Full Terms & Conditions of access and use can be found at
ISSN: 0262-6667 (Print) 2150-3435 (Online) Journal homepage: https://www.tandfonline.com/loi/thsj20
Historical trends and variability of meteorological droughts in Taiwan / Tendances historiques et variabilité des sécheresses météorologiques à Taiwan
SHIEN-TSUNG CHEN , CHUN-CHAO KUO & PAO-SHAN YU
To cite this article: SHIEN-TSUNG CHEN , CHUN-CHAO KUO & PAO-SHAN YU (2009) Historical trends and variability of meteorological droughts in Taiwan / Tendances historiques et variabilité des sécheresses météorologiques à Taiwan, Hydrological Sciences Journal, 54:3, 430-441, DOI:
10.1623/hysj.54.3.430
To link to this article: https://doi.org/10.1623/hysj.54.3.430
Published online: 19 Dec 2009.
Submit your article to this journal
Article views: 1075
View related articles
Citing articles: 39 View citing articles
Historical trends and variability of meteorological droughts in Taiwan
SHIEN-TSUNG CHEN, CHUN-CHAO KUO & PAO-SHAN YU
Department of Hydraulic and Ocean Engineering, National Cheng Kung University, No. 1, University Road, Tainan City, 701, Taiwan
yups@ncku.edu.tw
Abstract This work investigates historical trends of meteorological drought in Taiwan by means of long- term precipitation records. Information on local climate change over the last century is also presented.
Monthly and daily precipitation data for roughly 100 years, collected by 22 weather stations, were used as the study database. Meteorological droughts of different levels of severity are represented by the standard- ized precipitation index (SPI) at a three-monthly time scale. Additionally, change-point detection is used to identify meteorological drought trends in the SPI series. Results of the analysis indicate that the incidence of meteorological drought has decreased in northeastern Taiwan since around 1960, and increased in central and southern Taiwan. Long-term daily precipitation series show an increasing trend for dry days all over Taiwan. Finally, frequency analysis was performed to obtain further information on trends of return periods of drought characteristics.
Key words drought; trend; standardized precipitation index; climate variability; climate change; Taiwan
Tendances historiques et variabilité des sécheresses météorologiques à Taiwan
Résumé Cet article étudie les tendances historiques de la sécheresse météorologique à Taiwan sur la base de longues séries de précipitations. Les informations relatives au changement climatique au cours du siècle dernier sont également présentées. Les données de précipitations mensuelle et journalière sur environ 100 ans, collectées par 22 stations météorologiques, ont été utilisées comme base de l’étude. Les sécheresses météorologiques de niveaux de sévérité différents sont représentées selon l’indice standardisé de précipita- tion (ISP) à pas de temps trimestriel. De plus, la détection de rupture est utilisée pour identifier les tendances de sécheresse météorologique dans les séries d’ISP. Les résultats de l’analyse indiquent que l’incidence de la sécheresse météorologique a diminué dans le nord-est de Taiwan depuis 1960 environ, et augmenté dans le centre et le sud de Taiwan. Les longues séries de précipitation journalière présentent une tendance croissante pour les jours secs pour l’ensemble de Taiwan. Finalement, une analyse fréquentielle a été réalisée pour obtenir des informations supplémentaires sur les tendances des périodes de retour des caractéristiques de sécheresse.
