of Hydrology 1.1 to 1.19 by Pierrehumbert (“Short period rainfall intensity analy-
1999). As the CF is a simple multiplication factor applied to convert sis,” working paper, Bureau of Meteorology, Melbourne, Australia),
the fixed-interval annual maximum rainfall into the true-interval but the value 1.16, suggested in the Flood Estimation Handbook
annual rainfall, the frequency analysis result of true-interval rainfall (Institute of Hydrology 1999), was used as a representative value.
data is directly proportional to the CF. Thus, the design flood, Additionally, Dwyer and Reed (1995) suggested CFs with a range
from 1.15 to 1.16, and also mentioned that CF values could be af-
fected by weather conditions. Recently, however, the Hydrology
1 Report Series (Jakob et al. 2005) suggested the use of 1.15 as a
Professor, School of Civil, Environmental and Architectural Engineer-
ing, College of Engineering, Korea Univ., Seoul 136-713, Korea (corre- representative CF value, which had been determined by analyzing
sponding author). E-mail: [email protected] the rainfall data in seven major Australian cities (Melbourne, Sydney,
2
Research Assistant, School of Civil, Environmental and Architectural Brisbane, Darwin, Perth, Hobart, and Adelaide). Last, in New
Engineering, College of Engineering, Korea Univ., Seoul 136-713, Korea. Zealand, van Montfort (1990) suggested the use of a value of 1.137
3
Associate Researcher, Dept. of Safety and Environment Research, The
as a representative CF for the daily rainfall data, by summarizing the
Seoul Institute, Seocho-gu, Seoul 137-071, Korea.
Note. This manuscript was submitted on October 25, 2013; approved on CFs estimated monthly, using the generalized extreme value (GEV)
December 23, 2014; published online on February 20, 2015. Discussion and extreme value type-1 (EV1 )probability density functions.
period open until July 20, 2015; separate discussions must be submitted In Korea, most studies estimated CFs using minutely (i.e., at
for individual papers. This paper is part of the Journal of Hydrologic En- intervals of 1 min) observed rainfall data, along with hourly and
gineering, © ASCE, ISSN 1084-0699/04015018(9)/$25.00. daily data. The CFs estimated for 1, 3, 6, and 24 h and 1 day were

© ASCE 04015018-1 J. Hydrol. Eng.

J. Hydrol. Eng., 2015, 20(10): 04015018

 presented as 1.129, 1.033, 1.013, 1.005, and 1.161, respectively the CF, which may be kinds of empirical methods based on ob-
(Ministry of Construction and Transportation 2000). Here, it is served data, even though they did not derive the CF by directly
worthwhile to note that the CF for 1 day is higher than that of analyzing the observed data.
24 h. In fact, this result is reasonable as 1 day is composed of Weiss (1964) derived the CF by just applying the probability
one 1-day (time period between 0 and 24 o’clock), and 24 h is concept, but not considering any characteristics of rainfall. Several
composed of 24 1-h, which will be analyzed in more detail in assumptions were made, among which the most important one is
the following section. Kim et al. (1988) also presented the same that the rainfall intensity remains constant for the given duration of
CFs, by conducting a frequency analysis of the fixed- and true- rainfall. It is also an important assumption that the rainfall contin-
interval annual maximum rainfall data, using the EV1 distribution. ues only for the given duration. Any rainfall duration is possible,
Other studies have also determined CFs by directly comparing but the rainfall cannot be halted during the given duration. This also
fixed- and true-interval rainfall data (Cho et al. 2006; Moon et al. applies to rather long rainfall durations, like two or three days.
2008; Oh et al. 2008; Oh and Moon 2008). The analysis by Weiss (1964) can be explained as follows. To
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

