RADIO SCIENCE, VOL. 42, RS6004, doi:10.

1029/2006RS003535, 2007

Rainfall rate modeling and worst month statistics for millimetric

line-of-sight radio links in South Africa
Pius A. Owolawi1 and Thomas J. Afullo1
Received 12 June 2006; revised 18 December 2006; accepted 9 April 2007; published 30 November 2007.
[1] Rain attenuation has a serious impact on the availability and performance of radio
communication services operating in the frequencies above 10 GHz. In order to reliably
predict rain attenuation for a given location, there is a need for studying monthly and
seasonal rainfall variability, worst month and average worst month statistics of rain, and
rainfall rate distribution models for the site under study. In this paper, the tasks are
tackled by processing 5-year (a) rain rate data for selected sites in South Africa. The
dependence of the percentage of the average year (AY) on percentage of time of the
average worst month (AWM) of rainfall rate and the obtained rainfall rate distribution
models for eight regions in South Africa are assessed by examining the root-mean-square
(RMS), the average probability ratio (APR), and chi-square statistic (CHI) on the basis of
measured data and the International Telecommunication Union recommendation (ITU-R)
estimations.
Citation: Owolawi, P. A., and T. J. Afullo (2007), Rainfall rate modeling and worst month statistics for millimetric line-of-
sight radio links in South Africa, Radio Sci., 42, RS6004, doi:10.1029/2006RS003535.

1. Introduction calendar months can then be plotted together, and the

[2] The need to design radio communication systems
to meet performance and availability objectives in any
month is a task that must be accomplished. Since
propagation conditions vary considerably from month
to month and the monthly variability can change signif-
icantly from year to year, the propagation community
came up with the concept of the worst month, as defined
in International Telecommunication Union recommenda-
tion (ITU-R) P.581-2 [International Telecommunication
Union, 1994a].
[3] The worst month concept is relevant to the higher-
grade telecommunications services under which propaga-
tion engineers must design radio systems which ultimately
determine link availability. Worst month propagation data
is determined from extended measurement periods to
ensure that a representation of monthly variability is
observed, particularly for the tails of the propagation
distributions. Once several years of data are available,
the data for the individual calendar months from each year
(e.g., all the individual January data sets) are averaged.
The long-term averaged data for each of the twelve
1
School of Electrical, Electronic and Computer Engineering,
University of Kwazulu-Natal, Durban, South Africa.
Copyright 2007 by the American Geophysical Union.
0048-6604/07/2006RS003535
RS6004
envelope of this set of monthly distributions provides the
worst month distribution [Barclay, 2003]. This approach
is used to determine cumulative distributions of average
1-min rainfall rates for average year (AY) and average
worst month (AWM). Then the relations between them
are derived and studied with the ITU-R recommendation
for the stations in question.
[4] Rainfall rate data for a given time percentage are
required in the prediction of rain attenuation for the time
percentage. In accordance with recommendations ITU-R
P.618-6 [International Telecommunication Union, 1998]
and ITU-R P.530-8 [International Telecommunication
Union, 2000], the distribution of rainfall rate is deduced
from the observed rainfall data with 1-min integration
time [Lee et al., 2002]. Statistics of rainfall rates near and
above the Earths surface are needed in order to estimate
the percentage of time of absorption, or scattering of
radio waves that affect radio system design [Rice and
Holmberg, 1973].
[5] To estimate the percentage of time that such
phenomena are important to line-of-sight systems, the
radio system engineer needs prediction of the cumulative
time statistics of surface rainfall rates for particular areas
of interest. In this paper, the rainfall rate having an
integration time of 60 min is converted to ITU-R
recommended integration time of 1 min by using the
method given by Fashuyi et al. [2006], Segal [1986], and
Ajayi and Ofoche [1983]. Also, it summarizes a large
1 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Figure 1. Seasonal variation of cumulative rain rate distribution for South Africa.