Mots clefs sécheresse; tendance; indice standardisé de précipitation; variabilité climatique; changement climatique;
Taiwan
INTRODUCTION
A drought is a prolonged period of water deficit, and typically occurs when an area receives precipitation below usual levels for several months. Drought can be classified as meteorological, agricultural or hydrological, based on the different components of the hydrological cycle affected by a drought event. Meteorological drought is an extended period with precipitation below normal levels, and usually appears before other drought types. Meteorological drought in this work is defined using the standardized precipitation index (SPI) proposed by McKee et al. (1993). The SPI is easily applied and has other advantages, as discussed by Hayes et al. (1999). Therefore, the SPI is widely used to investigate drought characteristics. For example, Guttman (1998) compared the SPI with the Palmer drought index (PDI, Palmer, 1965), and demonstrated that SPI spectral characteristics do not vary among sites, whereas those of the PDI do. Guttman (1998) also showed that the SPI is an easily interpreted moving-average process, while the PDI has a complex structure with a long memory. Lloyd-Hughes & Saunders (2002) applied the SPI to analyse European droughts for 1901–1999, and indicated that the mean number and duration of extreme droughts were about six events and 27 months, respectively. Vicente-Serrano (2006) performed spatial and temporal analysis of droughts using the SPI on the Iberian Peninsula for 1910–2000, and identified the principal drought episodes; the most intense droughts were in the 1940s, 1950s, 1980s and 1990s. Moreover, the SPI has been used to investigate meteorological droughts in many
countries and regions, for example: Canada (Quiring & Papakryiakou, 2003); China (Wu et al., 2001); East Africa (Ntale & Gan, 2003); Greece (Tsakiris & Vangelis, 2004); India (Mishra &
Desai, 2005); Italy (Bonaccorso et al., 2003; Piccarreta et al., 2004); Spain (Lana et al., 2001);
Taiwan (Shiau, 2006); and the USA (Edwards & McKee, 1997; Ji & Peters, 2003).
Although the average precipitation in Taiwan is high, the country is prone to droughts due to the uneven temporal distribution of the precipitation and the small storage capacity of reservoirs.
A serious drought affected Taiwan in the spring of 2002, and had a massive impact on people, industry and agriculture throughout Taiwan. Huang & Yuan (2004) discussed the reservoir operation in response to the 2002 drought in northern Taiwan. The present study focuses on historical trends of meteorological drought all over Taiwan. Long-term monthly precipitation data were collected for Taiwan and the SPI series calculated to investigate historical trends and variability of meteorological droughts. Change-point detection was used to determine the most statistically significant change point in the SPI series. Trends were then identified by comparing drought characteristics before and after the change point. Additionally, available long-term daily precipitation data were used to identify any trend for dry days, which is also an indicator of meteorological drought. Finally, changes in the return period of drought characteristics were investigated, and trends and variability of meteorological droughts in Taiwan over the last century were identified.
CLIMATE AND LONG-TERM PRECIPITATION RECORDS IN TAIWAN
Taiwan is an island located in the northwestern Pacific Ocean, about 120 km off the southeastern coast of the Asian Continent. The climate is marine tropical, with a rainy season from May to October and a dry season from November to April. Four water resource regions in Taiwan are delineated according to river basins and climatic conditions by the Water Resources Agency (Fig. 1). Taiwan’s average annual precipitation is as high as 2510 mm. However, abundant precipi- tation has an uneven temporal distribution, especially in southern Taiwan, because typhoons deliver most of the annual precipitation over a period of a few days. In southern Taiwan, for example, 90% of annual rainfall occurs during the wet season, while only 10% falls during the dry season. Therefore, when typhoons bring less than normal amounts of precipitation during the wet season in a given year, drought is likely in the spring of the following year.
Northern Taiwan 5 Stations
Central Taiwan 7 Stations
Southern Taiwan 8 Stations
Eastern Taiwan 2 Stations
4 5
3 2 1
9 10 11 876
12 14 13
15 16 17
18
19
20 21
22
Fig. 1 Long-term raingauges in Taiwan.
This study collected long-term monthly precipitation records (>80 years) and station informa- tion in Taiwan. Preliminary inspection was made to remove suspicious data and stations, with location change and data missing longer than five years. Consequently, long-term data from 32 stations provide the study database. Among them, only 11 stations have complete, high-quality records, and other stations have missing records in a few months or years, particularly during the late 1940s. Missing data were reconstructed using the regression method with resorted available records from neighbouring raingauges (not restricted to the 32 stations with long-term records).
The quality-control criteria for reconstructing monthly precipitation data are as follows: (a) the simple regression formula made by one neighbouring gauge passes the F test with a significance level of 0.05, and the coefficient of determination R2 > 0.6; or, (b) the multiple regression formula made by a few neighbouring gauges passes the F test with a significance level of 0.05, and R2 > 0.6. Thus, reconstructed data of 11 raingauges passed quality control. Finally, only 22 stations with quality data provide long-term monthly precipitation records for this study. Table 1 lists the precipitation data, as well as regional precipitation data (northern, central and southern regions) calculated using the arithmetic averaging method. Among these, only eight raingauges have complete, high-quality long-term daily precipitation data for analysing changes in dry days.
