An Investigation of Changes in the New Thornthwaite Climate Classification Based on Temperature, Rainfall, and Evapotranspiration over Thailand, Using CMIP 5 for the Mid-21st Century Period

Abstract

This study aims (1) to study the trend and characteristics of monthly air temperature, monthly rainfall, and potential evapotranspiration (PET) in Thailand over the mid-21st century (2022–2060) period, and (2) to create a climate pattern map using the New Thornthwaite Climate Classification in Thailand over the same period under RCP4.5 and RCP8.5 scenarios using CSIRO-Mk3 in the CMIP5 dataset with Empirical Quantile Mapping (EQM) statistical downscaling. Spatial analyses of temperature and PET reveal significant warming trends, with temperatures rising by approximately 0.033 °C/year and PET rising by about 10 mm/year, especially in the Bangkok Metropolitan Region due to the urban heat island effect, with temperature values under RCP8.5 remaining consistently higher than those under RCP4.5. Rainfall projections show relatively stable spatial patterns across both scenarios, with higher concentrations along the Andaman coast, the eastern peninsula, and northeastern Thailand; these are areas influenced by the southwest monsoon and tropical cyclones. Central Thailand, however, exhibits persistently low rainfall, likely due to rain-shadow effects. PET patterns mirror early 21st-century observations, with the highest values projected in central Thailand and increasing trends under both scenarios, suggesting heightened drought risks. By 2060, The New Thornthwaite Climate Classification indicates that Moist climate zones are projected to disappear nationwide, with Semi-arid and Dry climates dominating under both RCP4.5 and RCP8.5 scenarios. Annual mean temperature will rise by 0.033 °C/year and PET by ~10 mm/year, while rainfall trends remain nearly stable. The classification’s reliance on minimal parameters—temperature, precipitation, and PET—provides a practical tool for climate monitoring and policy development.

1. Introduction

Climate change has been a global concern for over 65 years [1], largely driven by the industrial revolution, which shifted economies from agriculture to industry via combustion-based machinery and energy production [2,3]. While natural factors—terrestrial, astronomical, and extraterrestrial—affect climate [4], human activities have accelerated recent changes, leading to significant greenhouse gas emissions, including CO₂, N₂O, and CH₄ [5]. CO₂, in particular, increased from 260 ppm in 1750 to 420 ppm in 2020, contributing ~80% of observed warming since 1990 [6].Global temperatures have risen ~0.17 °C per decade, with spatially varying impacts shaping regional climates [7,8,9]. Climate change affects ecosystems and human activities differently across regions [10,11,12], consistent with the “dry gets drier, wet gets wetter” paradigm and Koutsoyiannis’ (2020) hypothesis on intensified hydrological cycles [13].

Climate classification aids in identifying regional climates and guiding land use and adaptation [14,15]. Köppen relies on temperature and rainfall, while the Thornthwaite system (TCC) incorporates potential evapotranspiration (PET) for assessing water balance impacts on vegetation and agriculture [16,17]. PET-based systems improve drought monitoring and are increasingly refined globally [18,19]. Thailand, located in the tropical zone, is a major agricultural country [20,21] experiencing rising summer temperatures [22], intense but less frequent rainfall [23], and increasing droughts [24], with expansion of the tropical wet and dry (Aw) zone [25]. Recent studies (e.g., Muhammad Iskandar et al., 2025; Amnuaylojaroen, 2021) [26,27] emphasize the need to update regional climate classifications using high-resolution downscaling in Southeast Asia, especially considering the 2023–2025 drought episodes and enhanced CMIP6 variability. However, few have focused on Thailand using the updated New Thornthwaite system with Empirical Quantile Mapping. This study addresses that gap and provides a simplified yet physically interpretable framework for policymakers to understand long-term climate risks.

Our previous study [28] established baseline climate zones in Thailand using observed temperature, precipitation, and PET (1987–2021) via the New TCC, supporting applications in mangrove management, drought assessment, pollution studies, energy planning, and adaptation strategies [29,30,31]. That study focused on spatial patterns, not future projections. The present study evaluates projected climate conditions for 2022–2060 using downscaled CMIP5 models to assess potential shifts in climate classifications.

Table 1. Comparison between the previous study (Phumkokrux & Trivej, 2024 [28]) and this study.

Aspect	Phumkokrux & Trivej (2024) [28]	This Study
Research focus	Investigation of baseline climate characteristics in Thailand.	Assessment of future climate shifts under climate change scenarios.
Time period	1987–2021 (observed data from TMD stations)	2022–2060 (projected data from CMIP5 models)
Climate variables	Temperature, precipitation, PET	Temperature, precipitation, PET (from climate model projections).
Climate classification	New Thornthwaite Climate Classification (TCC)	New Thornthwaite Climate Classification (TCC)
Statistical analysis	Mann–Kendall trend test, Sen’s slope, Pettitt test.	Mann–Kendall trend test, Sen’s slope, Pettitt test (applied to future projections).
Data source and resolution	Ground-based data from ~104 TMD stations.	Downscaled CMIP5 GCM outputs.
Main contribution	Provided the baseline spatial patterns and trends of climate classes in Thailand.	Provides projected shifts in climate classes under future scenarios.
Novelty	Established baseline climate zones	Offers future-oriented projections for climate adaptation and planning.

highlights conceptual and methodological differences, emphasizing the novelty of incorporating future scenarios, providing insights for adaptation, and long-term planning in Thailand [32,33,34].

This study aims to provide an improved understanding of Thailand’s mid-century climate trajectories using a simplified yet physically meaningful classification (New TCC). The results contribute to regional adaptation planning and national climate policy design.

2. Study Boundary

This study focuses on Thailand, located in Southeast Asia near the equator in the Northern Hemisphere, sharing borders with Myanmar, Laos, Cambodia, and Malaysia. The country spans 517,624 sq. km [35], with geographic coordinates ranging from 5°37′ N to 20°37′ N and 97°22′ E to 105°37′ E [36]. Thailand comprises 77 provinces and a population of 65,951,210, divided into five meteorological regions. Topography and climate vary widely, with elevations ranging from −77 to 2565 m above mean sea level (MSL), an average annual temperature of 20–30 °C, and annual rainfall of 2000–2500 mm [37].

The Thai Meteorological Department (n.d.) classifies the climate into three seasons: summer (February–May), rainy (June–September), and winter (October–January). The northern region is dominated by north–south mountain ranges and valleys, where the country’s highest peak and lowest temperatures can be found. The northeastern region features a sandstone-based plateau that frequently experiences meteorological droughts and borders Laos and Cambodia. The central region consists mainly of floodplains with abundant surface and groundwater resources that are connected to the Gulf of Thailand; this region, particularly Bangkok, records the highest annual temperatures. The eastern region, bordering Cambodia and the Gulf of Thailand, has undulating plains and low hills, and receives the highest amount of rainfall, due to the southwest monsoon. The southern region, a peninsula bordered by the Gulf of Thailand to the east and the Andaman Sea to the west, connects to Malaysia and is influenced by both the southwest and northeast monsoons, resulting in year-round high rainfall [25,37].

