Rainfall extremes under climate change in the Pasak River Basin, Thailand

The constant warming of global temperatures, rapid and unplanned urbanisation expansion, and land use changes have led to climate change, resulting in the increased intensity and frequency of some weather and climate extremes, particularly precipitation, droughts, tropical cyclones, and compound events (IPCC 2021a). Although these events have a low probability of occurrence, their impact is enormous. The number of people affected globally by climate-related disasters in the last 20 years, from 1995 to 2015, has been estimated at 4.16 billion, excluding fatalities (CRED/UNISDR 2016). Among these, 56% were affected by floods and 26% by droughts. Extreme weather events and occasional disasters have not only contributed to economic disruption and deaths but also damaged natural resources. The Assessment Report 5 (AR5) shows that the unprecedented occurrence of extreme events is likely to increase along with global warming. Over the past decade, there has been mounting evidence to support the unprecedented threat of weather disasters in many regions around the world. Consequently, there is a pressing need to monitor changes in extreme weather to improve the long-term planning of resource management and mitigation.

Most of the literature on monitoring extreme weather events is based on case studies of temperature and precipitation variability through climate extreme indices. It has been well-documented that studies of temperature extremes show relatively similar trends, rising in varying magnitudes throughout many regions (Limjirakan & Limsakul 2012; Adeyeri et al. 2019; IPCC 2021a; Khadka et al. 2021), while the trends in precipitation extremes show high variability in temporal and spatial patterns. Most of the studies (Limsakul & Singhruck 2016; Bhatti et al. 2020) on precipitation extremes have been conducted on a large scale with historical observed datasets used in analysis. However, less emphasis has been placed on the watersheds of tropical developing countries (IPCC 2021b), which tend to be susceptible to flooding, droughts, and occasionally tropical storms. The tropics experience high variability in rainfall. According to a study by Yao et al. (2010), most extreme precipitation occurs in summer and late autumn across South Asia but in all seasons across Southeast Asia. In addition, extreme precipitation in the Southeast Asia region shows a significantly positive trend across all seasons, except autumn. Naturally, the hydrological cycle is highly interconnected and sensitive to its components. Although considerable studies have been conducted in Thailand on precipitation extremes (Cooper 2019; Limsakul 2020; Amnuaylojaroen 2021), only a few focus on future trends of precipitation extremes under climate change on the watershed scale.

The results of these studies did not address the impact of extreme rainfall variability on the water resource infrastructures in the basin either. Quan et al. (2021) analysed the spatiotemporal variability of trends in extreme precipitation events in Ho Chi Minh City during 1980–2017. Although this study has given spatial distributions of trends in extreme precipitation, trends with spatial significance were not considered.

The output of the Global Climate Model (GCM) is the primary source of data for assessing future climate at the regional level (IPCC 2012). Over time, although the validity of future climate projections has been improved, they remain limited by future uncertainties. Therefore, when monitoring changes in projected extreme rainfall, the use of new future climate data from multiple climate models should be considered as well as new emission scenarios (Shrestha & Lohpaisankrit 2017). Many studies have adopted various climate change scenarios to extrapolate the future impacts of climate change. The scenarios presented in the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) have been widely applied to investigate extreme events in response to climate change (Adeyeri et al. 2019; Amnuaylojaroen 2021; Larbi et al. 2021). Few studies have been conducted using the new scenarios presented in the Sixth Assessment Report (AR6) of the IPCC published in 2021. These include the Shared Socioeconomic Pathways (SSPs), comprising a set of integrated scenarios combining SSP-based socioeconomic and energy-emissions-land use with forcing pathways achieved by 2100 (O'Neill et al. 2016). This new design covers a wide range of impact studies. Although the new GCMs of the Coupled Model Intercomparison Projects (CMIP) have higher resolutions compared with the previous models, they are not reliable for assessing the impact at the watershed scale (IPCC 2021b). Therefore, a downscaling technique and bias correction method are needed to reduce uncertainty from the scale differences and existing biases in the climate model outputs (Shrestha et al. 2014). The selection of either a downscaling technique or a bias correction method depends on the variables under consideration (Fang et al. 2015). So far, few studies have analysed the various downscaling techniques which focus on extreme events. However, Shrestha et al. (2014) found that the outcomes from the downscaling techniques are relatively similar. The literature suggests that the quantile mapping method performs well in correcting precipitation biases (Themeßl et al. 2012; Fang et al. 2015; Adeyeri et al. 2019).