Table 1 Information of stations and monthly precipitation.
District Station Record years Record duration (year)
Annual precipitation (mm/year)
Station altitude (m a.s.l.)
Period of missing data
(yyyy/mm) Note 1 1917−2005 89 2020.0 19.0 Complete * 2 1897−2005 109 2160.3 5.3 Complete * 3 1904−2005 102 2227.2 33.0 1945/08–1949/12
1955/08
1957/01–1957/03 4 1922−2005 84 1632.2 106.0 1945/01–1948/12
1997/08 North
5 1912−2005 94 1697.3 34.0 Complete * Region 1922−2005 84 1961.2
6 1904−2005 102 1967.3 337.0 1945/01–1947/12 1949/01–1949/05 7 1923−2005 83 1826.3 210.0 1945/01–1947/12
1955/01–1955/12
8 1897−2005 109 1710.0 84.0 Complete * 9 1923−2005 83 1409.3 9.0 1945/01–1947/12
10 1922−2005 84 1279.9 40.0 1945/01–1947/12 11 1922−2005 84 1353.4 16.0 1945/01–1946/12
1967/01–1967/12 1981/01–1981/12 Central
12 1922−2005 84 2037.7 111.0 Complete Region 1923−2005 83 1650.5
13 1904−2005 102 2506.3 120.0 1924/07–1924/08 1945/01–1946/04 1947/11–1947/12 1949/09–1949/10 14 1916−2005 90 1697.2 18.0 1945/10–1945/12
1951/03–1951/04
15 1900−2005 106 1730.4 8.1 Complete * 16 1924−2005 82 1825.3 5.0 Complete
17 1911−2005 95 2371.8 46.0 1945/04–1946/01 18 1916−2005 90 2589.7 40.0 Complete 19 1904−2005 102 1755.8 4.0 1945/01–1946/12 South
20 1900−2005 106 2154.8 22.1 Complete * Region 1924−2005 82 2081.7
21 1901−2005 105 1826.2 9.0 Complete * East
22 1901−2005 105 2044.3 16.1 Complete *
* indicates stations with long-term daily precipitation records.
METHODOLOGY
Meteorological drought and drought characteristics
The SPI at the three-month time scale is used to define meteorological drought in this work. The SPI computation is only based on precipitation and can be calculated for any time scale (e.g. 1, 3, 6 months). The first step in SPI calculation is to fit long-term precipitation data to a probability distribution. The cumulative probability of precipitation can then be derived from the probability distribution function. Next, this cumulative probability is transformed into a standard normal variable with a mean of zero and variance of 1. Last, the SPI takes on the value of a standard normal variable. A detailed description of the SPI calculation can be found in Lloyd-Hughes &
Saunders (2002). Negative SPI values indicate that precipitation is less than the median. A drought event is thus defined as a period in which the SPI is continuously negative. McKee et al. (1993) proposed four drought categories—mild, moderate, severe and extreme droughts—when the minimum SPI value of a drought event is less than 0, −1, −1.5 and −2, respectively. Figure 2 presents the drought events in each category and drought characteristics such as drought duration (D) and drought magnitude (M):
∑
=−
= D
i
i M
1
) (
SPI (1)
Therefore, each drought category can be described using three characteristics: drought frequency (F), drought duration (D) and drought magnitude (M).
–3 –2 –1 0 1 2 3
Mild drought
Moderate drought
Severe drought
Extreme drought
SPI D
∑
=−
= D
i
i M
1
) ( SPI
Fig. 2 Drought categories and characteristics.
