Spatio-temporal analysis of local climate variables 2.4
2.4.2 Hydro-climate trend analysis
According to the IPCC AR4 (Solomon et al. 2007), the projected climate simulations in Southeast Asia over the 21st-century result overall in an average increase of the mean annual precipitation by about 7%. More specifically, the rainfall in this region will be increasing in the wet, but decreasing in the dry season. However, there are large local variations in the change of climate in this region. For example, the rainfall potentially will decrease in the southern part of Southeast Asia, e.g. Indonesia (Boer and Faqih 2004) while it will increase in its northern part, e.g. Vietnam (Knutson et al. 2001, Walsh 2004).
a) Max temperature
b) Precipitation
c) Runoff
Dry Wet Dry
Winter
Winter Summer Rainy
Monsoon2 Monsoon1
Pre-monsoon Dry
max temperature (C)precipitation (mm/day)runoff (m3/s)
For the country Thailand itself, the analysis of meteorological data of the Thai Meteorological Department by Vongvisessomjai (2010) indicates decreasing trends in the annual rainfall and decreasing trends in the temperature for the 1951-2005 period. However, the author also concluded that these trends are only minor and depend on the region considered. Thus, for the whole eastern part of the country, where also this study area is located, the rainfall has a maximum decrease, while it is increasing in its northern part. The study of Vongvisessomjai (2010) also established that the mean minimum and maximum surface temperatures are overall slightly increasing.
In this section, the past and present-day variations in the hydro-meteorological time series are analyzed for trends and other periodicities. In fact, it is well known that any arbitrary (stochastic) time series can be decomposed into a trend, seasonal variation and random part:
where is the trend, is a periodic (seasonal) variation and is a stochastic (random) component that cannot be modelled (Hipel and McLeod 1994). One objective of time series analysis is to determine these components. Using the STL- (seasonal, trend, loess) filtering procedure (Cleveland et al. 1990), as implemented in the R-software environment, the climate time-series are decomposed into trend, seasonal variation and random part.
Figure 2.9 shows this decomposition for minimum temperature and precipitation at two climate stations. On may notice that the trends during the 1971-2006 time period are generally positive.
Furthermore, the periodic variation of the precipitation shows again the two-peak semi-annual seasonal pattern, as already emphasized in the previous section.
Figure 2.9. Additive decomposition of the time series of minimum temperature at station 48478 (left panel) and precipitation at station 48092 (right panel) into a trend, seasonal cycle and random noise.
The time series and anomalies of temperature, rainfall and streamflow are exhibited in Figure 2.10 to Figure 2.12 and demonstrate the variation of the climate series in the study area. The anomalies are computed by subtracting the long-year monthly average, which is taken over the total time period, from each monthly value. By doing so, the seasonal (monthly) variation in the
Addititive decomposition of minimum temperature Addititive decomposition of precipitation
(2.7)
time series is removed, so that the anomaly-climate series show the trends and variation of the monthly climate between 1971-2006 more clearly. One may notice that while there is a clear and continuous positive trend in the temperature anomalies (Figure 2.10) over this time period, the rainfall and streamflow anomalies (Figure 2.11 and Figure 2.12) oscillate around the monthly mean, with no visible trend. It may also be noted that while the temperature anomalies for years 1997-2003 in Figure 2.10 are positive, those of year 1999-2000 are negative. On the other hand, the precipitation anomaly for year 2000 (Figure 2.11) is positive. This climate character exhibits the outlier-behavior of the weather in 2000 as reported by TMD (2002).
Even though these anomaly climate series show visually nicely the seasonal trends in the climate series, more quantitative trend analyses can be endeavored. The easiest technique for a linear trend analysis is linear regression. Thus, to look for a linear trend in a time series , the latter is regressed on the monthly time scale , i.e.:
where and are the intersect and the slope, i.e. the trend of the fitted regression equation, respectively, and is the usual error to be minimized by least-squares.
