Introduction

Floods rank among the most destructive natural disasters and cause widespread damage to homes, infrastructure and livelihoods. Affected communities face economic losses, agricultural disruptions and long-term environmental degradation. As climate change intensifies and alters global rainfall patterns, both the frequency and severity of flood events are increasing. This makes flood risk management more urgent and complex. The Intergovernmental Panel on Climate Change (IPCC)1 reports that shifting precipitation patterns disrupt ecosystems, reduce agricultural productivity and raise disaster risks. These effects are especially critical in regions where seasonal rainfall supports economic stability, food security and hydrological balance2,3,4. Irregular rainfall and extreme events complicate water resource planning and urban flood control5. Assessing long-term variability and historical rainfall behavior is essential for flood prevention and climate adaptation6,7. Identifying flood-prone areas and analyzing rainfall trends is important for building early warning systems, reinforcing infrastructure resilience and securing sustainable water resources under changing climatic conditions.

Such climatic complexities are particularly relevant in tropical Southeast Asia, where rainfall is strongly influenced by monsoonal circulation, complex topography and regional climate variability. In Thailand, precipitation patterns are largely governed by two distinct monsoonal systems. The Southwest monsoon, which occurs from May to October, brings moisture from the Indian Ocean and Andaman Sea8,9 and often results in intense rainfall and widespread flooding10,11,12,13,14. The Northeast monsoon, from November to February, transports drier air from China and Mongolia8,15, although some regions continue to receive precipitation due to orographic effects and occasional tropical disturbances. These climatic mechanisms interact with Thailand’s diverse topography16,17, making the analysis of long-term rainfall trends essential for planning for water resource availability.

The impacts of rainfall variability are increasingly evident in upper northern Thailand, a region that constitutes the headwaters of several major river basins, including the Ping, Wang, Yom, and Nan Rivers. These rivers flow southward and converge in the central plains to form the Chao Phraya River, which traverses Bangkok and serves as the principal water source for the central region. This river network plays a vital role in sustaining agriculture, industry, and urban settlements. Consequently, hydrological alterations in upstream areas significantly influence both regional and downstream flood dynamics. In 2024, upper northern Thailand experienced its most severe flooding in the past fifty years18. The disaster was primarily driven by an anomalously strong La Niña event, which intensified rainfall and lowered temperatures across the country. Major rivers, including the Ping, Wang, Yom, and Nan Rivers, overflowed their banks, leading to widespread inundation across upper northen regions. The situation was further exacerbated by Typhoon Yagi, which struck in mid-September 2024, bringing additional heavy rainfall and prolonging the flooding. As a result, some locations experienced two to three flood events within a single month. In contrast to previous flood events that primarily affected downstream areas, the 2024 floods caused extensive damage within the northern region itself, highlighting a shift toward more localized hydrological extremes. Historical events also demonstrate the broader influence of rainfall variability in this region. In 2011, anomalously high early-season rainfall associated with a La Niña event led to cumulative precipitation between January and October exceeding the national average by approximately 35%. While the northern region experienced heavy rainfall and flash flooding, the most catastrophic impacts occurred in the central plains, particularly Bangkok, due to runoff from upstream basins. These contrasting cases underscore the importance of realizing long-term precipitation trends in upper northern Thailand. Despite recurring and intensifying events, few studies have systematically examined rainfall variability and extreme precipitation trends in this critical upstream region. While broader and nationwide analyses have been conducted19,20, and other studies have focused on specific river basins21,22 or urban areas23,24, localized investigations targeting upper northern Thailand remain limited. Given its hydrological significance and increasing exposure to climate extremes, a detailed analysis is essential to better awareness evolving precipitation patterns and to inform adaptive water and flood risk management strategies for both regional and downstream areas.

To investigate rainfall trends, a variety of statistical methods have been applied globally. These include both parametric and nonparametric approaches, each offering distinct advantages. Simple linear regression (SLR), a parametric method, is widely used to assess long-term monotonic trends in precipitation. It supports hypothesis testing and confidence interval estimation, but its reliability declines in the presence of outliers or when data deviate from normality25,26. To improve robustness, SLR is often combined with Sen’s slope estimator (SSE), which quantifies the magnitude of trends without assuming a specific distribution27.

