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

net/publication/295400559

Spatio-temporal variation of wet and dry spell characteristics of tropical

precipitation in Singapore and its association with ENSO

Article in International Journal of Climatology · February 2016

DOI: 10.1002/joc.4672

CITATIONS READS

54 1,036

3 authors:

Xin Li Ali Meshgi

Hohai University 10 PUBLICATIONS 410 CITATIONS
23 PUBLICATIONS 487 CITATIONS
SEE PROFILE
SEE PROFILE

Vladan Babovic
National University of Singapore
183 PUBLICATIONS 4,335 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Assessment of climate change impact on regional water resources based on nonstationary multi-dimensional downscaling methods View project

Data-driven water quality treatment management decision support systems View project

All content following this page was uploaded by Xin Li on 11 October 2017.

The user has requested enhancement of the downloaded file.

INTERNATIONAL JOURNAL OF CLIMATOLOGY
Int. J. Climatol. 36: 4831–4846 (2016)
Published online 19 February 2016 in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/joc.4672

Spatio-temporal variation of wet and dry spell characteristics

of tropical precipitation in Singapore and its association with
ENSO
Xin Li,* Ali Meshgi† and Vladan Babovic
Department of Civil and Environmental Engineering, National University of Singapore, Singapore

ABSTRACT: Climatology and statistical variability of wet and dry spell characteristics of precipitation provide an effective
tool to understand the impacts of natural climate variability such as El Niño/Southern Oscillation (ENSO) and climate change
on the water availability over a region. However, to date, detailed studies about precipitation spell characteristics in the tropical
urban context are still limited. Therefore, this study presents a comprehensive framework to characterize the wet and dry spells
of precipitation in a tropical urban city-state (i.e. Singapore), based on daily data from 26 precipitation stations. In addition,
14 wet and dry spell indices are defined, and spatial and temporal variations of the indices are analysed based on ordinary
kriging method and an improved Mann-Kendall (MK) test method, respectively. Lastly, the influence of ENSO condition on
the wet and dry spell indices is investigated utilizing a regional regression method. Results indicate that: (1) the region over
Singapore is predominantly characterized by 1-day wet and dry spells, while the main contributions to total duration of wet
and dry spells come from 3-4-day wet spells and 3–6-day dry spells, respectively. The major contribution to total accumulated
depth of wet spells comes from 3-4-day and 20–80 mm wet spells; (2) significant increasing trends are detected in general and
extreme characteristics of wet spells at annual time scale, however, these trends are not significant during the two monsoon
seasons. Spatial variation of wet and dry spell indices is also evident; (3) Singapore’s precipitation spells are influenced by
ENSO, but mainly during the southwest monsoon season. During the El Niño phase, precipitation spells generally become
drier. The predicted changes of wet and dry spell characteristics in the contexts of ENSO and climate change have implications
for water resources management in tropical urban cities.

KEY WORDS tropical urban environments; precipitation; wet and dry spell; trend analysis; El Niño/Southern Oscillation
Received 29 September 2015; Revised 24 December 2015; Accepted 4 January 2016

1. Introduction significance for practitioners in hydrology, ecology, agri-

culture and water resources management, etc.
Investigation of spatio-temporal variability of precipitation
Multitudes of studies have been conducted to explore
is of paramount significance to understand the impacts of
the variability of wet and dry spell characteristics. Among
natural climate variability and climate change on hydro-
all of these studies, Schmidli and Frei (2005) investigated
logic cycle and water resources. Therefore, it has been
the trends of heavy precipitation and wet and dry spells
received extended attention over the past few decades,
in Switzerland during the 20th century at seasonal time
especially in the context of global warming (Riley et al.,
scale based on daily precipitation data at 104 gauge sta-
1987; Sperber and Palmer, 1996; Turner et al., 1997;
tions. They defined a set of wet and dry spell indices, such
Cayan et al., 1998; Busuioc et al., 2001; Zou et al., 2005;
as mean dry(wet)-day persistence, max number of con-
Zhang et al., 2008; Niedźwiedź et al., 2009; Tabari and
secutive dry(wet) days and mean dry(wet)-spell length.
Talaee, 2011; Wang et al., 2011; Shen et al., 2012; Ris-
They found that spatially coherent and statistically sig-
bey et al., 2013; Wuebbles et al., 2014; Zhang et al., 2014;
nificant trends in the extreme indices only exist for win-
Chen and Sun, 2015). The day-to-day variability of precip-
ter and autumn. Bai et al. (2007) investigated the clima-
itation can be characterized by wet and dry spells, which
tology and trends of wet spells in China based on daily
are defined as episodes of consecutive rainy or non-rainy
precipitation from 594 meteorological stations, and found
days (Ratan and Venugopal, 2013). The climatology and
that the trends of wet spells exhibit significant regional
statistical variability of wet and dry spells are of great
differences. For instance, the annual number of days in
wet spells has slightly increased in most areas of west-
* Correspondence to: X. Li, Department of Civil and Environmental ern China, while significantly decreased over North China.
Engineering, National University of Singapore, Block E1A, #07-03, Singh and Ranade (2010) investigated the variation of wet
No. 1 Engineering Drive 2, Singapore. E-mail: xinli@u.nus.edu and dry spell characteristics across India using gridded
†
Present address: Public Utility Board (PUB), Singapore’s National daily rainfall for the period 1951–2007, demonstrating
Water Agency, 40 Scotts Road, Environmental Building, Singapore. that total duration of wet (dry) spells (TDWS, TDDS)