Drought return period
A return period is the average time interval between events of a certain intensity or greater. The return period, usually used as a design criterion in flood problems, is the inverse of annual probability that a specified value will be exceeded. Derivation of the return period is usually based on frequency analysis of annual maximum series; however, a drought event may last longer than one year and multiple droughts can occur in a given year. Shiau & Shen (2001) and Shiau (2003) developed theoretically the return period for univariate- and bivariate-distributed partial duration series, respectively, and applied this method to drought frequency analysis. Let T be the return period of a drought event with an intensity of X ≥ x; thus, T can be expressed as:
) ( 1
) (
x F
L T E
= − (2)
)
where L is the time period between any two consecutive drought events without considering intensity, E(L) is the expected value of L, and F(x) is the cumulative distribution function of X.
Therefore, denominator [1 – F(x)] in equation (2) is the probability of a drought event occurring that has intensity equal to or exceeding x.
Trend analysis
This work identified the trend for meteorological droughts and dry days using change point detection, which is a method for determining whether a change has occurred in a given time series.
Change point analysis can detect multiple changes. In this work, only the most significant change point is highlighted. Many change point detection approaches exist, and many works have discussed issues affecting the detection results, e.g. the impacts of serial and cross-correlation (Yue & Wang, 2002; Yue et al., 2003). This study applied a non-parametric statistical test, the Mann-Whitney-Pettitt (MWP) method developed by Pettitt (1979), to determine the change point.
The MWP statistic is briefly described as follows.
Consider a time series {X1, X2, …, Xn} with a length n. Let t be the time of the most likely change point. Two samples, {X1, X2, …, Xt} and {Xt+1, X t+2, …, Xn}, can then be derived by dividing the time series at time t. An index, Ut, is derived by:
∑ ∑ (
= =+
−
= t
i n t j
j i
t X X
U
1 1
sgn (3)
where sgn(x) = 1 if x > 0, sgn(x) = 0 if x = 0 and sgn(x) = –1 if x < 0. The most significant change point can be identified at time t where Ut is maximum. The approximated significance probability p(t) for a change point (Pettitt, 1979; Kiely, 1999) is given as:
⎟⎟⎠
⎜⎜ ⎞
⎝
⎛ +
− −
= 3 2
6 2
exp 1 )
( n n
t U
p t (4)
The change point is statistically significant at time t with a significance level of α when probability p(t) exceeds (1 – α).
TRENDS OF METEOROLOGICAL DROUGHTS Trends in SPIs
The SPI is calculated using monthly precipitation data to reflect rainfall surplus or deficit relative to median precipitation. Because the SPI in a 1-month time scale fluctuates between positive and negative values, detection of the start and end of a drought event is improper. Therefore, this work used the SPI with a 3-month time scale (SPI3) to identify drought events. The first step in calculating SPI3 is to compute the precipitation in a 3-month time scale using a moving average scheme. The 3-month precipitation at month t is calculated using rainfall data at month t – 2, t – 1 and t. Data for the 3-month precipitation series were fitted to a probability distribution. McKee et al. (1993) and Edwards & McKee (1997) suggested using a gamma distribution. This work examined the goodness of fit of a gamma distribution using the Kolmogorov-Smirnov test pertaining to precipitations at all stations and in all regions. Analytical results show that a gamma distribution fits all 3-month precipitation data with a significance level of 0.05. Finally, cumulative probability of the 3-month precipitation series, derived from the gamma distribution, can be transformed into the SPI3 series with a mean of zero and variance of 1.
The SPI3 trends are determined via change point detection. The MWP method identifies the most significant change points in the SPI3 series. Table 2 lists these change points and the approximated significance probability, p(t), for that corresponding change point. These change points are all statistically significant based on the significance level of 0.05. Change points of the
Table 2 Change point detection of SPI3 series.
District Station Change point
(yyyy/mm) p(t) value
1 1946/10 1.00
2 1969/08 1.00
3 1941/08 0.97
4 1969/07 0.98
North
5 1967/10 1.00
Region 1967/10 0.99
6 1945/12 1.00
7 1954/01 1.00
8 1953/12 0.99
9 1953/12 1.00
10 1950/10 1.00
11 1960/10 1.00
Central
12 1957/07 1.00
Region 1953/12 1.00
13 1954/04 1.00
14 1957/07 1.00
15 1957/07 1.00
16 1957/07 1.00
17 1954/04 1.00
18 1954/04 1.00
19 1961/11 1.00
South
20 1961/11 1.00
Region 1960/01 1.00
21 1976/08 0.95
East
22 1938/01 1.00
Note: All change points are statistically significant.