The impacts of climate change on the variability of precipitation extremes are likely to be superimposed on already problematic watersheds with complex land use, complex water resource systems, and some economic activities (e.g., agriculture) (Larbi et al. 2021). This is the reason for selecting the Pasak River Basin of Thailand as the case study area. The Pasak River Basin is at a high risk of flooding and experiences problems with land use change. The forest land area has gradually been converted to agricultural land and other development activities that could deteriorate water quality and the ability to use downstream water (World Bank Group 2011). According to the historical records on natural disasters, this river basin frequently suffers from floods and droughts. Such natural hazards indicate the effects of high rainfall variability. In addition, the Pasak River Basin is also the upstream watershed of the Chao Phraya River Basin, with the large multi-purpose Pasak Jolasid dam downstream. The operation of the Pasak Jolasid dam has a profound impact on downstream areas, especially flood protection (B. Tribune 2021). The operation could be adjusted to respond adaptively to extremes of high and low flow in the future due to the impacts of climate change on the likelihood of extreme rainfall events. Therefore, it is especially important to identify the historical extreme rainfall trends potentially leading to large future problems in the face of changing climate, as well as the future extreme trends under the hypothesised future development of anthropogenic climate change drivers along with socioeconomic development. Rarely do decision-makers in this area have such essential information on the current trends of extreme rainfall. These issues provide the motivation for this study.

This study is conducted to evaluate the possible temporal and spatial trends of extreme rainfall events in the future across the Pasak River Basin of Thailand using the most updated global climate dataset relative to the historical trends. The new CMIP6 climate dataset under two Shared Socioeconomic Pathways, namely SSP2-4.5 and SSP5-8.5, was selected to extrapolate the future climate scenarios to perceive the range of likely climate change, impacts, and possible responses. The resulting scientific information could be used to create an effective action plan for adaptation in response to change and help to guide watershed management on the path towards a sustainable future.

The PRB is in a tropical climate zone, influenced by the southwest and northeast monsoons. These monsoons induce the wet and dry seasons in this area. The wet starts from May to October, and the dry season from November to April. The average annual rainfall is about 1,200 mm (80% of the total rainfall occurs during the wet season), and the average annual temperature is approximately 27 °C, with April being the hottest. Tropical cyclones originating from the South China Sea and the Western North Pacific Ocean frequently move through this watershed during the monsoon season, especially in September and October. This river basin has always experienced floods and droughts caused by high variability in rainfall, directly affecting not only the livelihoods of agricultural communities but also the operation of the dam and existing infrastructure. The operation of the Pasak Jolasid dam is strategically significant in protecting Bangkok, Thailand's capital, from flooding.

Data used in this study consisted of daily rainfall time-series observations and future climate projections from global climate models.

Observed daily rainfall data were collected from the Thai Meteorological Department and Royal Irrigation Department. The ten stations were selected according to Table 1 after comparing the completeness and continuity of the recorded data among the rainfall stations distributed across the watershed. The baseline period of 1985–2014 was also selected based on the aforementioned results, and missing data was estimated using the normal ratio method. This method is commonly used in estimating missing rainfall values due to its simplicity and efficiency and has been used in Thailand (Okwala et al. 2020). Rainfall data in the study area fell within the criteria of the method where the surrounding gauges have the normal annual rainfall exceeding 10% of the gauge considered. Moreover, this method is considered to be suitable for tropical climates (Silva et al. 2007). The stations were grouped into two groups: high (EL. >100 m.MSL) and low (EL. <100 m.MSL) altitude stations, according to their elevation range.