A total of 104 meteorological stations under the Thai Meteorological Department were selected for integration with the climate simulation model. The study area is illustrated in Figure 1.

3. Data and Methodology

3.1. Temperature and Rainfall Prediction for the Mid-21st Century, Based on the Early 21st Century Data

The New Thornthwaite Climate Classification uses only temperature and rainfall data, from which potential evapotranspiration can be derived. Mid-21st century (2022–2060) projections of temperature and rainfall were obtained from four climate models in the CMIP5 series, representing four countries and continents: CanESM2 (Canadian Centre for Climate Modelling and Analysis, Canada, North America), CSIRO-Mk3 (Commonwealth Scientific and Industrial Research Organization/Queensland Climate Change Centre of Excellence, Australia, Oceania), MRI-CGCM3 (Meteorological Research Institute, Japan, Asia), and NorESM1-ME (Bjerknes Centre for Climate Research, Norwegian Meteorological Institute, Norway, Europe). These datasets, summarized in

Table 2. Climate model for historical weather variable prediction.

Climate Model	Country	Variable	Dataset
CanESM2	Canada	Temperature	tas_Amon_CanESM2_historical_r1i1p1_185001-200512
		Rainfall	pr_Amon_CanESM2_historical_r1i1p1_185001-200512
CSIRO-Mk3	Australia	Temperature	tas_Amon_CSIRO-Mk3-6-0_historical_r1i1p1_185001-200512
		Rainfall	pr_Amon_CSIRO-Mk3-6-0_historical_r1i1p1_185001-200512
MRI-CGCM3	Japan	Temperature	tas_Amon_MRI-CGCM3_historical_r1i1p1_185001-200512
		Rainfall	pr_Amon_MRI-CGCM3_historical_r1i1p1_185001-200512
NorESM1-ME	Norway	Temperature	tas_Amon_NorESM1-ME_historical_r1i1p1_185001-200512
		Rainfall	pr_Amon_NorESM1-ME_historical_r1i1p1_185001-200512

, were accessed via the World Data Center for Climate (WDCC) at https://www.wdc-climate.de/ui/q (accessed on 11 March 2024).

Simulated data were downscaled using the SD-GCM tool (version 2), provided by AgriMetSoft, Ames, IA, USA, using three statistical methods: Delta [38], Quantile Mapping (QM) [39], and Empirical Quantile Mapping (EQM) [40]. The most suitable model and downscaling method were selected by comparing simulated outputs with historical observations (1987–2005) from 104 meteorological stations. The period 1987–2005 was selected for both observation and simulation data, as the available historical datasets provide comparative information only up to 2005. Moreover, after examining the observation records, data collected after 1987 from the 104 selected meteorological stations were found to be more complete and more reliable than those from earlier years. The model performance was evaluated using a leave-one-out cross-validation (LOOCV) approach. In this method, each observation in the dataset is iteratively excluded from the training process and used as an independent test sample, while the remaining observations are employed to train the model. This process is repeated for all samples, and the overall accuracy is computed as the average of the individual validation results. LOOCV provides a nearly unbiased estimate of model performance, making it particularly suitable for datasets with a limited number of samples [41]. Model performance was evaluated using Mean Absolute Error (MAE) (Equation (1)) and Pearson correlation coefficient (Equation (2)). The combination yielding the lowest MAE and highest Pearson value was selected.

Subsequently, RCP4.5 and RCP8.5 datasets from the selected model were used to project mid-21st century temperature and rainfall using SD-GCM with the chosen downscaling method.

3.2. Trend and Characteristics of Monthly Air Temperature, Monthly Precipitation, and Evapotranspiration (PET) in Thailand for the Mid-21st Century

The New Thornthwaite Climate Classification prediction is classified using only air temperature (°C), rainfall (mm), and PET (mm) over the mid-21st century period (2022–2060), and these factors were simulated by the best selected CMIP5 model with appropriate statistical downscaling of the same 104 meteorological station locations covering Thailand, as shown in Figure 1b.

3.2.1. Data Variability

The coefficient of variation (%CV) method was selected to verify the precision of temperature and rainfall data by using a process from the normal distribution, as in Equation (1) [35,42]. A high %CV value indicates large data distribution, making it inappropriate for further use in the analysis; thus, the accepted value should not exceed 30% [43,44].

$$CV=100\times {\displaystyle \frac{\sigma }{\overline{Y}}}$$

(1)

where σ is the standard deviation of the data and $\overline{Y}$ is the mean of the data in N years.

3.2.2. Mean of Data Calculation in Time Series

The cumulative seasonal mean (CSM) technique was employed to calculate the average values for each temporal interval (decadal or centennial). These calculated means served as reference points to evaluate the stability or variation in the time series data [42,45,46,47], as defined in Equation (2).

$${CSM}_j={\displaystyle \frac{1}{j}}{\sum }_{i=1}^j{Y}_i, J=1, 2, 3,\dots , 4$$

(2)

where Y_i refers to temperature or precipitation or others in each year, while N is the number of years under the study (decade or century scale).

3.2.3. Homogeneity Test of Data and Abrupt Change Analysis

To assess data homogeneity and identify abrupt changes, Pettitt’s test—a straightforward, non-parametric statistical method—was utilized. This approach determines whether a dataset is uniform or contains a shift at an unknown time point (t), based on principles similar to the Mann–Whitney test. Under the null hypothesis (H₀), the data are considered homogeneous, while the alternative hypothesis (H_a) indicates the presence of a structural change at a specific time, supporting the identification of an abrupt transition. The formula used for Pettitt’s test is presented in Equation (3). In the case where the alternative hypothesis (H_a) is supported, the U_t,_T statistic—derived from the Mann–Whitney method—is subsequently applied across the entire dataset from time point 1 to T. The identification of a change point at time t is determined using Equations (4) and (5), with statistical significance evaluated according to Equation (6) [48,49,50]. The homogeneity assessment and abrupt change detection using Pettitt’s test were conducted via the XLSTAT software from student license (version 2023.2.1413 (1413)), provided by Lumivero, Denver, CO, USA, employing a 95% confidence level (p-value ≤ 0.05).

$$D_{ij}=sgn\left(x_i{-x}_j\right)=\left\{\begin{array}{c}-1,<\left(x_i{-x}_j\right)\\ 0,=\left(x_i{-x}_j\right)\\ +1,>\left(x_i{-x}_j\right) \end{array}\right\}$$

(3)

$$U_{t,T}={\sum }_{i=1}^t{\sum }_{j=t+1}^T{D}_{ij}$$

(4)

$$K_T={max}_1<t<T\left|U_{t,T}\right|$$

(5)

$$p=2·exp\left({\displaystyle \frac{-6K_T^2}{T^2+T^3}}\right)$$

(6)

where $x_i$ and $x_j$ are random variables from the data, while x_t follows x_t in time series. $U_{t, T}$ depends on $D_{ij}$, and $K_T$ refers to a change point at time t.

