Introduction

Global warming poses a real threat to the global population, as climate changes impact various sectors of the economy and daily human activities1,2. However, there are still climate change deniers, accompanied by public officials reluctant to prioritize investments, adaptations, and mitigation actions against climate change, considering the notable implications on biodiversity and ecosystems3,4. Climate change in Asia has been the focus of numerous studies due to the region’s high population density (56% of the world’s population)5,6, agricultural and fishery production1,4, and the occurrence of epidemics and pandemics7,8. Asia is highly vulnerable to climate changes, primarily due to variability in rainfall and rising temperatures9. For this reason, climate modeling, including Artificial Neural Networks (ANNs), has been increasingly utilized for investigations and research on climate changes.

For this study, we have added a Thailand-specific context to the ANN calibration process, detailing the model’s inputs, validation data, and adjustments made for local climate conditions. Mendes and Marengo (2010) explain the role of ANNs in downscaling, which allows for the integration of multiple predictors to fine-tune regional climate responses, essential for areas with unique monsoon-driven climates such as Thailand (Mendes & Marengo, 2010). Additionally, Boulanger et al. (2006) provide insights on the application of ANNs for climate projections, underscoring their utility in capturing non-linear patterns essential for regional variability (Boulanger et al., 2006).

Southeast Asia is highly threatened by climate change due to its topographic and socioeconomic characteristics. Thailand faces challenges such as vulnerability to sea level rise and extreme rainfall events, which affect coastal agricultural systems and raise concerns about increasing flood frequency in urban and rural areas. The country’s vulnerability to climate change is particularly concerning due to its extensive coastline and the severe effects predicted by climate models due to sea level rise10. Furthermore, climate change has intensified and increased the frequency of heatwaves, droughts, wildfires, coastal flooding, and irregular rainfall. These climatic factors, combined with other non-climatic factors such as changes in land use and occupation, exacerbate the situation in the country11,12,13,14,15.

Regional variability is influenced by local climate drivers such as topography, proximity to coastlines, and urbanization. For example, the higher drought risk in northern Thailand could relate to mountainous terrain that disrupts monsoon flow, as discussed by Rodchuen et al. (2020). Additionally, Limsakul & Singhruck (2016) and Duan et al. (2017) have shown that topographical variations contribute to localized drought risks in nearby regions, helping reinforce Thailand’s regional variability context in this study (Limsakul & Singhruck, 2016; Duan et al., 2017).

The effects of climate change on the supply chain and the global economy have been reported in recent years, particularly in East and Southeast Asia, impacting food availability (production and trade) and access, especially in terms of food security16. The Asian Development Bank (ADB) estimates that Southeast Asia could suffer more significant economic losses than other regions, potentially reducing the region’s GDP by 11% by the end of the century due to impacts in key sectors like agriculture, tourism, and fisheries17. Thailand plays a crucial role in the political and economic sectors of Southeast Asia18,19; however, the country’s vulnerability has increased due to the disorganized urbanization process of cities and the reduction of native vegetation20. Rapid urbanization intensifies the impacts of climate change by increasing heat emissions and blocking watercourses, making cities more susceptible to flooding21.

Also, we expanded on adaptation recommendations, focusing on specific policies relevant to Thailand, including flood defenses and agricultural adaptations tailored to Thailand’s climate conditions. Supratid & Aribarg (2022) have documented policy strategies that address climate risks in Southeast Asia, providing a basis for comparison and extension of our study’s recommendations (Supratid & Aribarg, 2022). Furthermore, Babel et al. (2011) present examples of agricultural adaptations in response to drought, particularly in rice farming, which is relevant given Thailand’s dependence on rice cultivation (Babel et al., 2011).

Climate change is affecting the activities and daily life of human beings on planet Earth22. Understanding these changes is crucial for preparation and adaptation to future risks23. The 2023 State of the Climate in Asia report highlights accelerated rates of key climate change indicators, such as surface temperature increase, glacier retreat, and sea level rise, which have major repercussions for societies, economies, and ecosystems in the region24. In the case of Thailand, some studies have been conducted on the impact of climate change19,20,23but a better understanding of regional and local effects is needed, based on specific information from new Global Circulation Models (GCMs)23.

By comparing CMIP6 data with regional climate models (RCMs) it enhances the robustness of our projections by establishing a benchmark based on observed data. Emmenegger et al. (2022) highlights the importance of such comparisons to validate projections and enhance their scientific rigor. We also included comparisons with long-term precipitation datasets from studies on regions like India, following methodologies similar to Cooper (2019) to better ground our findings in observed data trends (Emmenegger et al., 2022; Cooper, 2019).

These IPCC scenarios have been investigated in Thailand, for example, in relation to sugarcane production26, the water sector19,27, and rice and corn20,28, where the results show that such crops are subject to the effects of drought and increasing air temperatures. Climate change influences long-term trends in temperature patterns and rainfall regimes, as well as interannual variability and the increase in extreme events18,22.

The impacts of these conditions on agriculture are already noticeable, but there are still gaps in understanding how agricultural systems will be affected by climate change in the short and long term, and what implications these changes will have on the country’s major crops18,20,28, particularly those most vulnerable to such changes in the climate, in this case, the southern region of Thailand in relation to fruit cultivation.

The WMO’s 2023 report further emphasizes the growing severity of extreme climate events, such as heatwaves and their economic impacts, urging the need for more comprehensive climate services to support disaster risk reduction and adaptation efforts24. Additionally, air pollution and traffic congestion in urban areas contribute significantly to greenhouse gas emissions, intensifying the impacts of climate change29.

Therefore, the aim of this study is to provide a comprehensive analysis of drought and extreme precipitation events for Thailand, focusing on understanding the trends modeled by climates for future projections and implications for adapting to climate changes.

