You are on page 1of 9

Frequency Analysis in Precipitation Data

Exercise No. 3

David, Dianna Rose N.


Diaz, Rianne May M.
Gumaru, Jherdie Christopher D.
Nequinto, Maria Jose Aubrey A.
San Jose, Jose Mari E.

A laboratory report submitted


in partial fulfillment of the requirements in

ABE 71 C-1L

March 11, 2016


I. INTRODUCTION
Frequency analysis is an aid in determining the design discharge and design
rainfall. It is also used to calculate the frequency of other hydrologic (or even non-
hydrologic) events. In hydrology, frequency analysis is done to determine magnitude and
probability of occurrence of a given hydrologic event. This probability of occurrence and
magnitude are being correlated with each other to obtain a certain graph called the
frequency curve. The frequency curve is a graphic representation that shows the relative
performance of data being measured against each other. This frequency curve is also used
by engineers when designing infrastructures in relation with hydrology. Before coming
up with a frequency curve, data gathering or selection and determining statistical
parameters are done. Then, fitting of data with the use of different statistical distribution
and testing the significance of the results are what follow.
In performing frequency analysis and creating a frequency curve, different
programs can be used example of which is HEC-SSP. This software is developed by the
Hydrologic Engineering Center. It allows users to perform statistical analyses of
hydrologic data and can also be used in performing flood flow frequency analysis.
The applicability of frequency methods to the study of floods has been widely
recognized by numerous researchers in the field. Yet many others tend to criticize them
on the following grounds: (a) insufficient records on which to base any extrapolation; (b)
ignorance of the statistical laws on which extrapolations may be based; (c) homogeneity
of records (e.g. recorded flow may be the result of various incoherent climatological
mechanisms); (d) climatological changes over long periods (Wolf, 1966)

II. OBJECTIVES
Using the given data of one-day annual maximum rainfall, fitting of data and
testing of the significance of the results were done. Through this, the differences in the
three statistical distributions: Normal, Log, Normal and Log Pearson III Distributions
were observed. Also, the significance of the results had been obtained.

III. METHODOLOGY
One-day maximum rainfall data gathered by the UPLB-NAS using an 8-inch
standard rain gauge from year 1947-2014 were used in frequency data analysis. A
frequency curve with the line of best fit was determined using the three statistical
distributions: Normal, Log-Normal and Log-Pearson Type III, by Weibull Formula and by
eyeball estimation. Having the line of best fit, the magnitudes of the rainfall in year 2, 5,
10, 15, 20, 25 and 100 were obtained. The magnitudes from the 68-year period were
compared to the magnitude of a 17-year period (1998-2014) which has been analyzed
with the Gumbel Method by the Hydrometeorology Division of PAGASA. This data from
the 17-year period was also been analyzed using the Gumbel Method and had been
compared to the UPLB data from 1998-2014 with the same sets of return period.
IV. RESULTS AND DISCUSSION
Using HEC-SSP, different frequency curves with the Normal, Log Normal and Log
Pearson III distribution methods were obtained. Among the three distribution methods
performed for this exercise, the Log Pearson III Distribution as the most fit among them.
Looking at the graph of the General Frequency Analysis for Log Pearson III Distribution
(please refer to figure 3.1), it can be observed that the data points for the accounted years
were following the line of best fit. This can be further verified by comparing the graph of
the General Frequency Analysis for Normal Distribution and Log Normal Distribution in
the Appendix. From this line of best fit the magnitudes of 100-year and 2-year events
were obtained to compute for R, which was used to determine the minimum acceptable
number of years of record using the plot of the Mockus’ equation. It had been found out
that it was equivalent to 12 years.

Figure 3.1 General Frequency Analysis for Log Pearson III Distribution

In addition, with the frequency line of best fit, the rainfall magnitude in the return
period of 2, 5, 10, 15, 20, 25 and 100 years had been determined and summarized in table
3.1.