© 2016 Royal Meteorological Society

4832 X. LI et al.

decreases from 101 (173) to 45 (29) days, and the dura- association of the wet and dry spell indices with ENSO
tion of individual wet (dry) spell reduces from 12 (18) is explored.
to 7 (11) days following similar patterns. Zolina et al. In this article, the study area and the dataset are described
(2013) analysed the duration of wet spells in Europe in Section 2 while the methodologies applied in this study
over the period of 1950–2008 using daily rain gauge are introduced in Section 3. Section 4 provides the Results
data. Their results indicated that wet periods have become and discussion, including wet and dry spell characteris-
longer over most of Europe by about 15–20%. Singh tics, spatial and temporal variation of wet and dry spell
et al. (2014) investigated the variation of extreme wet indices, and regional regression analysis between Ocean
and dry spells during the South Asian summer mon- Niño Index and wet and dry spell indices. Lastly, Section
soon season based on 1∘ ×1∘ gridded daily precipita- 5 summarizes the main conclusions of this study.
tion data from India. They found statistically significant
increase in the frequency of dry spells and intensity of wet
spells, and statistically significant decrease in the intensity 2. Study area and dataset
of dry spells.
2.1. Study area
Most of the studies mentioned above only focused on
variability of durations of wet or dry spells, with limited Singapore is located off the southern tip of Malay Penin-
attention to accumulated depth of wet spells (ADWS). In sula, separated from Peninsular Malaysia by Johor Straits
addition, a majority of the studies only focused on either to the north and from Indonesia’s Riau Islands by the
general or extreme characteristics of wet or dry spells, Singapore Strait to the south. The city-state has an area
without comprehensive investigation of both aspects. In of 718 km2 . The country is highly urbanized, hence heat
fact, there are few studies which have attempted to char- island effect is also significant (Chow and Roth, 2006; Kar-
acterize wet and dry spells in detail, or investigated the dinal Jusuf et al., 2007; Priyadarsini et al., 2008).
variability of wet and dry spells based on a comprehensive Singapore is characterized by a tropical climate, with
list of indices considering both general and extreme char- abundant rainfall, high temperature and humidity all
acteristics, especially in the tropics. In this context, Ratan year round (Fong and Ng, 2012). Diurnal variations of
and Venugopal (2013) performed a detailed characteriza- temperature and precipitation are more prominent than
tion of wet and dry spells of global tropical rainfall and month-to-month variation (Fong and Ng, 2012). The
reported differences in wet and dry spell characteristics average annual rainfall is 2430 mm during the period of
between humid and arid regions. Utilizing two main char- 1980–2010, with the average of 51% rainy days (Manda-
acteristics, i.e. duration of wet or dry spells, and ADWS paka and Qin, 2013). Annual mean temperature in Singa-
of different durations, they characterized wet and dry pore is around 26–27.5 ∘ C (Chow and Roth, 2006). There
spells of global tropical rainfall based on satellite-based are two major monsoon seasons which govern the climate
daily rainfall observations. However, this characterization characteristics of Singapore, i.e. the northeast monsoon
approach was only based on duration, since the accumu- and the southwest monsoon. From the end of October till
lated depth was calculated from wet spells of certain dura- the early of March, the pressure gradient between the high
tions. A complete characterization of wet spells should pressure zone over Siberia and Tibet and the low pressure
be based on both duration and magnitude (accumulated of intertropical convergence zone (ITCZ) generates winds
depth of precipitation within a wet spell). Besides, clas- blowing from the northeast of Singapore, which is referred
sifications of wet and dry spells are also crucial to high- to as northeast monsoon (Fong and Ng, 2012; Beck et al.,
light extreme spells which are major concerns for water 2015). From June to September, the ITCZ locates to
resources management. the north of Singapore, similarly, the pressure gradient
Tropical precipitation process in urban cities is also com- between the high pressure zone over Australia and ITCZ
plex due to the complicity of convective system proper- generates winds blowing from southwest of Singapore,
ties, impacts of natural climate variability (such as ENSO) which is referred to as southwest monsoon (Fong and Ng,
and climate change, anthropogenic impact, etc. Tropical 2012; Beck et al., 2015). The proximity of the ITCZ to a
precipitation in urban cities highly varies in both spatial region may influence the precipitation amount it receives
and temporal dimensions. Characterization of wet and dry during the progression of ITCZ. However, for countries
spells of tropical precipitation in urban cities can help near the equator like Singapore, the ITCZ is never too far
authorities in understanding the variability of regional pre- away, as such, there are no distinct dry and wet seasons
cipitation, thus facilitate strategic planning in water-related (Fong and Ng, 2012). In addition, there is also a regional
problems. However, a review of the literature shows that weather system called ‘Sumatra Squalls’, which originates
to date, no detailed study has been done to (1) characterize over the Strait of Malacca and traverse Singapore from
the wet and dry spells and (2) investigate the variation of West to East, and tends to generate heavy torrential rainfall
wet and dry spell characteristics of tropical precipitation in particularly in the early mornings (Fong and Ng, 2012;
tropical urban cities. Beck et al., 2015).
Therefore, the aim of this article is to first provide a
detailed framework to characterize wet and dry spells of 2.2. Dataset
tropical precipitation in urban cities and then to investigate Daily precipitation data in Singapore are sourced from
the variability of wet and dry spells. In addition, the Meteorological Service Singapore (see http://www.

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

 SPATIO-TEMPORAL VARIATION OF PRECIPITATION SPELLS IN SINGAPORE 4833

Figure 1. Locations of the 26 precipitation stations in Singapore (Elevation derived from the 90 m SRTM DEM (Jarvis et al., 2008)).

weather.gov.sg/climate-historical-daily/), which is an Table 1. Temporal coverage and missing data percentage of the
operations division under National Environmental Agency 26 precipitation stations.
(NEA) of Singapore. To ensure the data quality, only sta-
Station Start year End year Missing data
tions with long temporal coverage and short missing percentage (%)
record length are used in this study. The temporal cov-
erage and missing data percentage of the resultant 26 S23 1980 2013 0.0
S24 1980 2013 0.0
stations (Figure 1) are provided in Table 1. The missing
S25 1980 2013 0.0
data are filled by the average of the precipitation values S07 1980 2013 0.0
of the same day from two nearest neighbouring stations S43 1980 2013 0.0
(Zhang et al., 2014). The choice of the number of nearest S06 1980 2013 0.1
neighbouring stations for gap filling is a compromise S31 1980 2013 0.1
between local sampling uncertainty and spatial variability S39 1980 2013 0.2
S40 1980 2010 0.3
of precipitation over Singapore (Mandapaka and Qin, S80 1991 2013 0.4
2013; Beck et al., 2015). As the missing record percent- S36 1980 2013 0.4
age is relatively low (i.e. less than 2% among 92% of the S29 1980 2013 0.4
precipitation stations), the uncertainty resulting from gap S64 1980 2013 0.7
filling process can be assumed not significant. Temporal S60 1980 2013 0.7
coverage of most of the stations (i.e. 85%) is between 1980 S63 1980 2013 0.8
S11 1980 2013 0.9
and 2013. The data lengths of all the stations are longer S66 1980 2013 1.1
than 30 years (except for S80), which is consistent with S72 1983 2013 1.2
World Meteorological Organization (WMO) standards for S50 1980 2013 1.2
climate-related studies. S44 1980 2013 1.3
The Ocean Niño Index (ONI) is used to quantify the S35 1980 2013 1.5
condition of El Niño/Southern Oscillation (ENSO). The S33 1980 2013 1.6
S46 1980 2013 1.6
ONI is a 3-month running mean of sea surface temper- S55 1980 2013 1.8
ature (SST) anomalies in the Niño 3.4 region (5∘ N–5∘ S S77 1983 2013 2.4
and 170∘ –120∘ W). NOAA’s Climate Prediction Centre S78 1980 2013 3.1
declares the onset of warm phase (El Niño) and cold phase
(La Niña) of ENSO based on a threshold of ±0.5∘ for ONI.
The ONI does not reflect a possible increase of SST due to
3. Methodology
global warming, because the monthly climatology and the
long-term trend are removed when computing the anoma- In the present study, a comprehensive framework is first
lies (Beck et al., 2015).The ONI data can be downloaded proposed to characterize wet and dry spells of precipitation
from National Oceanic and Atmospheric Administration in Singapore in detail. Spatial and temporal variations
of the United States (see http://www.cpc.ncep.noaa.gov/). of wet and dry spell characteristics are then investigated