In fact, most of the above-mentioned research studies were very make the explanation easier, let us consider the simplest case, that
empirical, based as they were on the analysis of observed data. is, the CF of the fixed-interval 1-h rainfall, when the hourly rainfall
Even though the definition of the CF is simple, its theoretical data are available. As the CF for the fixed-interval 1-h rainfall is
derivation and analysis are not. For example, the effect of various concerned, the true-interval annual maximum 1-h rainfall is
rainfall characteristics on the CF is unclear, such as the temporal assumed to continue only for 1 h. No rainfall occurs before or after
distribution and duration. It is unknown if the CF of a fixed-interval it. Also, the rainfall intensity remains constant during the true-
1-h rainfall and that of a fixed-interval 1-day rainfall should be the interval 1 h. Additionally, let us assume that the rainfall intensity
same. It is also ambiguous as to whether the CF should vary region- equals one. As the rainfall data are provided as fixed-interval 1-h
ally and seasonally. Unfortunately, any proper methodology or data (i.e., hourly data), it is unlikely that the true-interval 1 h for the
model that might be used to answer these questions is not available. maximum rainfall will agree with the fixed-interval 1 h. That is, the
The work by Weiss (1964) may be the only theoretical approach starting time point of the fixed-interval 1 h is a kind of random
for the derivation of the CF. He introduced a statistical concept to variable with respect to the true-interval 1 h. The starting time point
estimate the expected coverage of a fixed-interval by the same can be assumed to follow a uniform distribution.
length of a true-interval. The CF is nothing but the reciprocal of Now, the mean rainfall depth that the fixed-interval 1 h can have,
the expected coverage. However, his approach is fully dependent by overlaying it on the true-interval 1h, can be calculated by taking
on a simplified data structure. For example, he assumed that the the expected value. Simple calculation shows that the expected
rainfall continues only for the given duration, also, any length value is 0.5, thus the CF value becomes 2. However, this is the
of rainfall duration, such as several days or a week, is possible. result when considering an isolated fixed-interval 1 h. Even though
Because of this assumption, the CF of the fixed-interval 1-h rainfall the true-interval 1-h rainfall is isolated, the fixed-interval 1 h is
could be derived identical to that of the fixed-interval 1-day rainfall. sequential. Therefore, if the starting time point x is less than
He also assumed that the rainfall intensity remains constant during 0.5, more rainfall is recorded in the next fixed-interval 1 h;
the given duration. By introducing this assumption, he could similarly, if x is higher than 0.5, more rainfall is recorded in the
exclude the possibility of different temporal distribution of rainfall previous fixed-interval 1 h (Fig. 1). Weiss (1964) assumed that
by region. In conclusion, the CF derived by Weiss (1964) is the rainfall depth of 0.5 is always secured in any case, and that
unrelated to any of the rainfall characteristics that vary by region the mean rainfall depth should be expressed as the sum of the
or season. It is very surprising, however, that the CFs derived by rainfall depth 0.5 and the expected value for the zero-to-one half
Weiss are so similar to those estimated by the observed data or one half-to-one section (Fig. 2).
analysis. The expected value can be calculated as follows (Weiss 1964):
In this study, the Weiss approach to derive the CF was examined Z 1
in detail, and revised to consider the rainfall temporal distribution. E½Rfixed  ¼ 0.5 þ xdx ð1Þ
This process will show what the cause of the CF is and how the 0.5

rainfall temporal distribution affects the CF. As examples, several In fact, the above expression gives the value 0.875. Therefore,
simple distributions, such as triangular or pentagonal, were consid- The CF is determined as follows:
ered to derive the CF and to see the effect of the rainfall temporal
distribution on the CF. Also, several rainfall temporal distribution
models currently being used in the rainfall-runoff analysis, such as
the Huff method (Huff 1967), the Mononobe method (Jeong and
Yoon 2007), Yen-Chow method (Yen and Chow 1980), alternating
block method (Chow et al. 1988), and Keifer and Chu method
(Keifer and Chu 1957) were considered to derive the CF. The
resulting CFs were then compared with the CFs estimated by
analyzing the observed rainfall data, both in Korea and in several
other countries, such as the United States, the United Kingdom,
Australia, and New Zealand.

Weiss Model and Its Modification

Fig. 1. Schematic diagram explaining how the fixed-interval d-h rain-
The analysis by Weiss (1964) may be the only theoretical study
fall (dotted box) is recorded depending on its relative location [starting
related to the CF. It is possible to find some other studies using
time point x: (a) x < 0.5 and (b) x ≥ 0.5] to the true-interval d-h rainfall
the concepts of a probability density function (van Montfort
(solid box)
1990) and fractal theory (Dwyer and Reed 1994) to determine

© ASCE 04015018-2 J. Hydrol. Eng.

J. Hydrol. Eng., 2015, 20(10): 04015018

 1 1

Intensity
(mm/hr)
Rainfall
0.5 0.5

0 0
0 0.5 1 0 0.5 1
Starting Time (hour) (OR) Starting Time (hour)

Fig. 2. Weiss’ concept explaining how the rainfall intensity of fixed-interval 1-h rainfall changes depending on its relative location to the true-interval
1-h rainfall
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