sample of surface rainfall statistics and derives rainfall
rate distribution models for eight different regions in
South Africa. The results are studied with ITU-R models,
both at tropical and temperate regions.
2. Variability of Rainfall Rate Distribution
in South Africa
[6] There are four major seasons in South Africa,
namely, summer, autumn, winter and spring. The sum-
mer period (November March) is characterized by
higher rainfall rate than the other seasons [Fashuyi et
al., 2006; Republic of South Africa, 1989; Tyson et al.,
1976]. In the summer rainfall region, light orographic
rains are common along the windward slope of the
eastern escarpment of South Africa. Over most of the
summer rainfall season, however, violent convection
storms, accompanied by thunder, lightning, sudden
squalls, and often hail, are the source of most of the
rainfall. The period between June and August (the winter
period) is characterized by often long lasting and not
very intense rains except along mountains, where the
orographic effect may induce heavy showers [Republic
of South Africa, 1989]. Between the winter and summer
2 of 11
rainfall regions lies a transitional area where rain comes
in milder measures.
[7] Figure 1 shows four plots of seasonal variations of
rainfall rate, while Table 1 displays the rainfall rate at
0.1% exceedence level for twelve stations with the four
climatological seasons. The four stations were chosen on
the basis of vegetation classification and rainfall distri-
bution behavior. From Figure 1 we observe that the
seasonal variation of cumulative rainfall distribution
for Durban, Bloemfontein, Ulundi, Pietermaritzburg,
Vryheid, and Brandvlei shows a maximum rainfall rate
in summer and a minimum in winter. The opposite is the
case for Cape Town. Newcastle, Ladysmith, and Pretoria
have their maximum rainfall rate in spring (September
October) and minimum rainfall rate in winter. Finally,
both East London and Richards Bay have their maximum
rainfall rate in summer and their minima in spring and
autumn, respectively.
3. Statistics of the Worst Month
[8] Cumulative distributions of 1-min rainfall rate for
average year (AY) and average worst month (AWM) for
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Table 1. Seasonal Variation of Rain Rate for 12 Locations in error (ideally, close to 0). It should also give a c2 statistic
South Africa at 0.1% Exceedence that is lower than the threshold td for a defined
probability value d for N-1 degrees of freedom (DF).
Locations Summer Autumn Winter Spring
Typically, d is chosen to be 1% or 5%. In this work we
Durban 132.94 122.7 82.32 94.4 have chosen d = 1%.
Bloemfontein 129.54 60.07 48.76 93.49 [9] On the basis of seasonal, month-to-month, and
East London 132.09 85.18 88.89 77.65 year-to-year rain rate variability, the concept of the worst
Ulundi 207.55 67.02 62.07 156.19
Richards Bay 182.66 74.80 80.49 148.8
month comes in handy. This defines the worst conditions
Newcastle 124.42 60.07 48.76 175.53 for which propagation engineers must design the radio
Ladysmith 98.93 52.93 40.16 175.53 link, which ultimately determines link availability. To
Pietermaritzburg 199.85 97.12 49.28 123.56 address the issue of variability, ITU propagation working
Pretoria 106.09 64.06 54.99 123.56 groups came up with the concept of worst month, as
Vryheid 98.93 40.16 34.58 84.25
Cape Town 33.44 73.83 86.11 47.70 defined in recommendation ITU-R P.581-2 [International
Brandvlei 76.7 61.07 17.2 52.93 Telecommunication Union, 1994a]. The ITU-R defines
worst month as that the worst month of a year for a
reselected threshold for any performance degrading
mechanism, be that month in a period of twelve
consecutive calendar months for all threshold levels
the twelve stations are obtained by statistical processing [International Telecommunication Union, 1994a]. In
of rainfall rate data. The relationship between the per- another ITU-R recommendation a model for conversion
centage of time of average year (PAY) and average worst of the average annual time fraction of excess to the
month (PAWM) is optimized along with obtained rainfall average annual worst month time fraction of excess is
rate distribution models by using the root-mean-square given [International Telecommunication Union, 2003].
(RMS) error, average probability ratio (APR), and the [10] Figure 2 shows samples of plots of cumulative
chi-square statistic (c2). The RMS error is defined by distribution of monthly variability of average rainfall rate
[Timothy et al., 2002] for four of the 12 sites listed in Table 1. The month-to-
v month variability was observed for monthly cumulative
u
u X N  2 distributions of 1-min rainfall rate. For the percentage of
RMS t1=N Xest;i  Xmea;i 1 time at 0.1% exceedence level the maximum difference
i1 between rainfall rates reaches 184.97 mm/h. The
maximum average monthly rainfall rate observed is
Here N is the number of data points, and Xest and Xmea 211.38 mm/h for the month of February in Ulundi, and the
are the estimated and measured quantities, respectively. minimum average monthly rainfall rate is 7.69 mm/h for
APR is given by [Timothy et al., 2002] the month of June in Newcastle. In a calendar year the
N  month with maximum average monthly rainfall rate
1 X Pest;i
APR 2 spreads between January and March when summer is
N i1 Pmea;i pronounced. The exception is Cape Town which has its
maximum average monthly rainfall rate in August and
Here Pest and Pmea denote the cumulative probabilities of minimum in December. The winter rainfall region is a
estimated and measured quantities, respectively. Both relatively small area along the Capes western and
RMS and APR enable us to optimize the proposed southwestern coasts and has a rainfall regime of Medi-
relations and models while the chi-square (CHI) statistic terranean type with a conspicuous winter maximum. In
is used to confirm the acceptance or rejection of the null addition, the western Cape region gets most of its rainfall
hypothesis of the relations/models in question. The in winter, while the rest of the country is generally a
statistic, c2, is defined as [Downie and Heath, 1965] summer rainfall region. One of the few factors that
 2 contributes to this variation is the striking contrast
X N
Xest;i  Xmea;i
2
c 3 between temperatures on the countrys east and west
i1
Xest;i coasts, due to the warm Agulhas and cold Benguela
currents that sweep the coastlines.
The above equation specifies a probability, or sig- [11] Figure 3 shows sample plots of the ITU-R recom-
nificance level, d. This probability defines, for a given mended rain zone, AY, AWM, and proposed rain zone
application, what constitutes a significant enough devia- for the selected sites. There are two ITU-R recommenda-
tion from expected or modeled behavior to justify tions, ITU-R P.837-1 [International Telecommunication
rejection of the hypothesis or model. A good model Union, 1994b] and ITU-R P.837-4 [International
should give an APR close to unity and a minimum RMS Telecommunication Union, 1994c], examined here.