Materials and methods

Study area

Thailand is located between the coordinates 5°27’ and 20°27’ N and 97°27’ and 105°37’ E, with an area of 517,624 km², divided into three regions: North, Central, and South (Fig. 1). The coastline extends for 3,148 km, of which about 2,055 km are along the Andaman Sea and 1,093 km along the Gulf of Thailand19. There are 76 provinces, with the provinces divided into 878 districts in Thailand. Thailand is the third largest country in Southeast Asia, it is surrounded by Myanmar, Laos, Cambodia, and Malaysia on the west side, east side, southeast side, and south side, respectively (Phumkokrux & Trivej, 2024). The climate of Thailand can be classified into various types, including equatorial, tropical rainforest, tropical wet and dry or savanna, and tropical monsoon climates30.

The Thai Meteorological Department divides the year into three seasons: summer (February to May), rainy (June to September), and winter (October to January), with an average annual rainfall of 2000–2500 mm (TMD, 2024). The average minimum temperature is 20 °C, occurring between November and February, while the maximum temperature of 35 °C occurs between March and June (Shrestha et al., 2017; Kiguchi et al., 2021). The average annual evapotranspiration (ET) ranges from 1300 ± 140 mm. The interannual variation in ET reflects the land surface response to meteorological systems (southwest – SW and northeast – NE monsoons) and changes in land use/land cover (LULC), with the main categories being agricultural land, forest land, urban and built-up areas, water bodies, and others. (Kiguchi et al., 2021; Phumkokrux & Trivej, 2024).

Fig. 1
figure 1

Study area Thailand with elevation (m) (A), land use and occupation (B), precipitation (mm) (C) and air temperature (°C), respectively.

Full size image

All the maps in Fig. 1 were created using QGIS software version 3.32.2-LIMA, available at: https://www.qgis.org/project/visual-changelogs/visualchangelog332/. The following plugins were utilized: HCMGIS - Basemaps, Download OpenData, Batch Converter, VN-2000 Projections, and Field Calculation Utilities, in version 24.10.20; more information is available at: https://plugins.qgis.org/plugins/HCMGIS/. As for Fig. 1A, it was produced from the Copernicus digital elevation model (DEM) with 90-meter resolution, representing the Earth’s surface. It is available at: https://portal.opentopography.org/raster? opentopoID=OTSDEM.032021.4326.3 . . This product was obtained from the WorldDEM data of radar satellite during the TanDEM-X mission. As for Fig. 1B, a script for land use and land cover was utilized in Google Earth Engine (GEE) software at https://earthengine.google.com/. The MOD13Q1 product was obtained from MODIS satellite, collection 006, with spatial resolution of 250 m and temporal resolution of 16days - https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD13Q1. Averages were taken for the period between 2000 and 2023. As per Fig. 1C, a land use and land cover script was used in Google Earth Engine (GEE) software at https://earthengine.google.com/. Total accumulated precipitation data of ECMWF’s 5th version (ERA5) were used with a spatial resolution of 0.25° x 0.25° and a temporal resolution of 1 h. More information exists at https://cds.climate.copernicus.eu/datasets/reanalysis-era5-single-levels? tab=download . Refer to Fig. 1(D), a land use and land cover script was also used in Google Earth Engine (GEE) software and the mean temperature at 2 m of ERA5 were also employed like Fig. 1C.

Data

CMIP6 models

The Coupled Model Intercomparison Project Phase 6 (CMIP6) represents a significant advance over its previous phase (CMIP5), incorporating a larger number of modelling groups, future scenarios, and experiments. It provides global gridded climate data, projected on a regular latitude-longitude grid, with horizontal resolution varying among models. The vertical resolution encompasses single and pressure levels from 1 to 1000 hPa, allowing for a detailed analysis of the atmosphere. The data cover different periods, depending on the experiments. Historical experiments span from 1850 to 2014, while experiments based on Shared Socioeconomic Pathways (SSPs) range from 2015 to 2100, enabling the investigation of different emission trajectories and their impacts. Temporal resolutions vary, and can be monthly, daily, or fixed, which allows for analyses at different time scales31.

In addition to temperature and precipitation, CMIP6 provides a wide range of climate variables, including humidity, wind, solar radiation, cloud cover, and aerosols, allowing for a comprehensive analysis of the climate system32. The SSP scenarios explore different socioeconomic trajectories and their impacts on greenhouse gas emissions, providing insights into how different policies and mitigation actions can influence future climate. The models incorporate various climate forcings, such as greenhouse gases, aerosols, solar and volcanic activity, which influence the Earth’s energy balance and are essential for simulating climate change realistically31,32. Complex climate processes, such as atmospheric and oceanic circulation, the hydrological cycle, the cryosphere, and the carbon cycle, are represented in the models, allowing for the simulation of interactions between different components of the climate system.

It is crucial to recognize that CMIP6 climate projections are subject to uncertainties, which can arise from different sources, such as the representation of climate processes in models, emission scenarios, and natural climate variability32. CMIP6 represents a crucial advance in climate modelling, providing valuable data and information for understanding climate change and its impacts. The use of high-resolution models and the analysis of different emission scenarios allow for a more accurate assessment of the risks and opportunities associated with climate change, contributing to the development of effective adaptation and mitigation strategies.

Pre-processing of CMIP6 data

The pre-processing of CMIP6 data, based on the IPCC-AR6 climate models, is essential as the raw data from climate models may contain inconsistencies, errors, and different formats, making them difficult to analyze. Pre-processing standardizes and homogenizes these data, thus ensuring quality and facilitating their use in subsequent analyses35. Liang et al. (2022) emphasized the application of multivariate linear regression models to adjust and refine future climate projections, highlighting the importance of advanced statistical techniques in enhancing model reliability and reducing uncertainties in projections36,37.