Rtrue 1 States derived by analyzing the observed data (Yoo and Jun
CF ¼ ¼ R1 ≈ 1.143 ð2Þ
E½Rfixed  0.5 þ 0.5 xdx 2010), even though no effect from regionally and seasonally
different rainfall characteristics was considered. Thus, it is under-
Here, Rfixed is the rainfall depth that the fixed-interval 1 h has, by standable to doubt how the CFs would be, if considering the rainfall
covering the true-interval 1-h rainfall Rtrue , and E½Rfixed  is the temporal characteristics.
expected value derived. This derived CF can also be said to be Before considering the rainfall characteristics, it is also impor-
the mean length of the true-interval rainfall, covered by the tant to notice the problem of calculating the expected value that the
fixed-interval 1 h, as the rainfall intensity is assumed constant. true-interval rainfall overlays with the fixed- interval one. In Weiss
This concept of CF calculation is not limited to the hourly data, (1964), if x is less than or equal to 0.5 at a fixed-interval, more
but applicable to any fixed-interval data, like minutely and daily rainfall would be recorded in the next fixed-interval, so it was as-
rainfall data. sumed that the rainfall amount 0.5 is secured in any case, and the
As a more complex case, let us consider the CF for the fixed- remaining expected value was estimated for x is greater than 0.5. It
interval 12-h rainfall, when the hourly rainfall data are available, In is also the same if x is greater than or equal to 0.5 at a fixed-interval.
this case, it is also assumed that the rainfall intensity is constant As more rainfall would be recorded in the previous fixed-interval,
(i.e., one, as in the previous case) during the true-interval 12 h. This it was assumed that the rainfall amount 0.5 is secured, and the
12-h rainfall is also an isolated one. Now, simply considering the remaining expected value was estimated for x is less than 0.5.
data structure, it is obvious that the fixed-interval 12 h and the true- However, this is not true. For example, if x is 0.0 (that is, the
interval 12 h overlay in 11 out of the 12 h. The remaining 1 h is fixed-interval time coincides with the true-interval time), the rain-
exactly the same as the previous case. Therefore, the expected value fall in the fixed-interval 1 h becomes one. If x is 0.2, the rainfall in
of rainfall of the fixed-interval 12-h rainfall becomes as follows: the fixed-interval 1 h drops to 0.8, and the smallest value 0.5 is
 Z 1  obtained only when x is 0.5. After this point, the value increases
E½Rfixed  ¼ 11 þ 0.5 þ xdx ð3Þ again, and becomes one when x is 1.0 (Fig. 3).
0.5 Therefore, the expected value for the fixed-interval 1 h must be
calculated as follows:
More generally, by the unit of the available data, the mean depth Z 0.5 Z 1 Z 1
of the fixed-interval n-unit rainfall can be expressed as follows:
E½Rfixed  ¼ ð1 − xÞdx þ xdx ¼ 2 × xdx ð6Þ
 Z 1  0 0.5 0.5
E½Rfixed  ¼ ðn − 1Þ þ 0.5 þ xdx ð4Þ
0.5 The above expression gives 0.75. Thus, the general expression
of CF must be revised as follows:
and, as the depth of true-interval n-unit rainfall becomes n, the CF
of the fixed-interval n-unit rainfall is determined as follows: n n
CF ¼ R1 ¼ ð7Þ
ðn − 1Þ þ ð2 × 0.5 xdxÞ n − 0.25
n n
CF ¼ R1 ¼ R1
ðn − 1Þ þ ð0.5 þ 0.5 xdxÞ n − 0.5 þ 0.5 xdxÞ By applying the above expression, the CF values for hourly
¼
n
ð5Þ data can be obtained, e.g., CF ¼ 1.333 for 1-h rainfall,
n − 0.125
This is the result obtained by using Weiss (1964). Because of the
assumptions introduced, the CF is dependent only on the data struc- 1
ture. The same CF is obtained if the data structure is identical. Data
Intensity
(mm/hr)

structure here indicates the time unit of available data, and the time
Rainfall

unit of rainfall for the determination of CF. For example, to estimate 0.5
the CF of daily data, when hourly data are available, the time unit of
available data becomes hour, and the time unit for CF becomes day.
Thus, when estimating with hourly data (1 h ¼ 1 unit), CF ¼ 0
1.143 in the case of 1-h rainfall (n ¼ 1), CF ¼ 1.022 in the case 0 0.5 1
of 6-h rainfall (n ¼ 6), and CF ¼ 1.005 in the case of 24-h rainfall Starting Time (hour)
(n ¼ 24). When estimating with daily data, CF ¼ 1.143 is obtained
Fig. 3. Schematic diagram explaining how the rainfall intensity of
for daily rainfall, which is identical to the case of hourly data.
fixed-interval 1-h rainfall changes depending on its relative location
As mentioned above, the CFs derived by Weiss (1964) are very
to the true-interval 1-h rainfall
similar, with only a few percent error, to the CFs in the United

© ASCE 04015018-3 J. Hydrol. Eng.

J. Hydrol. Eng., 2015, 20(10): 04015018

 CF ¼ 1.043 for 6-h rainfall, and CF ¼ 1.011 for 24-h rainfall. unrealistic. More realistic models for the rainfall temporal distribu-
When estimating with daily data, CF ¼ 1.333 is obtained for tion will be reviewed in the following section.
daily rainfall, which is identical to the case of hourly data. From
the above results, it seems that the result in this study is inferior to
Derivation of CFs
the result of applying the original Weiss model. However, this is
attributable to the fact that no rainfall characteristics were applied,
and it was assumed that the rainfall intensity was constant for Case 1
the given duration. Under this assumption, the result derived Case 1 represents instantaneous rainfall (i.e., in the form of impulse),
in this study can be said to be more accurate than the result of of which the duration is close to zero. This is a virtual case that can-
the original Weiss model, as it is more faithful to the given not exist in reality. In this case, regardless of the position of x in the
assumption. 0–1-h section, the rainfall secured by the fixed-interval 1 h becomes
one. This indicates that the expected value for the fixed-interval 1 h is
always one. Therefore, the CF becomes always constant at 1.0, re-
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

gardless of the data used, such as the hourly or daily rainfall.