Global climate models of CMIP6 were used to project rainfall data for the future period 2023–2100. The global climate model outputs from CMIP6 and ScenarioMIP experiments were obtained from the Earth System Grid Federation (ESGF) via https://esgf-node.llnl.gov/projects/cmip6/. Lawin et al. (2019) recommended using a multi-model ensemble approach to improve the robustness of future climate projections. Therefore, in this study, the results of five GCMs from CMIP6 were selected from the top three GCM efficacy rankings for mainland Southeast Asia and Thailand according to Iqbal et al. (2021) and Khadka et al. (2022), as summarised in Table 2. Daily precipitation (pr) of the GCM outcomes was examined to assess the impact of climate change across the PRB under historical and future scenarios. The 30-year historical period from 1985 to 2014 was defined as the baseline. For the future period from 2023 to 2100, two Shared Socioeconomic Pathways, SSP2-4.5 and SSP5-8.5, were considered in relation to the regional changes to assess the potential impacts of climate change in the 21st century. The future period was divided into three timeframes: near (2023–2048), mid (2049–2074), and far-future (2075–2100). The SSP2-4.5 scenario represents the moderate forcing level with non-extreme land use and aerosol pathways, whereas the SSP5-8.5 scenario exhibits the highest emission, capable of generating a forcing level of 8.5 W/m² in 2100.

The gridded-based data from these five GCMs were directly downscaled into time-series at the meteorological and rainfall stations in the PRB. However, the downscaled time-series at the stations, from the coarse resolution of GCMs to finer resolution at station points, may have some biases from both systematic and temporal-spatial scales since these are unreliable for assessing watershed impact. Therefore, a statistical bias correction method is used post-processing to reduce such bias. The downscaled dataset was adjusted for comparison with observations at the stations. The transfer function used for correcting the biases was applied to correct the future dataset for the period 2023–2100.

Although several studies (Fang et al. 2015; Shrestha et al. 2017; Mendez et al. 2020; Enayati et al. 2021) have proven that the quantile mapping method can correct the mean, standard deviation, quantile, and frequency of wet days, its performance may vary depending on the topographical area and the weather characteristics. Therefore, the effectiveness of quantile mapping is basically evaluated in terms of statistical indices (Table 3), namely the percent bias (PBIAS), root mean square error (RMSE), and normalised standard deviation (⁠⁠) to validate the accuracy between observed and corrected GCMs data during the baseline period (1985–2014). If the values of these indices are closer to the perfect fit than before bias correction, it indicates that there are better correlations. The statistical relationship between the GCM outputs and observations during that period will be applied for future rainfall projections by assuming that the statistical relationships observed in the past will continue to hold in the future.

Table 3

Statistical indices used for performance evaluation of the bias correction method

Statistic Index .	Unit .	Equation .	Range .	Perfect Fit .
Percent Bias (PBIAS)	Percent (%)			0
Root Mean Square Error (RMSE)	Millimetre (mm)			0
Normalised standard deviation (⁠⁠)	Dimensionless (−)			+1

Statistic Index .	Unit .	Equation .	Range .	Perfect Fit .
Percent Bias (PBIAS)	Percent (%)			0
Root Mean Square Error (RMSE)	Millimetre (mm)			0
Normalised standard deviation (⁠⁠)	Dimensionless (−)			+1

Note: obs refers to the observed data, mod refers the GCM data, and n refers the number of data.

View Large

Concerning the potential impacts of climate change on extreme events at the watershed scale, the nine extreme rainfall indices (Table 4), a subset of 27 extreme indices, recommended by the World Meteorological Organization-Commission for Climatology (WMO-CCI)/World Climate Research Programme (WRCP)/Climate Variability and Predictability (CLIVAR) project's Expert Team on Climate Change Detection and Indices (ETCCDI), was selected for comparative purposes with previous studies and other areas and based on available data that better describe the climatic behaviour of Thailand's climate. Moreover, they also can provide indicators for identifying risks of frequent extreme events in Thailand such as floods and droughts in terms of intensity, frequency, and severity. The selected indices were calculated from daily rainfall data using RClimDex software. RClimDexQC was used to control the data quality and homogeneity of historical data before calculating the extreme indices. The software can detect unreasonable values and then automatically replace their values with −99.9 or NA (not available).