3.2.4. Trend Analysis

The Mann–Kendall trend test is a non-parametric method recommended by the World Meteorological Organization (WMO), and is widely applied for detecting trends in hydrological and meteorological datasets, without the need for data to follow a normal distribution. This method is represented in Equations (7)–(10). In addition, Sen’s slope is often used as a robust estimator for identifying trend magnitude and direction over time between two variables in a linear form, with minimal sensitivity to outliers or measurement errors [51,52,53,54,55,56,57,58]. The mathematical expressions for Sen’s slope are provided in Equations (11) and (12).

$$S={\sum }_{i=j}^{n-1}{\sum }_{j=i++1}^{n}sgn(x_j-x_i)$$

(7)

$$sgn\left(x_j-x_i\right)=\left\{\begin{array}{c}+1 , \left(x_j-x_i\right)>0 \\ 0 , \left(x_j-x_i\right)=0\\ -1 , \left(x_j-x_i\right)<0 \end{array}\right\}$$

(8)

$$Var\left(S\right)={\displaystyle \frac{n\left(n-1\right)\left(2n+5\right)-{\sum }_{i=1}^m{t}_i\left(i\right)(i-1)(2i+5)}{18}}$$

(9)

$$Z=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{Var\left(S\right)}}}\\ 0, S=0\\ {\displaystyle \frac{S+1}{\sqrt{Var\left(S\right)}}}, S<0\end{array}\right.$$

(10)

where S indicates the direction of the trend, with positive values suggesting an upward trend, negative values indicating a downward trend, and a value of zero implying no observable trend. The variance of S, denoted as var(S), is used to compute the Z-score, which helps determine the strength and direction of the trend. A higher Z-score reflects a significant increasing trend, whereas a lower Z-score signifies a decreasing trend.

$$Q_i={\displaystyle \frac{(x_j-x_i)}{j-i}}, i=\mathrm{1,2},3,\dots ,N$$

(11)

$$Q_{med}=\left\{\begin{array}{c}{Q}_{\left[{\displaystyle \frac{N+1}{2}}\right]}, if N=odd\\ {\displaystyle \frac{Q_{\left[\frac{N}{2}\right]}+Q_{\left[\frac{N+1}{2}\right]}}{2}}, if N=even\end{array}\right.$$

(12)

3.3. The New Thornthwaite Climate Classification for the Mid-21st Century

The Thornthwaite Climate Classification (THC) is a system designed to assess climate conditions relevant to agriculture and vegetation. First introduced in 1943 by Charles Warren Thornthwaite [16], it was based on a moisture index originally developed by Karl Linsser [59]. The method later evolved to incorporate water balance metrics and class intervals. Due to its complexity, Carter and Mather (1966) simplified the system by integrating the moisture index and water balance into a more practical model based on temperature, precipitation, and solar radiation [60].

Willmott and Feddema (1992) [61] and Feddema (2005) [62] further refined the method by introducing the New Thornthwaite Moisture Index (TMI), which requires only monthly rainfall (mm) and temperature (°C), thus enhancing its applicability. Since daily solar radiation data are often unavailable, Al-Sudani (2019) proposed latitude- and month-specific adjustment factors (Adj) to estimate solar influence (

Table 3. Adjustment factors in the TMI equation, separated by latitude and month (Adj) [63].

Lat./Month	January	February	March	April	May	June	July	August	September	October	November	December
60° N	0.54	0.67	0.97	1.19	1.33	1.56	1.55	1.33	1.07	0.84	0.58	0.48
50° N	0.71	0.84	0.98	1.14	1.28	1.36	1.33	1.21	1.06	0.90	0.76	0.68
40° N	0.80	0.89	0.99	1.10	1.20	1.25	1.23	1.15	1.04	0.93	0.83	0.78
30° N	0.87	0.93	1	1.70	1.14	1.17	1.16	1.11	1.03	0.96	0.89	0.85
20° N	0.92	0.96	1	1.05	1.09	1.11	1.10	1.07	1.02	0.98	0.93	0.91
10° N	0.97	0.98	1	1.03	1.05	1.06	1.05	1.04	1.02	0.99	0.97	0.96
00°	1	1	1	1	1	1	1	1	1	1	1	1
10° S	1.05	1.04	1.02	0.99	0.97	0.96	0.97	0.98	1	1.03	1.05	1.06
20° S	1.10	1.07	1.02	0.98	0.93	0.91	0.92	0.96	1	1.05	1.09	1.11
30° S	1.16	1.11	1.03	0.94	0.89	0.85	0.87	0.96	1	1.07	1.14	1.17
40° S	1.23	1.15	1.04	0.93	0.83	0.78	0.80	0.98	0.99	1.10	1.20	1.25
50° S	1.33	1.19	1.05	0.98	0.75	0.68	0.70	0.82	1.97	1.13	1.27	1.36

) [63]. TMI values range from −1 to 1, and are calculated using Equations (13)–(16) and

Table 4. Moisture Index for New Thornthwaite Climate Classification [62].

Moisture Type	Moisture Index (TMI)	Color in Map
Saturated	0.66 to 1.00
Wet	0.33 to 0.66
Moist	0.00 to 0.33
Dry	−0.33 to 0.00
Semi-Arid	−0.66 to −0.33
Arid	−1.00 to −0.66

. A TMI of −1 indicates no rainfall, whereas 1 indicates no PET, and 0 reflects a balance between rainfall and PET.

$$TMI=\left[\begin{array}{c}1-{\displaystyle \frac{PET}{Pr}};Pr>PET\\ 0;Pr=PET=0\\ {\displaystyle \frac{Pr}{PET}}-1;Pr\le PET\end{array}\right]$$

(13)

$$PET=\left(16·{\left({\displaystyle \frac{10T_m}{I}}\right)}^{\propto }\right)·{Adj}_i; in \mathrm{m}\mathrm{m} per month$$

(14)

$$I={\sum }_{i=1}^{12}{\left({\displaystyle \frac{T_{m_i}}{5}}\right)}^{1.514}$$

(15)

$$\propto =\left(6.75·{10}^{-7}\right)I^3-\left(7.71·{10}^{-5}\right)I^2+\left(1.792·{10}^{-2}\right)I+0.49239$$

(16)

In this context, PET indicates the monthly potential evapotranspiration measured in mm. Similarly, Pr stands for the amount of monthly precipitation in mm. The Heat Index, represented by I, is calculated based on the twelve-monthly mean temperatures ($T_{m_i}$), where $T_{m_i}$ is expressed in °C. The Adj_i accounts for adjustments that vary with latitude and daily solar duration, as detailed in Table 3.

Although air temperature has traditionally served as a thermal indicator, Thornthwaite (1948) pointed out that temperature by itself cannot adequately delineate thermal zones [16]. Thornthwaite further argued that PET better captures the evaporative demand of an ecosystem. Therefore, annual PET values (${PET}_{Ann}$) were selected for the calculation, including monthly PET (${PET}_m$) as defined in Equation (17), to establish its Thermal Index [62]. Based on this index, climate regimes were grouped into six distinct classes, as in

Table 5. Thermal Index for New Thornthwaite Climate Classification [62].