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

4834 X. LI et al.

using ordinary kriging method and an improved MK test, Further analysis is performed based on classifications of
respectively. A regional regression approach is further wet and dry spells. Dry spells are classified into four cate-
applied to explore the association between ONI and the gories based on their durations, i.e. 1-day, 2-day, 3–6-day
seasonal wet and dry spell indices. Lastly, the significance and >6-day dry spells, which generally correspond to
of the dependence between ONI and the seasonal wet and 0–25, 25–50, 50–90 and >90% percentiles of the dura-
dry spell indices is tested utilizing a non-parametric Monte tion distribution, respectively. Similarly, wet spells are
Carlo method (Beck et al., 2015). classified into 1-day, 2-day, 3–4-day and >4-day cate-
gories based on the corresponding percentiles of the dura-
3.1. Characterization of wet and dry spells tion distribution. Apart from the classification based on
A wet (dry) spell is defined as the number of consecutive duration, we also classify wet spells into four categories
rainy (non-rainy) days (Ratan and Venugopal, 2013). In based on magnitude, i.e. 0–5 mm (slight), 5–20 mm (mod-
the present study, a rainy day is defined as the day with erate), 20–80 mm (heavy) and >80 mm (extreme) wet
precipitation equal to or greater than 1 mm per day, which spells based on the corresponding percentiles of the mag-
is consistent with the study of Fong and Ng (2012). The nitude distribution (i.e. 0–25, 25–50, 50–90 and >90%
basic characteristics used to quantify wet and dry spells percentiles).
are (1) duration of a wet (dry) spell, d, defined as a
period of d consecutive rainy (non-rainy) days and (2) 3.2. Spatial and temporal variations of wet and dry spell
magnitude of a wet spell, m, which corresponds to the indices
accumulated depth of precipitation within a wet spell. It
To cover the general and extreme characteristics of wet
is necessary to emphasize that the basic characteristics
and dry spells, 14 wet and dry spell indices are defined
defined here are for single wet or dry spell. This should be
in this study. Details of these indices are listed in Table 2.
differentiated from TDWS or TDDS, or ADWS during a
certain period. The first five indices in Table 2 are general characteristics
As the precipitation in Singapore highly varies in space of wet and dry spells, such as total duration, frequency,
(Mandapaka and Qin, 2013; Beck et al., 2015), the anal- ADWS, etc., while the remaining nine indices are defined
ysis is carried out station by station. Quantities of inter- to explore the extreme characteristics, such as maximum
est for dry spells are the relative frequencies (RF) of dry duration, frequencies of extreme wet and dry spells and
spells of certain durations and fractional contributions to accumulated depth of extreme wet spells.
total duration (FCTD) of all the dry spells. The calcu- The original non-parametric MK test (Mann, 1945;
lations of FCTD and RF are carried out as follows: if Kendall, 1955) has been widely used in detecting trends
ndd is the number of d-day dry spells during a period, in hydro-meteorological time series in which the assump-
they contribute (ndd × d) days to TDDS during that period. tions of uniform variance and Gaussian residuals may
not be satisfied (Schmidli and Frei, 2005). However, if a
) ∑
dmax
(
Hence, d × ndd ∕ (d × ndd ) (where dmax is the largest positive autocorrelation exists within the time series, the
d=1 MK test rejects the null hypothesis more often than spec-
dry spell duration) is the fractional contribution of this ified by the significance level (Von Storch and Zwiers,
d-day dry spell to total duration of all the dry spells. RF 1995). Considering the impact of autocorrelation on the
∑
dmax
apparent level of significance of trend, a pre-whitening
is computed by ndd ∕ ndd .
approach which removes the serial correlation with the
d=1
Basic characteristics of wet spells include duration and assumption of a AR(1) model is usually utilized. This
magnitude. These two different measures cannot be mixed approach is efficient only if the resultant residuals are
because a wet spell with long duration does not lead to independent, while in other cases, the serial correlation
large magnitude and vice versa. FCTD for wet spells is may still remain in the residuals. Given this issue, Zhang
defined as the fractional contribution of the total duration et al. (2000) proposed an iterative procedure, which was
of wet spells with specific duration (magnitude) to total further refined by Wang and Swail (2001). The refined
duration of all the wet spells. Another quantity of interest iterative-based MK test has been shown to be robust and
for wet spells is the fractional contribution of the accumu- efficient in detecting trends in hydro-meteorological time
lated depth of wet spells with certain magnitude (duration) series (Zhang and Zwiers, 2004; Zhai et al., 2005). There-
to total accumulated depth of all the wet spells (FCAD). fore, this improved Mann-Kendall (IMK) test utilizing iter-
As the magnitude of wet spells is difficult to characterize ative procedure (Wang and Swail, 2001) is employed to
as discrete numbers as duration, we use intervals [a, b) to detect trends in the time series of wet and dry spell indices.
illustrate wet spells of different magnitudes. For instance, For the complete details of this methodology, a reference
the interval [0, 5) exemplifies that the magnitude of wet is made to the study of Wang and Swail (2001). For signif-
spell is between 0 and 5 mm. The FCAD of the wet spells icance test, the 5% significance level is set as the default
with magnitude specified by [a, b) is the fraction of sum of unless otherwise stated.
the wet spell magnitudes bounded by [a, b) to total accu- As the entire set of wet and dry spell indices is defined
mulated depth of all the wet spells. The calculations of RF based on precipitation, spatial variation of wet and dry
and FCTD for wet spells are similar, hence are not elabo- spell indices can be expected at different time scales due
rated here. to the high spatial variability of rainfall in Singapore. In

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

 SPATIO-TEMPORAL VARIATION OF PRECIPITATION SPELLS IN SINGAPORE 4835

Table 2. List of wet and dry spell indices used in this study.

