Combined use of graphical and statistical approaches for rainfall trend analysis in the Mae Klong River Basin, Thailand

Rainfall observation is a prerequisite for hydrological modeling of any river basin. Identification of rainfall trends is important in water resources planning and management. Rainfall is the primary source of agricultural food production and regulates our ecosystems. A decrease in rainfall can lead to serious water shortage and drought and, consequently, to a decreased crop yield and food insecurity (Cengiz et al. 2020). Ample quantities of freshwater resources availability are critical for life on Earth. Climate change, economic growth, and increase in population have increased the challenges of decision-makers to properly allocate water among different water use sectors (Khalil et al. 2018a). Planning of water resource systems is based on climate information. Water resource managers are mostly concerned about climate change and its impacts on the availability of water for different water use sectors.

Identifying trends in rainfall patterns helps in assessing the impact of climate change on precipitation. It provides valuable information for understanding how rainfall patterns are shifting over time and whether there are any significant changes. By detecting long-term trends, hydrologists can anticipate changes in water availability, plan for future water demands, and develop strategies to mitigate the effects of droughts or floods. Trend detection in hydro-meteorological variables has been studied by many researchers globally (Luis et al. 2000; Kumar et al. 2010; Longobardi & Villani 2010; Nenwiini & Kabanda 2013; Sayemuzzaman & Jha 2014; Zeleňáková et al. 2014; Merabtene et al. 2016; Wang et al. 2020). Trend detection in the time series data can be done by both parametric and non-parametric statistical techniques. The non-parametric techniques have been preferred. The Mann–Kendall (MK) test, a non-parametric test, has been frequently applied for the detection of trends in the time series data of precipitation and streamflow (Sayemuzzaman & Jha 2014). One benefit of the MK test is that the data need not conform to any specific distribution. The MK and Spearman's Rho (SR) tests are applicable under restrictive assumptions such as the independent structure of time series, normality of the distribution, and length of the data. The innovative trend analysis (ITA) method can be used to avoid such restrictive assumptions and the pre-whitening approach required by MK and SR tests for time series data with significant serial correlation is not necessary (Şen 2012). The ITA method can identify the hidden trends in the precipitation time series that cannot be detected using statistical methods (Cengiz et al. 2020). The ITA method has received increasing attention for trends analysis in hydro-meteorological variables (Dabanlı & Şen 2018; Wu et al. 2018; Zhou et al. 2018; Ali et al. 2019; Gedefaw et al. 2019; Alifujiang et al. 2020; Cengiz et al. 2020; Wang et al. 2020). Graphical (such as the ITA method) and statistical approaches, when combined, yield more influential and comprehensive information for inference than relying purely on statistical results (Onyutha 2016).