Thermal Index	Annual PET (mm)	Color in Map
Torrid	>1500.0
Hot	1200.0 to 1500.0
Warm	900.0 to 1200.0
Cool	600.0 to 900.0
Cold	300.0 to 600.0
Frigid	0.0 to 300.0

.

$${PET}_{Ann}={\sum }_i^{12}\left({PET}_m\right); in \mathrm{m}\mathrm{m} per month$$

(17)

Traditional climate classifications often distinguish “dry” and “wet” seasons based on “winter” and “summer.” However, Carter and Mather (1966) and Feddema (2005) questioned the applicability of these definitions near the equator, where such distinctions are unclear. They proposed identifying actual periods of wetness and dryness as a more accurate approach to capturing climate variability [60,62].

The New Thornthwaite Climate Classification, as refined by Carter and Mather (1966) [60] and presented by Feddema (2005) [62], builds on Thornthwaite’s original 1948 model. It uses the Thornthwaite Moisture Index (TMI) to represent climate variability based on precipitation and potential evapotranspiration (PET). Annual variability is assessed through the range of monthly TMI values (Equation (18)), with classification categories outlined in

Table 6. Climate variability level for New Thornthwaite Climate Classification [62].

Climate Variability	Annual TMI Range	Color in Map
Low	0.0 to 0.5
Medium	0.5 to 1.0
High	1.0 to 1.5
Extreme	1.5 to 2.0

.

$$Climate Variability = {TMI}_{\left(\max \right)}-{TMI}_{\left(\min \right)}; in each year$$

(18)

Climate variability might result from fluctuations in rainfall, temperature-related factors, or a combination of both. Determination of underlying causes helps us to better understand this variability, and this was calculated as in Equation (19), resulting in

Table 7. Climate variability cause for New Thornthwaite Climate Classification [62].

Cause	Annual Pr Range/Annual PET	Color in Map
Precipitation	<0.5
Combination	0.5 to 2.0
Temperature	>2.0

. A broad gap between the highest and lowest rainfall values suggested that rainfall was the main contributor to seasonal changes. On the other hand, if the variation in PET was relatively small, it indicated that temperature had a more significant role in shaping the seasonality [62].

$$Variability Cause = {\displaystyle \frac{{Pr}_{\left(max\right)}-{Pr}_{\left(min\right)}}{{PET}_{\left(max\right)}-{PET}_{\left(min\right)}}}; in each year$$

(19)

3.4. Spatial Distribution Analysis

Spatial distribution was analyzed using GIS-based mapping techniques. Shape files for Thailand’s national boundaries and meteorological stations were obtained from OCHA (https://data.humdata.org/dataset/geoboundaries-admin-boundaries-for-thailand, accessed 16 October 2024) [64] and the Thai Meteorological Department (TMD) (http://climate.tmd.go.th/content/file/75, accessed 6 October 2022) [65], respectively. Ordinary kriging interpolation was applied with 400-cell resolution using data from 12 surrounding stations. Output maps were exported at 500 dpi. Although Thailand’s official land area is 517,624 km² [35,66], the interpolated raster area was slightly reduced to 516,329.20 km² (−0.25%).

4. Results and Discussion

4.1. Temperature and Rainfall Value Predictions for Thailand for the Mid-21st Century

Four CMIP5 climate models, each from a different continent, were evaluated for their forecasting suitability in Thailand using Mean Absolute Error (MAE) and Pearson correlation coefficient. As shown in

Table 8. Model accuracy investigation for temperature and rainfall simulation in Thailand.

Stat.	Model	Country	Temperature			Rainfall
			Delta	EQM	QM	Delta	EQM	QM
MAE	CanESM2	Canada	1.32	1.08	1.16	79.19	87.20	85.17
	CSIRO-Mk3	Australia	1.57	0.84	0.87	75.66	85.90	83.40
	MRI-CGCM3	Japan	1.68	0.85	0.89	99.10	83.28	89.00
	NorESM1-ME	Norway	1.58	0.87	0.87	93.84	80.38	79.04
Pearson	CanESM2	Canada	0.572	0.604	0.572	0.544	0.516	0.538
	CSIRO-Mk3	Australia	0.740	0.743	0.740	0.599	0.501	0.525
	MRI-CGCM3	Japan	0.718	0.722	0.718	0.462	0.504	0.472
	NorESM1-ME	Norway	0.736	0.734	0.736	0.595	0.577	0.533

, the CSIRO-Mk3 model (developed by the Commonwealth Scientific and Industrial Research Organization (CSIRO), Australia) showed the best performance with the lowest MAE and highest Pearson values for both temperature and rainfall. Among the downscaling methods, Empirical Quantile Mapping (EQM) proved most effective for temperature (±0.84 °C, Pearson = 75.66), while the Delta method was more suitable for rainfall (±0.743 mm, Pearson = 0.599). The Empirical Quantile Mapping (EQM) method adjusted model outputs to the observed 1987–2005 TMD station data. Each grid cell was bias-corrected monthly to preserve intra-annual variability, with extreme rainfall events retained via generalized Pareto tail fitting. A leave-one-out cross-validation confirmed the bias-corrected outputs maintained consistent mean and variance values across subsets [40,41]. Therefore, CSIRO-Mk3 was selected for future projections, using EQM for temperature and Delta for rainfall.

4.2. Temperature Prediction for the Mid-21st Century Period and Its Characteristics

The CSIRO-Mk3 dataset under RCP4.5 and RCP8.5 scenarios, as presented in

Table 9. Description of CSIRO-Mk3 dataset for temperature and rainfall forecasting of Thailand for the mid-21st century period.

CSIRO-Mk3	Variable	Dataset
Rcp45	Temperature	tas_Amon_CSIRO-Mk3-6-0_rcp45_r1i1p1_200601-210012
	Rainfall	pr_Amon_CSIRO-Mk3-6-0_rcp45_r1i1p1_200601-210012
Rcp85	Temperature	tas_Amon_CSIRO-Mk3-6-0_rcp85_r1i1p1_200601-210012
	Rainfall	pr_Amon_CSIRO-Mk3-6-0_rcp85_r1i1p1_200601-210012

, was utilized in combination with the Empirical Quantile Mapping (EQM) statistical downscaling method to extract the average annual temperature for 104 meteorological stations across Thailand during the mid-21st century period (2022–2060), using the SD-GCM Version 2 software tool [67].

During the study period, Thailand’s average annual temperature ranged from 20.5 to 29.8 °C (RCP4.5) and 20.6 to 30.0 °C (RCP8.5). Monthly temperatures rose sharply from February, peaking in April (22.1–31.0 °C for RCP4.5; 22.2–31.1 °C for RCP8.5), before declining during the rainy season and reaching their lowest points in December–January (18.1–29.0 °C for RCP4.5; 18.3–29.3 °C for RCP8.5). Across all months, RCP8.5 showed slightly higher values than RCP4.5 (Figure 2a and Figure 3).

