IOP Conference Series: Earth

and Environmental Science

PAPER • OPEN ACCESS You may also like

- Drought-induced regime shift and
Exploring rainfall variabilities using statistical resilience of a Sahelian ecohydrosystem
Valentin Wendling, Christophe Peugeot,
functional data analysis Angeles G Mayor et al.

- Simulation-based inference on virtual brain

models of disorders
To cite this article: N A Mazelan and J Suhaila 2023 IOP Conf. Ser.: Earth Environ. Sci. 1167 012007 Meysam Hashemi, Abolfazl Ziaeemehr,
Marmaduke M Woodman et al.

- Functional Data Analysis for Extracting the

Intrinsic Dimensionality of Spectra:
Application to Chemical Homogeneity in
View the article online for updates and enhancements. the Open Cluster M67
Aarya A. Patil, Jo Bovy, Gwendolyn Eadie
et al.

This content was downloaded from IP address [Link] on 26/11/2024 at 12:36

STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

Exploring rainfall variabilities using statistical functional data

analysis

N A Mazelan1 and J Suhaila1,2*

1
Department of Mathematical Sciences, Faculty of Science, Universiti Teknologi
Malaysia, 81310, Johor Bahru, Johor, Malaysia
2
UTM Centre for Industrial and Applied Mathematics, Ibnu Sina Institute for
Scientific and Industrial Research, Universiti Teknologi Malaysia, 81310, Johor
Bahru, Johor, Malaysia
*Corresponding E-mail: suhailasj@[Link]

Abstract. Functional data analysis (FDA) has been widely applied in various scientific
fields, including climatological, hydrological, environmental, and biomedical. The
flexibility of the FDA in incorporating temporal elements into the statistical analysis
makes the method highly demanded compared to the conventional statistical approach.
This study introduces FDA methods to investigate the variations and patterns of rainfall
throughout Peninsular Malaysia, which includes 16 rain gauge stations in Peninsular
Malaysia from 1999 to 2019. A descriptive statistic of the functional data depicted the
mean and variation of the rainfall curve over time, while the functional principal
component analysis measured the temporal variability of the rainfall curve. According
to the findings, the first and second principal components accounted for 87.4% of all
variations. The first principal component was highly characterised by the stations over
the eastern region during the northeast monsoon since the highest variability was
observed from November to January. On the other hand, the stations impacted by the
inter-monsoon season were best described by the second principal component. Based
on the factor scores derived from the functional principal component, those rain gauge
stations with comparable features were then clustered. Overall, the results showed that
the rainfall pattern is strongly influenced by their geographical and topographical
features and the seasonal monsoon effect.
Keywords: Functional data analysis; Functional principal component analysis; Rainfall
variability.

1. Introduction
Rainfall events are typically defined by a finite number of discrete values that comprise multivariate
data that may or may not contain all of the information accessible in a smoothing curve. A modern
statistical method such as functional data analysis (FDA) has been used to capture the intriguing patterns
that can be found in rainfall data. The FDA technique is used to convert discrete data into a smoothing

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

curve or function. The FDA builds up a smoothing curve that can be examined at any time interval from
a series of observations at a discrete time interval. The smoothness and differentiability of the smoothing
curve of functional data distinguish it from conventional multivariate data. The conventional
multivariate approach fails to capture the underlying process in rainfall data. If a functional concept is
ignored, then additional information related to the functional nature of the rainfall dataset is not properly
utilized.
Functional data analysis techniques have been successfully carried out by several researchers in
various fields [1–6]. Suhaila and Yusop [7] used a functional analysis of variance to explain the
geographical and temporal variability in rainfall. Their study brought attention to the variations and
distinctions in rainfall profiles. On the other hand, Hael et al. [8] used FDA on projected data in the Taiz
region to analyse the temporal variations which help planners manage rainwater conservation.
Functional principal component analysis (FPCA) is the most popular tool in FDA. It is used to reduce
the dimensionality of functional data, capturing the largest amount of variance and exhibiting various
aspects of the underlying data. The FPCA has been used to interpret lactate curves [9], analyse kinematic
data [10], gene classification [11], explore variations in glomerular filtration rate curves [12], and model
aircraft degradation [13]. Suhaila [14] recently applied the FPCA techniques to determine the variations
of the multivariate El Niño Southern Oscillation Index (MEI). Their findings suggested that the two
principal components likely represented the variations of El Niño and La Niña episodes.
The conventional statistical method allows the expression of data variability as either a single value
or as multivariate values that indicate the standard deviation or the coefficient of variation. However,
there are no explanations for the temporal aspects. Therefore, this study intends to evaluate the changes
and variabilities of rainfall data in terms of smoothing curves via functional data analysis. The time when
the high or low variability occurrences took place will be analysed based on the smoothing curves. The
FDA method allows for the assessment of temporal aspects. One could argue that the FDA is more
willing to provide information that isn't accessible through conventional statistical techniques.
Additionally, it is simpler to comprehend rainfall behaviours and their temporal variation when rainfall
variabilities are visualised in smoothing curves. FDA is a relatively new technique in climate studies.
The results of this study should give a thorough analysis of the rainfall pattern and aid in future
forecasting.