Trends of observed and projected extreme rainfall indices were subject to station-based analysis using the Mann-Kendall test, recommended by the WMO. It is a non-parametric test that can be applied to any data distribution (Mann 1945; Kendall 1948) and a robust statistical tool for time-series trend analysis even in cases of missing or outlier values (Limjirakan & Limsakul 2012; Cui et al. 2017). Therefore, this statistical test has been widely used to detect meteorological and hydrological time-series trends in many regions (Khatiwada et al. 2016; Masud et al. 2016; Lawin et al. 2019; Ekwueme & Agunwamba 2021; El Kasri et al. 2021; Khadka et al. 2021; Larbi et al. 2021; Quan et al. 2021; Mann & Gupta 2022). For the hypothesis process in this statistical trend analysis, the null hypothesis (H₀) refers to the existence of no monotonic trend, while the alternative hypothesis (H_a) is defined as the existence of a monotonic trend (Larbi et al. 2021). The significance level (α) of 0.05 was considered for the rejection of the null hypothesis where the p-value is less than or equal to α. The direction of the trend was expressed in terms of positive or negative values. Positive values indicate increasing trends and negative values indicate decreasing trends. However, the Mann-Kendall test can only define the direction of the trends but is unable to quantify their magnitude (Ojo & Ilunga 2018). For this reason, the non-parametric Sen's slope is used to estimate the magnitude or true slope of the trends (Sen 1968), expressed as the changes per year. The Sen's slope calculates the median value of the slopes in all data pairs, where the time-series can be assumed as a linear trend (Ojo & Ilunga 2018; Khadka et al. 2021).

In this study, the inverse distance weighted technique (IDW) was used to estimate the spatial distribution of trends in extreme rainfall indices for ungauged locations across the PRB. This technique has been widely used to assess spatial distribution in the context of rainfall (Adeyeri et al. 2019; Cooper 2019; Larbi et al. 2021). The IDW estimates any ungauged location from the surrounding gauged points, assuming that each measured point has a local influence and decreases with distance. Therefore, it puts more weight on the point closest to the predicted location, and the weight is reduced as a function of distance (esri n.d.). Spatial interpolation of extreme rainfall trends using the IDW technique under the baseline and future climate scenarios was performed by ArcGIS software.

Low-altitude areas are shown to experience more extreme rainfall than high-altitude areas, with the exception of consecutive wet days (CWDs) and consecutive dry days (CDDs) in both the baseline and future projections. Considering the median values, the rise in these indices (R95p, R99p, RX1day, RX5day, PRCPTOT, SDII, and R20) corresponds to a decline in elevation ranging from 13.36 to 64.76%. On the other hand, CWD and CDD decrease slightly as altitude declined, with an average rate of 3.52%.

Considering each index in detail over the past and future periods, changes in very wet days (R95p) and extreme wet days (R99p) are expected to decrease in the near and mid-future before rising dramatically in the far-future relative to the baseline. Likewise, the number of days with very heavy rainfall each year (>20 mm), defined as R20, tends to be less than in the past for the future period in both scenarios. When comparing future scenarios, the number of days with the highest radiative forcing level (SSP5-8.5) is estimated to be greater than under the SSP2-4.5 medium forcing level.

In the case of intensity indices, RX1day and RX5day of the observation datasets showed higher variability than the projected datasets for either SSP2-4.5 or SSP5-8.5. These are indicated by the wide range of interquartile and long length of whiskers shown in Figure 4(c) and 4(d). RX1day and RX5day are likely to decrease by an average of 56 and 30%, respectively in the future. The annual total rainfall exceeding 1 mm (PRCPTOT) tends to fluctuate over the timeframes, potentially affecting the long-term management of water resources and water availability in the watershed. In contrast, the annual rainfall intensity on wet days (SDII) is expected to decline significantly, averaging 50% from the baseline. This means that the number of rainy days is more likely to increase than the decrease in total precipitation.