Spatially, both scenarios showed similar patterns, with the highest temperatures in central Thailand, particularly the Bangkok Metropolitan Region, which is home to 10.9 million people (16.5% of the national population) [68]. This urbanized, industrial area tends to trap heat, especially due to its minimal forest cover [69,70,71]. Temperatures decreased toward the northern region, influenced by mountainous terrain [72] and higher latitudes, which lowers solar incidence angles [73]. Although spatial trends were similar, RCP8.5 showed significantly higher temperatures, especially in the north and northeast, likely due to unstable northeast monsoons in winter and uncertain cyclone activity during the rainy season [74]. These areas also lie farther from the sea and are at higher elevations, limiting heat retention and ventilation [75]. The cumulative seasonal means (CSMs), shown by the pink line in Figure 4, were approximately 28.0 °C and 28.2 °C for RCP4.5 and RCP8.5, respectively. Seasonally, the CSM values were ~29.0 °C (summer), 28.5 °C (rainy), and 26.6 °C (winter) for RCP4.5, and 29.2 °C, 28.5 °C, and 26.8 °C for RCP8.5.

Temperature variability was assessed using the coefficient of variation (%CV) (Figure 3,

Table 10. Important statistical data on temperature, rainfall, and PET for RCP45 and RCP85.

Temperature—RCP45
Season	Min	Max	CSM.	S.D.	Kendall’s tau	p-value	Sen’s slope	%CV	Pettitt	Changing period
	(°C)	(°C)	(°C)				(°C/year)
Summer	28.0	30.2	29.0	0.5	0.517	<0.0001	0.040	1.64	<0.0001	2035
Rainy	28.0	28.8	28.5	0.2	0.660	<0.0001	0.020	0.71	<0.0001	2037
Winter	24.6	28.4	26.6	0.9	0.449	<0.0001	0.042	3.48	0.000	2047
Annual	27.3	28.9	28.0	0.5	0.614	<0.0001	0.033	1.62	<0.0001	2540
Temperature—RCP85
Season	Min	Max	CSM.	S.D.	Kendall’s tau	p-value	Sen’s slope	%CV	Pettitt	Changing period
	(°C)	(°C)	(°C)				(°C/year)
Summer	27.9	30.2	29.2	0.5	0.682	<0.0001	0.030	1.83	<0.0001	2043
Rainy	28.0	28.9	28.5	0.3	0.695	<0.0001	0.015	0.88	<0.0001	2041
Winter	25.5	28.3	26.8	0.7	0.528	<0.0001	0.054	2.50	<0.0001	2040
Annual	27.3	29.0	28.2	0.4	0.719	<0.0001	0.033	1.46	<0.0001	2040
Rainfall—RCP45
Season	Min	Max	CSM.	S.D.	Kendall’s tau	p-value	Sen’s slope	%CV	Pettitt	Changing period
	(mm)	(mm)	(mm)				(mm/year)
Summer	57.1	426.7	199.2	75.0	0.063	0.578	0.807	37.6	0.692	no
Rainy	676.0	1138.6	897.2	108.4	0.414	0.000	5.721	12.1	0.005	2034
Winter	63.9	364.5	174.7	76.6	−0.104	0.358	−1.259	43.9	0.979	no
Annual	935.9	1654.3	1271.1	159.1	0.263	0.019	5.334	12.5	0.105	no
Rainfall—RCP85
Season	Min	Max	CSM.	S.D.	Kendall’s tau	p-value	Sen’s slope	%CV	Pettitt	Changing period
	(mm)	(mm)	(mm)				(mm/year)
Summer	83.8	336.3	189.0	74.7	0.158	0.161	1.338	39.5	0.960	no
Rainy	682.3	1041.2	860.4	89.7	0.096	0.397	1.283	10.4	0.443	no
Winter	29.5	330.8	165.2	79.9	0.074	0.514	0.715	48.3	0.182	no
Annual	921.3	1485.4	1214.7	156.9	0.163	0.147	4.108	12.9	0.583	no
PET—RCP45
Season	Min	Max	CSM.	S.D.	Kendall’s tau	p-value	Sen’s slope	%CV	Pettitt	Changing period
	(mm)	(mm)	(mm)				(mm/year)
Summer	765.8	1024.9	873.7	62.8	0.495	<0.0001	3.952	7.2	<0.0001	2035
Rainy	713.8	810.9	768.0	23.1	0.646	<0.0001	1.690	3.0	<0.0001	2037
Winter	407.1	669.6	534.5	69.5	0.420	0.000	3.848	13.0	<0.0001	2047
Annual	1959.1	2464.2	2176.0	134.6	0.628	<0.0001	9.927	6.2	<0.0001	2047
PET—RCP85
Season	Min	Max	CSM.	S.D.	Kendall’s tau	p-value	Sen’s slope	%CV	Pettitt	Changing period
	(mm)	(mm)	(mm)				(°C/year)
Summer	714.9	1033.7	889.4	68.0	0.625	<0.0001	4.628	7.6	<0.0001	2039
Rainy	708.6	826.1	776.4	29.6	0.690	<0.0001	2.334	3.8	<0.0001	2041
Winter	461.0	679.6	552.3	52.1	0.503	<0.0001	3.143	9.4	<0.0001	2540
Annual	1938.0	2496.9	2221.8	135.1	0.722	<0.0001	10.361	6.1	<0.0001	2540

). Annual %CV ranged from 1.03 to 2.69% (mean = 1.62%) for RCP4.5 and 1.01 to 2.50% (mean = 1.46%) for RCP8.5. Higher %CV values were found in high-altitude or inland areas with large elevation differences, where rapid heating and cooling lead to greater variability [76]. In contrast, lower %CVs occurred in low-latitude, flat, or coastal areas, where land–sea breezes and humidity help moderate temperature changes [77,78]. Furthermore, the seasonal %CV values for summer, rainy, and winter were 1.64, 0.71, and 3.48 for RCP4.5, and 1.83, 0.88, and 2.50 for RCP8.5, respectively. The highest values occurred during winter, especially in the northern and northeastern regions, due to the unstable northeast monsoon increasing temperature variability [79]. In contrast, the lowest %CV values were recorded in the rainy season, indicating relatively stable temperatures. This stability results from low-level clouds reducing daytime shortwave radiation by reflecting sunlight, and limiting nighttime longwave radiation loss, thus maintaining consistent temperatures. The effect is particularly evident in the northeastern region, which is often influenced by tropical cyclones [80,81,82,83].

Data homogeneity and abrupt change analyses using the two-tailed Pettitt test at the 95% significance level revealed significant change points (p < 0.0001) at all 104 stations, indicating heterogeneous temperature data. Figure 4 illustrates these changes, with green and red lines representing the means before (µ₁) and after (µ₂) the shift. The years with detected changes were 2035 (summer), 2037 (rainy), 2047 (winter), and 2540 (annual) for RCP4.5, and 2043, 2041, 2040, and 2040 for RCP8.5, respectively. This divergence may stem from aerosol forcing and cloud feedback differences between RCP4.5 and RCP8.5. RCP4.5 assumes earlier aerosol reduction, enhancing shortwave radiation and leading to faster early-century warming, whereas RCP8.5 sustains higher aerosol levels delaying temperature amplification until the late-century period [84,85,86].