Nonparametric methods, particularly the Mann-Kendall (MK) test, are commonly adopted for hydro-meteorological trend detection. The MK test is robust to missing values and non-normally distributed data, making it suitable for irregular climate records28,29. Several studies have demonstrated its usefulness in examining long-term changes in rainfall extremes under different climatic and geographic contexts. For instance, Laz et al.30 and Hajani et al.31 focused on sub-daily and seasonal rainfall extremes in urban and regional catchments, while Adamowski and Bougadis32 applied the MK test to identify trends in annual maximum rainfall. Similarly, Bonaccorso et al.33 and Liuzzo and Freni34 examined changes in intense rainfall events over Mediterranean basins, and Burn et al.35 evaluated rainfall variability and extremes in high-latitude mountainous environments. Although the MK test is widely used, its sensitivity to autocorrelation can lead to inaccurate inferences, particularly in climate datasets with serial dependence. To address this limitation, the modified Mann-Kendall (MMK) test was developed, which adjusts the variance to improve reliability under autocorrelated conditions36. Meanwhile, SSE remains a reliable nonparametric measure of trend magnitude, especially for datasets containing outliers, which makes it particularly suitable for hydrometeorological data analysis37,38,39.

Innovative trend analysis (ITA) has emerged as a complementary approach for trend detection. Unlike conventional statistical methods, ITA does not require assumptions about data distribution, variance homogeneity, or independence. It effectively distinguishes between monotonic and nonmonotonic trends, enabling the detection of subtle, nonlinear variations and hidden patterns in precipitation data that might otherwise be overlooked27,40,41,42,43. Its ability to assess trends across low, medium, and high data ranges further enhances its utility in rainfall-runoff modeling, streamflow analysis, and climate-related studies38,44,45,46,47.

While long-term rainfall trends offer insights into climate shifts, the analysis of extreme rainfall events is equally crucial. One widely used index is the 1-day maximum precipitation (RX1Day), which captures the highest daily rainfall in each year and is essential for assessing flood risk and climate impacts. Recent studies have reported increasing RX1Day trends in many regions, including southeastern Europe48, the southern Levant49, and Bhutan50, emphasizing the global relevance of extreme event analysis for water management and adaptation planning.

Even though recurring flood events and growing rainfall variability in upper northern Thailand, few studies have systematically investigated long-term trends and extremes in this hydrologically critical region. Positioned upstream of the Chao Phraya River basin, the area plays a key role in shaping downstream flood dynamics and national water security. To address this research gap, this study applies SLR, MK/MMK tests together with SSE, and ITA to examine rainfall trends in upper northern Thailand. Extreme precipitation is assessed using the RX1Day index, analyzed through the same suite of methods. The primary objective is to deliver a comprehensive evaluation of seasonal, annual, and extreme rainfall variability, providing essential insights for flood risk management, climate adaptation, and sustainable water resource planning in this increasingly vulnerable region.

Material and methods

Study area

The study area, upper northern Thailand, includes the provinces of Chiang Mai, Chiang Rai, Lampang, Lamphun, Mae Hong Son, Nan, Phayao, Phrae, and Uttaradit, covering approximately 106,027 \(km^2\) (see Fig. 1). This region features mountainous terrain, river basins, and diverse climatic conditions, making it highly reliant on seasonal rainfall for agriculture, hydropower, and water supply. Four major rivers, the Ping, Wang, Yom, and Nan, originate here and collectively form the headwaters of the Chao Phraya River, playing a critical role in sustaining both urban and rural communities. The region is influenced by two primary monsoon systems: the southwest monsoon (mid-May to mid-October), which brings intense rainfall and often causes flooding in low-lying areas, and the northeast monsoon, which brings cooler, drier conditions with occasional light to moderate rainfall, particularly in areas affected by orographic lifting. The interaction between monsoon activity and complex topography results in pronounced spatial and temporal variations in rainfall, making the area particularly vulnerable to extreme weather events such as floods, droughts, and landslides.

Fig. 1
figure 1

Geographic distribution of rain gauge stations across the nine stations of upper northern Thailand.