With respect to the duration-based indices, CWDs analysed from the ensemble GCMs dataset showed substantial increases of 296–327% from the baseline for high and low-altitude areas. In contrast, CDDs analysed from the ensemble GCMs dataset showed similarly significant reductions of 54% in both high and low-altitude areas across the timeframes. The pattern of changes between observation and ensemble GCMs data for these indices is similar to those revealed by Imbulana et al. (2018), who attempted to analyse patterns of future extreme rainfall in Sri Lanka's Mahaweli River Basin using CMIP5 GCMs. Noticeably, the observed extreme rainfall indices had a larger range of variation than that predicted by climate models such as RX1day, RX5day, SDII, and CDD. These underestimated results may be related to the coarse spatial resolution of the GCMs (Kitoh et al. 2009). It has been suggested that changes in extreme rainfall indices in response to human-induced global warming predicted from the GCMs dataset may be underestimated (IPCC 2012).

Although the ability of new global climate models as well as the development and application of climate scenarios have improved in predicting the physical climate system, uncertainties remain regarding the projected results. One particular problem is the link between global climate models and regional and/or watershed climates. In this study, the performance of statistical bias correction was evaluated to check the validity of biased GCMs with the observations. However, some bias still exists in the rainfall projections that affect the generated rainfall extremes. In addition, this study applied an unweighted average multi-model since it is the simplest approach for minimising modelling uncertainties. This could be why the projected rainfall extremes seem to be underestimated. Accordingly, further research could involve the evaluation of individual climate models to rank GCMs and select the most appropriate for the PRB. The weighted average multi-model ensemble of the ranked model could provide optimised model performance. However, each method has its strengths and weaknesses for assessing climate impacts in a specific area.

The variation in the statistically significant trends of the extreme indices during the baseline period indicates heterogeneity, particularly the RX5day and CWD. R95p, PRCPTOT, SDII, and R20 show upward trends while R99p, RX1day, CDD reveal downward trends. The increasing trends of the aforementioned indices align with the findings of Limsakul & Singhruck (2016), who studied long-term trends and the variability of extreme precipitation in Thailand from 1955 to 2014. Remarkably, they also observed that the area with a highly significant increase was near the site of the Pasak Jolasid reservoir. Hence, it can be concluded that the PRB experiences more extreme events in wet climates than dry climates, particularly in the lower plain areas. According to the study of Quan et al. (2021), increasing trends in the frequency and intensity of rainfall extremes and the duration of wet days may be a consequence of the occurrence of tropical storms during the monsoon season. However, it is important to remember that the ETCCDI indices are designed to assess moderate extremes for estimating a few weather-induced extreme events every year rather than high-impact events (WMO 2009). The upward trends of R95p, PRCPTOT, SDII, and R20 also provide evidence to implicate flooding in the period under consideration. The assumption is based on the occurrence of historical flood events in the watershed from 2005 to 2014 recorded by the Geo-Informatics and Space Technology Development Agency (GISTDA).

Similarly, the future extreme rainfall indices were also predicted using different assumptions on the potential impact of climate change. Under the medium emission scenario SSP2-4.5 (Figure 5(a.2)–5(i.2)), trends in all extreme rainfall indices except for CDD are likely to uniformly increase across the watershed. However, the magnitude of changes in the increasing indices and decreasing CDD is moderate compared with the baseline. Indices such as R95p and PRCPTOT rose markedly with peak rates of 13.39 and 33.01 mm/year, respectively, for the baseline, while increasing at the highest rate of 4.62 and 5.61 mm/year in the same area near the Pasak Jolasid reservoir. CDD is also expected to decline at a comparatively lower rate than the baseline. The spatial trends of the SSP2-4.5 scenario suggest that the entire PRB will be subject to an increase in wet climates in the future, especially in the lower part of the basin.

Regarding the high emission scenario SSP5-8.5 (Figure 5(a.3)–5(i.3)), the spatial trends of extreme rainfall indices mainly reveal considerable increases at a higher rate than SSP2-4.5. The trend patterns of the indices for this scenario are quite steady across the basin, except for RX1day. The variation in RX1day (highest rainfall amount in a one-day period) can have a significant impact on society (Limsakul & Singhruck 2016). CWD projected under SSP5-8.5 is found to reverse the trend with SSP2-4.5 and the baseline (low-lying area). The overall trend in extreme rainfall indices during either the SSP2-4.5 or SSP5-8.5 scenario suggests that low-lying areas will experience more frequent and severe extreme rainfall events than elsewhere. This means the nearby dam will inevitably have a long-term effect. Indeed, it is well-known that existing infrastructures in Thailand have been designed on the basis of historical weather or extreme events, and watershed management also relies on this information. The implication is that the current design criteria and master plan for water management should be reconsidered to ensure its adaptive capacity to climate change. Therefore, this study attempts to shed light on the potential variability in extreme rainfall events relating to climate change that could exacerbate natural climate variability in the future.