3 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Figure 2. Cumulative distribution of average monthly rain rate for four stations in South Africa.

Durban and Pretoria showed different rain zones while Table 2. Figure 4 shows samples of modeled and ITU-R
Cape Town and Brandvlei are the same for both ITU-Rs. cumulative distributions of rainfall rates for average year
For example, in Durban the rainfall rates at 0.1% for the and average worst month for four stations. Table 3 shows
ITU-R P.837-1, ITU-R P.837-4, proposed zone, AY, and a comparison between the derived model and ITU-R
AWM are 15 mm/h, 22 mm/h, 65 mm/h, 68.98 mm/h, model percentage of time of AY and AWM. The RMS,
and 22.7 mm/h, respectively. The observed discrepancy APR, and CHI values give the best value for the relations
could lead to poor link performance and availability for in all the sites. The RMS and APR show that the
the site. modeled relations are appropriate for obtaining relations
between percentage of time of AY and AWM. Note
that the threshold td for the chosen d = 1% are as
4. Relations Between AY and AWM follows: td = 112.3 for 80 DF, 100 for 70 DF, 76 for
[12] The rain intensity data collected over a 5-year (a) 50 DF, 63.7 for 40 DF, and td = 50.9 for 30 DF. In Table 3
period were processed to obtain cumulative distributions the value of N ranges between 30 and 80. As expected,
of average 1-min rainfall intensities for AY and AWM. from Table 3 we observe that the AY and AWM models
The obtained dependence of percentage of time of the give a closer fit to the measured data than the ITU-R
average year PAY on percentage of time of the average curves for AY and AWM. This is immediately apparent
worst month PAWM for the same average 1-min rainfall from the CHI and RMS values.
rate is shown by equations (4) and (5) which is validated
by using the correlation coefficient R2. 5. Rain Rate Distribution Models
W
PAY YPAWM % 4 [13] Designing line-of-sight (LOS) or satellite link
systems needs yearly rainfall rate statistics for several
B percentages of time in the locations of interest. Thus it is
PAWM APAY % 5 necessary to estimate, within an average year, the time
percentage during which attenuation due to rain is
Here Y, W, A, and B are coefficients of equations (4) and
(5). The values of the coefficients are shown in the