CFs for Some Simple Temporal Distributions of
Rainfall Case 2
In Case 2, for the 0–0.5-h section, the rainfall amount that is lost
Simple Distributions Considered (that is, not considered in the fixed-interval data) by the movement
Weiss (1964) theoretically derived the CF by assuming that rain- of x from the origin to the right is 4x3 . When x is at zero (that is, the
fall intensity remains constant for the given duration of rainfall. fixed-interval 1 h and the true-interval 1 h coincides), the rainfall
amount secured by the fixed-interval 1 h becomes one. However, if
In other words, the temporal distribution of rainfall in Weiss’ ap-
x ¼ 0.2, the rainfall amount secured by the fixed-interval 1 h drops
proach was a uniform distribution. However, the real temporal
to 0.968. The lowest value 0.5 is obtained when x ¼ 0.5. After this
distribution of rainfall is not uniform, but appears in various and
point, the value increases again, and becomes one when x ¼ 1.
complicated forms, depending on areas, seasons, and the storm
Fig. 5 shows the behavior of the change of rainfall amount secured,
events themselves.
depending on the location of x.
In this study, a total of eight simple distributions, including the
Therefore, the expected value for the fixed-interval 1 h must be
triangular, tetragonal and pentagonal, were considered for the
calculated as follows:
derivation of CF (Fig. 4). The purpose of deriving the CF for these
 Z 0.5 
simple distributions is to show how the temporal distribution of
rainfall affects the value of CF. Even though the temporal distribu- E½Rfixed  ¼ 2 × 0.5 − 4x3 dx ð8Þ
0
tions considered in this section are all simple ones, some of them
are still used for rainfall-runoff analysis, like the Yen and The above expression gives 0.875. Therefore, CF is determined
Chow method (Yen and Chow 1980), thus they may not be fully as follows:

3 2 2

0 0 0
0 1 0 1 0 1 0 1
(a) (b) (c) (d)

2 2-a 2-a 2-a

a a a

0 0 0 0
0 1 0 1 0 1 0 1
(e) (f) (g) (h)

Fig. 4. Simple temporal distributions of rainfall considered in this study: (a) Case 1—impulse; (b) Case 2—quadratic function; (c) Case 3; (d) Case 4;
(e) Case 5—triangular; (f) Case 6—pentagonal; (g) Case 7; (h) Case 8—tetragonal

© ASCE 04015018-4 J. Hydrol. Eng.

J. Hydrol. Eng., 2015, 20(10): 04015018

 n n
1 CF ¼ R ¼ R
ðn − 1Þ þ ð1 − 2 00.5 2x2 dxÞ n − 2 00.5 2x2 dx

Intensity
n

(mm/hr)
Rainfall
¼ ð15Þ
0.5 n − 0.167
By applying the above expression, the CF values for hourly data
0 can be obtained, e.g., CF ¼ 1.2 for 1-h rainfall, CF ¼ 1.029 for
0 0.5 1 6-h rainfall, and CF ¼ 1.007 for 24-h rainfall. When estimating
Starting Time (hour) with daily data, CF ¼ 1.2 is obtained for the daily rainfall, which
is identical to the case of hourly data.
Fig. 5. Same as Fig. 3 (Case 2)
Case 4
For Case 4, the rainfall amount that is lost by the movement of x
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

Rtrue 1 from the origin to the right is x2 . When x ¼ 0, the rainfall amount
CF ¼ ¼ R 0.5 3 ≈ 1.143 ð9Þ secured by the fixed-interval 1 h becomes one. However, unlikepthe
E½Rfixed  1 − 2 0 4x dx ffiffiffi
above cases, the lowest value 0.5 is obtained when x ¼ 1= 2.
After this point, the value increases again, and becomes one when
If the duration of the data used to estimate the CF of rainfall with
x ¼ 1.
fixed-interval n equals one, the expected value becomes as follows:
Therefore, the expected value for the fixed-interval 1 h must be
 Z 0.5  calculated as follows:
E½Rfixed  ¼ ðn − 1Þ þ 1 − 2 4x3 dx ð10Þ Z 1=pffiffi2 Z 1
0
E½Rfixed  ¼ ð1 − x2 Þdx þ pffiffi x2 dx ð16Þ
0 1= 2
As the true-interval rainfall for the duration of n equals n, the
general equation of CF is derived as follows: The above expression gives 0.805. Therefore, the CF is deter-
n n mined as follows:
CF ¼ R 0.5 3 ¼ R 0.5 3
ðn − 1Þ þ ð1 − 2 0 4x dxÞ n − 2 0 4x dx Rtrue
CF ¼ ≈ 1.243 ð17Þ
n E½Rfixed 
¼ ð11Þ
n − 0.125
If the duration of the data used to estimate the CF of rainfall with
By applying the above expression, the CF values for hourly data fixed interval n ¼ 1, the expected value becomes as follows:
can be obtained, e.g., CF ¼ 1.143 for 1-h rainfall, CF ¼ 1.021 for Z 1=pffiffi2 Z 1
6-h rainfall, and CF ¼ 1.005 for 24-h rainfall. When estimating E½Rfixed  ¼ ðn − 1Þ þ ð1 − x2 Þdx þ pffiffi x2 dx ð18Þ
with daily data, CF ¼ 1.43 is obtained for daily rainfall, which 0 1= 2
is identical to the case of hourly data.
As the arbitrary time rainfall for the duration of n is n, the
Case 3 general equation of CF is derived as follows:
For Case 3, for the 0–0.5-h section, the rainfall amount that is lost n n
by the movement of x from the origin to the right is 2x2 . When CF ¼ R 1=pffiffi2 R1 ¼
n − 0.195
ð19Þ
ðn − 1Þ þ ð1 − x Þdx þ 2 pffiffi 2
x dx
x ¼ 0, the rainfall amount secured by the fixed-interval1 h becomes 0 1= 2
one. However, if x ¼ 0.2, the rainfall amount secured by the
fixed-interval 1 h drops to 0.92. The lowest value 0.5 is obtained By applying the above expression, the CF values for hourly data
when x ¼ 0.5. After this point, the value increases again, and can be obtained, e.g., CF ¼ 1.243 for 1-h rainfall, CF ¼ 1.034 for
becomes one when x ¼ 1. 6-h rainfall, and CF ¼ 1.034 for 24-h rainfall. When estimating
Therefore, the expected value for the fixed-interval 1 h must be with daily data, CF ¼ 1.243 is obtained for the daily rainfall, which
calculated as follows: is identical to the case of hourly data.
 Z 0.5  Case 5
E½Rfixed  ¼ 2 × 0.5 − 2x2 dx ð12Þ For Case 5, when compared with Case 4, the rainfall amount se-
0
cured is symmetric about x ¼ 0.5, and the same results are derived
as Eqs. (16)–(19) for Case 4. In other words, when the temporal
The above expression gives 0.83. Therefore, the CF is deter-
distribution corresponding to Case 5 was applied, the CFs for
mined as follows:
1-, 6-, and 24-h rainfall are 1.243, 1.034, and 1.008, respectively.
Rtrue 1 Furthermore, the CF estimated on the basis of daily data is also
CF ¼ ¼ R ≈ 1.2 ð13Þ 1.243, which is identical to the result of Case 4.
E½Rfixed  1 − 2 00.5 2x2 dx
Case 6
If the duration of the data used to estimate the CF of rainfall with For Case 6, similar to Cases 2 and 3, for the 0–0.5-h section, the
fixed interval n equals one, the expected value becomes as follows: rainfall amount that is lost by the movement of x from the origin to
 Z 0.5  the right is ½ð2 − 2aÞx þ ax. When x ¼ 0, the rainfall amount se-
E½Rfixed  ¼ ðn − 1Þ þ 1 − 2 2x2 dx ð14Þ cured by the fixed-interval 1 h becomes one. The rainfall amount
0 secured by the fixed-interval 1 h can be determined in the range
0 < a < 1, and this value becomes the lowest at 0.5, when
As the arbitrary time rainfall for the duration of n is n, the x ¼ 0.5. After this point, the value increases again, and becomes
general equation of CF is derived as follows: one when x ¼ 1.

