Future Changes in Precipitation Extremes over South Korea Based on Observations and CMIP6 SSP Scenarios

Abstract

This research assesses four Shared Socioeconomic Pathway (SSP) scenarios (SSP1-2.6, SSP2-4.5, SSP3-7.0, SSP5-8.5) concerning precipitation quantiles across Korea, utilizing the CMIP6 multi-model ensemble comprising 23 General Circulation Models alongside observational data to project future changes. Precipitation quantiles, derived from regional frequency analysis conducted at 615 sites, are calculated as annual averages for the period from 2015 to 2024. Each SSP scenario is evaluated for its spatial distribution through the application of observational data and chi-square tests, with the results indicating that the SSP3-7.0 ensemble most accurately reflects the current quantile estimates derived from observational data. Furthermore, interannual precipitation quantiles are projected to extend to the year 2100 to discern long-term trends within each reproducible period. It is anticipated that precipitation associated with the 100-year reproducible period will increase by over 20% in most regions across the nation by the century’s end, with this increase becoming more pronounced in accordance with the severity of the pathway. These findings underscore significant increases in extreme rainfall events under high-emission scenarios and highlight the critical need for the integration of enhanced flood mitigation, water resource management, and climate adaptation strategies within Korea’s planning framework.

1. Introduction

Global precipitation patterns have experienced significant changes as a result of climate change, particularly in areas that are susceptible to extreme precipitation events, which have been notably affected [1,2]. South Korea is particularly susceptible to environmental risks because of its distinctive topographical and climatic attributes. The precipitation patterns in Korea are predominantly influenced by the East Asian monsoon, which accounts for approximately 60–70% of the annual rainfall during the summer months (June to August). This phenomenon considerably increases the likelihood of extreme precipitation events [3,4].

Recent calamities in South Korea have underscored the region’s susceptibility to environmental adversities. Notably, severe rainfall occurrences during the summers of 2020 and 2024 precipitated extensive flooding, leading to considerable damage to infrastructure, substantial agricultural losses, and unfortunate fatalities [5,6,7,8,9,10]. Between 8 and 10 August 2022, the metropolitan area experienced unprecedented rainfall, with rates reaching 141.5 mm per hour and cumulative totals exceeding 500 mm, resulting in significant flooding in Seoul. This event represents one of the highest hourly precipitation levels documented in the city since the 1920s [11]. The Seoul Metropolitan Government has declared its intention to revise the flood prevention standards for the city, increasing the target from the current threshold of 95 mm per hour, which corresponds to a 30-year return period, to a minimum threshold of 100 mm per hour, reflecting a 50-year return period, following the flooding event of 2022. Furthermore, as part of a comprehensive mid- to long-term strategy, the government plans to enhance infrastructure aimed at mitigating damage, including the fortification of sewer systems and the installation of rainwater pumping stations, in anticipation of potential heavy rainfall events reaching up to 110 mm per hour, indicative of a 100-year return period [12].

As noted by Maidment [13], “hydrologic frequency analysis is primarily used to estimate the magnitude of events that have a specified probability of being exceeded in any year”. This information forms the basis for the design and risk analysis of water control structures such as dams, spillways, culverts, storm sewers, and channels. In South Korea, the likelihood of precipitation events occurring with frequencies of 10, 30, 50, and 100 years is assessed and incorporated into the design considerations based on the dimensions and significance of hydro-infra structures [14,15]. Nonetheless, alterations in precipitation patterns resulting from climate change pose significant challenges to conventional quantile estimation techniques, which are predicated on the assumption of hydroclimatic stationarity. Furthermore, numerous studies have suggested that the frequency and severity of extreme weather events may increase due to the climate change [15,16,17,18].