2. Data and Study Area

The daily rainfall data at 16 rain gauge stations was obtained from the Malaysian Meteorological
Department from 1999 to 2019. Based on their geographical locations, all rain gauge stations were
divided into the Northwest, Eastern, Central, Southwest, and Western regions. The climate of Peninsular
Malaysia is significantly influenced by the monsoons, particularly the Southwest Monsoon (SWM),
which runs from May to August, and the Northeast Monsoon (NEM), which runs from November to
February. The two inter-monsoons, from March to April and September to October, are also predicted
to bring heavy rains. During the NEM season, the eastern region frequently receives more rain than those
sheltered by the Main Range Titiwangsa. The Main Range Titiwangsa divides Peninsular Malaysia's
eastern and western parts over 483 km. The locations of the stations under study are shown in Figure 1.

2
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

Figure 1. The location of the studied stations.

3. Methodology
In this section, the idea of constructing a series of basis functions to transform discrete rainfall data into
a smoothing curve will be presented. The variation in the rainfall data will next be investigated using a
functional principal component analysis and a functional summary of functional data with curve
visualization.

3.1. Smoothing Rainfall Data

Consider observed discrete rainfall data, Xi = ( x1 ( t1 ) , x2 ( t2 ) ,..., xi ( tT ) ) for i = 1,2,…,n; j=1,2,…,T
where T =365 days, n is the number of rain gauge stations and xi (t j ) is the amount of rain at day tj of
the i-th stations. These values are converted into smoothing curves yi (t ) which can be expressed
linearly as a combination of the basis function;
K

yi ( t ) =  d k k ( t ) (1)
k =1

where dk is the basis coefficient, Bk (.) is the known basis function, and K denotes the maximum size of
basis functions. The periodic function is represented as a Fourier basis which can be written as

ŷi ( t ) = d o + d1 sin ( t ) + d 2 cos ( t ) + d 3 sin ( 2t ) + d 4 cos ( 2t ) + ... (2)

which is defined by the basis 0 (t ) = 1, 2r −1 (t ) = sin (rt ), 2r (t ) = cos(rt ). The constant  is
related to the period T by the relation  = 2 T . By minimizing the sum of squared residuals, the
expansion of the coefficients dk are calculated using the least squares approach, as in
SSE =  x − y (t ( )) ,i = 1, 2,..., n
2
T
(3)
i j j
j =1

The errors of the curves can be minimized by using the penalized sum square of error (PSSE), which
is given as
PSSE =   x j − y (t j ) +    D 4 y ( t )  dt
2 2
(4)
 j

where  is the parameter for smoothing while  D 4 y ( t )  measuring the curvature in y at t.

2

3
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

The smoothing parameter  will be chosen based on the minimum generalized cross-validation
(GCV),
 n  SSE 
GCV ( ) = (5)
 n − df ( )  n − df ( ) 
  

where df ( ) are the equivalent degrees of freedom of the smoothing operator.

3.2 Summary of Functional Data

The mean and variance function of curves are given, respectively as
N
1
 y(t ) =  yi (t )
N i=1
(6)

VAR ( t ) =   y ( t ) − y ( t )
2
n (7)
1 i
y
N  
i=1

where yi (t ) is the sample curves or functions and y (t ) is the sample mean.