© ASCE 04015018-5 J. Hydrol. Eng.

J. Hydrol. Eng., 2015, 20(10): 04015018

 Therefore, the expected value for the fixed-interval 1 h must be the quadrangular-shape temporal distribution in Fig. 4, the
calculated as follows: expected value for the range 0 < a < 1 can be calculated.
Z 0.5 This p value becomes
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi the lowest at 0.5, when x ¼
E½Rfixed  ¼ 2 × f1 − 2½ð2 − 2aÞx þ axgdx ð20Þ ð−a þ a2 − 2a þ 2Þ=ð2 − 2aÞ. After this point, the value in-
0 creases again, and becomes one when x ¼ 1.
Therefore, the expected value for the fixed-interval 1 h must be
If the value of a is assumed to be 0.25, 0.5, and 0.75, the ex- calculated as follows:
pected values for the fixed-interval 1 h become 0.824, 0.792, and
0.771, respectively. Therefore, the CFs in Case 6 are determined by pffiffiffiffiffiffiffiffiffiffiffi
Z ð−aþ a2 −2aþ2Þ
taking the reciprocal of the expected value, as with the above cases. ð2−2aÞ
The results are 1.231, 1.263, and 1.297, respectively. Furthermore, E½Rfixed  ¼ ½1 − ð1 − aÞx2 − axdx
0
the general equation of CF where the duration of fixed-interval and Z 1
true-interval rainfall is n can be determined by applying the same þ pffiffiffiffiffiffiffiffiffiffiffi ½ð1 − aÞx2 þ axdx ð22Þ
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

ð−aþ a2 −2aþ2Þ
methods described above. For example, the result of the case where ð2−2aÞ

a ¼ 0.5 is as follows:
n n Similar to Case 6, the expected values for the fixed-interval 1 h
CF ¼ R 0.5 ¼ ð21Þ
ðn − 1Þ þ 2 0 ð1 − x2 − 0.5xÞdx n − 0.208 can be calculated with given specific values of a, such as 0.25,
0.5, and 0.75. The calculation results are 0.782, 0.765, and 0.754,
and, accordingly, the CFs become 1.278, 1.307, and 1.326,
Case 7 respectively. For a ¼ 0.5, the general equation for CF, when
For Case 7, the rainfall amount that is lost by the movement of x the duration of fixed-interval and true-interval rainfalls is n, is
from the origin to the right is ½ð1 − aÞx þ ax. When considering as follows:

n n
CF ¼ R ðpffiffi5−1Þ=2 R1 ¼
n − 0.235
ð23Þ
ðn − 1Þ þ 2
ð1 − 0.5x − 0.5xÞdx þ pffiffi ð0.5x 2
þ 0.5xÞdx
0 ð 5−1Þ=2