The Shared Socioeconomic Pathways (SSPs) framework, developed for the IPCC Sixth Assessment Report, outlines scenarios based on diverse socioeconomic trajectories and greenhouse gas emission levels [19]. Despite their widespread use in projecting climate change at global and regional scales, significant uncertainties remain within individual models and scenarios. Employing a multi-model ensemble approach is widely acknowledged as an effective strategy to mitigate these uncertainties [15,17,20,21]. Although research into climate change impacts in Korea has expanded, significant gaps remain in understanding future extreme precipitation patterns. Most studies rely on a limited subset of CMIP6 models, potentially underestimating the full range of prediction uncertainties. In addition, comprehensive assessments of SSP scenarios are scarce, hindering the identification of socioeconomic pathways to prioritize under current conditions. Furthermore, the spatial distribution of precipitation quantiles has been largely overlooked, even though it plays a critical role in informing proactive climate change preparations and regional adaptation planning.

The primary aim of this research is to identify which Shared Socioeconomic Pathway (SSP) scenario most accurately reflects the existing patterns of extreme precipitation in South Korea, as well as to assess the implications for future climate adaptation strategies. The central scientific hypothesis posits that the current observed precipitation quantiles in Korea are already consistent with one of the more severe emission scenarios, indicating that the region’s climate system may be shifting towards a higher warming trajectory than previously expected. This study intends to evaluate this hypothesis by systematically comparing observed precipitation quantiles from 2015 to 2024 with projections derived from 23 models within the Coupled Model Intercomparison Project Phase 6 (CMIP6) across four SSP scenarios, thereby identifying the scenario that most effectively represents the current characteristics of extreme precipitation. Furthermore, this research aims to quantify anticipated changes in precipitation extremes and their spatial distribution through the year 2100, thereby providing critical information for the development of evidence-based climate adaptation strategies in South Korea.

2. Data and Methods

2.1. Study Area

Figure 1 illustrates the geographical context of South Korea, situated on the eastern edge of the Asian continent between 33° and 39° north latitude and 124° and 130° east longitude, alongside the spatial distribution of 615 precipitation observation sites with 26 hydrological homogeneous regions utilized in this analysis [22]. The Korean Peninsula is predominantly mountainous, with approximately 70% of its terrain characterized by elevated topography, while being surrounded by water bodies on three sides. This unique geographical configuration creates complex interactions between continental and maritime air masses, leading to distinctive precipitation regimes across relatively short distances.

The country’s climate is classified as humid continental (Dwa) according to the Köppen-Geiger classification system, characterized by four well-defined seasons with significant temperature variations throughout the year [23]. Mean annual temperatures range from 11 °C in northern and mountainous regions to 14 °C along southern coastal areas, reflecting the influence of latitudinal gradients and elevation differences. This temperature regime interacts with moisture-laden air masses to create distinct precipitation patterns that vary considerably across the peninsula’s complex topography.

Annual precipitation across South Korea exhibits substantial spatial and seasonal variability. Regionally, annual totals range from approximately 1000 mm in the northwestern inland areas to over 1800 mm in the southern and eastern coastal and mountainous regions, reflecting the influence of complex topography and prevailing atmospheric circulation patterns. Temporally, precipitation is strongly concentrated during the summer monsoon season (June–August), with 60–70% of the annual total occurring within this three-month period, primarily driven by the East Asian monsoon system [23,24]. This seasonal concentration, combined with spatially uneven distribution, often results in localized flooding, challenges in water resource allocation, and significant implications for infrastructure resilience and climate adaptation planning.

2.2. Dataset and Sources

This research employed two principal data sources: observational precipitation measurements specific to South Korea and climate model projections derived from the Coupled Model Intercomparison Project Phase 6 (CMIP6). Both data sources were used at the daily temporal scale of precipitation.

2.2.1. Observational Data

Comprehensive precipitation datasets were obtained from multiple authoritative sources in South Korea, including the Korea Meteorological Administration (KMA), the Korea Water Resources Corporation (K-water), and the Ministry of Environment (MOE). The observational dataset includes precipitation measurements collected from 615 observation sites located throughout South Korea, covering from 1961 to 2024. All observational data were accessed through the Water Resources Management Information System (WAMIS), which maintains standardized records of hydrometeorological variables across South Korea [24]. These datasets provide continuous, quality-controlled precipitation measurements that serve as the observational foundation for evaluating climate model performance and conducting bias correction procedures.