Table 3.1. Magnitudes of Corresponding Return Period from the Frequency Line of Best
Fit
RETURN
2 5 10 15 20 25 100
PERIOD
RAINFALL
175 240 270 280 285 290 350
(mm)
Referring to the 68 years of data, not a single typhoon was considered an extreme
event having a return period of at least 100 years since the greatest magnitude of
precipitations are 321.6, 308, and 305.8 mm which occurred in the years 1964, 2006, and
1960 respectively. Even in the years 2005 to 2014 where the 308 mm is the greatest
precipitation that occurred, no rainfall is considered an extreme event but are just
classified as normal strong events.
Given by table 3.2 the 24-hour maximum rainfall synthesized by the
Hydrometeorology Division of PAGASA from the years 1998 to 2014, the data is
compared to the values obtained in the frequency line of best fit.

Table 3.2. Computed extreme values of 24-hour maximum rainfall from PAGASA RIDF
curve
RETURN
2 5 10 15 20 25 100
PERIOD
RAINFALL
155.7 216.7 257.1 279.8 295.8 308.1 383.5
(mm)

It has been observed that in the return periods 2, 5, 10, and 15, the magnitudes
are greater than that of PAGASA’s then in return periods 20, 25, and 100, the magnitudes
became lower than that of PAGASA’s. The greatest difference in magnitude can be found
at the return period of 100 where the magnitude in the frequency line of best fit is 33.5
mm less than of PAGASA’s RIDF curve. The data from PAGASA are not representative of
the computed magnitudes of the 68-year event from the frequency line of best fit due to
the observed significant differences, but the magnitudes obtained with the use of Gumbel
Method which are presented in table 3.3 had small differences with the data synthesized
by PAGASA. These magnitudes were better than magnitudes from the frequency line of
best fit. Some were overestimate, but some were closer to the data from PAGASA. With
all these comparisons, the latter was more acceptable. Gumbel method was a better
method to use in computing for the magnitudes of a corresponding return period.

Table 3.3 Magnitudes of Corresponding Return Period from Gumbel Method


RETURN
2 5 10 15 20 25 100
PERIOD
RAINFALL
180.6 228.0 263.8 284.8 299.7 311.2 382.8
(mm)

With the same set of data from table 3.2 and the UPLB data from year 1998-2014,
comparison had been done with the use of Gumbel Method. As observed, all the
magnitude from both data really differ. These differences were significant because they
are observed in all the return period. And based on the Mockus plot, the 17 years of record
was adequate. Thus, the differences were really significant.

Table 3.3 Computed magnitudes of UPLB data from year 1998-2014 with the same
return period
RETURN
2 5 10 15 20 25 100
PERIOD
RAINFALL
190.4 240.6 278.2 300.2 315.8 327.9 403.1
(mm)

Comparing the UPLB data from 1998-2014 to the same data from 1947-2014, it
was observed that the short records were overestimate of the long records. It was
concluded that it was better to analyze the data from 68-year event and to compare it to
the data from PAGASA.

V. SUMMARY AND CONCLUSION


Flood frequency analysis had been carried out for the One-day maximum rainfall
totals from UPLB-NAS. The outcome of the analysis clearly reveals the good capability of
the Normal, Log-Normal, and Log-Pearson Type III function to plot the events with
probability and predict the magnitude of rainfall associated with the desired return
periods of 2, 5, 10, 15, 20, 25, and 100 years. (Table 3.1). Among the three distribution
methods performed for this exercise, the group chose the Log Pearson III Distribution as
the most fit among them through from the graph generated from the HEC-SSP.
Using the Gumbel method for the same sets of return periods, the magnitude of
events was compared. The Gumbel method has been simplified in such a manner that one
can obtain the magnitude of a given return period flood without recourse to looking at a
table and working out the value of the coefficient of variation of the given data. The results
obtained by the simplified version are compared with those obtained from using both the
original approach and those from modification of Gumbel’s method.
Carrying out Frequency Analysis of Precipitation Data using the different
distribution methods is very much important in terms of predicting reliable flood
frequency estimates are vital for floodplain management; to protect the public, minimize
flood related costs to government and private enterprises, for designing and locating
hydraulic structures and assessing hazards related to the development of flood plains
(Tumbare, 2000).