Case 8 CFs for Some Temporal Distribution Models of

Case 8 has the symmetrical form of Case 7 about x ¼ 0.5 h, as with Rainfall
the relationship between Cases 4 and 5. Thus, applying the method
for calculating the CF in Case 7 results in Eqs. (22) and (23). In In the previous section, the temporal distribution of rainfall was
other words, the CF calculation results, by applying the temporal simplified as triangular, tetragonal, and pentagonal, and the CFs
distribution corresponding to Case 8, also vary depending on the for those distributions were derived. In this part of the study, several
value of a, and the CF values for each case are identical to those in models for the temporal distributions of rainfall, such as the Huff
Case 7. method (Huff 1967), the Mononobe method (Jeong and Yoon
2007), the alternating block method (Chow et al. 1988), and the
Keifer and Chu method (Keifer and Chu 1957) were considered,
Comparison of Results to derive the CFs.
In this study, to derive the temporal distribution of rainfall for
In this study, the CFs for several simple temporal distributions of those models considered, it was first assumed, as an example, that
rainfall, such as triangular, tetragonal, and pentagonal, were derived the total rainfall amount is 400 mm, and the duration is 24 h. Then,
for comparison. The Weiss model with the uniform temporal
distribution of rainfall is one of the cases considered in this study.
The main purpose of comparing these different temporal distribu- 1.6
tions of rainfall is to see their effect on the CFs derived. Obviously,
more realistic temporal distribution of rainfall will provide more
realistic CFs, similar to those obtained by analyzing the observed 1.4
Uniform
rainfall data. Fig. 6 shows the comparison results of CFs derived in
this study.
The results in Fig. 6 show that the CF for the impulse rainfall
CF

1.2
is the smallest at 1.0, and that for the uniformly distributed
rainfall is the highest at 1.333. For other simple temporal
distributions, the CFs are derived in between these two Impulse
1
values. Especially, it should be noticed that when temporal
distribution of rainfall is not considered (or assumed to be
uniform, such as in the Weiss model), the CF derived could 0.8
be unrealistically higher than those of considering more realistic Case 1 Case 2 Case 3 Case 4 Case 6 Case 7
temporal distributions of rainfall. It is obvious that the effect of
Fig. 6. Comparison of CFs derived by considering the simple temporal
the temporal distribution of rainfall is very significant on the
distributions of rainfall
decision of the CF.

© ASCE 04015018-6 J. Hydrol. Eng.

J. Hydrol. Eng., 2015, 20(10): 04015018

 8 8

Rainfall Intensity (mm/hr)

Rainfall Intensity (mm/hr)

6 6

4 4

2 2
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

0 0
0 1 0 1
(a) Time (hour) (b) Time (hour)

8 8

Rainfall Intensity (mm/hr)

Rainfall Intensity (mm/hr)

6 6

4 4

2 2

0 0
0 1 0 1
(c) Time (hour) (d) Time (hour)

Fig. 7. Temporal distribution models of rainfall considered in this study: (a) huff; (b) mononobe; (c) alternating block; (d) Keifer and Chu (1957)

the temporal distributions were standardized between 0 and 1, so 1-day rainfall. Sometimes, probability density functions like the
that the CFs could be derived, just as in the previous section. For the GEV and EV1 were used to quantify the annual maximum data
Huff method, the temporal distribution of rainfall considered was for the comparison of fixed- and true-interval rainfall data (van
for the second quartile, which is most frequent in Korea. For the Montfort 1990). A recent study by Dwyer and Reed (1994) used
Mononobe method and the alternating block method, those derived fractal theory to calculate the CF.
in the Ministry of Construction and Transportation (2000) were Fig. 9 shows the comparison of the CFs for the United States
used. The equation for the rainfall intensity by Lee and Park (U.S.), the United Kingdom (U.K.), Australia (AU), and New
(1992) was used to apply the Keifer and Chu method. Additionally, Zealand (NZ). The CFs compared in this study are those mentioned
11.5 h was applied for the rainfall peak time for the Mononobe in the “Introduction.” In fact, the CFs of these countries are very
method and the alternating block method. The temporal distribu-
tions derived for those models considered are shown in Fig. 7.
With given temporal distributions, the calculation of the CFs is
straightforward, such as in the previous section. For the Huff 1.6
method, the CF was derived to be 1.232. For the Mononobe method
and the alternating block method, the CFs derived were 1.172 and
1.213, respectively. Last, the CF for the Keifer and Chu method was 1.4
Uniform
1.166. All the CFs derived are approximately 1.2, which are also
compared in Fig. 8.
CF

1.2

Comparison with Empirical CFs Worldwide

Impulse
1
In most studies on the CF, the fixed-interval hourly and minute-
based data are generally used for comparison. For example, the
fixed-interval annual maximum 1-h rainfall is compared with the 0.8
fixed-interval annual maximum 60-min rainfall, to derive the CF Case 1 Keifer & Chu Mononobe Blocking Huff

for the fixed-interval 1-h rainfall. Also, the fixed-interval annual

Fig. 8. Comparison of CFs derived by considering several temporal
maximum 1-day rainfall is compared with the fixed-interval annual
distribution models of rainfall
maximum 1,440-min rainfall, to derive the CF for the fixed-interval

© ASCE 04015018-7 J. Hydrol. Eng.