Full size image

Data set

Daily rainfall data (mm) from 1981 to 2021, covering a 41-year period, were obtained from the Thai Meteorological Department (TMD)51for nine rain gauge stations (Fig. 1). Less than 10% of daily rainfall data was missing at each station, primarily during the summer and winter seasons. Missing data were interpolated using the inverse distance weighting method combined with the k-nearest neighbors algorithm, following the approach of Chutsagulprom et al.52, which demonstrated acceptable accuracy for rainfall data reconstruction. The interpolation utilized observations from 67 meteorological stations distributed across the country. Rainfall data were categorized into seasonal and annual segments by the TMD51, and this classification was adopted in our analysis. The seasonal classification includes summer (February to May), the rainy season (June to October), and winter (November to January). Seasonal totals were computed by summing rainfall values within each season for every year, while annual totals represent the sum of daily rainfall across each calendar year. Consequently, there are 41 data points for each season and for annual precipitation, corresponding to the length of the study period. The spatial distribution of seasonal and annual rainfall in upper northern Thailand exhibits distinct regional patterns.

As shown in Table 1, the rainy season records the highest average rainfall among the three seasons. This is mainly due to the influence of the Southwest monsoon, which brings moist air from the Andaman Sea. Rainfall during this season ranges from approximately 750 to 1,300 mm. Chiang Rai receives the highest average rainfall at 1,261.13 mm, with a peak of 1,719.90 mm, and shows substantial variability (s.d. 222.11 mm), likely due to orographic effects from nearby mountain ranges.

Rainfall drops sharply during the summer and winter seasons. Winter, dominated by the Northeast monsoon that brings cool and dry air from southern China, is the driest season with rainfall averaging between 35 and 90 mm. Summer follows, with totals ranging between 230 and 250 mm. Minimum rainfall values confirm this pattern, as most stations recorded no rainfall during winter. Uttaradit shows the lowest variability in this season (s.d. 40.29 mm), indicating stable dry conditions51,53.

According to Table 1, annual rainfall across the nine stations ranges from about 1,000 to 1,700 mm. Chiang Rai stands out with a much higher annual average compared to other provinces. Annual variability also reflects regional differences. Uttaradit and Chiang Rai show the largest fluctuations, likely influenced by changes in monsoon intensity driven by ENSO events54. In contrast, Phrae and Lampang, with standard deviations around 224.36 mm, show more consistent rainfall patterns that suggest greater climate stability.

Table 1 Rainfall data statistics by season and annual.
Full size table

Simple Linear Regression (SLR)

The SLR is a parametric statistical method commonly utilized for the trend of long-term seasonal and annual patterns of rainfall. The regression formula proposed by Meshram et al.55 was used in this study. The positive and negative values of the slope of SLR (SL) represent increasing and decreasing trends, respectively. This methodology uses the equation

$$\begin{aligned} Y = \beta _0 + \beta _1 X + \epsilon , \end{aligned}$$

to define a linear relationship between time and rainfall. Here, Y stands for the amount of rainfall, X for time, \(\beta _0\) and \(\beta _1\) for the intercept and slope, respectively, and \(\epsilon\) for the error term56. In this study’s application, the test statistic t is computed by dividing the slope (\(\beta _1\)) by its standard error, and the resulting \(p-value\) is compared to a specified significance level, \(\alpha = 0.05\) and \(\alpha = 0.01\).

The Mann–Kendall (MK) Test

The MK, first introduced by Mann57 and later refined by Kendall58, is a widely used nonparametric statistical method for detecting trends in climatological and hydrological time series data. A major advantage of this test is its robustness against the influence of extreme values, making it resilient to outliers. The null hypothesis (\(H_0\)) assumes that the data points are independently distributed with no trend, whereas the alternative hypothesis (\(H_1\)) indicates the presence of a trend59.