4 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Figure 3. Plots of International Telecommunication Union recommendation (ITU-R), average

year (AY), average worst month (AWM), and proposed rain zone for South Africa.

significant, and consequently, the future link perfor- [17] The Salonen-Baptista model evolved because the
mance and availability. Rice-Holmberg rainfall rate model is not accurate
[14] There are several rainfall rate climatological models enough for high-availability systems. The Salonen-Bap-
for short integration time and long integration time statistics tista model attempted to find proper functions which
of precipitation. These include the Rice-Holmberg model describe rainfall rate distributions, both in tropical and
[Rice and Holmberg, 1973], the Salonen-Baptista model
[Poiares Baptista and Salonen, 1998], and the Moup-
fouma model [Moupfouma, 1987]. Table 2. Coefficients and Correlation Coefficients Validity
[15] The Rice-Holmberg rainfall rate model contains for Approximations (4) and (5) for Rain Rate Range 0.01  R
three basic parameters, namely [Rice and Holmberg,  212
1973] (1) the average annual rainfall, Mr, in mm;
(2) the ratio of thunderstorm rain to total rain, b; and Locations Y W A B R2
(3) the annual average number of days with rainfall
Durban 0.2217 0.9593 4.7465 1.0353 0.9931
0.25 mm, D. Bloemfontein 0.2188 0.9789 4.7097 1.0197 0.9981
[16] The ratio of thunderstorm rain to total rain b is not East London 0.2542 0.9978 3.9358 1.0007 0.9985
readily available from meteorological databases. Thus Ulundi 0.2252 0.9969 4.0414 1.0016 0.9985
Dutton et al. [1974] have presented a calculation method Richards Bay 0.2593 0.9631 4.0561 1.0373 0.999
Newcastle 0.2201 1.030 4.3085 0.9662 0.9958
for b from three parameters, which are available from Ladysmith 0.2448 0.8946 4.7797 1.1125 0.9952
long-term meteorological observations, namely (1) max- Pietermaritzburg 0.2635 0.9484 4.0414 1.0481 0.9985
imum monthly rainfall rate in 30 a; (2) average number Pretoria 0.2298 0.9848 4.4291 1.0128 0.9974
of thunderstorm days in a year; and (3) average number Vryheid 0.2161 0.963 4.8787 1.0357 0.9973
of rainy days in a year. Cape Town 0.2183 1.0197 4.4329 0.979 0.9983
Brandvlei 0.2016 1.0014 4.9488 0.9986 1.0000

5 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Figure 4. Plots of modeled and ITU-R cumulative distributions of rain rate for AY and AWM for
four sites in South Africa.

midlatitude climates. It was found that the probability rainfall rate cumulative distribution, which is accepted by
that the rainfall rate r is exceeded could be described well ITU-R for the prediction of rainfall rate cumulative
with the function [Poiares Baptista and Salonen, 1998] distribution for all climates. Provided the rainfall rate
1kr
R0.01 (mm/h) exceeded during 0.01% of time is known,
Pr Po e8r 1cr 6 Moupfouma showed that the prediction procedure of
rainfall rate cumulative distribution can be expressed as
The next step was to develop empirical functions to [Ajayi et al., 1996]
calculate parameters 8, k, and c, and the rain probability