2.2.2. Climate Model Data and Scenarios

Climate projection data were acquired from the Earth System Grid Federation (ESGF) platform, which provides access to GCM outputs produced by research institutions worldwide. The dataset comprises historical simulations (1985–2014) and future projections (2015–2100) of daily precipitation across four SSPs combined with Representative Concentration Pathways (RCPs): SSP1-2.6 (SSP126), SSP2-4.5 (SSP245), SSP3-7.0 (SSP370), and SSP5-8.5 (SSP585) [25]. These scenarios represent a comprehensive range of potential future pathways, from ambitious mitigation strategies (SSP126) to fossil-fuel intensive development (SSP585), allowing for robust assessment of precipitation response across varying degrees of radiative forcing. For each scenario, multiple GCM ensemble members were selected to capture model uncertainty and provide more reliable projections for South Korea’s unique geographical and climatological context.

2.2.3. Data Processing and Bias Correction

To address the spatial resolution limitations of GCMs (typically 100–500 km), which introduce significant uncertainties when applied to South Korea’s complex topography, we implemented a two-stage data processing approach. First, spatial disaggregation was performed using inverse distance weighting (IDW) interpolation to standardize the spatial resolution across different models. The IDW interpolation technique estimates values at locations where measurements are absent by calculating a weighted average of known data points. In this method, the weights assigned to each data point are inversely related to the distance from the interpolation point to the respective data points [26]. Jo et al. [27] compared various spatial statistical interpolation techniques to generate high-resolution gridded climate data across South Korea and concluded that IDW was the most suitable method for representing precipitation. The mathematical representation of IDW is articulated as follows:

$$Z\left(x_0\right)={\displaystyle \frac{{\sum }_{i=1}^n{w}_i·Z\left(x_i\right)}{{\sum }_{i=1}^n{w}_i}}$$

(1)

$$w_i={\displaystyle \frac{1}{d_i^2}}$$

(2)

where $Z\left(x_0\right)$ is the estimated value at the interpolation point, and $Z\left(x_i\right)$ is the known value at data point $i$, and $n$ is the number of known data points used in the interpolation. The weight ($w_i$) is calculated using Equation (2), depends on the Euclidean distance ($d_i^2$) between the interpolation point and data point $i$. This process converted the coarse-resolution GCM outputs into finer scales that are more suitable for regional analysis.

Following spatial disaggregation, non-parametric Quantile Mapping (QM) was applied to correct systematic biases in the climate model outputs. The QM methodology is grounded in the essential concept of adjusting the cumulative distribution function (CDF) of simulated data to align with that of observed data. This process ensures the retention of the relative frequency attributes inherent in the observational dataset, while simultaneously upholding the climate change signal present in the model forecasts [28]. The mathematical representation of this transformation is articulated as follows:

$$x_{corrected}=F_{obs}^{-1}\left(F_{mod}(x_{mod})\right)$$

(3)

where $F_{obs}$ and $F_{mod}$ represent the cumulative distribution functions (CDFs) of observations and model data, respectively, and $F_{obs}^{-1}$ denotes the inverse CDF of observations. Figure 2 presents a schematic diagram of the QM approach.

The non-parametric characteristics of this methodology set it apart from parametric bias correction techniques by eschewing assumptions regarding specific theoretical probability distributions. Empirical quantile mapping directly derives the transformation function from the empirical distributions of both observed and simulated data, thereby demonstrating considerable robustness in correcting precipitation variables that frequently display intricate, non-Gaussian distributional traits [30]. This approach is particularly advantageous for addressing extreme precipitation events, as it effectively preserves the tail behavior of the observed distribution while rectifying biases across the entire spectrum of precipitation intensities, ranging from dry days to extreme occurrences [31].