Wet and dry spell indices Description Units

TDDS Total duration of dry spells day
TDWS Total duration of wet spells day
ADWS Accumulated depth of wet spells mm
DSF Frequency of dry spells –
WSF Frequency of wet spells –
MDDS Maximum duration of dry spells day
MDWS Maximum duration of wet spells day
MMWS Maximum magnitude of wet spells mm
DSD90 Number of dry spells with duration exceeding 90% percentile of –
the long-term duration distribution
WSD90 Number of wet spells with duration exceeding 90% percentile of –
the long-term duration distribution
WSM90 Number of wet spells with magnitude exceeding 90% percentile –
of the long-term magnitude distribution
TDDS90 Total duration of the DSD90 dry spells day
TDWS90 Total duration of the WSD90 wet spells day
ADWS90 Accumulated depth of the WSM90 wet spells mm

(a) (b)

Figure 2. Boxplots of (a) relative frequencies (RF) of dry spells of different durations and (b) their fractional contributions to total duration (FCTD)
during the stations’ respective periods (lines in box denote medians and circles denote means).

this study, we only investigate the spatial variations of As the precipitation in Singapore highly varies in space
the long-term mean values of two typical wet and dry (Mandapaka and Qin, 2013; Beck et al., 2015), regres-
spell indices (i.e. TDDS and ADWS). To obtain the spatial sion analysis based on single stations may be influenced
distribution of the index values, ordinary kriging method is by local effects and sampling uncertainty and hence dif-
applied to interpolate among the 26 precipitation stations. fer from the regional signal (Beck et al., 2015). Regional
The detailed description of this methodology is given in regression is a good alternative to investigate the relation-
the study of Kitanidis (1997). ship between ENSO and seasonal wet and dry spell char-
acteristics from a regional perspective.
3.3. Regional regression analysis of ONI and seasonal As the difference in scales may cause problem in
wet and dry spell indices parameter estimation process based on least squared error
method (Beck et al., 2015), the wet and dry spell indices
Regional regression is an extension of single-site regres- of interest are first normalized. The normalization is con-
sion for simultaneously measured data at several locations. ducted by dividing index values by the station’s average
Both linear and nonlinear functions can be applied to con- of the index of interest. Take the ADWS as an example:
struct regional regression functions. Beck et al. (2015)
used a regional linear regression model to investigate ADWSi,j
NADWSi,j = ( ) (1)
the association of precipitation extremes and seasonal mean ADWSi
precipitation totals in Singapore with ONI, and the results
) 1∑
n
demonstrated that the regional linear regression model is ( i

efficient to establish the regression relationships. There- mean ADWSi = ADWSi,j (2)
ni j=1
fore, a regional linear regression model is applied in the
present study to explore the influence of ENSO condition where NADWSi,j is the normalized ADWS of station i in
on the seasonal wet and dry spell characteristics. year j. NADWSi,j is calculated by dividing the ADWS of

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

4836 X. LI et al.

(a) (b) (c)

Figure 3. Boxplots of (a) relative frequencies (RF) of wet spells of different durations, (b) their fractional contributions to total duration (FCTD) and
fractional contributions to total accumulated depth (FCAD) during the stations’ respective periods (lines in box denote medians and circles denote
means).

(a)

(b)

(c)

Figure 4. Boxplots of (a) relative frequencies (RF) of wet spells with different magnitudes, (b) their fractional contributions to total duration (FCTD)
and (c) fractional contributions to total accumulated depth (FCAD) during the stations’ respective periods. Magnitudes are partitioned at every 5 mm
(lines in box denote medians and circles denote means).

station i in year j (ADWSi,j ) by the average ADWS of approximating the significance level 𝛼 by the fraction of
station i during the ni years’ period. linear regression slopes calculated from shuffled data that
have a steeper slope (of the same sign) than the regression
3.4. Significance test in regional regression analysis calculated from the observed data (Beck et al., 2015). The
A non-parametric Monte Carlo method (Beck et al., 2015) detailed description of this methodology is given by Beck
is applied to test the significance of the dependence et al. (2015). The 5% significance level is applied to test
between the wet and dry spell indices and ONI. The con- the significance of the dependence between the wet and
cept of this method is as follows: assuming that a signif- dry spell indices and ONI.
icant dependence exists in the data, there is a probability
that it will be destroyed by shuffling the order of the obser-
vations of the independent variable. However, if the depen- 4. Results and discussion
dence is just due to sampling effect, there is a probability
that the dependence in the shuffled data is even more pro- 4.1. Wet and dry spell characteristics
nounced than that in the original data (Beck et al., 2015). Figure 2 shows the boxplots of the RF of dry spells of
Hence the significance test can be performed by compar- different durations and their FCTD during the stations’
ing the slopes of the two data sets, using a one-sided test, respective periods. As can be seen in Figure 2, 1-day dry

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

 SPATIO-TEMPORAL VARIATION OF PRECIPITATION SPELLS IN SINGAPORE 4837

(a) (b)

Figure 5. Boxplots of (a) relative frequencies (RF) of dry spells of durations of 1-day, 2-day, 3–6-day and >6-day and (b) their fractional contributions
to total duration (FCTD) during the stations’ respective periods (lines in box denote medians and circles denote means).

(a) (b) (c)

(d) (e) (f)

Figure 6. Upper row: boxplots of (a) relative frequencies (RF) of wet spells of durations of 1-day, 2-day, 3–4-day and >4-day, (b) their fractional
contributions to total duration (FCTD) and (c) fractional contributions to total accumulated depth (FCAD) during the stations’ respective periods.
Lower row: boxplots of (d) relative frequencies (RF) of slight (S), moderate (M), heavy (H) and extreme (E) wet spells, (e) their fractional
contributions to total duration (FCTD) and (f) fractional contributions to total accumulated depth (FCAD) during the stations’ respective periods
(lines in box denote medians and circles denote means).

spell dominates over the region of Singapore. However, decay with the increase of duration d (with an exception
2-day dry spell contributes the most (with a mean value of 2-day dry spell for FCTD). We also notice the existence
about 16%, ‘mean value’ is omitted hereinafter for sim- of spatial variability of RF and FCTD through the range of
plicity) to total duration of all the dry spells even though its the boxes in the boxplots (see Figure 2). This is consistent
occurrence is only half of the 1-day dry spell. The most fre- with the fact of high spatial variability of precipitation over
quent dry spells over the region of Singapore are 1–3-day relatively small region of Singapore (Mandapaka and Qin,
long dry spells and each of the spells contributes more than 2013; Beck et al., 2015).
10% in terms of RF. The main contribution to total dura- The same analysis is repeated for wet spells. Figure 3
tion of all the dry spells appears to come from 1–4-day shows RF of wet spells of different durations, their frac-
long dry spells (each of the spells contributes more than tional contributions to total duration and total accumulated
10%), the contribution of dry spells of duration of 15 days depth of all the wet spells. As with dry spells, 1-day wet
or longer is insignificant (less than 1%). As expected, the spells are the most frequent wet spells (occurring more
occurrence and fractional contribution of d-day dry spells than 50% in terms of RF), contributing more than 25% and

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

4838 X. LI et al.

(a) (b)

(c) (d)

(e) (f)

Figure 7. Trends of (a) TDDS, (b) ADWS, (c) DSF, (d) MDDS, (e) WSM90 , and (f) ADWS90 . The symbols represent the sign of the trend estimate
(blank up-triangle: positive trend; blank down-triangle: negative trend; filled up-triangle: significant positive trend at 5% significance level; filled
down-triangle: significant negative trend at 5% significance level; circle no trend. ns. means not significant in the legend).