Trend analyses using the Mann–Kendall test and Sen’s slope showed statistically significant warming trends across all seasons and stations. Kendall’s Tau values for RCP4.5 were 0.517 (summer), 0.660 (rainy), 0.449 (winter), and 0.614 (annual); for RCP8.5: 0.682, 0.695, 0.528, and 0.719 (all p < 0.0001). Spatial patterns (Figure 3) revealed stronger warming in higher latitudes, especially in inland areas, while coastal regions exhibited milder trends. The estimated warming rates (Figure 4; Table 10) were ~+0.040 °C/year (summer), +0.020 °C/year (rainy), +0.042 °C/year (winter), and +0.033 °C/year (annual) for RCP4.5, and +0.030 °C/year, +0.015 °C/year, +0.054 °C/year, and +0.033 °C/year for RCP8.5. Interestingly, RCP4.5 showed steeper mid-century warming, possibly due to nonlinear climate responses, regional feedbacks (e.g., clouds, humidity), and varying aerosol assumptions, consistent with Oleson et al. (2015) [84], Grandey et al. (2016) [85], and Copernicus Climate Change Service (n.d.) [87] also noted that RCP8.5 warming intensifies later in the century, while RCP4.5 increases earlier.

Phumkokrux and Trivej (2024) [28] observed similar spatial temperature patterns in Thailand during 1987–2021 compared to projections for 2022–2060 under both scenarios. However, projected mid-century temperatures were significantly higher, indicating a warmer future for the country [28].

4.3. Rainfall Prediction of the Mid-21st Century Period and Its Characteristics

The RCP4.5 and RCP8.5 scenarios from the CSIRO-Mk3 dataset (Table 9) were downscaled using the Delta method and observational data from 104 meteorological stations via SD-GCM Version 2 [67]. Projected total monthly rainfall for 2022–2060 showed annual values ranging from 826.2 to 4768.5 mm (RCP4.5) and 759.4 to 4302.0 mm (RCP8.5). Rainfall under RCP8.5 was lower at 95 stations (–466.5 to –4.9 mm), while only 9 stations—mainly in the southernmost and northeastern regions—showed slightly higher values (13.9 to 44.2 mm). The highest rainfall occurred during the rainy season (May–September), typically peaking in June, ranging from 676.0 to 1138.6 mm (RCP4.5) and 682.3 to 1041.2 mm (RCP8.5). The lowest rainfall was recorded from mid-winter to mid-summer. Throughout the year, rainfall trends under both scenarios were similar (Figure 2), with spatial patterns also largely consistent (Figure 5). Moderate rainfall was common in inland areas, while high rainfall prevailed along the Andaman coast and eastern seaboard—particularly in Chanthaburi and Trat—due to orographic effects from the southwest monsoon [88,89]. The northeastern border near Laos, especially around Nakhon Phanom, also received high rainfall linked to low-pressure systems and tropical cyclones [90]. Compared with the early 21st century (1987–2021), as reported by Phumkokrux and Trivej (2024), the spatial distribution remains similar [28]. However, overall rainfall during summer, winter, and annual periods is lower under both RCPs, while rainy season values remain relatively unchanged. Cumulative seasonal means (CSMs) in Figure 6 were about 1271.1 mm (RCP4.5) and 1214.7 mm (RCP8.5). Seasonal CSMs for RCP4.5 were 199.2 mm (summer), 897.2 mm (rainy), and 174.7 mm (winter); for RCP8.5, 189.0 mm, 860.4 mm, and 165.2 mm, respectively.

The coefficient of variation (%CV) was analyzed to assess rainfall variability, with results shown in Figure 5 and Table 10. Annual %CV patterns for RCP4.5 and RCP8.5 were similar, ranging from 12.58% to 29.09% (mean 12.5%) and 12.41% to 31.21% (mean 12.9%), respectively. High %CV values were frequent in the eastern and southern regions near the Gulf of Thailand and Andaman Sea, due to unstable southwest and northeast monsoons causing inter-annual rainfall variability [91], especially during winter. Elevated %CV was also found in northeastern border areas with Laos, influenced by unstable low-pressure I systems and tropical cyclones [92]. Seasonal %CVs were 37.6%, 12.1%, and 43.9% (summer, rainy, winter) for RCP4.5, and 39.5%, 10.4%, and 48.3% for RCP8.5, indicating the highest variability in winter and the lowest during the rainy season. Compared with Muhammad Iskandar et al. (2025) [26] Amnuaylojaroen (2021) [27], and Phumkokrux and Trivej (2024) [28] regarding past results, winter remains the most variable season, and %CV values have increased, suggesting greater future rainfall variability and uncertainty. Projected spatial %CV distributions differ from past patterns, with higher %CV along the northeast edges extending southward and notably higher values in the southern region.

Pettitt’s test (95% confidence) showed p-values > 0.05 for annual rainfall at 104 (RCP4.5) and 102 (RCP8.5) stations, indicating homogeneous data. Only the rainy season under RCP4.5 showed a significant abrupt change starting around 2034; no abrupt changes were found in other seasons or under RCP8.5. Compared to 1987–2021 [28], projected abrupt change periods are consistent with historical patterns, implying gradual rainfall changes without major fluctuations.

Mann–Kendall and Sen’s slope analyses indicated statistically significant increasing trends at 42 stations (RCP4.5) and 18 stations (RCP8.5). Kendall’s tau values for RCP4.5 were 0.063 (summer), 0.414 (rainy), –0.104 (winter), and 0.263 (annual); for RCP8.5, the values were 0.158, 0.096, 0.074, and 0.163, respectively. Only the rainy season and annual trends under RCP4.5 were significant at 95% confidence (p < 0.05).

Figure 5 shows the spatial distribution of annual Sen’s slope and Kendall’s tau for rainfall in Thailand. Under RCP4.5, increasing trends predominated at higher longitudes, especially the eastern and southern regions, with a statistically significant Kendall’s tau of 0.263 (p = 0.019). Under RCP8.5, Kendall’s tau was positive (0.163) but not significant (p = 0.147), notably in the southern region. This suggests weak rainfall increases with high variability under extreme scenarios like RCP8.5, likely due to complex atmospheric circulation changes affecting tropical monsoon systems [86,93,94]. Such variability causes irregular seasonal rainfall shifts, making long-term trends hard to detect, consistent with findings in strong monsoon regions [95]. Figure 6 shows increasing rates (mm/year) under RCP4.5 as +0.807 (summer), +5.721 (rainy), –1.259 (winter), and +5.334 (annual); and under RCP8.5 as +1.338, +1.283, +0.715, and +4.715, respectively. Rates under RCP8.5 were generally lower than those under RCP4.5, reflecting nonlinear, region-specific climate responses influenced by monsoon dynamics, ocean–atmosphere interactions, and topography [86,93,96]. Thus, rainfall increases under RCP8.5 are not necessarily greater than under RCP4.5.

Comparison with Phumkokrux and Trivej (2024) [28] shows similar increasing trends since the early 21st century, with slightly higher rates under both RCPs.