The empirical quantile mapping process is executed through the following steps: (1) ranking all values within both observed and modeled datasets to construct their respective empirical cumulative distribution functions (CDFs), (2) establishing a one-to-one correspondence between quantiles of equal probability in both distributions, and (3) applying this quantile-to-quantile transformation to rectify model biases [32]. In the context of precipitation data, particular attention is devoted to the treatment of dry days (defined as precipitation < 0.1 mm), wherein the frequency of wet days in the model is adjusted to align with that of the observations prior to implementing the quantile transformation on non-zero values.

This methodology effectively mitigates both location and scale biases inherent in climate model outputs while preserving the temporal sequencing and inter-variable relationships present in the original model simulations. Furthermore, it maintains the climate change signal by applying the bias correction transformation derived from the reference period to future projections, based on the assumption that model biases remain stationary over time [33]. All bias correction procedures were executed utilizing the ‘qmap’ package within the R programming environment (version 4.4.3) [34], resulting in regionally appropriate precipitation projections that retain the fundamental characteristics of both observed climatology and anticipated climate change signals.

2.3. Precipitation Quantile Estimation

Frequency analysis of hydrologic data entails the statistical estimation of the probability and size of historical severe events through the use of probability distributions. In this context, the observed precipitation quantiles (OBS) refer to the empirical quantiles derived from the annual maximum precipitation series at each monitoring station. Specifically, these quantiles represent the ordered annual maxima expressed in terms of non-exceedance probabilities. They serve as a reference point for the fitting and validation of theoretical models. This study applies to the Regional Frequency Analysis (RFA) method, which incorporates the Index Flood Method (IFM) based on the Generalized Extreme Value (GEV) distribution and the L-moments approach, as specified in the Standard Guidelines for Estimating Flood Quantiles in Korea [35]. The IFM posits that within hydrologically homogeneous regions, precipitation frequency distributions across different sites are identical, differing only by a site-specific scaling factor. This principle allows for the pooling of data from multiple stations within these regions, thereby enhancing the effective sample size and improving the statistical robustness of estimates for extreme precipitation events through the development of a regional growth curve based on normalized data.

The GEV distribution is a widely utilized probabilistic model in frequency analysis. Its probability density function (PDF) and CDF are given by the subsequent equations:

$$\mathrm{PDF}: f\left(x\right)={\displaystyle \frac{1}{\alpha }}{\left[1-{\displaystyle \frac{\beta \left(x-x_0\right)}{\alpha }}\right]}^{\left(1/\beta \right)-1}\exp \left[-{\left(1-{\displaystyle \frac{\beta \left(x-x_0\right)}{\alpha }}\right)}^{1/\beta }\right]$$

(4)

$$\mathrm{C}\mathrm{DF}: F\left(x\right)=\exp \left[-{\left(1-{\displaystyle \frac{\beta \left(x-x_0\right)}{\alpha }}\right)}^{1/\beta }\right]$$

(5)

where $x_0$, $\alpha$, and $\beta$ are the location, scale, and shape parameters, respectively.

2.4. Evaluation Metrics

To quantitatively assess the performance of precipitation quantile estimations across different climate scenarios, this study employed a suite of statistical evaluation metrics: bias, correlation coefficient (Corr), coefficient of determination (R²), and Root Mean Square Error (RMSE).

2.4.1. Bias

Bias represents the systematic difference between observed and simulated values. A positive bias signifies that the model tends to underestimate the observed data, whereas a negative bias indicates that the model tends to overestimate them. Bias is calculated as follows:

$$\mathrm{Bias}={\displaystyle \frac{{\sum }_{i=1}^n\left({\mathit{Sim}}_i-{\mathit{Obs}}_i\right)}{n}}$$

(6)

where ${\mathit{Obs}}_i$ and ${\mathit{Sim}}_i$ denote the observed and simulated values at time step $i$, respectively, and $n$ is the total number of observations. Bias serves as a crucial metric for evaluating the overall tendency of the model to overestimate or underestimate precipitation values.