P0 from the statistical meteorological data. This model is R0:01 1 b1 u1 ao
simple but not accurate for all locations because of a P R  r e 8
r1
limited data set, as well as its dependence on latitudes.
These factors make it unsuitable for our analysis. where ao = (R0.01 + 1)  loge(104) and r (mm/h)
[18] Moupfouma and Martin [Moupfouma and Martin, represents the rain rate exceeded during the fraction of
1993] improved on the weaknesses of the mentioned the time P. The parameter b1 is well approximated by the
models and came up with the following simple model following analytical expression [Ajayi et al., 1996]:
[Moupfouma and Martin, 1993]:  
r  R0:01 1r
eur b1 log 9
P R  r a ; r  2 mm=h 7 R0:01 R0:01
rb
The u1 value is analyzed by ITU-R Study Group 3 on the
with the parameters a, b, and u depending on the basis of available rain rate cumulative distribution data
integration time of the rain gauges used as well as the which leads to the following two relationships, depend-
climatology of the location. Moupfouma and Martin ing on the climate:
[1993] then proposed a new prediction procedure of

6 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Table 3. Comparison Between the Models, Measured Data, and ITU-R Cumulative Distributions of Rain Intensities for AY and
AWM for Eight Stations in South Africaa
Sites Model AY Model AWM ITU-R AY ITU-R AWM

RMS Durban 0.031 0.12 0.37 0.64

APR 1.007 0.95 0.86 0.95
CHI 0.24 3.75 8.35 8.82
RMS Bloemfontein 0.005 0.028 0.28 0.48
APR 1.029 1.034 1.041 0.885
CHI 1.029 0.031 3.92 2.57
RMS East London 0.055 0.23 1.015 1.52
APR 1.01 1.024 1.15 1.088
CHI 0.021 0.088 1.85 1.20
RMS Newcastle 0.004 0.022 0.138 0.22
APR 1.015 0.99 1.019 1.05
CHI 0.0085 0.04 2.45 1.36
RMS Pietermariztburg 0.016 0.081 0.22 0.43
APR 1.021 1.024 1.085 1.086
CHI 0.059 0.29 3.91 4.07
RMS Pretoria 0.0076 0.035 0.17 0.28
APR 1.017 1.019 1.048 0.971
CHI 0.021 0.095 3.066 1.92
RMS Cape Town 0.0083 0.038 0.14 0.23
APR 0.995 1.011 1.0124 1.242
CHI 0.009 0.041 1.56 0.92
RMS Brandvlei 0.0002 0.0008 0.029 0.045
APR 1.034 1.033 1.066 1.091
CHI 0.025 0.088 1.85 1.20
a
ITU-R is International Telecommunication recommendation. APR is average probability ratio. AY is average year. AWM is average worst
month. RMS is root-mean-square. APR is average probability ration. CHI is chi-square statistic.

[19] 1. For the tropical localities [Dutton et al., 1974], parameters approach of equation (7) was used to com-
"  # pute the distributions for the eight selected sites. The
loge 104 r Y results provided in Table 4 are a good fit of the rainfall
u1 exp l 10
R0:01 R0:01 rate probability distribution function observed in various
sites in South Africa. The values of constants a, b, and u,
Here l and Y positive are constants such that l = 1.066 shown in Table 4, were used in equation (7) to plot
and Y = 0.214. cumulative distributions for the eight sites, as displayed
[20] 2. For the temperate zones [Dutton et al., 1974], in Figure 5. The figure shows rainfall rate statistics for
2 3 selected sites in South Africa based on the models.
 4
[22] Mathematically, the derived single function em-
loge 10 6 1
u1 4  t 7 5 11 pirical model is as follows. If
R0:01 r
1 h R0:01 F r P R  r 12