The dependence of measurements across various argument values is summarised by the functional
covariance, which is computed for all s and t by

1 N( y ( s ) − y ( s ) )( y (t ) − y (t )).
 ( s,t ) = i i
(8)
N i =1

3.3. Functional Principal Component Analysis (FPCA)

The FPCA is used to reduce the dimensionality of the variables by establishing a set of new variables as
linear combinations of the original ones. Let zi (t ) be the center for functional datasets such as

zi (t ) = yi (t ) − y (t ) ,i = 1, 2,..., n (9)

where y (t ) is the mean function of ( yi (t ),..., yn (t )) . A FPCA is applied to zi (t ) ,i = 1, 2,..., n, produce

a small group of harmonic functions which draw attention to the principal source of variation in the data.
The first principal component w1 (t ) describes a weight function, zi (t ) that exists over the same range.
The first component gives the maximum variation in the functional principal component scores as


fi1 = w1 (t ) zi (t )dt (10)

 w ( t ) dt = 1 .
2
subject to the normalisation constraint 1

Repeat for the following components fik =  w (t )z (t )dt by maximising the variance of the
k i

corresponding scores under the constraints  wk ( s )wj ( s )ds = 0, k  2, k  j.

4
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

4. Results and Discussion

This section gives a summary statistic of the rainfall data at the stations under study. The process of
rainfall smoothing and its variations will be discussed based on the functional principal component
analysis.

4.1 Descriptive analysis of annual rainfall data

Table 1 displays the summary statistics of annual rainfall data for stations over the northwest, east,
central, southwest, and west regions of Peninsular Malaysia. Petaling Jaya, which is located in the
western region of Peninsular Malaysia, has the greatest annual mean rainfall with nearly 3384.41 mm
but the lowest variability. It appears that the annual rainfall values at Petaling Jaya did not exhibit
significant year-to-year variation. Sitiawan, at the high north of the western area, recorded the lowest
annual rainfall. The eastern stations recorded higher annual mean rainfall and a higher coefficient of
variation than other regions, as indicated in Table 1. Throughout the NEM season, the eastern region is
known for having more frequent and intense downpours than other areas. As a result, stations in the
eastern region of Peninsular Malaysia have higher annual mean rainfall than stations in other parts of
the country.

Table 1. Summary statistics of annual rainfall data for each studied station.

Location Coding Latitude Longitude Annual CV

Mean
Northwest Alor Setar NW1 6° 12' N 102° 15' E 2176.04 13.66
Region Bayan Lepas NW2 5° 18' N 100° 16' E 2367.96 17.49
Baling NW3 5° 68' N 100° 91' E 2232.14 16.89
Dungun E1 4° 46' N 103° 25' E 2872.19 21.16
Eastern Kota Bharu E2 6° 10' N 102° 18' E 2726.14 21.98
Region Kuantan E3 3° 46' N 103° 13' E 3015.48 19.31
Pasir Mas E4 06° 02' N 102° 07' E 2855.36 25.37
Central Tanah Rata C1 4° 28' N 101° 23' E 2604.46 15.48
Region
Pontian SW1 01° 29' N 103° 23' E 1921.30 16.34
Southwest Tangkak SW2 02° 16' N 102° 32' E 1947.81 16.76
Region Kluang SW3 02° 01 'N 103° 19' E 2147.85 17.54
Melaka SW4 2° 16 'N 102° 15' E 1978.54 12.79
Senai SW5 1° 38' N 103° 40' E 2554.30 15.57
Western Petaling Jaya W1 3° 06' U 101°39' T 3384.41 11.01
Region Lenggong W2 5° 06' N 100° 58' E 2105.07 15.92
Sitiawan W3 4° 13' N 101° 42' E 1825.10 15.08

4.2 Smoothing rainfall data using Fourier Basis function

A Fourier basis function was employed to describe the rainfall pattern since the climate in Malaysia is
influenced by the monsoon season. The rainfall data were smoothed using a Fourier basis function that
consists of sine and cosine functions, which can be expressed in terms of harmonics. Seven basis
functions were used to describe the rain patterns in the central and southwest regions, but only five were
needed for the northwest and west. On the other hand, the rainfall pattern in the eastern region required
the use of nine basis functions. The deviations for the three, five, and seven basis functions were still
significantly higher. When nine basis functions were applied, however, the deviation was significantly
decreased, and no more basis functions were needed. Consequently, nine basis functions, including four
harmonics, could accurately depict the rainfall distribution over the eastern stations. Due to the high
unpredictability and fluctuations in rainfall patterns observed at these rainfall stations, an increase in the
number of basis functions is essential.

5
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

Figure 2. The right panel shows the observed mean daily rainfall with its corresponding smoothing
curves for the regions and their rate of change in the left panel.