4.4. Potential Evapotranspiration (PET) Prediction of the Mid-21st Century Period and Its Characteristics

Potential evapotranspiration (PET) was calculated using Equation (14) based on temperature and rainfall from the CSIRO-Mk3 dataset’s RCP4.5 and RCP8.5 scenarios. Data for all 104 meteorological stations in Thailand were generated by the SD-GCM Version 2 tool with EQM downscaling [67]. Annual PET ranged from 1034.9 to 2783.5 mm (mean 2176.0 mm) under RCP4.5, and 1049.0 to 2848.7 mm (mean 2221.8 mm) under RCP8.5. Monthly PET, shown in Figure 2, rose from January, peaked in April, then declined and stabilized during the rainy season, dropping sharply in winter, mirroring temperature trends. PET values were consistently slightly higher under RCP8.5. Spatial PET patterns (Figure 7) were similar between scenarios, though drier areas were more extensive under RCP8.5, especially in the northeast and central regions near Bangkok, with lowest PET in the north, which is consistent with the temperature distribution. Cumulative seasonal means (CSMs) of PET (Figure 8) were 873.7 mm (summer), 768.0 mm (rainy), 534.5 mm (winter), and 2176.0 mm (annual) for RCP4.5, and 889.4, 776.4, 552.3, and 2221.8 mm, respectively, for RCP8.5. Pettitt’s test (95% significance) showed statistically significant abrupt change points (p < 0.0001) at all stations for annual PET under both scenarios. Change points occurred around 2035 (summer), 2037 (rainy), 2047 (winter), and 2047 (annual) for RCP4.5, and 2039, 2041, 2040, and 2040, respectively, for RCP8.5.

The coefficient of variation (%CV) for annual PET ranged from 3.8% to 10.0%, averaging 6.2% (RCP4.5) and 6.1% (RCP8.5), indicating low variability. Higher %CV occurred in central and northeast regions, and lower values in northernmost and southern areas (Figure 7, Table 10). Seasonal %CV showed the highest variability in winter (13.0% RCP4.5; 9.4% RCP8.5), followed by summer and rainy seasons. The rainy season had lowest variability due to higher humidity reducing evaporation [97]. Summer’s higher PET resulted from elevated temperatures and strong winds, while winter’s dry northeast monsoon winds still facilitated evaporation despite lower temperatures [98].

Trend analysis of PET using the Mann–Kendall test (Kendall’s tau) and Sen’s slope estimator showed increasing trends at all meteorological stations and seasons. Kendall’s tau values for RCP4.5 were 0.495 (summer), 0.646 (rainy), 0.420 (winter), and 0.628 (annual), and for RCP8.5, the values were 0.625, 0.690, 0.503, and 0.722, respectively; all values were significant at 95% confidence (p < 0.0001). Annual Sen’s slope and spatial Kendall’s tau for PET trends are shown in Figure 7. PET trends spatially resemble temperature trends, but RCP8.5 shows more extensive drier areas, especially along Thailand’s western border with Laos and southern regions near Malaysia. Wetter zones appear at higher latitudes, with lower temperatures limiting evaporation, and in southern Thailand near the Andaman Sea and Gulf of Thailand, where higher humidity and rainfall reduce PET [99]. Rates of PET increases (Figure 8, Table 10) are approximately +3.95 mm/year (summer), +1.69 (rainy), +3.85 (winter), and +9.93 (annual) for RCP4.5, and +4.63, +2.33, +3.14, and +10.36 mm/year for RCP8.5. These suggest PET is influenced more by temperature changes than rainfall.

Phumkokrux and Trivej (2024) [28] noted that early 21st-century PET patterns (1987–2021) resemble mid-21st-century projections (2022–2060) under both scenarios, but with higher PET values and trends expected in the future. Increased PET raises drought risks by amplifying atmospheric evaporative demand and soil moisture deficits, even without reduced rainfall. Drought indices like SPEI incorporate PET to capture this effect [100,101].

4.5. The Predicted New Thornthwaite Climate Classification of the Mid-21st Century Period

The predicted New Thornthwaite Climate Classification, based on Feddema (2005) [62] and the CSIRO-Mk3 dataset (RCP4.5 and RCP8.5), focused on temperature and precipitation to evaluate (1) Moisture Index, (2) Thermal Index, (3) climate variability level, and (4) variability-caused modifiers.

Climate types were determined using the 1948 Thornthwaite Moisture Index (TMI) [62] with temperature and rainfall data from CSIRO-Mk3 for the mid-21st century. TMI, calculated via Equations (13)–(16) and classified by Table 4, assessed humidity levels from combined temperature and rainfall effects. Figure 9 and

Table 11. New Thornthwaite Climate Classification for RCP4.5 and RCP8.5 results description.

Climate Types Based on the 1948 TMI.
Climate Type	RCP45		RCP85		Changed from RCP45 to RCP85
	(sq.km.)	(%)	(sq.km.)	(%)	(sq.km.)	(%)
Arid	0.0	0.0	0.0	0.0	0.0	0.0
Semi-Arid	231,165.1	44.8	263,080.9	51.0	31,915.7	6.2
Dry	247,154.1	47.9	227,730.0	44.1	−19,424.1	−3.8
Saturated	35,262.4	6.8	23,770.8	4.6	−11,491.6	−2.2
Wet	2747.6	0.5	1747.6	0.3	−1000.0	−0.2
Moist	0.0	0.0	0.0	0.0	0.0	0.0
Total	516,329.2	100.0	516,329.2	100.0	-	-
Thermal Index Based on the 1948 TMI
Thermal Index	RCP45		RCP85		Changed from RCP45 to RCP85
	(sq.km.)	(%)	(sq.km.)	(%)	(sq.km.)	(%)
Frigid	0.0	0.0	0.0	0.0	0.0	0.0
Cold	0.0	0.0	0.0	0.0	0.0	0.0
Cool	0.0	0.0	0.0	0.0	0.0	0.0
Warm	0.0	0.0	0.0	0.0	0.0	0.0
Hot	0.0	0.0	0.0	0.0	0.0	0.0
Torrid	516,329.2	100.0	516,329.2	100.0	0.0	0.0
Total	516,329.2	100.0	516,329.2	100.0	-	-
Climate Variability Level Based on Annual TMI Range.
Variability Level	RCP45		RCP85		Changed from RCP45 to RCP85
	(sq.km.)	(%)	(sq.km.)	(%)	(sq.km.)	(%)
Low	0.0	0.0	0.0	0.0	0.0	0.0
Medium	0.0	0.0	0.0	0.0	0.0	0.0
High	165,551.0	32.1	137,618.6	26.7	−27,932.4	−5.4
Extreme	350,778.2	67.9	378,710.6	73.3	27,932.4	5.4
Total	516,329.2	100.0	516,329.2	100.0	-	-
Climate Variability Cause Based on Annual Pr/PET Range.
Variability Cause	RCP45		RCP85		Changed from RCP45 to RCP85
	(sq.km.)	(%)	(sq.km.)	(%)	(sq.km.)	(%)
Rainfall	0.0	0.0	0.0	0.0	0.0	0.0
Combination	210,856.8	40.8	222,149.4	43.0	11,292.6	2.2
Temperature	305,472.4	59.2	294,179.8	57.0	−11,292.6	−2.2
Total	516,329.2	100.0	516,329.2	100.0	-	-

show predicted climate types for RCP4.5 (top left) and RCP8.5 (bottom left). Under RCP4.5, Dry (yellow) and Semi-Arid (orange) dominate, covering ~47.9% (247,154.1 km²) and 44.8% (231,165.1 km²) of Thailand, respectively. Saturated (green) occurs in ~6.8% (35,262.4 km²), mainly in northeastern border with Laos, southern eastern region near Chanthaburi mountains, and western southern coast along the Andaman Sea. Small Wet (blue) patches (~0.5%, 2747.6 km²) appear in high rainfall areas. RCP8.5 shows a similar pattern, but Semi-Arid expands to 51.0% (263,080.9 km²), while Dry, Saturated, and Wet types decrease to 44.1% (227,730.0 km²), 4.6% (23,770.8 km²), and 0.5% (1747.6 km²), respectively.