Here h and t are positive constants such that h = 4.56

and t = 1.03. Table 4. Estimated Values of a, b, and u for Eight Regions in
South Africa (2000 2004)
6. Rain Rate Distribution Models for Name a b u
Regions in South Africa Pietermaritzburg 0.014 0.7700 0.021
East London 0.017 1.529 0.021
[21] Rainfall rate measurements obtained for eight Bloemfontein 0.81 1.390 0.02018
regions in South Africa over a period of 5 a were Brandvlei 0.016 0.769 0.022
processed to derive regional empirical models at 1-min Cape Town 0.013 0.98 0.0276
integration time, as supported by ITU-R [International Newcastle 0.85 0.770 0.017
Radio Consultative Committee (CCIR), 1982; Interna- Pretoria 0.0117 0.768 0.0246
Durban 0.0115 0.770 0.0247
tional Telecommunication Union, 1997]. The three

7 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Figure 5. Rain statistics for selected regions in South Africa based on the proposed model.

is the distribution function of the rainfall rate r, it follows for radio wave attenuation prediction for both terrestrial
that line-of-sight and Earth-space links [International Radio
Consultative Committee (CCIR), 1982; International
F r 1  P R  r 13 Telecommunication Union, 1997]. The parameter b is
plotted against the value of rainfall rate exceeded 0.01%
 ur  of time for each location; the corresponding points can
1  a erb r>0
F r 14 be fitted by the relationships obtained for all the com-
0 r0 puted locations given in Table 3. By using linear regres-
Therefore the model of equation (7) has the following sion for different R0.01(mm/h) the values of b for
probability density function: different locations are obtained for South Africa
 eur    [Owolawi, 2006].
dF a rb u br r > 0 [27] The results show fairly good agreement with the
r f r 15 ITU-R (Moupfouma) model in some locations that have
dr 0 r0
high rain intensities, and very good agreement with
For equations (12) to (15) to hold, the follow conditions location that have low rain intensities, e.g., Brandvlei.
must be obeyed: It may be seen also that for higher rain rate (above
[23] 1. F(+1) = 1; 100 mm/h), there is a rather sizable difference between
[24] 2. u > 0 and b > 0; the predicted results and experimental values. This is not
[25] 3. As F(r) is always an increasing function, f(r) is surprising since the accuracy of the model proposed here
always positive, and therefore a > 0, since u > 0 and b > 0. depends on the conversion factor of the rain data,
especially for the high rain rate for which measurement
inaccuracies may occur [Moupfouma and Martin, 1993].
7. Evaluation of the b Parameters for All Once b is fixed, parameter a can be determined knowing
Regions any rain rate value R0.01 (mm/h) for any given time
percentage. Then using R0.01(mm/h) and substituting into
[26] The International Radio Consultative Committee
equation (15) one obtains
(CCIR) plenary assembly held in Geneva in 1982 gave
contour maps of rainfall intensity R0.01 (mm/h) exceeded  uR0:01
e
for 0.01% of the time for different hydrometeorological P R  R0:01 a 0:01% 16
Rb0:01
regions of the world [Owolawi, 2006; Tseng et al., 2005].
These values are highly interesting since they are used
8 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Figure 6. Plots of measured, proposed, and ITU-R models for four stations in South Africa.

Thus we have
 
a 104 Rb0:01 euR0:01 17
8. Plots of the Proposed and ITU-R Models
[28] Figure 6 shows plot of ITU-R rain rate values (for
both temperate and tropical regions), the measured data,
and the proposed model for each of the selected eight
stations. The statistical tools used to optimise the models
are the RMS, the APR, and the c2 statistic. Table 5
shows statistical study of ITU-R, measured, and pro-
posed models for the eight sites. As discussed before, the
c2 threshold values for the 30 80 degrees of freedom
used here vary from 50 to 112. Therefore a very good
agreement is observed between the proposed empirical
rain rate models and the measured data. We also observe
a fair agreement with the ITU-R tropical zone model.
This is so because most of South African locations are
subtropical [Fashuyi et al., 2006; Republic of South
Africa, 1989; Tyson et al., 1976]. However, the excep-
tions are found in Cape Town and East London, in which
the ITU-R temperate model is better than the tropical
model. In all cases, however, it is seen that the proposed
regional models perform better than the two ITU-R
models.
9. Conclusion
[29] In this study, the seasonal variability has been
investigated. It is observed that the rainfall season in
western Cape is during the winter period while the rest of
the country experiences their rainfall season during the
summer. This is because of the effect of temperature
contrast on the coasts of South Africa. Also, being in the
Southern Hemisphere, the seasons stand in opposition to
those of Europe and North America. As a result of
seasonal variability, the concept of worst month is
employed, and the relationship between the percentage
of time PAY and PAWM has been developed.
[30] The empirical model given by equation (7) with
the parameters a, b, and u displayed in Table 2 appears to
be more suitable for estimating rain intensity distribu-
tions for South Africa over the existing global ITU-R
rain rate distribution model. This is confirmed by the
RMS, APR, and CHI statistics, as observed from the plot
of the ITU-R, measured data, and empirical model. The
close agreement between the empirical model and the