The Mae Klong River Basin is one of the 25 major river basins in Thailand. Administratively, the Mae Klong River Basin can be divided into 25 districts belonging to eight provinces. Around 70% of the basin area is composed of three provinces, namely Ratchaburi, Samut Songkhram, and Kanchanaburi. The Electricity Generating Authority of Thailand (EGAT) has been authorized to control all structures of the three main dams and operate the water supply in reservoirs. However, coordination at the departmental level between the EGAT and the Royal Irrigation Department (RID) is carried out by creating a seasonal plan of agricultural water needs, which will be used for water release determination from upper reservoirs. Allocating water in the main irrigation system and satisfying downstream water needs at Mae Klong Dam are accomplished by the RID. Meanwhile, the farm irrigation system is managed by water user associations (WUAs), which the farmers may pay for the maintenance of the system. The use of groundwater in the basin is free by regulation and law, both in terms of control and price, under the responsibility of the Department of Groundwater (Biltonen et al. 2003). The upland in the basin is covered with forest that is well preserved. The lowland has agricultural activities which contribute more sediments in river flows. The Greater Mae Klong Irrigation Project (GMKIP) which is located in the lower region of the basin is the second largest irrigation project in Thailand. The Lampachi River Basin, which is also located in the lower region, has a low forest land ratio but a high agriculture land ratio. Even though the rainfall is not very high, the high drainage density yields more suspended sediments as compared to other areas of the basin. For proper soil erosion control in the basin, it is necessary to protect the forests in the uplands and conserve soil in the cultivated lands in the lowlands. This requires establishing an appropriate relation between rainfall, watershed characteristics, and activities in the land comprising forest and agriculture (Maita et al. 1998). As per the development plans, the total agricultural area is expected to increase in the future, which in addition to an increase in population and industrial activities, will result in an increase in water requirement in the basin. Presently, the basin is facing water shortage, especially in the dry season (Tospornsampan et al. 2005; Khalil et al. 2018a). In the future, since the situation of water resources will be very tight, it is crucial for water resources planners to get information about the river basin from statistical and/or rainfall-runoff analysis. This information may be specifically required for effective water resource management in ungauged basins (Sugiyama et al. 1998). The watershed contributing runoff to the Srinagarind Dam has less rainfall as compared to the watershed contributing runoff to the Vajiralongkorn (VJK) Dam; however, the Srinagarind Dam watershed is a more reliable source of water supply, especially during prolonged drought conditions on an annual scale (Sugiyama et al. 1998). Kwanyuen et al. (1999) identified rainfall as one of the major factors that influenced the quality of water for both irrigation and drainage canals. Shrestha (2014) reported a reduction in water availability for the Mae Klong River Basin in the first half of this century followed by an increase in the latter half during the dry season. However, in the wet season, water availability is expected to increase during all decades of the century. Sharma & Babel (2018) studied climate change impacts on the streamflow in the Mae Klong River Basin. The results indicated a decrease in streamflow in the dry season while the streamflow is expected to increase in the wet season in the future. Rojrungtavee (2009) reported an increased magnitude in peak rainfall with a shorter duration of rainfall in the basin. Deb et al. (2018) also reported rainfall anomalies that are in line with the results of Shrestha (2014) for the Mae Klong River Basin using data from five GCMs (Global Climate Models). Khalil et al. (2018b) studied seasonal and annual trends of rainfall and streamflow in the Mae Klong River Basin, Thailand. The results showed increasing rainfall trends in the wet season and decreasing rainfall trends in the dry season which are in line with Shrestha (2014) and Deb et al. (2018).

According to a survey of the literature, the Mae Klong River Basin's rainfall trends are not analyzed using the ITA approach. The comparison of the ITA method with the widely utilized MK and SR tests was a novel aspect of this research study. The main objectives of this study were (i) to determine the temporal and spatial distribution of rainfall trends for eight stations in the Mae Klong River Basin in Thailand using the MK and SR tests, (ii) to compare the results of the MK test and the SR test with the ITA method, and (iii) to determine the significance of monotonic trends at 5% significance level. The results of this study are expected to help policymakers shape better strategies for climate change mitigation in the basin.

The standard normal homogeneity test (SNHT) and the Buishand test were used to check the homogeneity of the rainfall data. These tests were conducted at a 5% significance level using the ‘trend’ R package (Pohlert 2018). In this study, for a data sample of 45, the critical value of the SNHT test statistic (To) is 8.33 (Alexandersson 1986) and the critical value of the Buishand test statistic (⁠⁠) is 1.55 (Buishand 1982). The results of the two tests for the mean annual rainfall data of the stations are given in Table 2. For both tests, all the test statistic values were less than the critical values indicating the rainfall data series as homogeneous.

Table 2

Homogeneity tests statistics for mean annual rainfall data (for homogeneous series, To <8.33, and <1.55)

Station .	SNHT statistic (To) .	Buishand test statistic (⁠⁠) .
Station 130013	6.65	1.29
Station 130042	6.78	1.18
Station 130053	2.82	0.85
Station 130211	4.45	1.10
Station 130221	2.46	1.14
Station 130571	3.10	1.11
Station 376401	2.51	1.26
Station 470161	4.69	1.15

Station .	SNHT statistic (To) .	Buishand test statistic (⁠⁠) .
Station 130013	6.65	1.29
Station 130042	6.78	1.18
Station 130053	2.82	0.85
Station 130211	4.45	1.10
Station 130221	2.46	1.14
Station 130571	3.10	1.11
Station 376401	2.51	1.26
Station 470161	4.69	1.15