The linear trend lines in Figure 2.10 to Figure 2.12 show the long-term behavior of the hydro-meteorological time series in the study region. In Table 2.5, the mean and range of the slopes b in Eq. (2.8) of the climate series for all climate stations are listed. As these slopes are generally positive, particularly, for the temperatures, there is enough evidence of a regional warming in the region over the recent decades. Although the rainfall series appears to exhibit also a small increasing trend (Figure 2.11), the corresponding slope values of Table 2.5 are less indicative.
This is also supported by the slopes of the runoff series, which are essentially zero, i.e. monthly streamflows have been more or less constant over the last few decades.
Another well-known technique to test a time series for a trend is the Mann-Kendall (MK) trend test, which was introduced by Mann (1945) and that has since then been extensively employed for environmental and hydrological time series (Hipel and McLeod 1994). In this study, seasonal Mann-Kendall trend test (Hirsch et al. 1982, Hipel and McLeod 1994) is applied to perform the trend analysis. The seasonal MK-test establishes trends for individual months or a group of months and is advocated for time series with strong seasonal variations, as is the case here.
The MK-test is a special case of the rank-correlation test, which is used to test if two time series Xi and Yi are independent (Hipel and McLeod 1994). The test is based on the computation of the following test statistics, the Kendall score (Kendall 1970)
where S is the Kendall score, Xi and Yi are the time series, and denotes the sign function.
For the special case of the MK-test, Xi denotes the times. Basically, Eq. (2.8) is equal to the difference between the number of concordant and discordant pairs in the Xi - and Yi -series, when these are ranked.
Eq. (2.8) is evaluated for each season or month of the year (i=1,..,s), with s, the number of seasons, and then the Kendall's seasonal rank correlation coefficient, is computed as (Hirsch et al. 1982):
(2.8)
∑
(2.9)
Figure 2.10. Monthly maximum and minimum temperatures and corresponding anomalies, relative to the 1971-2005 average at the 4 temperature stations in the study area with linear trend lines.
-6 -4 -2 0 2 4 6
Temperature anomaly (°C)
Year
X48459 X48461
X48477 X48478
max temp
-6 -4 -2 0 2 4 6
Temperature anomaly (°C)
Year
X48459 X48461
X48477 X48478
min temp
Figure 2.11. Monthly rainfall and rainfall anomaly, relative to the 1971-2005 average at the 24 precipitation stations in the study area with linear trend lines.
where (i = 1,..,s) are the Kendall scores and the denominators, which are related to bias-corrected standard deviations of (Hipel and McLeod 1994) for the ith season. As it can be shown theoretically that S or follows a normal distribution for already a small number of observations, the acceptance of rejection of the null hypothesis (no trend) H0 at the significance level α (usually α = 0.05) is done by comparing the in Eq. (2.10) with the corresponding z1- α/2 percentile of the standardized normal distribution, or equivalently, by computing the p-value.
The seasonal MK-test is done here in the R-programming environment
-20 -15 -10 -5 0 5 10 15 20
Monthly rainfall anomaly (mm/day)
Year
X478002 X459201 X459203 X48012
X48053 X48172 X48182 X48121
X9042 X9052 X9062 X9102
X9160 X48160 X48201 X48131
X48100 X48150 X48193 X48241
∑
∑ (2.10)
Figure 2.12. Monthly streamflows, with linear trend lines (upper panel) and anomalies, relative to the 1977-2006 average (bottom panel), at stream stations z4, z15 and z38.
Table 2.5. Linear trends in the observed climate time-series, determined by linear regression variable unit number
of series slope of trend (unit/year) average standard error
lowest Mean highest
Tmax °C 4 +0.03 +0.05 +0.07 1.07
Tmin °C 4 +0.01 +0.03 +0.07 1.93
PCP mm/day 24 -0.06 +0.01 +0.05 3.42
streamflow m3/s 3 -0.02 +0.00 +0.02 2.85
z4 trend z38 trend
z15 trend
-12 -10 -8 -6 -4 -2 0 2 4 6 8 10 12
Monthly streamflow anomaly (m3/s)
Year
z4 z15 z38
The significance test is performed as a two-tailed test, so that both positive and negative trends in the data series can be tested. In this study, the hypothesis of trends are test at the α = 0.05 significance level. This means in terms of the p-value that the Ho-hypothesis of no trend is rejected, whenever the p-value is less than α /2.