20% to total duration and total accumulated depth, respec- Figure 5 shows the boxplots of the RF of the classified
tively. Relatively frequent wet spells over the region of Sin- dry spells and fractional contributions to total duration of
gapore are 1–3-day long (each occurring more than 10% in all the dry spells during the stations’ respective periods. As
terms of RF). The main contributions to total duration and can be seen in Figure 5, although dry spell of duration of 1
total accumulated depth come from 1- to 4-day long wet day is the most frequent dry spell (with RF above 40%), it
spells (each contributing more than 10%). Contribution of only contributes approximately 15% to total duration. The
wet spells longer than 10 days is insignificant. Decays of main contribution to total duration of all the dry spells is
RH, FCTD and FCAD with the increase of duration are from 3- to 6-day long dry spells, contributing to around
observed for wet spells, as well as the spatial variability of 40%. It is interesting to note that dry spells with duration
wet spell characteristics over the region of Singapore. longer than 6 days contribute more (approximately 30%)
Figure 4 shows the RF of wet spells of different mag- to total duration than dry spells of duration of 1-day (about
nitudes, the fractional contributions to total duration and 15%) and 2-day (about 16%) even though its occurrence is
total accumulated depth. The magnitudes are partitioned the lowest (less than 10%). This indicates that extreme dry
at every 5 mm. As can be seen in Figure 4, 0–5, 5–10, spells play a significant role in TDDS. Moreover, high spa-
10–15 and 15–20 mm wet spells have RF around or above tial variability is observed from the fractional contribution
10%, and 0–5 mm wet spell contributes more than 10% of extreme dry spells to total duration. This might be
to total duration. RH and FCTD decay with the increase due to the uneven spatial distribution of extreme dry
of wet spell magnitudes, however, spikes of 30–40 mm spells caused by the high spatial variability of precipitation
wet spells are found in FCAD. It is necessary to emphasize over Singapore.
that the magnitudes in Figure 4(c) are partitioned at every Figure 6 depicts the boxplots of the RF of classified wet
5 mm, hence different spikes can be found if the interval spells and fractional contributions to total duration and
of the partition is different. Spatial variability in wet spell total accumulated depth of all the wet spells during the
characteristics in Figure 4 is also evident. stations’ respective periods. We can see from the upper

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

 SPATIO-TEMPORAL VARIATION OF PRECIPITATION SPELLS IN SINGAPORE 4839

(a) (b)

(c) (d)

(e) (f)

Figure 8. Trends of (a) TDDS, (b) ADWS, (c) DSF, (d) MDDS, (e) WSM90 , and (f) ADWS90 for Northeast Monsoon season. The symbols represent
the sign of the trend estimate (blank up-triangle: positive trend; blank down-triangle: negative trend; filled up-triangle: significant positive trend at
5% significance level; filled down-triangle: significant negative trend at 5% significance level; circle no trend. ns. means not significant in the legend).

row of Figure 6 that, as with dry spells, 1-day wet spells long as 51 days, and predominantly from 3- to 6-day long
are the most frequent (with RF above 50%) while 3–4-day dry spells over the region of Singapore. Dominant contri-
wet spells contribute the most to total duration and total butions to total duration and total accumulated depth of
accumulated depth (both around 30%). Unlike dry spells, wet spells come from wet spells of duration as long as
contributions of wet spells of different duration categories 24 days, and predominantly from 3- to 4-day long wet
are quite evenly distributed. The lower row of Figure 6 spells. In addition, heavy wet spells with a magnitude of
illustrates that heavy wet spells (magnitude of 20–80 20–80 mm dominate the region of Singapore in terms of
mm) dominate Singapore in terms of RF (about 40%), occurrence, contribution to total duration and total accu-
fractional contribution to total duration (about 45%) and mulated depth of wet spells. Extreme wet and dry spells
fractional contribution to total accumulated depth (about contribute a substantial percentage to total duration and
50%). Extreme wet spells (with magnitude larger than 80 total accumulated depth (for wet spells), even though their
mm) contribute about 40% to total accumulated depth, occurrences are relatively low.
and about 25% to total duration, even their occurrences
are only about 10%. Relatively high spatial variability 4.2. Spatial and temporal variation of wet and dry spell
is found from the fractional contributions of the extreme indices
wet spells (with extreme duration or magnitude) to total Figure 7 illustrates spatial distribution of the trends in
duration or total accumulated depth, but they are not as annual time series of a subset of the 14 wet and dry
obvious as that of dry spells. spell indices defined in Section 3.2. Figure 7 shows that
In summary, Singapore precipitation spells show pre- the entire Singapore is dominated by decreasing TDDS,
dominant occurrence of 1-day wet and dry spells. Con- and these trends in TDDS are statistically significant
tributions to TDDS come from dry spells of duration as at 5% significance level at eight stations located at the

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

4840 X. LI et al.

(a) (b)

(c) (d)

(e) (f)

Figure 9. Trends of (a) TDDS, (b) ADWS, (c) DSF, (d) MDDS, (e) WSM90 , and (f) ADWS90 for southwest monsoon season. The symbols represent
the sign of the trend estimate (blank up-triangle: positive trend; blank down-triangle: negative trend; filled up-triangle: significant positive trend at
5% significance level; filled down-triangle: significant negative trend at 5% significance level; circle no trend. ns. means not significant in the legend).

eastern and northern part of the region. Figure 7 also shows significant increase of ADWS90 . This indicates that the
that 19 stations are characterized by statistically signifi- occurrence of extreme wet spells (wet spells with extreme
cant increasing ADWS, accounting for 73% of the total magnitude) tends to increase over the region. In addition,
number of precipitation stations. Most stations are iden- contributions of these events to total accumulated depth
tified by increasing dry spell frequency (DSF), neverthe- also rise at annual time scale. All these effects may have
less, only one station located at the northwestern corner implications for urban water management in Singapore.
of the region is characterized by statistically significant The results show that the entire region of Singapore is
increasing DSF. dominated by increasing ADWS, with the occurrence of
In terms of trends in extreme wet and dry spells, extreme wet spells and their contributions to total accumu-
Figure 7(d) illustrates that the northeastern part of the lated depth rising at annual time scale. Similar results are
region is dominated by decreasing maximum duration found by Beck et al. (2015) that yearly precipitation totals
of dry spells (MDDS), whereas north, south and west of in Singapore significantly increase as well as the hourly
the region are dominated by increasing MDDS. Only one and daily precipitation extremes. It could be interesting to
station located at the north-western corner is identified by establish whether these trends exist during the two mon-
statistically significant increasing MDDS. It is interesting soon seasons, i.e. northeast monsoon season and south-
to note that 25 stations are characterized by increasing west monsoon season. Figure 8 shows trend test results in
trends in WSM90 , and 15 stations (accounting for 60% the same subset of the 14 wet and dry spell indices dur-
of the total number of the stations) are dominated by ing the northeast monsoon season. As with annual time
statistically significant increasing WSM90 . Besides, all scale, most of the stations (above 75% of the total number
of the stations are dominated by increasing trends in of the precipitation stations) are dominated by decreas-
ADWS90 , and 14 stations (accounting for more than 50% ing TDDS, and increasing ADWS and DSF even though
of the total number of the stations) show statistically none of them is statistically significant. Similar results