Moreover, climate model data covers large geographical areas and long periods, resulting in extensive and complex datasets. Pre-processing may include selecting relevant variables, reducing spatial or temporal resolution, and interpolating to fill gaps, making the data suitable for specific analyses. Emmenegger et al. (2022) offered insights on how to assess rainfall relationships using CMIP6 models, underscoring the importance of observational data in the validation and calibration of climate models, which are essential for correcting biases and homogenizing data. Another reason for pre-processing is the need to perform comparisons between different climate models or between observations and models, which often require adjustments to ensure that the data are comparable and that the uncertainties associated with the models are considered38.

The pre-processing of CMIP6 data involved applying statistical techniques to correct biases or spurious trends in the model data, thus improving the reliability of climate projections. Based on this, (i) we selected CMIP6 models suitable for the region of interest, considering factors such as historical performance, spatial resolution, and temporal coverage; (ii) we addressed inconsistencies and adjusted heterogeneous data to ensure compatibility and uniformity among the selected models; and (iii) we carried out dimensionality reduction using dimensionality reduction techniques.

Extreme climate indices

In the study, the following indices were used: (i) Consecutive dry days (CDD), (ii) Maximum number of consecutive summer days (CSU), (iii) Consecutive wet days (CWD), (iv) Warm spell duration index (WSDI), and (v) Maximum number of consecutive wet days (WW)33,34. Several research have highlighted the utility of these indices in modeling climate change scenarios and assessing vulnerability to extreme climate conditions.

Artificial neural networks

In the study, Artificial Neural Networks (ANNs) and the backpropagation method were used to create an ensemble of CMIP6 models. Backpropagation is a fundamental algorithm for training ANNs. It is used to adjust the network weights based on the error between the model’s predictions and the actual values of the training data39,40,41.

The backpropagation process starts with the neural network’s weights initialized randomly or using a specific initialization strategy. In this phase, a feedforward pass sends the input data through the neural network, layer by layer, until the outputs are produced. During feedforward, each neuron receives weighted inputs from the previous layer, applies an activation function to the weighted sum of these inputs, and passes the result to the next layer40,42.

Error calculation is based on comparing the outputs produced by the network with the real values of the training data, where the difference between the network’s predictions (outputs of the last layer) and the real values is determined using an error function, such as mean squared error (MSE) or cross-entropy43. Propagate the calculated error from the output layer back to the hidden layers and then to the input layer, adjusting the weights along the way. For each layer, calculate the partial gradients of the error with respect to the layer’s weights using the chain rule. Update the network weights to minimize the error. This is done by moving in the opposite direction of the gradient, a quantity proportional to the learning rate40.

Use the gradients calculated during backpropagation to update the network’s weights to minimize the error. The learning rate determines the size of the steps taken along the gradient. A very high value can lead to oscillations or convergence to a local minimum, while a very low value can result in slow training or stagnation. Repeat steps 2 to 5 for the entire training dataset (or in mini batches, for stochastic batch training) until the network’s error reaches an acceptable level or until a maximum number of iterations is reached. After training, evaluate the network’s performance on a validation set. If necessary, adjust the network’s hyperparameters (such as the learning rate, number of layers, number of neurons per layer etc.) and repeat the training process42,44.

A common specific initialization strategy is the Xavier initialization, also known as Glorot initialization. This strategy was proposed by Xavier Glorot and Yoshua Bengio in their study titled: “Understanding the difficulty of training deep feedforward neural networks” in 201045. The basic idea behind Xavier initialization is that the weights should be initialized in a way that keeps the variance approximately constant across all layers of the network. This helps to prevent gradients from diminishing too quickly or exploding during backpropagation45.

To initialize the weights according to Xavier initialization, use the following formula Eq. (1):

$$\:VAR\left(W\right)=\frac{2}{{n}_{in}+{n}_{out}}$$
(1)

where W is a weight matrix of a layer, \(\:{n}_{in}\) is the number of neurons in the input layer, and \(\:{n}_{out}\) is the number of neurons in the output layer. After calculating the desired variance, the weights can be initialized from a normal distribution with zero mean and variance based on \(\:{n}_{out}\).

Let’s consider a neural network with a single hidden layer to simplify, but the same principle applies to neural networks with multiple hidden layers.

Suppose we have a neural network with:

L layers (counting the input layer as L = 0),

\(\:{n}_{i}\) neurons in layer l,

\(\:{w}_{ij}^{\left(i\right)}\) represents the weight of the connection between neuron j in layer l-1 and

neuron i in layer ,

\(\:{b}_{iji}^{\left(i\right)}\) is the bias for neuron i in layer l,

\(\:{z}_{i}^{\left(l\right)}\) is the weighted sum of the inputs to neuron i in layer l,

\(\:{a}_{i}^{\left(l\right)}\) is the output (or activation) of neuron i in layer l,

\(\:{\delta\:}_{i}^{\left(l\right)}\) is the error for neuron i in layer l.

Let’s describe the backpropagation process46:

For each layer l = 1, 2, ……, L, we calculate the activations\(\:{a}_{i}^{\left(l\right)}\) for all neurons, according to Eq. (2):

$$\:{z}_{i}^{\left(l\right)}={\sum\:}_{j}{w}_{ij}^{\left(l\right)}\:{a}_{j}^{(l-1)}+{b}_{i}^{\left(l\right)},$$

\(\:{a}_{i}^{\left(l\right)}=g({z}_{i}^{\left(l\right)}\) ), (2)

where, g() is the activation function.

The Error Calculation at the Output Layer is based on a function like MSE43 or cross-entropy, given by Eq. (3):

$$\:{\delta\:}_{i}^{\left(L\right)}=\frac{\partial\:e}{\partial\:{z}_{i}^{\left(L\right)}}$$
(3)

,

where, is the error function and \(\:{z}_{i}^{\left(L\right)}\) is the net input before the activation function of neuron in the output layer.