Figure 2 illustrates the smoothing curves for each region based on the required number of basis
functions. The right panel of Figure 2 shows a bimodal rainfall pattern for the northwest, southwest,
west, and central regions. During the inter-monsoon, the first peaks in these four regions occurred
between April and May. It was demonstrated that the first peak was lower than the second peak. The
temporal delay between the northwest and the other three regions was very different during the second
peak, though. While rainfall peaked in November for the other three regions, it did so in September and

6
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

October for the northwest region. Rainfall decreased after the second peak and hit its lowest level in
early February. But for the western regions, the lowest rainfall occurs in July. The left panel of Figure 2
shows the rate of change in rainfall. Positive changes in rainfall were observed from February to April
and August to September for the west, southwest, northwest, and central regions, where the season
changes from SWM and NEM to inter-monsoon, respectively. On the other hand, negative changes can
be seen at the end and the beginning of the year.
Figure 2 demonstrates how the pattern of the rainfall stations in the eastern area differs from those
in the other four regions. The range of the average daily rainfall is 5 to 25 mm. The rainfall has been
seen to follow a unimodal trend, with November and December seeing the highest peak. Rainfall levels
are low from mid-May to mid-September, increasing in October until it reaches its peak and drastically
decreasing after mid-February. Based on the rate of change, a significant change in rainfall occurred
during the NEM flow from October to early February, but only small tweaks occurred from mid-
February to October.

4.3 Summary of functional data

Figure 3 illustrates the descriptive statistics of functional mean and functional variation for 16 stations
from 1999 to 2019. It reveals that the highest mean rainfall was observed in November during the NEM
season. Conversely, the lowest mean rainfall was recorded between February and March and July and
August. The NEM brings heavy rain and stormy seas, and it occasionally causes flooding in the eastern
part of Peninsular Malaysia. Rainfall variability is high between the end of November and early January,
while other months have moderate fluctuations. The high rainfall variability reflects the influence of the
northeast monsoon. During this period, the cold surge from northern Asia impacted most of the southern
part of the South China Sea, including Malaysia. In addition, the Intertropical Convergence Zone
(ITCZ), a cloud band that circles the globe around the equator and is frequently wetter than the southwest
monsoon, impacts the area. Hence, a high rainfall peak was observed from November to January.

Figure 3. The smoothing mean and variance curves for all regions.

4.4 Estimating the variation of functional data via functional principal component analysis
Functional principal component analysis was carried out to identify the main sources of variation in the
rainfall data. For the first two principal components, the variance rates were 70.1% and 17.3%,
respectively. The overall variation in the rainfall data was explained by these two components, which
accumulated to 87.4%. The first principal component (PC1) that showed the largest variation was seen
in November, as shown in Figure 4. Contrarily, the second principal component (PC2) revealed that
April had the greatest fluctuating rainfall values.

7
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

Figure 4. Centered Harmonic Functions.

A high correlation between the principal component and the given variable was demonstrated by the
high loading score for that variable. Based on the score values of PC1, the highest score values were
dominated by the eastern stations, as depicted in Figure 5(a). It shows that PC1 was associated with the
eastern region of Peninsular Malaysia during the NEM. Rains during the inter-monsoon season from
April to May and September to October played a significant role in forming the second component, PC2.
Stations in the northwesterly and western areas dominated it, as shown in Figure 5(b). The interval
between monsoons, or the "grey period," is when the southwest and northeast monsoons alternately
change, which is referred to here as the inter-monsoon. During the southwest and northeast monsoon
shifts, the weather plays havoc and suddenly switches from a hot, sunny day to a violent thunderstorm.
The weather is cloudy and wet most of the time during the period.

(a) (b)
Figure 5. Smoothing curves with the highest loading for the (a) first principal component and
(b) second principal component.

The scores of the first two principal components were then mapped onto Figure 6. The figure shows
some exact groupings that reveal the most important type of variation in the rainfall curve. These stations
are grouped based on their high PC1 scores: Kuantan, Dungun, Pasir Mas, and Kota Bharu. On the other
hand, based on the shape of their curves, Alor Setar, Bayan Lepas, and Baling could be grouped together
as a single cluster. In Figure 6, Petaling Jaya stands out as a distinct cluster since it has the highest PC2
score, suggesting that the area may have a distinct rainfall shape from other stations. Also, Tanah Rata
in the central part of Peninsular Malaysia and Senai in the southwest could each have their cluster.

8
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

Figure 6. Map of scores for PC1 and PC2.