9 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Table 5. Comparison Between Measured Data With ITU-R Rep. 721-1, in Documents of the XVth Plenary Assembly,
and Proposed Models for Eight Sample Stations in South Africa Doc. 5/1029/E (Sept. 28, 1981), Geneva, Switzerland.
International Telecommunication Union (1994a), Recommenda-
Site Model RMS Error APR c2
tion 581-2, The Concept of Worst Month, Geneva, Switzerland.
Durban ITU-R temperate 0.51 0.85 16.46 International Telecommunication Union (1994b), Recommen-
ITU-R tropics 0.22 0.76 4.96 dation 837-1, Characteristics of Precipitation for Propaga-
proposed model 0.098 0.88 1.011 tion Modeling, Geneva, Switzerland.
Bloemfontein ITU-R temperate 0.19 0.71 3.76 International Telecommunication Union (1994c), Recommen-
ITU-R tropics 0.17 0.70 3.45
proposed model 0.14 1.09 1.36 dation 837-4, Characteristics of Precipitation for Propaga-
East London ITU-R temperate 0.12 0.61 3.15 tion Modeling, Geneva, Switzerland.
ITU-R tropics 0.21 0.62 5.11 International Telecommunication Union (1997), Recommen-
proposed model 0.14 0.99 1.50 dation 618-5, Propagation Data and Prediction Methods
Newcastle ITU-R temperate 0.29 0.80 8.71
Required for the Design of Earth-Space Telecommunications
ITU-R tropics 0.87 0.69 1.49
proposed model 0.042 0.83 0.61 Systems, Geneva, Switzerland.
Pietermaritzburg ITU-R temperate 0.39 0.72 12.34 International Telecommunication Union (1998), Recommenda-
ITU-R tropics 0.19 0.60 4.98 tion 618-6, Propagation data and prediction methods required
proposed model 0.024 1.062 0.090 for the design of Earth-space telecommunication systems, in
Pretoria ITU-R temperate 0.21 0.59 5.44
ITU-R tropics 0.19 0.54 4.86 Propagation in Non-Ionized Media, pp. 1 26, Geneva,
proposed model 0.027 1.18 0.14 Switzerland.
Cape Town ITU-R temperate 0.104 0.67 1.76 International Telecommunication Union (2000), Recommenda-
ITU-R tropics 0.16 0.73 2.35 tion 530-8, Propagation data and prediction methods re-
proposed model 0.031 1.01 0.52
quired for the design of terrestrial line-of-sight systems,
Brandvlei ITU-R temperate 0.083 0.74 1.28
ITU-R tropics 0.030 1.079 0.186 Geneva, Switzerland.
proposed model 0.007 1.078 0.025 International Telecommunication Union (2003), Recommenda-
tion 841-3, Conversion of Annual Statistics to Worst Month
Statistics, Geneva, Switzerland.
Lee, J.-H. et al. (2002), Conversion of rain rate distribution for
measured data confirms the acceptability of the proposed various integration time, paper presented at the XXVIIth
empirical model for accurate rainfall rate distribution General Assembly of the International Union of Radio
prediction for radio system designer in South Africa. Science, Maastricht Exhib. and Congr. Cent. (MECC),
Maastricht, Netherlands, 17 24 August.
References Moupfouma, F. (1987), More about rainfall rate and their pre-
diction for radio system engineering, IEE Proc., 134, 527
Ajayi, G. O., and E. B. C. Ofoche (1983), Some tropical rainfall 537.