J. Hydrol. Eng., 2015, 20(10): 04015018

 1.6
total of eight simple distributions, such as triangular, tetragonal,
and pentagonal, were considered and derived the CF. Also, several
rainfall temporal distribution models currently being used in the
1.4 rainfall-runoff analysis in Korea were considered to derive the
Uniform
CF. The resulting CFs were then compared with the CFs estimated
by analyzing the observed rainfall data, both in Korea and in several
other countries, such as the United States, the United Kingdom,
CF

1.2
Australia, and New Zealand. The findings from this study can
be summarized as follows:
Impulse First, it was found that the temporal distribution of rainfall has a
1
very significant impact on the determination of the CF. Among
those simple temporal distributions considered in this study, the
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

CF of the impulse rainfall was the smallest at 1.0, but that of

0.8 the uniformly-distributed rainfall was the highest at 1.333. For
Case1 US NZ AU KR UK other temporal distributions, the CFs were derived in between these
Fig. 9. Comparison of empirical CFs for the daily rainfall used in two values. Overall, the CF was estimated smaller as the temporal
several countries worldwide distribution became more peaked.
Second, the CFs derived for the temporal distribution models,
currently used in the rainfall-runoff analysis in Korea, also showed
the same trend, that is, the CF for a peaked distribution model was
Table 1. Comparison of All CFs of Daily Rainfall Data Considered in This estimated smaller than that for a flat distribution model. The CFs
Study
derived for the Huff method and the alternating block method were
Case CF Remarks rather higher at 1.232 and 1.213, respectively, and those for the
Case 1 1 Impulse Mononobe method and the Keifer and Chu (1957) method were
U.S. 1.13 U.S. Weather Bureau (1956) smaller at 1.172 and 1.166, respectively.
New Zealand 1.137 van Montfort (1990) Finally, it was found that the empirically-derived CFs used in
Case 2 1.143 Quadratic function Korea, the United States, the United Kingdom, Australia, and
Australia 1.15 Jakob et al. (2005) New Zealand are very similar to each other, with their mean of
Korea 1.161 MOCT (2000) 1.146 and range of just 0.037. However, the mean of CFs estimated
Keifer and Chu method 1.166 Keifer and Chu (1957) for the temporal distribution models was higher than that of the
United Kingdom 1.167 Dwyer and Reed (1994)
empirical values, also their range was much wider. Among simple
Mononobe method 1.172 —
Case 3 1.2 Triangular temporal distributions considered in this study, the CF of the quad-
Alternating block method 1.213 — ratic function (1.143) was most similar to the empirical values.
Huff method 1.232 Huff (1967) Also, among the temporal distribution models considered in this
Cases 4 and 5 1.243 Triangular study, the CF of the Keifer and Chu (1957) method derived the most
Case 6 (a ¼ 0.5) 1.263 Pentagonal similar value, 1.166, to the values used worldwide.
Cases 7 and 8 (a ¼ 0.5) 1.307 Tetragonal As the design flood is ultimately proportional to the CF, its im-
— 1.333 Uniform pact on the design of hydraulic structures cannot be underesti-
mated. Additionally, as shown in this study, the CF is strongly
related with the rainfall temporal distribution. Thus, the temporal
similar to each other, with their average CF of 1.146. This is also distribution model used in a country could be evaluated using the
very similar to the CF calculated by considering the temporal dis- CF derived empirically. Their consistency may confirm the validity
tribution of Case 2, which is 1.143. However, in Korea (KR), the of both the CF and the rainfall temporal distribution model. How-
CF was differently estimated with different durations, i.e., the CF ever, most of the rainfall temporal distribution models are found not
for the hourly rainfall data is different from that of the daily rain- to be consistent with the empirical CFs used worldwide. A bit more
fall data. peaked model for rainfall temporal distribution is needed to resolve
Finally, all the temporal distributions examined in this study and this inconsistency.
their CFs are summarized in Table 1. As shown in this table, the CF
estimated from the Keifer and Chu method, among the temporal
distribution models examined in this study, was most similar to Acknowledgments
the empirical CF. Furthermore, the CFs estimated from the
Mononobe method and alternating block method were not signifi- This research was funded partially by Basic Science Research
cantly different from the empirical CFs. Among the simple Program through the Korea Research Foundation (KRF-2008-
distributions considered, the CF estimated for the quadratic func- 313-D01083) and partially by the National Research Foundation
tion symmetrical around the center of rainfall duration (Case 2), of Korea (NRF) through the Ministry of Education, Science and
was most similar to the CF derived using the observed data. Technology (No. 2010-0014566).

Conclusions References
In this study, the Weiss approach to derive the CF was examined Cho, H., Um, M.-J., Cho, W., and Cho, J. Y. (2006). “Adjustment factors of
and revised to consider the rainfall temporal distribution. Through precipitation using one-minute data in Seoul.” Proc., 2006 Korea Water
this process, it could be shown what the cause of the CF is and how Resources Association Symp., Korea Water Resources Association,
the rainfall temporal distribution affects the CF. As examples, a Seoul, Korea, 1506–1510.

© ASCE 04015018-8 J. Hydrol. Eng.