The results of the Mann-Kendall test applied to the local climate time-series in the study area are listed in Table 2.6. The values in the table indicate the number of stations with a positive or negative trend for a given climate variable. One may notice from the table that the precipitation has a positive trend for only a few observing stations, unlike the two temperatures which exhibit clear positive trends for all stations. The results of this analysis concur with the historic trends of the 20th-century climate variations of the study of Vongvisessomjai (2010), who found both positive and negative trends for the precipitation for the various regions of Thailand. However, in contrast to the general trends found here for the eastern seaboard where only two precipitation stations exhibit decreasing trends, those of Vongvisessomjai found for the few precipitation stations in the region are consistently negative.
Table 2.6. Number of climate stations which exhibit significant positive (+) or negative (-) trends on the monthly and the four-season scale during the 1971-2006 time period, based on the seasonal Mann-Kendall trend test.
total n
significant monthly trend
(n)
significant specific seasonal trend (n)
variable 1) dry 2)
pre-monsoon
3)
monsoon_1
4) monsoon_2
+ - + - + - + - + -
PCP 24 9 2 0 1 5 1 3 2 6 1
WetDay 24 5 10 1 8 4 3 5 7 5 9
Tmax 4 4 0 4 0 4 0 4 0 4 0
Tmin 4 4 0 3 0 3 0 2 0 3 0
HMD 4 0 2 0 2 0 0 0 1 1 1
SLR 2 0 1 0 0 0 2 0 0 0 0
Stream 3 1 0 0 0 1 1 1 0 0 0
The linear regression trend lines shown in Figure 2.10 to Figure 2.12 represent the long-term trends in the climate time-series, with the slopes quantifying the magnitudes of the linear change rates.
Figure 2.13 illustrates the long-term annual change rates for the various climate variables separately for each month of the year. One can notice that the temperatures have been consistently increasing in every month over the past 1971-2006 time period. On the other hand, rainfall and streamflow have only been increasing in the middle of the years, i.e. in the wet season, but are decreasing at the beginning and end of the year, when the dry season takes place.
These already occurring change rates of the precipitation will most likely lead to more extreme wet and dry spells in the future.
In fact, this can also been seen from the averages of the decadal climate change rates listed in Table 2.7 which demonstrate, for example, that the percentage of wet days has been strongly decreasing during the recent past, which means the rain intensities must have gotten stronger.
The rise of the annual temperatures over the last decades brought utmost hot summers, particularly, in the dry and pre-monsoon seasons, whereas the cold winters were fading away.
Since the temperature is one of the factors determining evaporation, more water has been lost from the storage, i.e. the risk of water resource shortages has increased in recent times, as witnessed, for example, by the water crisis in the EST just a decade ago in years 2005-2006.
Figure 2.13. Boxplots of rate of change of monthly maximum and minimum temperature, precipitation and streamflow at all measuring stations during years 1971-2006.
Table 2.7. Linear decadal annual and four-season (dry season (Oct-Dec), pre-monsoon1 (Jan-Mar), pre-monsoon2 (Apr-Jun) and monsoon (Jul-Sep) trends (change rates) in the various hydro-climate time series during 1971-2006.
variable unit avg.
trend
(unit/decade)
average seasonal trend (unit/decade)
dry pre-monsoon monsoon_1 monsoon_2
Tmax C +0.46 +0.51 +0.45 +0.52 +0.41
Tmin C +0.27 +0.34 +0.42 +0.22 +0.19
HMD % -0.32 -0.73 +0.02 -0.38 -0.18
SLR MJ/m2 -0.30 -0.32 -0.79 +0.05 +0.01
PCP mm/day +0.11 -0.12 +0.14 +0.21 +0.25
%Wet %(day/day) -0.90 -2.30 -0.10 -0.70 -0.30
Streamflow m3/s +0.04 +0.14 +0.47 -0.16 -0.01