View Large

The MK test requires the rainfall time series to be serially independent. If the data has serial correlation (also called autocorrelation), then the significance level of the MK test will be either underestimated or overestimated depending on whether the serial correlation is negative or positive. If the lag-1 correlation coefficient is not significant at the 5% level, then the MK test is applied to the original data series. Several methods have been proposed for the elimination of serial correlation in the time series data such as pre-whitening (von Storch 1995), variance correction (Hamed & Rao 1998), and the trend-free pre-whitening (TFPW) approach (Yue et al. 2002a). The TFPW procedure provides a better assessment of the significance of the trends for serially correlated data (Yue et al. 2002a, 2003; Zhang & Lu 2009) than the other approaches and has been applied by many researchers (Burn et al. 2004; Novotny & Stefan 2007; Wu et al. 2008; Kumar et al. 2009; Zhang & Lu 2009; Shadmani et al. 2012; Blain 2013; Ahmad et al. 2015). In this study, the data series with a significant autocorrelation (at a 5% significance level) was corrected by the TFPW approach.

The value of Z_MK is the MK test statistic that follows a normal distribution with mean 0 and variance 1. The trend test can be done at a selected significance level (denoted by α). The null hypothesis of no trend is rejected when ⁠, and the time series has a statistically significant trend. The value of can be obtained from the standard normal distribution table. In this study, the MK test is applied for the detection of rainfall trends that are statistically significant at α = 0.1% (99.9% confidence intervals), α = 0.1% (99% confidence intervals), α = 5% (95% confidence intervals), and α = 10% (90% confidence intervals). At the 0.1%, 1, 5, and 10% significance levels, the null hypothesis (no trend) is rejected if ⁠, ⁠, ⁠, and ⁠, respectively.

The Excel template MAKESENS 1.0 (MK test for trend and Sen's slope estimates) developed by the Finnish Meteorological Institute, Finland was used for detecting and estimating trends in the monthly, seasonal, and annual time series of rainfall data. In MAKESENS, the two-tailed test is used for four different significance levels α: 0.001, 0.01, 0.05, and 0.1 (Salmi et al. 2002). This tool has been used by many researchers for the trend analysis of hydro-meteorological variables (Burgmer et al. 2007; Santos & Fragoso 2013; Bernsteinová et al. 2015; Soni & Singh 2017; Tan et al. 2017). TeREsA software (Travaglini et al. 2016), which is a toolbox in R for environmental analysis, was used for plotting the spatial distribution of rainfall trends in the Mae Klong River Basin.

The lag-1 autocorrelation coefficient for the monthly, seasonal, and annual rainfall data for all stations was calculated using the function Acf in R programming language. The function Acf computes (and by default plots) an estimate of the autocorrelation function of a univariate time series. The lag-1 autocorrelation coefficient values for the monthly, seasonal, and annual time series are given in Table 3. The rainfall series for which the autocorrelation coefficient was significant at a 95% confidence level was ‘pre-whitened’ using the TFPW method suggested by Yue et al. (2002a) before the application of non-parametric tests using the ‘modifiedmk’ R package (Patakamuri 2020).