The MK statistic (S) is computed by comparing all possible pairs of n data points defined by

$$\begin{aligned} S = \sum _{i=1}^{n-1} \sum _{j=i+1}^{n} \text {sgn}(X_j - X_i), \end{aligned}$$

where \(X_i\) and \(X_j\) are individual data points. The sign function is given by:

$$\begin{aligned} \text {sgn}(X_j - X_i) = {\left\{ \begin{array}{ll} 1 & \text {if } X_j - X_i > 0, \\ 0 & \text {if } X_j - X_i = 0, \\ -1 & \text {if } X_j - X_i < 0. \end{array}\right. } \end{aligned}$$

The variance of S, denoted as \(\text {VAR}(S)\), is calculated using:

$$\begin{aligned} \text {VAR}(S) = \frac{n(n-1)(2n+5) - \sum _{p=1}^{m} t_p (t_p - 1)(2t_p + 5)}{18}, \end{aligned}$$

where n is the number of observations, m is the number of distinct values in the dataset (excluding duplicates), and \(t_p\) represents the frequency of each unique value p (see60). The standardized test statistic \(Z_s\) is then computed as:

$$\begin{aligned} Z_s = {\left\{ \begin{array}{ll} \frac{S - 1}{\sqrt{\text {VAR}(S)}} & \text {if } S > 0, \\ 0 & \text {if } S = 0, \\ \frac{S + 1}{\sqrt{\text {VAR}(S)}} & \text {if } S < 0. \end{array}\right. } \end{aligned}$$

A negative \(Z_s\) value indicates a downward trend, while a positive value suggests an upward trend (see61). The statistical significance of the trend is assessed using a two-tailed test at a significance level of \(\alpha = 0.05\) and \(\alpha = 0.01\). If \(|Z_s| < |Z_{1-\alpha /2}|\), the null hypothesis (\(H_0\)) is accepted, implying no significant trend. Conversely, if \(|Z_s| > |Z_{1-\alpha /2}|\), \(H_0\) is rejected, indicating a statistically significant trend. The critical values, derived from the standard normal distribution, are 1.96 for \(\alpha = 0.05\) (95% confidence) and 2.58 for \(\alpha = 0.01\) (99% confidence). The \(p-value\) corresponding to \(Z_s\) further determines trend significance, with \(H_0\) rejected if the \(p-value\) is below the chosen significance level \(\alpha\).

Modified Mann-Kendall (MMK) Test

As reported in Hamed and Rao36 and Yue and Wang61, serial correlation in time series data can significantly affect the variance of S. Positive serial correlation inflates the variance, increasing the likelihood of false positives (Type I errors), while negative correlation reduces variance, potentially leading to false negatives (Type II errors). To mitigate these effects, the effective sample size approach was introduced as a modification36,62. The adjusted variance of the MK statistic, \(\text {VAR}^*(S)\), is given by:

$$\begin{aligned} \text {VAR}^*(S) = \frac{n}{n^*} \cdot \text {VAR}(S), \end{aligned}$$

where n is the actual sample size, and \(n^*\) is the effective sample size, calculated as:

$$\begin{aligned} n^* = \frac{n}{1 + 2 \sum _{k=1}^{n-1} (1-\frac{k}{n}) \cdot r_k}. \end{aligned}$$

Here, \(r_k\) represents the lag-k rank correlation. In a practical application, lag-1 correlation (\(k=1\)) is commonly used to adjust MK test statistics for serially correlated data63.

The Sen’s Slope Estimator (SSE)

The SSE is a nonparametric method for estimating trend magnitude in time series data, introduced by Şen64. This approach has gained widespread adoption in environmental and hydrological studies for quantifying monotonic trends, particularly when datasets exhibit nonnormal distribution or contain outliers. The method serves as an essential complement to the MK test by providing a quantitative estimate of the trend slope, thus enabling more precise interpretation of temporal changes. The estimator’s robustness derives from its calculation of the median slope across all pairwise differences in the data, rendering it particularly effective for applications in hydrology, meteorology, and climate studies65. Given a time series \(X = {X_1, X_2, \dots , X_n}\) comprising n data points, the slope Q between any two data points \(X_i\) and \(X_j\) is determined by:

$$\begin{aligned} Q_{ij} = \frac{X_j - X_i}{j - i} \quad \text {for } i < j. \end{aligned}$$

The Sen’s slope, \(\hat{Q}\), is subsequently derived as the median of all individual slopes \(Q_{ij}\):

$$\begin{aligned} \hat{Q} = \text {median}(Q_{ij}). \end{aligned}$$

This median value provides the optimal estimate of the trend magnitude, wherein a positive \(\hat{Q}\) indicates an increasing trend and a negative \(\hat{Q}\) denotes a decreasing trend66. The statistical significance of the \(\hat{Q}\) can be evaluated in conjunction with the MK test results. When the trend provides significance according to the MK test, the SSE provides a substantive quantitative measure of this trend. This integrated approach has become fundamental in environmental studies, particularly in assessments of climate change impacts, water quality analyses, and streamflow investigations60.