SPI3 series at stations in northern Taiwan are in the range of 1941–1969, while that for regional precipitation occurs in 1967. Change points in central Taiwan vary during 1945–1960, and the change point of regional precipitation is in 1953. In southern Taiwan, change points vary in 1954–
1961, with a change point of regional precipitation in 1960. Two stations in eastern Taiwan have change points in 1976 and 1938. There is a temporal agreement between change points from north to south. Generally, change points of the SPI3 series are around 1960, within a decade before or after. Analytical results for change points of the SPI3 series (monthly precipitation) are consistent with those of annual precipitation identified by Yu et al. (2006), in which change points are also around 1960.
Trends of drought characteristics
Meteorological drought characteristics were then estimated pertaining to the former and the latter SPI3 series split by the change point. Trends for drought characteristics for individual sites are generally consistent with those for water resource regions. In short, drought characteristics in the northern, central and southern regions and for stations 21 and 22 are presented. Figure 3 shows the changes in drought characteristics in frequency (time/year), duration (month) and magnitude (–
Σ
SPI3). Trends in these regions can be divided into two groups (Fig. 3). Trends in northern Taiwan and at station 22 are similar, and can be viewed as Group I. Conversely, trends in the central and southern regions and at station 21 are comparable, and can be grouped as Group II.Group I covers northeastern Taiwan, while Group II includes central and southern Taiwan.
The frequency of mild droughts in Group I increased, while that of moderate, severe and extreme droughts decreased. Drought duration and magnitude in Group I exhibited a decreasing
0.0 0.5 1.0 1.5
Frequency (time/year) North Taiwan Before After
Mild Moderate Severe Extreme 0 5 10 15
Duration (month)
Mild Moderate Severe Extreme 0 5 10 15 20
Magnitude (-∑SPI3)
Mild Moderate Severe Extreme
0.0 0.5 1.0 1.5
Frequency (time/year) Central Taiwan Before After
Mild Moderate Severe Extreme 0 5 10 15
Duration (month)
Mild Moderate Severe Extreme 0 5 10 15 20
Magnitude (-∑SPI3)
Mild Moderate Severe Extreme
0.0 0.5 1.0 1.5
Frequency (time/year) South Taiwan Before After
Mild Moderate Severe Extreme 0 5 10 15
Duration (month)
Mild Moderate Severe Extreme 0 5 10 15 20
Magnitude (-∑SPI3)
Mild Moderate Severe Extreme
0.0 0.5 1.0 1.5
Frequency (time/year) Station 21 Before After
Mild Moderate Severe Extreme 0 5 10 15
Duration (month)
Mild Moderate Severe Extreme 0 5 10 15 20
Magnitude (-∑SPI3)
Mild Moderate Severe Extreme
0.0 0.5 1.0 1.5
Frequency (time/year) Station 22 Before After
Mild Moderate Severe Extreme 0 5 10 15
Duration (month)
Mild Moderate Severe Extreme 0 5 10 15 20
Magnitude (-∑SPI3)
Mild Moderate Severe Extreme
Fig. 3 Changes to drought characteristics.
trend, except for those of severe drought. Overall, the trend of meteorological droughts in Group I decreased over the last century. As for Group II, drought frequency has no clear trend. However, drought duration and magnitude in Group II increased, especially for extreme drought. Generally, the trend of meteorological droughts in Group II has increased over the last century. Conclusively, the incidence of meteorological droughts has decreased in northeastern Taiwan and increased in central and southern Taiwan. These trends are in good agreement with precipitation trends investigated by Hsu & Chen (2002) and Yu et al. (2006), who concluded that the amount of precipitation in Taiwan increased in northern Taiwan and decreased in southern Taiwan over the last century.