No trends for rainfall were detected for the month of January in any of the observation stations. For the month of March, Station 376401 located in the upper region of the basin indicated a statistically significant increasing trend at 99% confidence level with a trend slope of 0.941 mm/year. The SR test also showed an increasing trend for Station 376401 in March. The SR test statistic was statistically significant at 95% confidence level with a value of 3.216 (⁠⁠). In April, six out of eight stations showed increasing trends while two stations showed decreasing trends (Station 130013 and Station 130053). The increasing trend for Station 376401 is significant at 99% confidence level with a slope of 1.277 mm/year. The increasing trends of Station 130221 and Station 130571 were significant at 95% confidence level with a slope of 1.857 mm/year and 1.176 mm/year, respectively. The SR test also showed statically increasing trends for Stations 130221, 130571, and 376401 with test statistic values of 2.179, 2.257, and 2.628, respectively. In May and June, which are the dry seasons, five out of eight stations showed a decreasing trend. The months of August, September, and October have shown a similar trend with rainfall trends increasing for 50% (4 out of 8) of the stations while decreasing for the remaining 50% stations. In September, Station 376401 showed a statistically significant increasing trend with 5% significance level having a slope of 1.681 mm/year and a Spearman Rho test statistic value of 2.445. The rainfall trends for five stations decreased in November. Station 470161 had a statistically significant decreasing slope of −0.845 mm/year at 95% confidence level whereas the value of the Spearman Rho test statistic was 2.049. No trends were detected for December except for Station 130571, which had a statistically significant decreasing trend at 95% confidence level with a slope of −0.006 mm/year while the Spearman Rho test statistic value was 2.632. The variation in the rainfall trends could be due to the geographical locations (latitude and longitude) of the rainfall stations and topography of the study area.

For the month of January, the range of the ITA slope varied from −0.179 to 0.339. An increasing trend was indicated by five stations while three stations showed decreasing trends. These trends were found to be significant at 95% confidence level for six stations. For the month of March, the range of the ITA slope varied from −0.522 to 1.730. Significant increasing trends were found for six stations. Two stations (Station 130211 and Station 470161) indicated decreasing trends (also indicated by the MK and SR tests) of which the trend for Station 130211 was significant. In July, 75% (six out of eight) of the stations showed increasing trends. All the trends (increasing and decreasing) were found to be significant. The decreasing trends of Stations 130042 and Station 130571 are in line with MK and SR tests. In September, seven stations showed significant increasing trends while the decreasing trend of Station 130042 was not significant. For the month of November, the ITA method showed significant decreasing trends for six stations, whereas the MK and SR tests indicated decreasing trends for five stations.

For the dry season using the MK test, 63% of the stations (five out of eight) showed increasing trends while three stations showed decreasing rainfall trends. The increasing trend of Station 376401 was significant at 95% confidence level with an increasing slope of 4.277 mm/year. The increasing trend of Station 130571 was significant at 90% confidence level with a slope of 2.387 mm/year. The SR test indicated increasing trends for five stations in the dry season. The increasing trends of Stations 130571 and 376401 were significant at 95% confidence level with Spearman Rho test statistic values of 2.104 and 2.548, respectively. The ITA method showed significant increasing trends for six stations in the dry season as shown in Figure 7. The slope of ITA varied from −1.645 to 5.397. Station 130053 and Station 130221 indicated decreasing trends.

In the wet season using the MK test, 50% (four out of eight) of stations showed increasing trends while 50% of stations indicated decreasing rainfall trends. The increasing trend of Station 376401 was significant at 99% confidence level with a slope of 6.221 mm/year. The decreasing trend of Station 130042 was significant at 90% confidence level with a slope of −5.560 mm/year. The SR test showed increasing trends for 50% (four out of eight) stations in the wet season. The increasing rainfall of Station 376401 was significant at 95% confidence level with a test statistic value of 2.818. The upper region of the Mae Klong River Basin showed increasing trends while the middle and lower regions showed mixed seasonal rainfall trends. Station 376401 and Station 130053, which are located in the upper region, showed increasing rainfall trends in the wet season. For the wet season using the ITA method, 50% of the stations had increasing trends as shown in Figure 8. The slope of ITA ranged from −3.05 (Station 130042) to 6.182 (Station 376401). The indication of decreasing trends for these four stations (Stations 130013, 130042, 130571, and 470161) are consistent with the MK and SR tests.

On an annual scale using the MK test, 75% of the stations (six out of eight) showed increasing trends while two stations showed decreasing trends as shown in Figure 6. The increasing trend of Station 376401 was statistically significant at 99.9% confidence level with an increasing slope of 9.614 mm/year. The decreasing trend of Station 130042 was significant at 90% confidence level with a slope of −6.533 mm/year. The SR test showed increasing trends for five stations on an annual scale. The increasing trend of Station 376401 was significant at 95% confidence level with a test statistic value of 3.539. The ITA method indicated increasing trends for seven stations as shown in Figure 9. All increasing trends were significant at 95% confidence level, except for Station 470161. The slope of ITA varied from −2.09 to 11.579. The decreasing trend of Station 130042 is consistent with the MK and SR test results. A comparison between statistical and ITA methods for rainfall trends on seasonal and annual scales for the eight rainfall stations is depicted in Table 7.