Backpropagation is calculated for each layer l – L – 1, L – 2, …., 1, according to Eq. (4):

$$\:{\delta\:}_{i}^{\left(L\right)}=\frac{\partial\:E}{\partial\:{z}_{i}^{\left(l\right)}}={\sum\:}_{k}\frac{\partial\:E}{\partial\:{z}_{k}^{(l=1)}}\frac{{z}_{k}^{(l=1)}}{\partial\:{a}_{i}^{\left(l\right)}}\frac{\partial\:{a}_{i}^{\left(l\right)}}{\partial\:{z}_{i}^{\left(l\right)}}$$
(4)

,

where, is the index of neurons in layer l + 1.

Update the weights and biases using gradient descent, as per Eq. (5) and Eq. (6):

$$\:{w}_{ij}^{\left(l\right)}={w}_{ij}^{\left(l\right)}-\alpha\:\frac{\partial\:E}{\partial\:{w}_{ij}^{\left(l\right)}}$$
(5)

,

$$\:{b}_{ij}^{\left(l\right)}={b}_{ij}^{\left(l\right)}-\alpha\:\frac{\partial\:E}{\partial\:{b}_{ij}^{\left(l\right)}}$$
(6)

,

Where is the learning rate.

These equations represent the backpropagation process40, where the gradients of the error with respect to the weights and biases are calculated and used to update the parameters of the neural network, allowing it to learn from the training data.

Projections and CMIP6 models

The CMIP6 data are essential for the Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC). They are primarily used to address outstanding scientific questions, enhance understanding of the climate system, provide estimates of future climate changes and related uncertainties, supply data for climate change adaptation, examine climate predictability, and assess the accuracy of models in simulating the recent past. The “experiments” of CMIP6 refer to three main categories of simulations47:

Historical experiments: cover the period from 1850 to 2014, during which modern climate observations exist. These experiments demonstrate how the GCMs (Global Climate Models) behave with respect to past climate and provide a reference for comparison with future projections.

Climate projection experiments: follow the combined pathways of the Shared Socioeconomic Pathway (SSP) and the Representative Concentration Pathway (RCP). They span from 2015 to 2100 and provide different trajectories for future climate forcing48. Additionally, the CMIP6 data now offers the option of spatial and/or temporal subsets to meet data requests, a new feature that depends on computing processes carried out on the ESGF, where the data are located.

The projection of the variables used from 6 GCMs of the CMIP6 was analyzed in three representative scenarios of the SSP: (i) Historical (20th century); (ii) intermediate (RCP-4.5); and (iii) high (RCP-8.5) forcing. RCP-4.5 is the closest representation of current efforts undertaken by nations to curb greenhouse gas emissions (GHG) in the climate system48.

In the process of analyzing complex data (Fig. 2), as those related to climate change, accuracy and reliability are paramount. To handle the variety of available models, we adopt a robust and systematic approach, ensuring consistent and reliable results. Initially, all six selected models undergo an ensemble process known as Aggregation of IPCC-AR6 Models. This technique allows for combining the forecasts from different models, capturing a broader range of perspectives, and reducing potential individual biases. This first aggregation stage is essential for obtaining a comprehensive and representative view of the phenomenon under study. However, accuracy alone is not sufficient. To ensure that the results are further refined and reliable, we apply the Back-propagation method to Artificial Neural Networks (ANN). This process of backpropagation allows for adjusting the weights and connections within the network, optimizing it to better fit the input data and produce more accurate and normalized outputs.

Fig. 2
figure 2

Processing of the models through the Alluvial Diagram of IPCC-AR6 Phases.

Full size image

Finally, to consolidate and further refine our forecasts, we conduct another stage of aggregation through the final ensemble. This additional step of model combination is crucial for integrating all processed information and generating a robust and reliable final forecast. Thus, by adopting this comprehensive and carefully crafted approach, we ensure that our analyses and forecasts are supported by a rigorous process, resulting in valuable and reliable insights for understanding and addressing the complexities of climate change.

Results and discussion

Climatic indices

Figure 3 shows the CMIP6 model simulation scenarios related to the CDD index. Note that Figs. 3, 4, 5, 6 and 7 were prepared using OpenGrADS software, version 2.2.1.oga.1 that was available at: http://opengrads.org/. The CDD (Maximum number of consecutive days with daily precipitation < 1 mm) is an effective measure in assessing extreme rainfall and droughts50,51,52,53. The spatial results of the CDD showed significant regional differences across the three adopted scenarios. In the Historical scenario (Fig. 2a), the highest CDD values (> 35 consecutive dry days) were found in northern (N) and northeastern (NE) Thailand near the Tibetan Plateau, where meteorological drought is frequent in this area (TMD, 2024; Phumkokrux & Trivej, 2024), followed by the Central region, which corresponds to the plains leading to the Gulf of Thailand. In contrast, the southern region, characterized by a peninsular shape, has the highest precipitation due to the southwest (SW) monsoon – (TMD, 2024; Phumkokrux & Trivej, 2024), with lower CDD values (< 10 consecutive dry days).

In the R4.5 scenario, a meridional pattern of CDD appears, with a gradual increase in consecutive dry days from east (E) to west (W) across the North and Central regions of Thailand (Fig. 1), with values ranging between 25 and 35 days. This result is similar to Duan et al.51, who used observed data from 718 rain gauge stations in China from 1961 to 2012 to examine the spatiotemporal characteristics and seasonality of CDD, a result comparable to this study’s findings for the Historical scenario. Once again, southern Thailand maintained the lowest CDD values (< 10 days), similar to the Historical scenario. In the R8.5 scenario, there was a significant increase in CDD values in northern and central Thailand, ranging from 30 to 35 days, followed by an increase in some locations in the southern region, with values between 5 and 15 days, differing from the Historical and R4.5 scenarios.