VI. REFERENCES
Tumbare, M. J. 2000. Mitigating floods in Southern Africa. Paper presented at the 1st
WARSFA/WaterNet Symposium: Sustainable Use of Water Resources, 1-2
November, Maputo
Wolf, P.O. (1966) Comparison of Methods of Flood Estimation. Paper 1, Session A,
Proceedings of the Symposium organized by the Institution of Civil Engineers on
River Flood Hydrology, pp. 1-23.

VII. APPENDIX
Table 3.4 One-day maximum rainfall data from the 8-inch standard rain gauge at UPLB-
NAS
Rank P (%) MAGNITUDE (MM) LOG (magnitude)
1 1.449275 321.6 2.507316
2 2.898551 308 2.488551
3 4.347826 305.8 2.485437
4 5.797101 299.8 2.476832
5 7.246377 293.2 2.467164
6 8.695652 288.3 2.459845
7 10.14493 276.6 2.441852
8 11.5942 267.7 2.427648
9 13.04348 257.8 2.411283
10 14.49275 254.6 2.405858
11 15.94203 253 2.403121
12 17.3913 249.4 2.396896
13 18.84058 232.8 2.366983
14 20.28986 232.2 2.365862
15 21.73913 228.7 2.359266
16 23.18841 217.9 2.338257
17 24.63768 212.1 2.326541
18 26.08696 211.3 2.324899
19 27.53623 209.8 2.321805
20 28.98551 203.2 2.307924
21 30.43478 200.8 2.302764
22 31.88406 196.5 2.293363
23 33.33333 196 2.292256
24 34.78261 195.7 2.291591
25 36.23188 195.4 2.290925
26 37.68116 195.3 2.290702
27 39.13043 190.3 2.279439
28 40.57971 187.2 2.272306
29 42.02899 185.4 2.26811
30 43.47826 182.9 2.262214
31 44.92754 180.3 2.255996
32 46.37681 178.3 2.251151
33 47.82609 173.2 2.238548
34 49.27536 166.9 2.222456
35 50.72464 166.1 2.22037
36 52.17391 158.2 2.199206
37 53.62319 157.8 2.198107
38 55.07246 157.7 2.197832
39 56.52174 154.4 2.188647
40 57.97101 150.6 2.177825
41 59.42029 150.5 2.177536
42 60.86957 149.9 2.175802
43 62.31884 145 2.161368
44 63.76812 144.5 2.159868
45 65.21739 143.6 2.157154
46 66.66667 143.5 2.156852
47 68.11594 141.5 2.150756
48 69.56522 137.9 2.139564
49 71.01449 137.2 2.137354
50 72.46377 136.6 2.135451
51 73.91304 135.4 2.131619
52 75.36232 135.1 2.130655
53 76.81159 133.4 2.125156
54 78.26087 116.3 2.06558
55 79.71014 115.1 2.061075
56 81.15942 114.6 2.059185
57 82.6087 103.9 2.016616
58 84.05797 103.1 2.013259
59 85.50725 101.6 2.006894
60 86.95652 98.2 1.992111
61 88.4058 93 1.968483
62 89.85507 86.6 1.937518
63 91.30435 83.8 1.923244
64 92.75362 80.6 1.906335
65 94.2029 73.2 1.864511
66 95.65217 66.6 1.823474
67 97.10145 62.2 1.79379
68 98.55072 47 1.672098
MEAN (M) 174.5985 2.207654
STANDARD DEVIATION
(S) 66.28771 0.181793
SKEWNESS (g) 0.332141 -0.63534

Table 3.5 Results of Normal Distribution


VALUES (MM) P (%)
M+S 240.8862 15.9
M 174.5985 50
M -S 108.3108 84.1

Table 3.6 Results of Log Normal Distribution


VALUES (MM) P (%)
MxS 245.1583 15.9
M 161.3072 50
M/S 106.1356 84.1

Table 3.7 Results of Log Pearson III Distribution


P (%) k X (MM)
99 -2.77938 50.3943
50 0.105008 168.5557
20 0.857 230.914
10 1.193992 265.8968
4 1.513864 303.9929
2 1.699856 328.606
1 1.853848 350.4856
0.5 1.984194 370.1401

Figure 3.2 General Frequency Analysis for Normal Distribution

Figure 3.3 General Frequency Analysis for Log Normal Distribution

You might also like