Overall, the results showed that stations with significant trends based on ITA methods were more numerous than those based on MK and SR tests. There could be several reasons. The ITA method is designed to be sensitive in identifying and analyzing trends within a given time series. It incorporates visual inspection and numerical calculations to determine the trend type and slope. Unlike the MK and SR tests, the ITA method compares last parts of the record with earlier periods within the time series itself, allowing for the appreciation of trend variations within the record. Additionally, the method uses the number of crossings along the trend line to identify surplus and deficit parts of the time series with respect to the trend line. This sensitivity allows for a more comprehensive understanding of the trend patterns within the data (Şen 2017b).

The rainfall trends increased on seasonal and annual scales for the basin. For the MK test, the dry season has more increasing rainfall trend as compared to the wet season with a slope of 0.708 mm/year. On an annual scale, the rainfall has an increasing trend with a slope of 0.843 mm/year. SR test results were similar to the MK test results for the entire basin on the annual time scale. The ITA representation of trend analysis for seasonal and annual rainfall is shown in Figure 10. Significant increasing trends at 95% confidence level were observed for seasonal and annual rainfall. The values of the ITA slope ranged from 1.206 (wet season) to 2.922 (annual rainfall). A comparison between statistical and ITA methods for rainfall trends on seasonal and annual scales for the whole basin is depicted in Table 10.

As shown in Figure 2(b), the upper region of the basin had more mean annual rainfall contributing runoff to two main reservoirs, Srinagarind and Vajiralongkorn dams. The downstream release from both the dams is re-regulated by two diversion dams, Tha Thung Na and Mae Klong Dams, due to which the lower region of the basin with relatively lower mean annual precipitation had no potential droughts. Water is supplied to the GMKIP from the Mae Klong Dam. This irrigation demand was 6,219 MCM/year during the period 2000–2015. The increasing rainfall trends in the basin have pointed to the ample availability of water resources in the basin for meeting inside water demands as well as outer basin transfer to MWA Bangkok and to the neighboring Tha Chin Basin during the dry season. For the period 2000–2015, this diverted water to the Tha Chin Basin and MWA was 849 and 352 MCM/year, respectively (Khalil et al. 2018a).

Both the MK and SR tests indicated similar trends in the monthly, seasonal, and annual rainfall data. The ITA method showed comparable findings in seasonal and annual rainfall. Increasing rainfall trends were found for 50% of the stations in the wet season which could continue in the future as reported by other studies (Rojrungtavee 2009; Shrestha 2014; Deb et al. 2018). Station 376401 which is located in the upper region of the basin (Figure 2(a)) showed increasing rainfall trends in both dry and wet seasons, which are in line with the findings of Manee et al. (2015). Also, for Station 376401, increasing trends were indicated by all tests on an annual scale. Since runoff in this region contributes inflows to the Srinagarind Dam, it implies an increasing trend for inflows to the dam. Manee et al. (2015) reported increasing trends for inflows to the Srinagarind Dam on seasonal and annual scales.