The study by Limsakul and Singhruck52 revealed an overall pattern of dry days via CDD for the period 1955 to 2014, indicating statistically significant upward trends in Central, Northern, and parts of Southern Thailand. Previously, Sharma & Babel (2014) also identified that CDD was increasing in Northern Thailand based on extreme daily precipitation from 1961 to 2002 for western Thailand via the Mann-Kendall (MK) test, results similar to those from the CMIP6 models.

Rojpratak and Supharatid (2022) used CMIP5 models, adopted by the IPCC, via ensemble to evaluate precipitation extremes in Thailand based on ten climate indices, with baseline scenarios (observation period, 1980–2005) and target projection years (2010–2099). The CDD in Thailand varied from 27 to 112 consecutive dry days. The areas with significant increases were in the Northwest, North, Northeast, and Central regions, while the lowest CDD records occurred in the South, a result similar to that obtained in this study with CMIP6 models. CMIP5 models projected more consecutive dry days across the country, especially in the West, Central, and South, except in the Northeast, a result contrary to that of the CMIP6 models, which did not show an increase in CDD for the southern region.

It is worth noting that the more recent CMIP6 models use a combination of SSPs and RCPs, which better characterize future scenarios (Eyring et al., 2016). CMIP6 has been improved in various aspects, such as higher horizontal resolution, better representation of synoptic processes, and better agreement with global energy balance estimates, yielding satisfactory results in studies of climate extremes (Srivastava et al., 2020). Khadka et al. (2021) evaluated the ability of CMIP5 and CMIP6 models to simulate large-scale atmospheric circulations for the Southeast Asian summer monsoon. The results indicate a better representation of annual precipitation cycles as well as spatial patterns by CMIP6 models compared to CMIP5, particularly since CMIP6 models simulate the onset, retreat, and duration of the rainy season closer to observations, followed by the convergence of summer monsoon moisture flow. The CMIP6 models outperformed CMIP5 models in simulating precipitation and large-scale circulation due to the higher spatial resolutions of CMIP6, a greater number of vertical levels, and better atmospheric and land surface parameterization.

Regardless of the scenarios obtained by the CMIP6 models, southern Thailand remains humid due to complex interactions between climate models in the Indo-Pacific sector, combined with orographic effects52, followed by the intrusion of moist air from sea breeze circulations, which in turn reduces water vapor inland54,55. In contrast, in the northern and central regions, where CDD significantly increased across adopted scenarios, these drought-prone areas are expected to experience greater severity and frequency of drought in the future (Rojpratak and Supharatid, 2022). Therefore, understanding water stress conditions or drought-prone areas is essential for agriculture in Thailand, given its heavy reliance on rainfall (Kiguchi et al., 2021).

Fig. 3
figure 3

CDD Index (days) applied to Thailand for (a) the Historical, (b) R4.5, and (c) R8.5 scenarios.

Full size image

Figure 4shows the CMIP6 model simulation scenarios related to the WSDI index. The WSDI is an annual count of days with at least six consecutive days where the maximum temperature (TX) exceeds the 90th percentile52. The spatial results of the WSDI were significant across the three adopted scenarios. In the historical scenario (Fig. 4a), there were no significant changes in the three regions of Thailand (Fig. 1), except in the N and NE sectors, with a variation of 3 to 9 warm spell days.

In the R4.5 scenario (Fig. 4b), there was an intensification of the WSDI, particularly in the Northern and Central regions, with 12 to 16 warm spell days, followed by the Southern region with less variation in the WSDI (between 5 and 9 warm spell days). Notably, the highest WSDI values were observed in northeastern Thailand. Regarding the R8.5 scenario (Fig. 4c), all regions of Thailand experienced a significant increase in the WSDI (> 15 warm spell days), except for Southern Thailand. The R8.5 scenario results from CMIP6 models were similar to those from the Nested Regional Climate Model applied in Northern Thailand using WSDI for the R8.5 scenario (2020–2029), which indicated an increase in high-temperature days with adverse impacts on agricultural production, particularly rice and maize (Amnuaylojaroen & Parasin, 2022).

Previously, Shrestha & Roachanakanan (2021) evaluated changes in extreme temperatures and precipitation using 14 climate indices in the Lower Songkhram River Basin (Northern Thailand) with three regional climate models (RCMs) selected from CMIP5 under two scenarios (RCP 4.5 and RCP 8.5) for three periods: the 2030s (2020–2044), 2055 (2045–2069), and 2080 (2070–2094), compared to the baseline climate of 1980–2004. In both scenarios, heat events are projected to increase upstream and downstream of the basin, with implications for water resource planning and management.

According to Limjirakan and Limsakul56, Thailand recorded a significant temperature increase across the country over the past four decades, accompanied by extreme events associated with the daily distributions of minimum and maximum temperatures. Kachenchart et al.57showed that city-induced warming contributed to a 40.5% increase in air temperature in Thailand, highlighting how urbanization influences temperature changes and precipitation extremes in Thailand52,58,59. Previously, Manomaiphiboon et al.60 evaluated future temperature changes in Thailand for the 21st century (2031 to 2070) under three IPCC emission scenarios (A2, A1B, and B1) using the RegCM3 model with a resolution of 20 km, based on results from the ECHAM5/MPI-OM model experiments. In all scenarios considered, temperature is expected to increase in Thailand by 0.4 to 3.3 °C, being more pronounced in A2 and A1B in the last decade (2061 to 2070). The Southern sub-region is the least impacted in terms of average temperature, similar to the R4.5 and R8.5 scenario results via CMIP6 adopted in this study.