Trends for dry days
Dry days provide information for understanding of meteorological droughts. Among the stations with long-term precipitation records, eight maintained by the Central Weather Bureau have long- term complete daily precipitation data. Therefore, these data are used to investigate changes in number of dry days using the MWP method. Various thresholds can be used for daily precipitation to define a dry day, or non-rainy day. This work adopts the threshold of precipitation ≤ 0.6 mm/d as the definition for a dry day; this definition is commonly used in the water resources sector in Taiwan. Table 3 lists the trends for dry days. All stations exhibited an increasing trend for number of dry days, while data for two stations were not statistically significant. The increase of dry days ranges from 10 to 25 d. The change points in number of annual dry days are around 1960 and 1970, and are comparable with those of the SPI3 series. Table 4 lists the changes in maximum number of consecutive dry days. Changes are statistically significant at four stations. These sig- nificant changes for maximum consecutive dry days are all increasing at 3–11 d. Overall, ana- lytical results show that the number of dry days and maximum consecutive dry days increased in Taiwan over the last century.
Table 3 Trends for dry days.
Dry days (d/year):
Station Change point
(year) p(t) value
Before
change point After
change point Change
1 1975 0.91 229 235 +6
2 1961 1.00 222 232 +11 5 1986 0.90 260 271 +11 8 1953 1.00 267 277 +10
15 1961 1.00 281 295 +15
20 1961 1.00 254 270 +15
21 1978 1.00 253 271 +18
22 1975 1.00 215 240 +25
Note: Statistically significant values are in bold.
Table 4 Trends for maximum consecutive dry days.
Maximum consecutive dry days (d/year):
Station Change point
(year) p(t) value
Before
change point After
change point Change
1 1953 0.86 20 23 +3
2 1957 0.99 17 20 +3
5 1945 0.55 30 28 −2
8 1981 0.71 43 51 +8
15 1958 0.99 48 59 +11
20 1954 0.99 30 37 +7
21 1942 0.88 25 23 −2
22 1977 1.00 17 22 +5
Note: Statistically significant values are in bold.
Change of return period of drought characteristics
This section discusses changes in the return period for drought duration and magnitude. Drought duration, as used in this work, is a discrete variable, and is fitted as a geometric distribution by Kendall & Dracup (1992) and Mathier et al. (1992). However, if drought duration is considered a continuous variable, it can be described by an exponential distribution (Zelenhasic & Salvai, 1987;
Shiau, 2006). On the other hand, drought magnitude is usually fitted as a gamma distribution (Zelenhasic & Salvai, 1987; Mathier et al., 1992; Shiau, 2006). This work applied the Kolmogorov-Smirnov test to examine the goodness of fit of the distributions, and confirmed that
0 4 8 12 16 20 Drought Duration (month)
0 20 40 60 80 100
Return Period (year)
North Taiwan Before After
0 5 10 15 20 25
Drought Magnitude (∑SPI3) 0
20 40 60 80 100
Return Period (year)
North Taiwan Before After
0 4 8 12 16 20
Drought Duration (month) 0
20 40 60 80 100
Return Period (year)
Central Taiwan Before After
0 5 10 15 20 25
Drought Magnitude (∑SPI3) 0
20 40 60 80 100
Return Period (year)
Central Taiwan Before After
0 4 8 12 16 20
Drought Duration (month) 0
20 40 60 80 100
Return Period (year)
South Taiwan Before After
0 5 10 15 20 25
Drought Magnitude (∑SPI3) 0
20 40 60 80 100
Return Period (year)
South Taiwan Before After
0 4 8 12 16 20
Drought Duration (month) 0
20 40 60 80 100
Return Period (year)
Station 21 Before After
0 5 10 15 20 25
Drought Magnitude (∑SPI3) 0
20 40 60 80 100
Return Period (year)
Station 21 Before After
0 4 8 12 16 20
Drought Duration (month) 0
20 40 60 80 100
Return Period (year)
Station 22 Before After
0 5 10 15 20 25
Drought Magnitude (∑SPI3) 0
20 40 60 80 100
Return Period (year)
Station 22 Before After
Fig. 4 Changes to return period of drought characteristics.
1945 1946 1947 2001 2002 2003 Change Point: Jan. 1960
D= 10
M= 9.7 D= 13
M= 12.1
-3 -2 -1 0 1 2 3
SPI3
Fig. 5 Two extreme drought events in southern Taiwan.