From the spatial distribution of stations, it is observed that the annual rainfall trends increase in the upper and middle regions of the Mae Klong River Basin while they decrease in the lower region of the basin. The increased rainfall trends in the wet season and on an annual scale in the upper region of the basin could have implications for inflows to the two main dams (Srinagarind and Vajiralongkorn). It could suggest that the dams should be operated in a manner to optimize the hydropower production and downstream release of water during the dry season to meet the water demands. The RID has recently planned water supply to Uthai Thani Province from the Srinagarind Dam at 1892 MCM/year (Khalil et al. 2018a). Station 130571, which is located in the Lam Taphoen River Basin (Figure 1), has an increasing rainfall trend on an annual scale. The Lam Taphoen River drains to the Khwae Yai River downstream of the Tha Thung Na (TN) Dam. A greater portion of the area of Lam Taphoen is dependent on rain-fed agriculture. Station 470161, which is located in the Lampachi River Basin, showed decreasing rainfall trends in the wet season and on an annual scale. The Lampachi River Basin has a serious problem of soil erosion due to which it has no medium or large-scale irrigation projects (Biltonen et al. 2003). Trend analysis for the whole of the Mae Klong River Basin showed increasing trends in seasonal (dry and wet) and annual rainfall. This could help in the preservation of forests, which cover 68.13% of the area, and in increasing water availability for irrigation, as agriculture accounts for 22.90% of the area in the basin.

Rainfall provides the primary input to water resources of any basin. The current study evaluated trends in monthly, seasonal, and annual rainfall for the Mae Klong Basin located in the western region of Thailand during 1971–2015. The MK test was used to detect the trend and the Sen's slope method was used to determine the magnitude of the trend. The SR test was also used for the detection of trends. The TFPW approach was used to correct the autocorrelation in the time series. The results obtained from the statistical tests were compared with that of the ITA method. The major conclusions are given as follows:

• (1)

  The ITA method offers a more modern, simple, easy-to-interpret, and effective approach to trend analysis compared to classical methods such as the MK and SR tests. The ITA method allows for the comparison of recent parts of the time series with earlier periods within the same record. Unlike the MK and SR tests, it does not rely on restrictive assumptions like the independent structure of the time series or normality of the distribution. Furthermore, the ITA method allows for the calculation of trend magnitude (slope) without the need for regression approaches or additional assumptions.

• (2)

  For station-based trend analysis using the MK test, five stations showed increasing trends in the dry season while for the wet season, 50% (four out of eight) of the stations showed increasing rainfall trends. Station 376401, which is located in the upper region of the Mae Klong River Basin, showed a statistically significant upward trend at 95% confidence level with a slope of 4.277 mm/year in the dry season while a significant upward trend at 90% confidence level with a slope of 6.221 mm/year in the wet season. On an annual scale, 75% of the stations exhibited upward rainfall trends. The SR test indicated similar trends for both seasonal and annual rainfall. The ITA method showed significant increasing trends for six stations in the dry season while four stations (50%) showed increasing trends in the wet season. For annual rainfall, the ITA method showed increasing trends for seven out of eight rainfall stations.

• (3)

  Trend analysis for the entire Mae Klong River Basin showed increasing rainfall trends in both seasonal and annual data. Sen's slope estimates for dry and wet seasons were 0.708 mm/year and 0.292 mm/year, respectively, while 0.843 mm/year for the annual data. The SR test also showed increasing rainfall trends for seasonal and annual rainfall. The ITA method also showed significant increasing trends for both dry and wet seasons and annual rainfall. The values of the ITA slope ranged from 1.206 (wet season) to 2.922 (annual rainfall).

• (4)

  Increasing rainfall trends both on seasonal and annual scales have supported the fact of outer basin water transfer to the neighboring Tha Chin Basin during the dry season and to the MWA for water supply in Bangkok. As water in the basin is supplied both inside and outside to meet water demands, it could imply to operate the dams considering suitable reservoir operating policies in the basin for efficient water resource management.

Trend detection in rainfall data provides valuable insights into climate change, water resource management, infrastructure planning, and risk assessment. It helps in making informed decisions and developing strategies to adapt to changing rainfall patterns. Identifying trends in rainfall data also aids in assessing the risk of extreme weather events such as heavy rainfall or prolonged droughts. This information is essential for developing early warning systems and implementing appropriate measures to mitigate the impacts of such events. The results of this study can help water resource managers and local stakeholders to understand the variability of rainfall for better assessment and planning of water resources in the basin.

The author is thankful to the Royal Irrigation Department (RID) and Thai Meteorological Department (TMD) for providing data for this study.

No funding was received for this study.

A.K. contributed to the study conception and design, material preparation, data collection and analysis, and wrote the manuscript.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.