rate characteristics at Ile-Ife for microwave and millimeter Moupfouma, F., and L. Martin (1993), Point rainfall rate cumu-
wave application, J. Clim. Appl. Meteorol., 23, 562 567. lative distribution function valid at various locations in the
Ajayi, G. O., S. Feng, S. M. Radicella, and B. M. Reddy (Eds.) world, Electron. Lett., 29(7), 1503 1504.
(1996), Handbook on Radiopropagation Related to Satellite Owolawi, P. A. (2006), Rain rate and rain drop size distribution
Communications in Tropical and Subtropical Countries, models for line-of-sight millimetric systems in South Africa,
pp. 7 14, ICTP, Trieste, Italy. M. Sc. diss., Univ. of Kwazulu-Natal, Durban, South Africa.
Barclay, L. (Ed.) (2003), Propagation of Radiowaves, 2nd ed., Poiares Baptista, J. P. V., and E. T. Salonen (1998), Review of
chap. 4, pp. 47 72, Inst. of Electr. and Electron. Eng., Lon- rainfall rate modeling and mapping, paper presented at the
don, UK. URSI-F Open Symposium, CLIMPARA98, Ottawa, Canada,
Downie, N. M., and R. W. Heath (1965), Basic Statistical 27 29 June.
Methods, 2nd ed., pp. 160 175, Harper and Row, New Republic of South Africa (1989), South Africa 1989 1990:
York. Official Yearbook of the Republic of South Africa, 15th ed.,
Dutton, E. J., H. T. Dougherty, and R. F. Martin (1974), Pre- pp. 1 22, Bur. for Inf., Dept. of Foreign Affairs, Pretoria,
diction of European rainfall and link performance coeffi- South Africa.
cients at 8 To 30 GHz, Rep. AD/A-000804, Natl. Tech. Inf. Rice, P. L., and N. R. Holmberg (1973), Cumulative time sta-
Serv., U.S. Dept. of Comm., Springfield, Va. tistics of surface-point rainfall rates, IEEE Trans. Commun.,
Fashuyi, M. O., P. A. Owolawi, and T. J. Afullo (2006), Rain- 21, 1131 1136.
fall rate modeling for LOS radio systems in South Africa, Segal, B. (1986), The influence of raingage integration time on
Trans. S. Afr. Inst. Electr. Eng., 97(1), 74 81. measured rainfall-intensity distribution functions, J. Atmos.
International Radio Consultative Committee (CCIR) (1982), Oceanic Technol., 3, 662 671.
Attenuation by precipitation and other atmosphere particles,

10 of 11
RS6004 OWOLAWI AND AFULLO: RAINFALL RATE AND WORST MONTH STATISTICS RS6004

Timothy, K. I., J. T., Ong, and E. B. L. Choo (2002), Raindrop
size distribution using method of moments for terrestrial and
satellite communication applications in Singapore, IEEE
Trans. Antennas Propag., 50(10), 1420 1424.
Tseng, Chi-Huei, Chen Kun-Shan and Chu Chih-Yuan (2005),
Variability analysis of Ka band rain attenuation in Taiwan,
paper presented at Progress in Electromagnetics Research
Symposium 2005, Zhejiang Univ., Hangzhou, China,
23 26 August.
Tyson, P. D., R. A Preston-Whyte, and R. E. Schulze (1976),
The Climate of the Drakensberg, Natal Town and Reg.
Plann. Rep. 31, pp. 50 60, Town and Reg. Plann. Comm.,
Pietermaritzburg, South Africa.

T. J. Afullo and P. A. Owolawi, School of Electrical,
Electronic and Computer Engineering, University of Kwazulu-
Natal, Durban 4000, South Africa. (afullot@ukzn.ac.za)

11 of 11