Fig. 4
figure 4

WSDI Index (days) applied to Thailand for (a) the Historical, (b) R4.5, and (c) R8.5 scenarios.

Full size image

Figure 5 shows the CMIP6 model simulation scenarios related to the CSU index. The spatial results of the CSU indicate no changes across the three scenarios used in the study, regardless of the projections adopted by CMIP6. In all scenarios—Historical (Fig. 5a), R4.5 (Fig. 5b), and R8.5 (Fig. 5c)—values ranged between 70 and 90 days. Spatially, there were no differences across Thailand’s regions, only in the border regions, which contradicts previous results by Limsakul and Singhruck52, where the CSU index showed a reduction across Thailand from 1955 to 2014.

It is noteworthy that station-derived data were used and underwent quality control and data homogeneity checks. The study, based on data from the CLIVAR project provided by the World Meteorological Organization (WMO), showed a significant decrease across Thailand of 0.27 days per decade. This decrease observed in Thailand was also seen in the global average of CSU from 1971 to 200361.

Fig. 5
figure 5

CSU Index (days) applied to Thailand for (a) the Historical, (b) R4.5, and (c) R8.5 scenarios.

Full size image

Figure 6 shows the CMIP6 model simulation scenarios related to the CWD index (Maximum number of consecutive days with daily precipitation > 1 mm). Unlike the CSU index, the spatial results of CWD showed significant changes across the three adopted scenarios. In the historical scenario (Fig. 6a), CWD changes were observed among Thailand’s three regions (Fig. 1), forming a moisture gradient from the North and Northeast, followed by the Central region, with intervals between 15 and 30 rainy days, and particularly the South (> 35 rainy days). Rojpratak and Supharatid (2022) used CMIP5 models via ensemble to assess precipitation extremes in Thailand based on ten climate indices, in contrast to our study, which uses CMIP6 models and ANN to calculate the ensemble with only five climate indices. Similarity in results shows that the lowest CWD values are located in the Northeast, while the highest values are in the South.

It is worth noting that northern Thailand is influenced by (SW) monsoons53, which, in turn, cause flooding and coincide with consecutive rainy days. Additionally, physiographic effects, where mountainous regions meet flat areas, also influence consecutive rainy days62. In the R4.5 and R8.5 scenarios, there was a significant increase in CWD, especially in Central and Southern Thailand (between 20 and 40 rainy days), in contrast to the NE sectors (between 10 and 25 rainy days).

A recent study on climate change-induced precipitation for three future periods using three dynamically downscaled regional climate models (RCM1–RCM3) based on two scenarios (RCP 4.5 and 8.5), covering the near future (2022–2040), mid-future (2041–2060), and distant future (2061–2099) in Central Thailand (Pathumthani - north of Bangkok), indicated a higher susceptibility to future floods, with higher CWD (increasing from 7 days to 21–22 rainy days) – (Thanvisitthpon et al., 2024).

Previous studies by Alexander et al.61and Limsakul and Singhruck52indicated a significant decrease in CWD in Thailand, contrasting with CMIP6 model results, which showed a wet gradient across all proposed scenarios for the Central and Southern regions and a significant decrease in the N and NE sectors. This differs from the study by Amnuaylojaroen63, which was based on the critical scenario (RCP 8.5) using the Nested Regional Climate Model (NRCM). The NRCM, based on the WRF model and forced by the Community Climate System Model version 4 (CCSM4) for initial and boundary climate conditions, showed a significant reduction in CWD in most areas, especially in southern Thailand, with values between − 10 to −30% for the period 2020–2029. The CWD results found in the literature show significant variation compared to those obtained in this study. It is known that models simulate lower precipitation compared to observations, as highlighted by Asadieh and Krakauer (2015). They found that global and continental averages of extreme precipitation data point to a higher value than that obtained from climate models, which in turn indicates that all climate models underestimate the annual maximum daily precipitation.

Fig. 6
figure 6

CWD Index (days) applied to Thailand for (a) the Historical, (b) R4.5, and (c) R8.5 scenarios.

Full size image

Figure 7 shows the CMIP6 model simulation scenarios related to the WW index (Maximum number of consecutive wet days). The spatial results of WW were highly significant across the three adopted scenarios. In the historical scenario (Fig. 7a), values varied significantly in the North, Central, and particularly NE and South of Thailand, with values below 0.6 days.

In the R4.5 scenario (Fig. 7b), there was a significant increase in WW values compared to the historical scenario, particularly between the North and Central regions, with the formation of a WW gradient ranging from 3.6 to 6 rainy days, followed by the N and NE sectors, ranging from 2.4 to 4.2 rainy days, with the highest variation occurring in Southern Thailand, from 1.2 to 9 rainy days. In the R8.5 scenario (Fig. 7c), there was an intensification of WW across all regions of Thailand (ranging from 3 to 9.6 rainy days). The spatial distribution of extreme climate indices by Rojpratak and Supharatid (2022) showed that wet areas occur in the South and East, while some dry areas are located in the North, Northeast, and Central regions.

Fig. 7
figure 7

WW Index (days) applied to Thailand for (a) the Historical, (b) R4.5, and (c) R8.5 scenarios.

Full size image

Comparing CMIP6 data with regional climate models (RCMs) enhances the robustness of our projections by establishing a benchmark based on observed data. Emmenegger et al. (2022) highlights the importance of such comparisons to validate projections and enhance their scientific rigor. We also included comparisons with long-term precipitation datasets from studies on regions like India, following methodologies similar to Cooper (2019) to better ground our findings in observed data trends (Emmenegger et al., 2022; Cooper, 2019).

Supratid & Aribarg (2022) have documented policy strategies that address climate risks in Southeast Asia, providing a basis for comparison and extension of our study’s recommendations (Supratid & Aribarg, 2022). Furthermore, Babel et al. (2011) present examples of agricultural adaptations in response to drought, particularly in rice farming, which is relevant given Thailand’s dependence on rice cultivation (Babel et al., 2011).