Innovative trend analysis (ITA)

The innovative trend analysis (ITA) method, introduced by Şen7, is a visual and nonparametric approach for detecting and evaluating trends in time series data. It classifies trends at two levels: general and detailed. The general classification distinguishes between increasing, decreasing, and no trend. The detailed classification provides further insight by identifying whether the trend is monotonic or nonmonotonic, resulting in five possible categories: monotonic increasing, monotonic decreasing, nonmonotonic increasing, nonmonotonic decreasing, and no trend.

ITA begins by dividing the time series into two equal halves, which are then sorted in ascending order and plotted against each other. The points are visualized relative to a 45-degree reference line, which serves as the key interpretative guide. In the detailed classification, monotonic trends are characterized by a consistent directional shift: if almost all points lie above the reference line, the trend is classified as monotonic increasing; if almost all lie below, it is monotonic decreasing. In contrast, nonmonotonic trends exhibit fluctuations around the reference line while maintaining an overall increasing or decreasing tendency. A nonmonotonic increasing trend generally moves upward with intermittent declines, whereas a nonmonotonic decreasing trend trends downward with occasional rises. When the data points are symmetrically distributed around the reference line, the pattern indicates no significant trend, reflecting random variation without a clear directional change.

Beyond visual classification, ITA can incorporate numerical trend estimation by computing the slope (\(S\)) of the trend based on the mean values of the first and second halves of the time series67:

$$\begin{aligned} S = \frac{2(\bar{y}_2 - \bar{y}_1)}{n}, \end{aligned}$$
(1)

where \(\bar{y}_1\) and \(\bar{y}_2\) are the mean values of the first and second halves of the time series, respectively, and \(n\) is the total number of data points. The statistical significance of \(S\) can be assessed using confidence limits (CL), computed as:

$$\begin{aligned} \text {CL} = z_{1-\alpha /2} \times \frac{\sigma _D}{\sqrt{n}}, \end{aligned}$$
(2)

where \(z_{1-\alpha /2}\) is the critical value from the standard normal distribution at the chosen confidence level, \(\sigma _D\) is the standard deviation of differences between the two halves, and \(n\) is the sample size. If \(S\) lies beyond the lower or upper confidence limits, the null hypothesis of no significant trend is rejected at the \(\alpha\) significance level.

In this study, we modify the ITA method by replacing the 1:1 reference line with a CL line. The trend classification is then based on the proportion of data points that fall above or below the 95% confidence limit. Specifically, if 90% of the data points lie above the CL line, the trend is classified as monotonic increasing; if 90% lie below, it is considered monotonic decreasing. Similarly, if 55% of the data points are above the CL line, the trend is classified as nonmonotonic increasing, while 55% below indicates a nonmonotonic decreasing trend. Data distributions that fall outside these limits are interpreted as exhibiting no significant trend. Additionally, we implement a three-part classification approach (low, medium, and high) following the methodology proposed by Hoang et al.47, to further characterize the magnitude of rainfall trends.

Results

This section presents two primary findings: long-term rainfall patterns derived from seasonal and annual precipitation data, and trends in extreme rainfall events as indicated by the RX1Day index. By integrating these datasets, the analysis aims to enhance better awareness of regional precipitation behavior and to identify areas where extreme events correspond with long-term trends, which may indicate areas that are more vulnerable to severe hydrological events.

Long-term rainfall pattern

This subsection presents an overview of rainfall trend results across nine stations in upper northern Thailand, based on seasonal and annual precipitation data. Table 2 presents the estimated trend values, \(\hat{Q}\) from SSE and \(\beta _1\) from SRL. The corresponding ITA results are illustrated in Figs. 2, 3, 4, 5.