Table 5 The most severe droughts in southern Taiwan after the change point.
Event Duration (month) Magnitude
1 23 26.8
2 21 20.6
3 17 16.2
4 17 10.9
5 12 18.3
2002 Drought 13 12.1
Note: Values greater than that of the 2002 drought are given in bold.
an exponential distribution and gamma distribution accurately describe drought duration and drought magnitude for examined data.
Figure 4 shows trends and changes to drought duration and magnitude before and after the change point. Return periods of drought events of equal characteristics (duration and magnitude) increase after the change point for northern Taiwan and Station 22, but decrease for central and southern Taiwan and Station 21, e.g. Fig. 5 shows two extreme drought events in southern Taiwan, one before and one after the change point. The 1946 drought is the most serious drought during the former period (36 years) before the change point; this drought lasted 10 months and had a magni- tude of 9.7. Both drought duration and magnitude attain maximum values. According to the fitted probability distributions, the 1946 drought has return periods of 23 years (duration) and 54 years (magnitude). Such a drought after the change point has return periods of only 7 years (duration) and 6 years (magnitude). Figure 5 also shows another extreme drought (the 2002 drought) with a duration of 13 months and magnitude of 12.1. Table 5 shows that four events are longer than 13 months, and another four events have drought magnitudes of >12.1. Therefore, the 2002 drought ranks fifth during the latter period after the change point. Based on the fitted probability distribution, the 2002 drought has return periods of 13 years (duration) and 9 years (magnitude).
The 2002 drought was more severe than the 1946 drought, but was a more usual event in recent years.
CONCLUSIONS
This work investigated historical trends of meteorological droughts in Taiwan over the last century based on long-term monthly and daily precipitation data. The SPI3 series, as calculated from 3-month precipitation, was used to define droughts into four classes: mild, moderate, severe and extreme droughts. The MWP method was applied to identify the change point in the SPI3 series and reveal trends. Analytical results show that change points of the SPI3 series are around 1960,
and are similar to those based on annual precipitation series. Trends pertaining to individual sites are commonly comparable to those pertaining to neighbouring sites. Thus, trends of drought characteristics can be spatially divided into northeastern Taiwan (Group I) and central and southern Taiwan (Group II). Generally, the incidence of meteorological droughts in Group I decreased and that in Group II increased. As for dry days, however, statistically significant changes demonstrate that the number of dry days increased. Moreover, this work applied frequency analysis to drought duration and magnitude, and obtained drought frequency curves before and after the change point. Subsequently, return periods of historical drought events show that southern Taiwan has suffered more severe meteorological droughts in recent years than before.
Acknowledgements The authors thank the National Science Council of the Republic of China, Taiwan (Project no. NSC95-2625-Z-006-016) and National Cheng Kung University (Project no.
R046) for partly supporting this study.
REFERENCES
Bonaccorso, B., Bordi, I., Cancelliere, A., Rossi, G. & Sutera, A. (2003) Spatial variability of drought: an analysis of the SPI in Sicily. Water Resour. Manage. 17, 273–296.
Edwards, D. C. & McKee, T. B. (1997) Characteristics of 20th century drought in the United States at multiple time scales.
Climatology Report no. 97-2, Department of Atmospheric Science, Colorado State University, USA.
Guttman, N. B. (1998) Comparing the Palmer Drought Index and the standardized precipitation index. J. Am. Water Resour.
Assoc. 34(1), 113–121.
Hayes, M. J., Svoboda, M. D., Wilhite, D. A. & Vanyarkho, O. V. (1999) Monitoring the 1996 drought using the standardized precipitation index. Bull. Am. Met. Soc. 80(3), 429–438.
Hsu, H. H. & Chen, C. T. (2002) Observed and projected climate change in Taiwan. Met. Atmos. Phys. 79, 87–104.
Huang, W. C. & Yuan, L. C. (2004) A drought early warning system on real-time multireservoir operations. Water Resour. Res.
40, W06401, doi:10.1029/2003WR002910.