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

 SPATIO-TEMPORAL VARIATION OF PRECIPITATION SPELLS IN SINGAPORE 4841

(a) (b)

(c) (d)

(e) (f)

Figure 10. Spatial distribution of the long-term mean of TDDS and ADWS at annual and seasonal time scales (northeast monsoon season and
southwest monsoon season). (a) TDDS_mean_annual, (b) ADWS_mean_annual, (c) TDDS_mean_NE, (d) ADWS_mean_NE, (e) TDDS_mean_SW
and (f) ADWS_mean_SW

are found for WSM90 and ADWS90 , where 22 stations respectively. However, most of them are not statistically
are dominated by non-significant increase of WSM90 and significant. For WSM90 and ADWS90 , 17 stations are
24 stations are characterized by non-significant increase dominated by increasing WSM90 , whereas 23 stations
of ADWS90 . All these suggest that extreme wet spells are characterized by increasing ADWS90 . However, only
tend to occur more frequently and their contributions to one station shows significant increasing ADWS90 . Similar
total accumulated depth rise during the northeast mon- results are found for other wet and dry spell indices (i.e.
soon season, despite of the fact that these trends are not TDWS, WSF, MDWS, MMWS, DSD90 , WSD90 , TDDS90
statistically significant. and TDWS90 ), because most of them are not significant,
The spatial distribution of the trend test results for the thus are not shown here.
same subset of wet and dry spell indices during the south- Figure 10 shows the spatial distributions of the long-term
west monsoon season is summarized in Figure 9. There mean values of TDDS and ADWS at both annual and
are no general trends in TDDS over Singapore. However, seasonal time scales (northeast monsoon season and south-
all the stations are dominated by increasing ADWS with west monsoon season). The map is based on the values of
four stations showing statistically significant increase of the 26 precipitation stations and interpolated by ordinary
ADWS. Unlike annual time scale and northeast monsoon kriging method. Spatial variations of the long-term mean
season, more than 60% of stations are dominated by values of TDDS and ADWS at both annual and seasonal
decreasing DSF, which indicates that the occurrences of time scales can be found over Singapore. For the long-term
dry spells decline during the southwest monsoon season, mean values of TDDS, the maximum values (about 228
even though most of the detected negative trends are not days at annual time scale, 76 days and 79 days during
statistically significant. Nearly half of the stations are the northeast monsoon season and southwest monsoon
identified by increasing MDDS and decreasing MDDS, season, respectively) are observed in the eastern part, and

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

4842 X. LI et al.

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 11. Dependence between Ocean Niño Index and the 14 wet and dry spell indices at each station during the Northeast Monsoon season.
(a) TDDS, (b) TDWS, (c) ADWS, (d) DSF, (e) WSF, (f) MDDS, (g) MDWS, (h) MMWS, (i) DSD90 , (j) WSD90 , (k) WSM90 , (l) TDDS90 , (m)
TDWS90 and (n) ADWS90.

the minimum values (about 204 days at annual time scale, spells are also increasing significantly. Similar results are
65 days and 73 days during the northeast monsoon season found during the northeast monsoon season and southwest
and southwest monsoon season, respectively) are mainly monsoon season. However, the detected trends are not
distributed in the western and northern part. In contrast, statistically significant over the region of Singapore dur-
the maximum values of the long-term mean of ADWS ing the two monsoon seasons. This phenomenon could
(about 2760 mm at annual time scale, 1044 mm and 765 be due to the fact that changes in precipitation depth and
mm during the northeast monsoon season and southwest occurrence of extreme wet spells at annual time scale are
monsoon season, respectively) are observed in the western mitigated at seasonal time scale.
and northern part, and the minimum values (about 2166 The detected trends imply that wet spells in Singa-
mm at annual time scale, 879 mm and 572 mm during pore will become wetter and more extreme if the trends
the northeast monsoon season and southwest monsoon observed in the past 30 years continue in the future.
season, respectively) are mainly distributed in the eastern Similar results are shown by Beck et al. (2015) that
part of the island. Spatial variations of other wet and dry yearly precipitation totals in Singapore have increased
spell indices are also evident, due to space limitation, we significantly over the past 30 years, as well as the
skip here the detailed results. hourly and daily precipitation extremes. This could be
In summary, the region over Singapore shows statisti- attributed to a combined effect of global warming, anthro-
cally significant increase of ADWS at annual time scale. pogenic influences, natural climate variability (such as
The occurrence and accumulated depth of extreme wet ENSO), etc.

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

 SPATIO-TEMPORAL VARIATION OF PRECIPITATION SPELLS IN SINGAPORE 4843

(i) (j)

(k) (l)

(m) (n)

Figure 11. Continued

Table 3. Statistics of regional regression functions between the normalized wet and dry spell indices during the two monsoon seasons
and the averaged ONI of the same period.

Response variable (normalized) Slope Intercept Significance

TDDS 0.007 (0.123) 1.000 (0.995) 0.381 (0)
TDWS −0.012 (−0.206) 1.000 (1.009) 0.330 (0)
ADWS 0.002 (−0.301) 1.000 (1.013) 0.480 (0)
DSF −0.042 (−0.131) 0.999 (1.006) 0.003 (0)
WSF −0.045 (−0.131) 0.999 (1.006) 0.002 (0)
MDDS 0.049 (0.322) 1.001 (0.986) 0.250 (0)
MDWS 0.017 (−0.169) 1.000 (1.007) 0.349 (0.004)
MMWS 0.066 (−0.218) 1.002 (1.009) 0.202 (0.001)
DSD90 0.041 (0.421) 1.001 (0.982) 0.274 (0)
WSD90 0.044 (−0.328) 1.001 (1.014) 0.210 (0.003)
WSM90 −0.009 (−0.517) 1.000 (1.022) 0.447 (0)
TDDS90 0.069 (0.567) 1.002 (0.976) 0.185 (0)
TDWS90 0.051 (−0.385) 1.001 (1.016) 0.200 (0.001)
ADWS90 0.038 (−0.576) 1.001 (1.025) 0.337 (0)
Bold texts indicate the regressions are significant at 5% significance level. The values without parentheses are calculated for northeast monsoon
season and the values in parentheses are calculated for southwest monsoon season.