Table 2 Values of \(\beta _1\) and \(\hat{Q}\) for seasonal and annual rainfall (mm/year).
Full size table

Summer trends

During summer, most stations exhibited positive values for both \(\hat{Q}\) and \(\beta _1\) were mostly positive, indicating an overall upward tendency in rainfall. However, trends were detected by \(\hat{Q}\) at three stations without corresponding significance in the SLR results.

A significant upward trend at the 1% level was observed in Lampang (\(\hat{Q} = 2.138\)), while Nan (\(\hat{Q} = -1.215\)) and Uttaradit (\(\hat{Q} = -2.735\)) exhibited significant downward trends at this level. According to ITA, trends were detected at four of the nine stations: Lampang, Lamphun, Mae Hong Son, and Uttaradit. Monotonic increasing trends were observed in Lampang and Lamphun, while Mae Hong Son and Uttaradit showed nonmonotonic and monotonic decreasing trends, respectively. In the high-value range, Chiang Rai appeared above the upper CL line. Although no statistically significant trend was detected at this station, the elevated values suggest a potential for heavier summer rainfall. Consistently, ITA results matched those of MK/MMK in Lampang and Uttaradit, although SLR detected no significant trends in this season.

Rainy trends

During the rainy season, significant increasing trends were detected in Lamphun (\(\hat{Q} = 3.249\)) and Phrae (\(\hat{Q} = 4.348\)) at the 1% significance level, while a similar trend was observed in Chiang Mai at the 5% significance level (\(\hat{Q} = 2.266\)). Phrae was the only station where a statistically significant upward trend was identified using SLR (\(\beta _1 = 4.551\), 5% significance level). Through ITA, monotonic increasing trends were observed in Chiang Mai and Phrae, while Lamphun and Phayao displayed nonmonotonic increasing patterns. Importantly, Phrae showed consistent results across all methods, indicating strong agreement. In ITA, rainfall values at Phrae exceeded the CL line, further confirming a robust increasing trend during the rainy season.

Winter trends

During the winter season, no statistically significant trends were detected at any station using MK/MMK or SLR. However, ITA identified a nonmonotonic decreasing trend in Chiang Mai, suggesting a gradual decline in winter rainfall that was not captured by the other methods. This finding aligns with regional climatic conditions, as the southwest monsoon typically withdraws before winter, and the northeast monsoon brings relatively dry air to northern inland areas like Chiang Mai, resulting in limited precipitation during this season.

Annual trends

In annual precipitation trends, a significant upward trend was identified in Chiang Rai (\(\hat{Q} = 3.819\), 5% significance level) and Mae Hong Son (\(\hat{Q} = 5.214\), 1% significance level) using MK/MMK. An increasing trend was detected in Phrae (\(\beta _1 = 5.241\), 5% significance level) through SLR. Using ITA, increasing trends were observed in six stations, including Chiang Mai, Chiang Rai, Lampang, Lamphun, Phayao, and Phrae. Specifically, 100% of the rainfall data in Lamphun exceeded the upper CL line, suggesting a strong upward trend and a heightened likelihood of extreme precipitation events.

Fig. 2
figure 2

The results of applying the ITA method for precipitation in summer.

Full size image
Fig. 3
figure 3

The results of applying the ITA method for precipitation in rainy.

Full size image
Fig. 4
figure 4

The results of applying the ITA method for precipitation in winter.

Full size image
Fig. 5
figure 5

The results of applying the ITA method for precipitation in annual.

Full size image

Trends in extreme rainfall (RX1Day)

To assess extreme rainfall characteristics, this section analyzes RX1Day based on daily precipitation records. RX1Day trends were analyzed using SLR, SSE associated with MK/MMK tests, and ITA, which allowed for an evaluation of extreme precipitation trends. The statistical parameters, including \(\beta _1\), \(\hat{Q}\), and the trend from ITA, were presented in Table 3. The \(\beta _1\) and \(\hat{Q}\) values were closely aligned across all stations, suggesting consistency between SLR and SSE. Positive slopes were identified in five of the nine stations using SLR and SSE. Statistically significant increasing trends were observed in Lamphun and Phayao. In Lamphun, both methods detected trends at the 5% level, while in Phayao, SSE detected a trend at the 1% level and SLR at the 5% level. These findings support previous observations that indicate elevated flood risk in these areas. ITA offered further insight by distinguishing between monotonic and nonmonotonic trends, which conventional methods may overlook. Monotonic increases were found in Lamphun and Phayao, while Chiang Mai and Lampang showed nonmonotonic increases, suggesting irregular fluctuations in extreme rainfall. A monotonic decreasing trend was found only in Chiang Rai, which consistently recorded higher RX1Day values than other stations. No significant trends were detected in the remaining stations.