The Predicted Thermal Index, based on annual PET following Thornthwaite’s 1948 method [62] via Equation (17), is classified into six classes (Table 11) and mapped in Figure 9 (top right for RCP4.5, bottom right for RCP8.5). Average annual PET exceeds 1500 mm, approx. 2176.0 mm for RCP4.5 and 2221.8 mm for RCP8.5. Consequently, the entire country falls into the torrid category under both scenarios, reflecting persistently high temperatures [86].

The predicted climate variability level was assessed using the Thermal Moisture Index (TMI) range, calculated as the difference between annual maximum and minimum TMI at each station (Equation (18)) and classified into four categories (Table 11). Spatial distributions are shown in Figure 9 for RCP4.5 (first row, third column) and RCP8.5 (second row, third column). Most of Thailand falls into the extreme variability category, covering about 67.9% (350,778.2 km²) under RCP4.5 and 73.3% (378,710.6 km²) under RCP8.5. The central region shows high variability, covering 32.1% (165,551.0 km²) for RCP4.5 and 26.7% (137,618.6 km²) for RCP8.5. Without emission controls, climate variability is expected to increase, reaching extreme levels across many areas soon [86].

The causes of predicted climate variability were assessed via the relationship between annual rainfall and PET ranges (Equation (19), Table 11), with spatial distributions shown in Figure 9 for RCP4.5 (first row, fourth column) and RCP8.5 (second row, fourth column). Results indicate that temperature primarily influences climate variability in most northeastern, central, eastern, and southern areas, covering 59.2% (305,472.4 km²) under RCP4.5 and 57.0% (294,179.8 km²) under RCP8.5. Conversely, northern and some other regions are affected by combined rainfall and temperature influences, representing 40.8% (210,856.8 km²) for RCP4.5 and 43.0% (222,149.4 km²) for RCP8.5. Thus, temperature remains the main driver of climate variability, though under RCP8.5, rainfall’s role increases [93,94].

The predicted mid-21st century New Thornthwaite Climate Classification (Figure 9) differs from early 21st century observations (1987–2021) reported by Phumkokrux & Trivej (2024) [28]. Dry and semi-arid types dominate, with semi-arid type expanding under RCP8.5, replacing wet areas. Moist zones are projected to disappear under both scenarios. The Thermal Index remains torrid, consistent with earlier conditions. Predicted climate variability shows an expansion of high-level variability zones, replacing previously widespread extreme-level zones. Finally, the combined rainfall and temperature influence is projected to become the main driver of variability, contrasting past dominance by temperature alone.

The projected patterns identified in this study are broadly consistent with regional-scale projections across Southeast Asia and global assessments reported by the IPCC AR6. Similar warming magnitudes and spatial heterogeneity have been observed in neighboring countries such as Vietnam, Cambodia, and Malaysia, where enhanced surface temperatures and declining wet-season rainfall are linked to weakened monsoon circulation and rising aerosol concentrations. Studies using CMIP6 ensembles (e.g., Tangang et al., 2023 [95]; Li et al., 2024 [19]) also highlight intensified dry season warming and increased inter-annual variability in the Indochina Peninsula, supporting the present results. The slightly faster mid-century warming under RCP4.5 compared to RCP8.5 aligns with findings from Oleson et al. (2015) [84] and Grandey et al. (2016) [85], which attribute early century warming to aerosol reduction and radiative forcing effects. These comparisons confirm that Thailand’s projected climatic shifts reflect broader regional tendencies toward aridity and hydroclimatic instability across Southeast Asia.

5. Conclusions

This study projects Thailand’s mid-21st century climate using CSIRO-Mk3 under RCP4.5 and RCP8.5 scenarios. Results reveal continuous warming (+0.033 °C/year), stable rainfall, and rising PET trends (~+10 mm/year). By 2060, Semi-arid and Dry climates dominate nationwide, while Moist zones vanish completely. These shifts signify intensifying aridity and drought risks, emphasizing proactive adaptation strategies for Thailand’s agricultural and water sectors.

Annual rainfall under both scenarios displayed similar spatial patterns with no abrupt changes and minor trend differences. RCP8.5 generally projected lower rainfall, especially along the western southern coast (the Andaman Sea) and the southeastern region (Chanthaburi and Trat), influenced by the southwest monsoon. High rainfall was projected in northeastern areas, particularly around Nakhon Phanom, likely due to tropical storms. Central Thailand exhibited low rainfall due to rain-shadow effects and distance from the coast. These patterns align with early 21st-century observations [38], but mid-century simulations showed higher temperature and lower rainfall trends, indicating worsening climate impacts.

PET patterns under both RCPs resembled those in the early 21st century, with highest values in central Thailand decreasing northward. Abrupt increases and upward trends, mainly driven by rising temperatures, were more pronounced under RCP8.5, suggesting increased evaporative demand and drought risk.

The New Thornthwaite Climate Classification predicted dominance of Semi-arid and Dry climates with torrid Thermal Index in central Thailand, transitioning to wetter types in limited coastal and northeastern zones influenced by monsoon and tropical cyclones. Compared to the early 21st century, Moist types disappeared, indicating increased aridity. Most of Thailand showed high to extreme climate variability driven mainly by temperature rises. Both temperature and rainfall impact on variability are expected to grow, partly due to a weakening southwest monsoon [102,103]. ENSO events may further intensify these changes [104,105]. These evolving patterns highlight the need for ongoing climate monitoring in Thailand.

The Thornthwaite Classification, relying on temperature, precipitation, and PET with a simplified approach, offers practical utility for local authorities to assess and respond to climate change. This study focused on these core variables to characterize climate change and drought, though future research should incorporate additional atmospheric factors and policy considerations influencing Thailand’s climate dynamics.

This study integrates the New Thornthwaite Climate Classification with statistically downscaled CMIP5 projections for Thailand, providing insights for agricultural zoning, water management, and climate adaptation. Limitations include reliance on a single GCM (CSIRO-Mk3) and a lack of ensemble-based uncertainty evaluation. Future research should extend to CMIP6 datasets and include multiple models to better capture uncertainty and projection robustness.