In addition, spatial variations of wet and dry spell indices monsoon season at the 26 precipitation stations and the
are also evident, which can be explained by the high spatial averaged ONI of the same period. The straight lines
variability of precipitation over the island. indicate the regional regression functions, whereas the
equations are provided in the legends of the figures. The
4.3. Regional regression analysis of ONI and wet and basic statistics for regional regression functions are listed
dry spell indices in Table 3. As can be seen from Figure 11 and Table 3,
Figure 11 provides an overview of scatterplots between the none of the wet and dry spell indices reacts to ONI except
normalized wet and dry spell indices during the northeast for dry and wet spell frequencies (DSF and WSF) during

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

4844 X. LI et al.

(a) (b)

(c) (d)

(e) (f)

(g) (h)

Figure 12. Dependence between Ocean Niño Index and the 14 wet and dry spell indices at each station during the southwest monsoon season.
(a) TDDS, (b) TDWS, (c) ADWS, (d) DSF, (e) WSF, (f) MDDS, (g) MDWS, (h) MMWS, (i) DSD90 , (j) WSD90 , (k) WSM90 , (l) TDDS90 , (m)
TDWS90 and (n) ADWS90 .

the northeast monsoon season. A higher ONI during the monsoon season will lead to less extreme wet spells
northeast monsoon season will lead to less frequent dry (MMWS), the occurrence of wet spells with duration
and wet spells. exceeding the 90% percentile of the duration distribution
Figure 12 summarizes all the scatterplots of the normal- (WSD90 ) and their contribution to TDWS90 will be lower.
ized wet and dry spell indices versus the average ONI dur- Besides, when the ONI is higher during the southwest
ing the southwest monsoon season. The results (Figure 12 monsoon season, the occurrence of wet spells with magni-
and Table 3) show that there are significant dependences tude exceeding the long-term 90% percentile of the mag-
between ONI and the entire set of wet and dry spell nitude distribution (WSM90 ) will be lower and their contri-
indices. A higher ONI during the southwest monsoon sea- bution to total accumulated depth of wet spells (ADWS90 )
son will lead to less frequent wet and dry spells (WSF and will also be lower. During southwest monsoon season,
DSF), longer TDDS, shorter TDWS and lower ADWS. maximum duration of wet spells (MDWS) and the occur-
In addition, higher ONI during the southwest monsoon rence of extreme wet spells (WSM90 ) become lower when
season will result in higher occurrence of dry spells ONI is higher. All these results prove that a higher ONI
with duration exceeding the long-term 90% percentile during the southwest monsoon season will lead to drier
of the duration distribution (DSD90 ) and larger contribu- conditions. This finding is consistent with the results found
tion to total duration (TDDS90 ), as well as more extreme by Beck et al. (2015) that seasonal precipitation totals are
dry spells (MDDS). At the same time, as indicated in lower during the El Niño phase for southwest monsoon
Table 3 and Figure 12, a higher ONI during the southwest season in Singapore.

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

 SPATIO-TEMPORAL VARIATION OF PRECIPITATION SPELLS IN SINGAPORE 4845

(i) (j)

(k) (l)

(m) (n)

Figure 12. continued

The dependences of wet and dry spell indices on ENSO spatial and temporal variation of wet and dry spell
can be expected because several studies have reported the characteristics based on 14 indices. The association of wet
connection between ENSO and variability of precipitation and dry spell indices with ENSO is also explored. Some
in the South East Asia region (Kripalani and Kulkarni, interesting conclusion can be drawn as follows:
1997; Kane, 1999; Xu et al., 2004; Beck et al., 2015),
especially the study performed by Beck et al. (2015) which 1 Singapore is predominantly characterized by 1-day wet
is exactly targeted at Singapore. As shown by Beck et al. and dry spells. However, the main contribution to TDWS
(2015), precipitation activities in Singapore are influenced and TDDS comes from 3-4- (3–6) day wet (dry) spell.
by ENSO, but not during the northeast monsoon season. The main contribution to total ADWS comes from
However, our finding indicates that during the northeast 3-4-day and 20–80 mm wet spells in terms of dura-
monsoon season, wet and dry spell frequencies are still tion and magnitude, respectively. Besides, extreme wet
impacted by ENSO. or dry spells contribute a substantial percentage to total
The association between precipitation regime and ENSO duration and total accumulated depth (for wet spells),
depends on the preferential atmospheric flow direction even though their occurrences are relatively low. This
(Beck et al., 2015). It seems that ENSO mainly influences gives practical implications for water resources manage-
the atmospheric circulation during the southwest monsoon ment in Singapore.
season over Singapore. To uncover the underlying reason, 2 Spatial variability of wet and dry spell characteristics is
the impact of ENSO on the trade winds strength and the found over the region of Singapore. This is related to
moisture content in the arriving air masses during the the fact that precipitation in Singapore highly varies in
two monsoon seasons over Singapore and the surrounding space.
areas would need to be investigated, which is out of the 3 Statistical significant increases of total ADWS, with ris-
scope of this study. ing occurrence of wet spells of extreme magnitude, and
increasing corresponding contribution to total accumu-
lated depth are found over most of the precipitation sta-
tions in Singapore at annual time scale. However, these
5. Summary and conclusions
trends are not statistically significant at 5% significance
In this article, we characterize the wet and dry spells of level at most of the stations during both the northeast
tropical precipitation in Singapore, and investigate the monsoon season and southwest monsoon season.