Regional variability is influenced by local climate drivers such as topography, proximity to coastlines, and urbanization. For example, the higher drought risk in northern Thailand could relate to mountainous terrain that disrupts monsoon flow, as discussed by Rodchuen et al. (2020). Additionally, Limsakul & Singhruck (2016) and Duan et al. (2017) have shown that topographical variations contribute to localized drought risks in nearby regions, helping reinforce Thailand’s regional variability context in this study (Limsakul & Singhruck, 2016; Duan et al., 2017).

Variability of extreme indices

The study analyzed climate projections for three regions of Thailand, based on the RCP4.5 and RCP8.5 scenarios from the IPCC through climate modeling (Table 1). For this, five extreme climate indices were analyzed as per Sect. 2.2.2. In the Northern region, under the RCP4.5 scenario, the results showed an average of 0.98 and a median of 0.86 CDD. The duration of the maximum number of consecutive days was approximately 2.53 with a median of 2.81 days. The average for CWD was 0.59 days, followed by WSDI with an average of 13.38 days, and WW had an average of 3.50 days.

Table 1 IPCC CMIP6 models used in the study49.
Full size table

In the Central region, also for the same projections, there was an average of 1.01 CDD and a median of 0.85 days. The duration of CSU was approximately 2.92 with a median of 3.03 days. The average of CWD is 0.71 days, while WSDI presents an average of 12.89 days. Extreme WW has an average of 3.53 days. In the Southern region, the projections indicated an average of 0.62 and a median of 0.55 days. The average duration of CSU was approximately 3.01 with a median of 3.04 days. The average of CWD was 0.86 days, followed by WSDI with an average of 8.23 days, and WW had an average of 3.93 days.

The statistical analysis of the data concerning the three regions of Thailand showed that the country will face different climate patterns under the RCP4.5 scenario. Notably, the Northern region showed greater variability in all analyzed climate variables, like what was investigated previously52,59, unlike the Southern region which tends to have less variability, yet with a significant increase in the duration of the rainy season and in maximum precipitation intensity58,63. The Central region displayed an intermediate pattern between the other two mentioned regions. These findings emphasize the need for adaptation and mitigation to face future climate challenges in Thailand. Previously, Kiguchi et al.19showed that there is a modification of climate conditions across Thailand, primarily in relation to the precipitation regime and temperature patterns64,65.

In the RCP8.5 scenario, a significant difference occurred compared to the RCP4.5 scenario, where the Northern Region is expected to have an average dry day duration of approximately 1.03 days, with a range from 0.12 to 2.75 days. The median was 0.99, indicating a symmetrical distribution66. Regarding CSU, the average was 2.60 days, with a range from 0.97 to 3.07 days, and the median was 2.89 days. The CWD had an average of 0.54 days, ranging from 0.02 to 2.13 days, with a median of 0.42 days. The WSDI averaged 19.97 days, with extreme values exceeding 68 days, and a median of 14.98 days. The WW was 3.97 days, with maximum values close to 14.16 days.

For the Central Region, the results showed significant differences, for example, the CDD was 1.04 days, ranging from 0.06 to 2.99 days with a median of 0.94 days. For CSU, the average was 2.94 days, with a range from 1.76 to 3.07 days, and a median of 3.03 days. The CWD had an average of 0.71 days, ranging from 0.00 to 2.30, and a median of 0.44 days. The WSDI averaged 20.21 days, with extreme values reaching up to 72.47 days and a median of 12.81 days. WW had an average of 4.26, with maximum values close to 19.88 days65.

For the Southern Region, under the RCP8.5 scenario, the CDD was 0.64 days, ranging from 0.04 to 1.91 days with a median of 0.55 days. The CSU was 3.00 days, with a range from 2.34 to 3.07 days and a median of 3.05 days. The CWD was 0.91 days, ranging from 0.10 to 2.23 days with a median of 0.87 days. The WSDI averaged 13.74 days, with extreme values reaching up to 59.33 days, and a median of 8.37 days. Lastly, for this scenario, the WW was 5.39 days, with maximum values reaching up to 26.70 days.

The study using the Beijing Climate Center (BCC) Climate System Model version 1.1 (BCC-CSM1.1) (280 km resolution) and BCC-CSM1.1 m (110 km resolution) simulated extreme climate events in China over the past 40 years. However, the BCC-CSM1.1 m did not show an improvement in simulating CDD, in contrast to the results obtained for all of Thailand using CMIP6 models, which indicated regions with both increases and decreases in dry days. CWD, however, is more sensitive to model resolution, yielding better results in the simulation. CWD will decrease on the Tibetan Plateau and in northeastern and Central China, and increase in other areas, particularly under the RCP8.5 scenario via BCC-CSM1.1 m, contrary to the results obtained in the study that indicated changes across all of Thailand.

Simulations from EURO-CORDEX, under two scenarios for various climate indices related to extreme precipitation conditions in Greece during the period from 1971 to 2100, revealed a trend toward drier conditions, especially under RCP8.5. Significant trends were found for both dry and wet periods. Consecutive dry days are expected to increase via CDD by 20–50% (2041–2070) and 40–80% (2071–2100), respectively. Annual CDD showed upward trends in all cases, with the largest changes expected in southern parts of Greece. In contrast, CWD was observed in decline.

Studies on the climate of Thailand have indicated an upward trend in the average of maximum surface temperatures over the last decade in various regions, with the northern region seeing an increase of 0.164 °C, the central region by 0.137 °C, the eastern region by 0.127 °C, the northeast by 0.194 °C, and the southern region by 0.068 °C67. Moreover, the analysis of climate variables in Thailand points to changes in rainfall patterns, increases in temperature, and variations in aerosol optical depth across different seasons68.