Time-series analysis revealed notable interannual variability, reflecting broader climatic influences. Table 3 shows clear upward slopes in Lamphun and Phayao, which are confirmed by the statistically significant results across all methods.

Table 3 Trend analysis of RX1Day using SSE, SLR, and ITA methods.
Full size table

Discussion

Rainfall data from nine stations in upper northern Thailand were analyzed using SLR, MK/MMK with SSE, and ITA. The results summarized in Table 4 classified a trend as statistically significant when at least two out of the three methods detected it, which helps ensure reliability against random fluctuations and methodological bias.

Seasonal and annual precipitation patterns across upper northern Thailand reflect the influence of diverse climatic drivers. These trends are shaped by the dynamics of the Southwest Monsoon and the region’s complex topography. In summer, rainfall increased in Lampang but decreased in Uttaradit. The declining trend in Uttaradit may be linked to weakened early-season monsoonal flow or delayed onset, possibly driven by ENSO-related shifts in regional circulation or increased evapotranspiration due to rising temperatures. During the rainy season, increasing trends in Chiang Mai, Lamphun, and Phrae align with intensified moisture transport from the Andaman Sea and the Gulf of Thailand. Orographic effects amplify this pattern as moist winds ascend mountainous terrain, enhancing precipitation. These findings correspond with previous windward cluster classifications that highlight stronger rainfall regimes along wind-exposed slopes in northern Thailand. The consistent increase across all three methods in Phrae suggests an elevated risk of flash flooding, potentially due to greater convective activity and more frequent monsoonal surges. In contrast, winter showed no significant trends, which is expected as the Northeast Monsoon typically suppresses rainfall during this period. However, a weak but statistically significant trend detected in Chiang Mai may result from topographically trapped residual moisture or urban-induced microclimatic changes. These findings demonstrate the importance of using trend detection methods that can identify subtle and localized variations, particularly during transitional seasons.

On an annual scale, rainfall increased in Chiang Rai, Lamphun, and Phrae, suggesting a long-term rise in regional precipitation. These patterns emphasize the spatial and seasonal variability of rainfall in northern Thailand, where precipitation is primarily governed by monsoonal systems during summer and the rainy season.

Table 4 Summarization trends of three methods MK/MMK with SSE, SRL, and ITA across nine regions in upper northern Thailand.
Full size table

When extreme precipitation was examined using the RX1Day index (Table 3), the three methods produced varying results. The significant increases trend observed in Lamphun and Phayao confimed by all methods, suggesting an elevated risk of extreme rainfall events that could lead to flash flooding. By integrating the RX1Day results with seasonal and annual analyses, Lamphun clearly emerged as a high-risk area. Although Phrae did not exhibit a statistically significant trend in RX1Day, it was the only station where all three methods consistently detected a significant increase in rainfall during the rainy season. This consistency highlights the need for targeted flood mitigation measures in the province.

Tables 3 and 4 present differences between seasonal rainfall trends and the RX1Day index. In some regions, seasonal rainfall increased significantly, whereas RX1Day exhibited no clear trend, indicating a rise in moderate rainfall events rather than extreme single-day occurrences. Despite the lack of statistically significant RX1Day trends in Chiang Mai and Chiang Rai, both stations warrant continued monitoring. Chiang Mai exhibited an upward trend during the rainy season, whereas Chiang Rai showed a consistent annual increase and ranked among the provinces with the highest total rainfall (see Table 1). These indicators may signal an increased flood risk in both areas. In contrast, Uttaradit exhibited a declining summer rainfall trend. This variation in rainfall patterns may be influenced by differences in elevation, regional wind systems, and geographic features. The declining summer trend in Uttaradit raises concerns about seasonal drought and its implications for agriculture and water availability. This observation is consistent with the findings of Moazzam et al.68, who identified a long-term risk of water scarcity in the region. The lack of significant winter trends again points to the relative meteorological stability of that season.