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

4846 X. LI et al.

4 Significant dependences are found between ENSO and Mann HB. 1945. Nonparametric tests against trend. Econometrica 13:
245–259.
the wet and dry spell indices, but mainly during south- Niedźwiedź T, Twardosz R, Walanus A. 2009. Long-term variability of
west monsoon season in Singapore. All of the 14 indices precipitation series in east central Europe in relation to circulation
react actively during the southwest monsoon season, patterns. Theor. Appl. Climatol. 98: 337–350.
Priyadarsini R, Hien WN, David CKW. 2008. Microclimatic modeling
while only wet and dry spell frequencies react to ENSO of the urban thermal environment of Singapore to mitigate urban heat
during the northeast monsoon season. As indicated dur- island. Sol. Energy 82: 727–745.
ing the southwest monsoon season, a higher ONI will Ratan R, Venugopal V. 2013. Wet and dry spell characteristics of global
tropical rainfall. Water Resour. Res. 49: 3830–3841.
lead to drier conditions in Singapore. During the north- Riley GT, Landin MG, Bosart LF. 1987. The diurnal variability of
east monsoon season, a higher ONI will result in less precipitation across the central Rockies and adjacent Great Plains.
wet and dry spells. Mon. Weather Rev. 115: 1161–1172.
Risbey JS, McIntosh PC, Pook MJ. 2013. Synoptic components of
5 The predicted changes of the wet and dry spell char- rainfall variability and trends in southeast Australia. Int. J. Climatol.
acteristics of precipitation in Singapore have practi- 33: 2459–2472.
cal implications for water resources management, flood Schmidli J, Frei C. 2005. Trends of heavy precipitation and wet and
dry spells in Switzerland during the 20th century. Int. J. Climatol. 25:
defence and drought mitigation, ecological design, etc. 753–771.
Future work will be carried out on the anthropogenic Shen Z, Chen L, Liao Q, Liu R, Hong Q. 2012. Impact of spatial rainfall
impact, such as urban heat island effect, change of land variability on hydrology and nonpoint source pollution modeling.
J. Hydrol. 472: 205–215.
cover and land use, etc., on wet and dry spell character- Singh N, Ranade A. 2010. The wet and dry spells across India during
istics of precipitation. 1951–2007. J. Hydrometeorol. 11: 26–45.
Singh D, Tsiang M, Rajaratnam B, Diffenbaugh NS. 2014. Observed
changes in extreme wet and dry spells during the South Asian summer
monsoon season. Nat. Clim. Chang. 4: 456–461.
Acknowledgements Sperber K, Palmer T. 1996. Interannual tropical rainfall variability
in general circulation model simulations associated with the Atmo-
The authors thank Mr. Chin Hong Ong from National spheric Model Intercomparison Project. J. Clim. 9: 2727–2750.
Environmental Agency of Singapore for his help in data Tabari H, Talaee PH. 2011. Temporal variability of precipitation over
Iran: 1966–2005. J. Hydrol. 396: 313–320.
processing. Turner J, Colwell S, Harangozo S. 1997. Variability of precipitation over
the coastal western Antarctic Peninsula from synoptic observations.
J. Geophys. Res. Atmos. 102: 13999–14007.
References Von Storch H, Zwiers FW. 1995. Misuses of Statistical Analysis in
Climate Research. In Analysis of Climate Variability, von Storch H,
Bai A, Zhai P, Liu X. 2007. Climatology and trends of wet spells in Navarra A (eds). Springer-Verlag Berlin Heidelberg: New York, NY,
China. Theor. Appl. Climatol. 88: 139–148. pp. 11–26, doi: 10.1007/978-3-662-03167-4_2.
Beck F, Bárdossy A, Seidel J, Müller T, Sanchis EF, Hauser A. 2015. Wang XL, Swail VR. 2001. Changes of extreme wave heights in North-
Statistical analysis of sub-daily precipitation extremes in Singapore. ern Hemisphere oceans and related atmospheric circulation regimes.
J. Hydrol.: Reg. Stud. 3: 337–358. J. Clim. 14: 2204–2221.
Busuioc A, Chen D, Hellström C. 2001. Temporal and spatial variability Wang A, Lettenmaier DP, Sheffield J. 2011. Soil moisture drought in
of precipitation in Sweden and its link with the large-scale atmospheric China, 1950–2006. J. Clim. 24: 3257–3271.
circulation. Tellus A 53: 348–367. Wuebbles D et al. 2014. CMIP5 climate model analyses: cli-
Cayan DR, Dettinger MD, Diaz HF, Graham NE. 1998. Decadal vari- mate extremes in the United States. Bull. Am. Meteorol. Soc.
ability of precipitation over western North America. J. Clim. 11: 95: 571–583.
3148–3166. Xu Z, Takeuchi K, Ishidaira H. 2004. Correlation between El
Chen H, Sun J. 2015. Changes in Drought Characteristics over China Niño–Southern Oscillation (ENSO) and precipitation in South-east
Using the Standardized Precipitation Evapotranspiration Index. Asia and the Pacific region. Hydrol. Process. 18: 107–123.
J. Clim. 28: 5430–5447. Zhai P, Zhang X, Wan H, Pan X. 2005. Trends in total precipitation
Chow WT, Roth M. 2006. Temporal dynamics of the urban heat island and frequency of daily precipitation extremes over China. J. Clim. 18:
of Singapore. Int. J. Climatol. 26: 2243–2260. 1096–1108.
Fong M, Ng LK. 2012. The weather and climate of Singapore. Meteoro- Zhang X, Zwiers FW. 2004. Comment on “Applicability of prewhitening
logical Service Singapore: Singapore. to eliminate the influence of serial correlation on the Mann-Kendall
Jarvis A, Reuter HI, Nelson A, Guevara E. 2008. Hole-filled SRTM for test” by Sheng Yue and Chun Yuan Wang. Water Resour. Res. 40:
the globe Version 4. http://srtm.csi.cgiar.org (accessed 18 December W03805, doi: 10.1029/2003WR002073.
2015). Zhang X, Vincent LA, Hogg W, Niitsoo A. 2000. Temperature
Kane R. 1999. El Niño timings and rainfall extremes in India, southeast and precipitation trends in Canada during the 20th century.
Asia and China. Int. J. Climatol. 19: 653–672. Atmosphere-Ocean 38: 395–429.
Kardinal Jusuf S, Wong NH, Hagen E, Anggoro R, Hong Y. 2007. The Zhang Q et al. 2008. Spatial and temporal variability of precipitation
influence of land use on the urban heat island in Singapore. Habitat maxima during 1960–2005 in the Yangtze River basin and possible
Int. 31: 232–242, doi: 10.1016/j.habitatint.2007.02.006. association with large-scale circulation. J. Hydrol. 353: 215–227.
Kendall MG. 1955. Rank Correlation Methods. Griffin: London. Zhang Q, Peng J, Singh VP, Li J, Chen YD. 2014. Spatio-temporal
Kitanidis P. 1997. Introduction to Geostatistics: Applications in Hydro- variations of precipitation in arid and semiarid regions of China:
geology. Cambridge University Press. the Yellow River basin as a case study. Glob. Planet. Change.
Kripalani R, Kulkarni A. 1997. Rainfall variability over south-east 114: 38–49.
Asia-connections with Indian monsoon and ENSO extremes: new Zolina O, Simmer C, Belyaev K, Gulev SK, Koltermann P. 2013.
perspectives. Int. J. Climatol. 17: 1155–1168. Changes in the duration of European wet and dry spells during the
Mandapaka PV, Qin X. 2013. Analysis and characterization of prob- last 60 years. J. Clim. 26: 2022–2047.
ability distribution and small-scale spatial variability of rainfall in Zou X, Zhai P, Zhang Q. 2005. Variations in droughts over China:
Singapore using a dense gauge network. J. Appl. Meteorol. Climatol. 1951–2003. Geophys. Res. Lett. 32: L04707, doi: 10.1029/
52: 2781–2796. 2004GL021853.

© 2016 Royal Meteorological Society Int. J. Climatol. 36: 4831–4846 (2016)

View publication stats