This study seeks to project tendencies of rainfall extremes across the Pasak River Basin in terms of frequency and severity under the impacts of climate change. The elevation-dependent trend of extreme rainfall indices represents an interesting development in this study. However, rainfall variability patterns depend on a number of factors, such as the influence of large-scale circulation and oceanic and associated drivers, which were not considered in this study. The approach used in this study can be applied to other watershed with different physical and geographical characteristics but a complete and homogeneous long-term observation record for rainfall data and proper measurement station distribution is essential.

In this study, changes in extreme rainfall events were monitored and predicted through index-based trend analysis. Nine rainfall indices were selected to describe the characteristics of extremes in terms of intensity, frequency, and persistence of the PRB. The Mann-Kendall test and Sen's slope estimator were used to track changes in the nine indices. The observed daily rainfall at the ten stations was used to calculate the indices during the baseline period to understand past conditions before taking further steps to project the future climate. The observed stations range in altitude from 48.00 to 386.12 m.MSL. To predict possible future extreme rainfall events and their variability, an ensemble of five GCMs (EC-Earth3, EC-Earth3-Veg, EC-Earth3-CC, MRI-ESM2-0, and NorESM2-MM) of CMIP6 under SSP2-4.5 and SSP5-8.5 scenarios was used to extrapolate future rainfall to reduce the level of uncertainty. The grid-based GCMs were extracted to the station points. However, the extracted GCMs still had biases, and these were removed using the empirical quantile mapping method. The datasets were validated by statistical indicators against the observed data. The predicted rainfall time-series were then calculated as the extreme indices for the period 2023–2100. The IDW technique was applied to visualise the spatial trends. The results revealed that most of the rainfall indices in low-altitude areas were higher than in high-altitude areas, except for the duration-based indices CWD and CDD. The observed extreme rainfall showed significantly greater variation compared with that predicted by the climate models. Very high greenhouse emissions contributed to greater uncertainty of future extreme rainfall in plain areas than in high-altitude areas. The monitoring of extreme rainfall variability revealed that in the past, the PRB experienced extreme rainfall volatility, with different trends and levels of statistical significance in each area. However, considering the trends across the basin, the PRB was more likely to experience wet climates than dry climates. Concerning future climate possibilities, trends in extreme rainfall projected under the SSP2-4.5 intermediate GHG emission SSP2-4.5 scenario are likely to increase gradually and uniformly across the basin relative to the baseline. Nevertheless, these tend to be more intense in low-lying areas due to the increasing trends in intensity and duration indices (RX1day, RX5day, and CWD). In the worst case of a very high GHG emission scenario (SSP5-8.5), significant increasing trends in intensity and frequency are likely with higher rates of change compared to the SSP2-4.5 scenario, while the duration indices CWD and CDD show non-significant decreasing trends in the entire basin. Remarkably, the spatial trends over the past and future periods highlighted significant increasing trends in the area of the Pasak Jolasid reservoir. The results of this study will benefit policymakers by placing them in a position to reduce future climate vulnerabilities and provide information to build local adaptation strategies in response to long-term climate change. The findings from this study suggest that the optimisation of reservoir operation rule curves is required to take into account the effects of climate change in further studies.

The authors would like to acknowledge the Thai Meteorological Department (TMD) and Royal Irrigation Department (RID) of Thailand for providing data for this study. We would also like to appreciate the reviewers and their comments on this manuscript, as well as the editors.

Piyanuch Nontikansak: conceptualisation, methodology, analysis, investigation, and writing of original draft manuscript. Sangam Shrestha: supervision, review, and editing. Mohana Sundaram Shanmugam, Ho Huu Loc, and Salvatore G.P. Virdis: review and comments.

No competing financial or non-financial interests involved are disclosed for this paper.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.