J. Hydrol. Eng., 2015, 20(10): 04015018

 Chow, V. T., Maidment, D. R., and Mays, L. W. (1988). Applied hydrology, 1998 Korean Society of Civil Engineers Symp., Korean Society of Civil
McGraw-Hill, New York, 572. Engineers, Seoul, Korea, 216–219.
Dwyer, I. J., and Reed, D. W. (1994). “Effective fractal dimension and Lee, W. H., and Park, S. D. (1992). “A unification of the probable rainfall
corrections to the mean of annual maxima.” J. Hydrol., 157(1–4), intensity formula at Seoul.” KSCE J. Civ. Eng., 12(4), 135–143.
13–34. Miller, J. F., Frederick, R. H., and Tracey, R. J. (1973). “Precipitation-
Dwyer, I. J., and Reed, D. W. (1995). “Allowance for discretization in frequency atlas of the western United States.” NOAA Atlas 2, National
hydrological and environmental risk estimation.” Rep. No. 123, Institute Weather Service, National Oceanic and Atmospheric Administration,
of Hydrology, Wallington, U.K. U.S. Dept. of Commerce, Washington, DC.
Fowler, H. J., Ekströmb, M., Kilsbya, C. G., and Jonesb, P. D. (2005). “New MOCT (Ministry of Construction and Transportation). (2000). “Research
estimates of future changes in extreme rainfall across the UK using survey report about development of management technique in
regional climate model integrations. 1. Assessment of control climate.” water resources: Vol. 1 the probable isohyetal chart in Korea.” Seoul,
J. Hydrol., 300(1–4), 212–233. Korea.
Frederick, R. H., Myers, V. A., and Auciello, E. P. (1977). “Five- to Moon, Y.-I., Oh, T.-S., Oh, K.-T., and Jun, S.-Y. (2008). “Conversion
60-minute precipitation frequency for the eastern and central United factor estimates between the rain data per minute and fixed-time-
Downloaded from ascelibrary.org by Science Library of Korea Univ on 02/24/16. Copyright ASCE. For personal use only; all rights reserved.

States.” NOAA Technical Memorandum NWS HYDRO-35, National interval.” Proc., 2008 Korean Society of Hazard Mitigation Symp.,
Oceanic and Atmospheric Administration, National Weather Service, Korean Society of Hazard Mitigation, Seoul, Korea, 679–682.
Silver Spring, MD. Oh, T. S., and Moon, Y.-I. (2008). “Conversion factor calculation of annual
Hershfield, D. M. (1961). “Rainfall frequency atlas of the United States for maximum precipitation in Korea between fixed and sliding durations.”
durations from 30 minutes to 24 hours and return periods from 1 to KSCE J. Civ. Eng., 28(5B), 515–524.
100 years.” U.S. Weather Bureau Technical Paper No. 40, U.S. Dept. Oh, T. S., Oh, K. T., Moon, Y.-I., and Park, R. K. (2008). “Conversion
of Commerce, Washington, DC. factor estimation of temporal time by calibration for minutely rain
Huff, F. A. (1967). “Time distribution of rainfall in heavy storms.” Water data using hourly rain data.” Proc., 2008 Korea Water Resources
Resour. Res., 3(4), 1007–1019. Association Symp., Korea Water Resources Association, Seoul, Korea,
Huff, F. A., and Angel, J. R. (1992). “Rainfall frequency atlas of the 1215–1219.
Midwest.” Illinois State Water Survey Bulletin 71, Midwest Climate U.S. Weather Bureau. (1956). “Rainfall intensities for local drainage design
Center Research Rep. 92-03, Illinois State Water Survey, in western United States.” Technical Paper No. 28, U.S. Dept. of
Champaign, IL. Commerce, Washington, DC.
Institute of Hydrology. (1999). Flood estimation handbook, Institute of van Montfort, M. A. J. (1990). “Sliding maxima.” J. Hydrol., 118(1–4),
Hydrology, Wallingford, U.K. 77–85.
Jakob, D., Taylor, B., and Xuereb, K. (2005). “A pilot study to explore Weiss, L. L. (1964). “Ratio of true to fixed-interval maximum rainfall.”
methods for deriving design rainfalls for Australia. Part 1.” J. Hydraul. Eng., 90(HY1), 77–82.
HRS Rep. No. 10, Australian Government Bureau of Meteorology, Yen, B. C., and Chow, V. T. (1980). “Design hyetographs for small drainage
Melbourne, Australia, 5–14. structures.” J. Hydraul. Div., 106(HY6), 1055–1076.
Jeong, J. H., and Yoon, Y. N. (2007). Design practices for water resources, Yoo, C., and Jun, C. (2010). “A short note on the conversion factor of
Goomi Press, Seoul, Korea. fixed- to true-interval precipitation.” Mag. Korea Water Resour. Assoc.,
Keifer, C. J., and Chu, H. H. (1957). “Synthetic storm pattern for drainage 43(9), 123–129.
design.” J. Hydraul. Div., 83(4), 1–25. Young, C. B., and McEnroe, B. M. (2003). “Sampling adjustment factors
Kim, K.-H., Kim, Y. S., Lee, J. W., and Kim, S. (1988). “Conversion factor for rainfall recorded at fixed time intervals.” J. Hydrol. Eng., 10.1061/
of maximum precipitation between fixed and sliding durations.” Proc., (ASCE)1084-0699(2003)8:5(294), 294–296.

© ASCE 04015018-9 J. Hydrol. Eng.

View publication stats J. Hydrol. Eng., 2015, 20(10): 04015018