4. Conclusion
In this study, functional data analysis was used to convert rainfall data from each region into smoothing
curves that can be analysed at any time interval. Additional information, such as the changes in rainfall,
could be estimated based on the derivative of the smoothing Fourier function. A positive or negative
change could be observed from the smoothing curve over the time interval. The functional findings of
this study showed that more basis functions are needed than in the other regions to fully represent the
variation in the pattern of rainfall over the eastern region. The highest rainfall is recorded during the
NEM season, while the lowest rainfall occurs between mid-May and September during the southwest
monsoon. In contrast, the northwest regions got less rain during the months when the northeast monsoon
flowed, while the second inter-monsoon was the wettest time.
The rainfall patterns over Peninsular Malaysia could be adequately described by the first two
principal components, which represented 87.4% of the total variation. According to the first principal
component, the northeast monsoon season occurs most frequently in the east. In contrast, the first inter-
monsoon season occurs most commonly in the regions represented by the second principal component.
Additionally, the topographical and geographical implications will significantly impact Peninsular
Malaysia's rainfall patterns. The current work is limited to exploring and profiling rainfall patterns.
However, the findings from this study can provide additional information for those who work with the
water management system as an alternative way to describe rainfall events, such as in terms of rainfall
variability, rate of change, and potential anomalies. In addition, further research should consider
inferential aspects in the analysis, such as comparing rainfall patterns using the functional analysis of
variance, building up the relationship using the functional linear model, and forecasting rainfall values
using functional time series.

References
[1] Alaya MAB, Ternynck C, Dabo-Niang S, Chebana F and Ouarda TBMJ 2020 Change point detection of flood events
usinga functional data framework Adv Water Resour 137 103522
[2] de Pinedo AR, Couplet M, Iooss B, Marie N, Marrel A, Merle E and Sueur R 2021 Functional Outlier Detection by Means
of h-Mode Depth and Dynamic Time Warping Appl. Sci. 11, 11475

9
STACLIM-2022 IOP Publishing
IOP Conf. Series: Earth and Environmental Science 1167 (2023) 012007 doi:10.1088/1755-1315/1167/1/012007

[3] Dai W, Mrkvička T, Sun Y and Genton MG Functional outlier detection and taxonomy by sequential
transformationsComput. Stat. Data Anal 149 106960
[4] Ghumman AR, Rauf AU, Husnain Haider H and Shafiquzamman M 2020 Functional data analysis of models for
predictingtemperature and precipitation under climate change scenarios J. Water Clim. Chang 11. 4 1748-1765
[5] Wang D, Zhong Z, Bai K and He L 2019 Spatial and Temporal Variabilities of PM2.5 Concentrations in China
UsingFunctional Data Analysis Sustainability 11 1620
[6] Chebana F, Dabo-Niang S and Ouarda TBMJ 2012 Exploratory functional flood frequency analysis and outlier
detectionWater Resour. Res. 48 W04514
[7] Suhaila J and Yusop Z 2017 Spatial and temporal variabilities of rainfall data using functional data analysis Theor. Appl.
Climatol. 129 229–242
[8] Hael MA 2021 Modeling of rainfall variability using functional principal component method: a case study of Taiz region,
Yemen Model. Earth Syst. Environ. 7, 17–27
[9] Newell J, McMillan K, Grant S and McCabe G 2006 Using functional data analysis to summarise and interpret
lactatecurves Comput. Biol. Med. 36(3) 262-275
[10] Ryan W, Harrison A and Hayes K 2006 Functional data analysis of knee joint kinematics in the vertical jump Sports
Biomech. 5(1) 121-138
[11] Song JJ, Deng W, Lee H-J and Kwon D 2008 Optimal classification for time-course gene expression data using functional
data analysis Comput Biol Chem. 32(6) 426-432
[12] Dong JJ, Wang L, Gill J and Cao J 2018 Functional principal component analysis of glomerular filtration rate curves after
kidney transplant Stat. Methods. Med. Res. 27(12) 3785-3796
[13] Zhang B, Zheng K, Huang Q, Feng S, Zhou S and Zhang Y 2020 Aircraft engine prognostics based on informative sensor
selection and adaptive degradation modeling with functional principal component analysis Sensors 20(3) 920
[14] Suhaila J 2021 Functional Data Visualization and Outlier Detection on the Anomaly of El Niño Southern Oscillation
Climate 9 (118)

Acknowledgments
The authors would like to acknowledge the financial support from the Ministry of Education Malaysia
under the Fundamental Research Grant Scheme (FRGS/1/2020/STG06/UTM/02/3) and the Universiti
Teknologi Malaysia for the funding under UTMER (QJ130000.3854.19J58).

10