According to Thaiwater.net18, the observed increase in rainfall corresponds to a heightened risk of flooding and flash floods. Records from 2020 to 2024 indicate a continued pattern, with some locations experiencing multiple flood events within a single year. Chiang Rai, Chiang Mai, Lamphun, Lampang, and Phrae were among the most frequently affected areas. Chiang Mai and Chiang Rai experienced flooding nearly every year, consistent with their increasing trends in the rainy season and annual precipitation. Historical records also indicate that Chiang Rai typically experienced two to three flood events per year, aligning with the highest rainfall totals recorded at the Chiang Rai station for both annual and rainy season periods. Meanwhile, Phayao and Lamphun consistently experienced intense rainfall episodes each year. In Phayao, flash floods occurred in several areas, corresponding with statistically significant increases in RX1Day values recorded at this station. Lamphun consistently experienced very heavy rainfall, frequent flooding, and flash floods, supported by significant upward trends identified in both the RX1Day index and rainy season and annual precipitation analyses. These long-term trends were notably evident in 2024, with Chiang Mai experiencing its most significant flood event in fifty years, directly reflecting the area’s increasing rainy season rainfall trend. Chiang Rai reported seven flooding events between July and October 2024, reinforcing its long-term upward trend in annual precipitation and further confirming its classification as a high-risk, flood prone area.

These results make evident the growing intensity of extreme rainfall events in upper northern Thailand. They reinforce the urgent need for enhanced flood preparedness, adaptive infrastructure, and climate-resilient planning. This aligns with the findings of Leščešen et al.48, who reported statistically significant increases in extreme rainfall and suggested a possible linkage to increasing flood risk. Furthermore, the findings underscore the importance of spatially targeted monitoring strategies, particularly in provinces such as Lamphun and Phrae, which appear increasingly vulnerable to hydrological extremes. While some areas exhibited clear upward trends, others such as Uttaradit showed signs of declining rainfall during critical periods. This divergence indicates the need for differentiated policy responses that address both flood hazards and drought risks under changing climatic conditions.

Conclusion

This study analyzed rainfall and extreme precipitation trends across nine stations in upper northern Thailand using three complementary methods including SLR, MK/MMK with SSE, and ITA. The results reveal both increasing and decreasing trends that vary by season and location. In particular, significant summer trends were observed in Lampang and Uttaradit, which showed upward and downward trends, respectively. During the rainy season, increasing rainfall was found in Chiang Mai, Lamphun, and Phrae, while no significant trends were detected in winter. On an annual scale, Chiang Rai and Phrae exhibited increasing trends. Regarding extreme precipitation, the RX1Day index showed statistically significant upward trends in Lamphun and Phayao evident in all methods, suggesting an elevated flood risk in these areas. These observed patterns align well with historical flood records, as discussed in the previous section, reinforcing the reliability of the analytical results and underscoring their value for improving flood preparedness, optimizing water resource planning, and developing effective climate adaptation strategies.

Future studies should expand the spatial scope beyond the nine stations in upper northern Thailand to provide a more comprehensive understanding of regional rainfall trends. Incorporating additional extreme precipitation indices beyond RX1Day, such as RX5Day (maximum five-day precipitation), CWD (consecutive wet days), CDD (consecutive dry days), R95p and R99p (rainfall exceeding the 95th and 99th percentiles), and SDII (simple daily intensity index), would enhance the assessment of precipitation extremes by capturing intensity, frequency, and duration dimensions. This would allow for a more holistic evaluation of changes in rainfall behavior. Moreover, increasing the temporal resolution through hourly or sub-daily datasets and integrating satellite-based precipitation products could improve trend detection accuracy, particularly in data-scarce areas. These improvements would contribute to more reliable climate assessments and support effective strategies for flood risk management, sustainable water use, and agricultural adaptation in the face of evolving precipitation patterns.