Ji, L. & Peters, A. J. (2003) Assessing vegetation response to drought in the northern Great Plains using vegetation and drought indices. Remote Sens. Environ. 87, 85–98.
Kendall, D. R. & Dracup, J. A. (1992) On the generation of drought events using an alternating renewal-reward model.
Stochast. Hydrol. Hydraul. 6(1), 55–68.
Kiely, G. (1999) Climate change in Ireland from precipitation and streamflow observations. Adv. Water Resour. 23(2), 141–151.
Lana, X., Serra, C. & Burgueno, A. (2001) Patterns of monthly rainfall shortage and excess in terms of the standardized precipitation index for Catalonia (NE Spain). Int. J. Climatol. 21(13), 1669–1691.
Lloyd-Hughes, B. & Saunders, M. A. (2002) A drought climatology for Europe. Int. J. Climatol. 22, 1571–1592.
Mathier, L., Perreault, L., Bobée, B. & Ashkar, F. (1992) The use of geometric and gamma-related distributions for frequency analysis of water deficit. Stochast. Hydrol. Hydraul. 6(4), 239–254.
McKee, T. B., Doesken, N. J. & Kleist, J. (1993) The relationship of drought frequency and duration to time scales. In: Proc.
Eighth Conf. on Applied Climatology (17–22 January 1993, Anaheim, California.), 179–184.
Mishra, A. K. & Desai, V. R. (2005) Drought forecasting using stochastic models. Stochast. Environ. Res. Risk Assess. 19, 326–339.
Ntale, H. K. & Gan, T. Y. (2003) Drought indices and their application to East Africa. Int. J. Climatol. 23(11), 1335–1357.
Palmer, W.C. (1965) Meteorological Drought. Research Paper no. 45, US Weather Bureau, Washington, DC, USA.
Pettitt, A. N. (1979) A non-parametric approach to the change-point problem. Appl. Statist. 28(2), 126–135.
Piccarreta, M., Capolongo, D. & Boenzi, F. (2004) Trend analysis of precipitation and drought in Basilicata from 1923 to 2000 within a southern Italy context. Int. J. Climatol. 24, 907–922.
Quiring, S. M. & Papakryiakou, T. N. (2003) An evaluation of agricultural drought indices for the Canadian prairies. Agric.
For. Met. 118, 49–62.
Shiau, J. T. (2003) Return period of bivariate distributed extreme hydrological events. Stochast. Environ. Res. Risk Assess. 17, 42–57.
Shiau, J. T. (2006) Fitting drought duration and severity with two-dimensional copulas. Water Resour. Manage. 20, 795–815.
Shiau, J. T. & Shen, H. W. (2001) Recurrence analysis of hydrological droughts of differing severity. J. Water Resour. Plan.
Manage. 127(1), 30–40.
Tsakiris, G. & Vangelis, H. (2004) Towards a drought watch system based on spatial SPI. Water Resour. Manage. 18, 1–12.
Vicente-Serrano, S. M. (2006) Spatial and temporal analysis of droughts in the Iberian Peninsula (1910–2000). Hydrol. Sci. J.
51(1), 83–91.
Wu, H., Hayes, M. J., Weiss, A. & Hu, Q. (2001) An evaluation of the standardized precipitation index, the China-Z Index and the statistical Z-Score. Int. J. Climatol. 21, 745–758.
Yu, P. S., Yang, T. C. & Kuo, C. C. (2006) Evaluating long-term trends in annual and seasonal precipitation in Taiwan. Water Resour. Manage. 20(6), 1007–1023.
Yue, S., Pilon, P. & Phinney, B. (2003) Canadian streamflow trend detection: impacts of serial and cross-correlation. Hydrol.
Sci. J. 48(1), 51–63.
Yue, S. & Wang, C. Y. (2002) Applicability of prewhitening to eliminate the influence of serial correlation on the Mann- Kendall test. Water Resour. Res. 38(6), 1068, doi:10.1029/2001WR000861.
Zelenhasic, E. & Salvai, A. (1987) A method of streamflow drought analysis. Water Resour. Res. 23(1), 156–168.
Received 16 June 2008; accepted 29 December 2008