Recent studies emphasize the various climatic impacts observed in these regions due to global climate changes. For example, Northern Thailand experiences significant seasonal variations that influence both agricultural production and biodiversity19,20. Central Thailand, though less varied, faces challenges related to water management and the effects of urban heat59,62. Southern Thailand, known for its tropical climate, exhibits different patterns due to its geographical isolation and maritime influences, impacting local ecosystems and coastal processes52.

Therefore, it is understood that the varied impacts of climate change across the country require specific adaptation strategies for each region. The incorporation of comprehensive climate modeling data and the use of robust predictive methodological frameworks, as suggested by the recent CMIP6 scenarios, assist in projecting climate conditions with greater accuracy31,32.

These results highlight the significant variations in climate patterns among the regions of Thailand under different climate scenarios. Increases in temperature and changes in precipitation distribution can have significant impacts on agriculture, water resources, and local ecosystems. These data underscore the importance of adaptation and mitigation strategies to address the challenges posed by climate change in different parts of the country19,64,65.

Regarding the percentages for the 20th century, as a means of identifying climate changes, Table 2 presents the values of climate indices for three regions (North, Central, and South) and the average of these values for the 20th century alone. In the North, the CDD was low (0.46%), with a lower incidence of dry days. Conversely, the WSDI was high (24.78%), indicating intense Warm Spell Duration Index events. The other indices, CSU, CWD, and WW, had values close to zero, showing little significance.

Table 2 Percentages (%) for each region of extreme climate indices, with increases (positive) or decreases (negative) for the historical scenario (20th century).
Full size table

In the Central region, the CDD was moderate (0.57%), with a moderate incidence of dry days, while the CSU was negative (−0.42%), indicating a decrease in the Maximum Number of Consecutive Summer Days. The CWD was high (2.55%), suggesting a greater presence of wet days. The WSDI and WW also show significant values (3.90% and 0.92%, respectively), indicating a moderate intensity of wet events and a moderate amount of precipitation.

In contrast, in the South, the values were stark; for example, the CDD was very high (67.82%), indicating an increase in dry days, while the WSDI was extremely high (81.23%), suggesting a very high intensity of wet events. The CSU, CWD, and WW are also considerable (0.33%, 20.09%, and 9.70%, respectively), showing a moderate number of rainy days, a high presence of wet days, and a significant amount of precipitation.

The average values across the three regions show that, in general, there is a moderate incidence of dry days (22.95%) and a low number of rainy days (−0.05%). The CWD is moderate (7.52%), while the WSDI is relatively high (36.63%), indicating a considerable intensity of wet events. The average amount of precipitation was 3.57%. These data suggest significant climatic variability between the regions, with the South experiencing drier and wetter conditions, while the North experiences a more stable climate, with less variation between dry and wet days.

Conclusion

This study employs an ensemble of six IPCC-CMIP6 climate models, leveraging the backpropagation method to assess extreme climate indices across multiple future projections for Thailand. The findings underscore substantial regional disparities in climate impacts across Thailand’s diverse geographical landscape, particularly under the R4.5 and R8.5 scenarios. The North and Northeast regions are projected to experience an increase in drought events, posing significant threats to agriculture, the predominant economic activity in these areas. Additionally, a pervasive rise in air temperature across all regions is expected, which, coupled with increased rainfall in the South, will impact multiple sectors essential to Thailand’s economic stability, particularly tourism, industry, and agriculture.

Detailed regional analyses reveal an increasing number of dry days moving from the North to Central regions under R4.5 and R8.5, suggesting that these regions will face intensified water scarcity challenges. Conversely, the Southern region, though less affected by drought, is projected to encounter more frequent and intense rainfall events. These shifts in precipitation patterns signal critical implications for water resource management and flood preparedness, especially in areas where tourism is a vital part of the economy. The heightened frequency of maximum daily extremes in temperature and precipitation, observed across all projections, further emphasizes the urgency for Thailand’s policymakers to develop and implement adaptive strategies tailored to the unique vulnerabilities of each region.

Thailand’s projected climate trajectory, as well as that of neighboring Southeast Asian nations, signals a future marked by considerable increases in both temperature and extreme weather events25. These projections underscore governance challenges related to both historical and anticipated climate shifts under scenarios R4.5 and R8.5. The Northern region emerges as a climate vulnerability hotspot, exhibiting the most significant variability across climate variables. This heightened vulnerability, coupled with the socio-economic challenges of the Northern region, suggests that the impacts of climate change will disproportionately affect its already vulnerable communities, exacerbating socio-economic inequities. Meanwhile, Southern Thailand, while experiencing relatively less variability, is expected to undergo a prolonged rainy season coupled with peak rainfall intensities. Such changes could disrupt tourism - a sector crucial for the regional economy - underscoring the need for targeted mitigation strategies aimed at sustaining economic resilience and minimizing climate-related disruptions.

In summary, this study highlights the need for a multi-sectoral approach to climate resilience, especially for industries like environmental management, water resources, agriculture, and tourism, which will be directly impacted by these regional climate shifts. The findings emphasize the critical importance of protecting populations at high climate risk, particularly those in socially vulnerable areas, as climate impacts in tropical regions are expected to exacerbate existing socio-economic disparities. Adaptive responses tailored to Thailand’s diverse regional needs are imperative, requiring coordinated efforts across local, national, and regional governance levels. This study not only provides valuable insights for Thailand’s strategic planning but also underscores the broader implications for Southeast Asia, advocating for region-specific adaptation and mitigation strategies to manage the differential impacts of climate change. As Thailand and its neighbors brace for increasingly frequent and severe climate extremes, proactive adaptation and resilience-building strategies are essential to ensuring sustainable socio-economic development across this climatically vulnerable region.