2.4.2. Correlation Coefficient

The correlation coefficient quantifies the strength and direction of the linear relationship between observed and simulated values. Its value ranges from −1 to 1, where values closer to 1 indicate a strong positive correlation, and values closer to −1 indicate a strong negative correlation. The correlation coefficient is calculated as follows:

$$\mathrm{Corr}={\displaystyle \frac{{\sum }_{i=1}^n\left({\mathit{Obs}}_i-\stackrel{-}{\mathit{Obs}}\right)({\mathit{Sim}}_i-\stackrel{-}{\mathit{Sim}})}{\sqrt{{\sum }_{i=1}^n{\left({\mathit{Obs}}_i-\stackrel{-}{\mathit{Obs}}\right)}^2}\sqrt{\sum _{i=1}^n{({\mathit{Sim}}_i-\stackrel{-}{\mathit{Sim}})}^2}}}$$

(7)

where $\stackrel{-}{\mathit{Obs}}$ and $\stackrel{-}{\mathit{Sim}}$ are the mean values of the observed and simulated data, respectively. A higher correlation coefficient indicates a stronger agreement between the observed and simulated datasets.

2.4.3. Coefficient of Determination

The R² represents a dimensionless statistical parameter that quantifies the degree of correspondence between predicted values and the observed variability within the empirical data. It ranges from 0 to 1, with values closer to 1 indicating a stronger fit between the observed and simulated datasets. The R² is mathematically defined according to the following expression:

$${\mathrm{R}}^2={\displaystyle \frac{{\left[{\sum }_{i=1}^n\left({Obs}_i-\overline{Obs}\right)({Sim}_i-\overline{Sim})\right]}^2}{\left[{\sum }_{i=1}^n{\left({Obs}_i-\overline{Obs}\right)}^2\right][{\sum }_{i=1}^n{({Sim}_i-\overline{Sim})}^2]}}$$

(8)

where $\stackrel{-}{\mathit{Obs}}$ and $\stackrel{-}{\mathit{Sim}}$ represent the mean value of the observed and simulated data, respectively. The value of ${\mathrm{R}}^2$ ranges from 0 to 1, where a value closer to 1 indicates that a larger proportion of variance in the observed data is explained by the model, demonstrating a strong agreement between observed and simulated values. Conversely, a value closer to 0 suggests that the model explains little of the variability, indicating poor model performance.

2.4.4. Root Mean Square Error

The RMSE represents a widely used statistical metric that quantifies the magnitude of differences between predicted and observed values by measuring the square root of the average squared deviations. It provides an absolute measure of model performance in the same units as the original data, making it particularly valuable for assessing the accuracy of precipitation forecasts. The RMSE is mathematically defined according to the following expression:

$$\mathrm{RMSE}=\sqrt{{\displaystyle \frac{1}{n}}\sum _{i=1}^n{\left({Obs}_i-{Sim}_i\right)}^2}$$

(9)

where ${Obs}_i$ represents the observed value, ${Sim}_i$ denotes the model-simulated value at time step $i$, and $n$ is the total number of observations. The RMSE values range from 0 to infinity, where a value of 0 indicates perfect agreement between observed and simulated data, representing ideal model performance. Lower RMSE values indicate better model accuracy, as they reflect smaller average deviations between predictions and observations. Conversely, higher RMSE values suggest larger prediction errors and poorer model performance. The RMSE is particularly sensitive to outliers and extreme values, making it especially useful for evaluating model performance in capturing precipitation extremes, which are critical for hydrological and climate impact assessments.

2.5. Chi-Square Test

The Chi-square (${\mathsf{\chi }}^2$) test represents a statistical methodology employed to evaluate the congruence between empirical observations and theoretical expectations by quantifying the cumulative discrepancy between these distributions. It is commonly used to test the goodness-of-fit for categorical or binned data; however, it can also be applied to continuous data by categorizing the data into suitable intervals. The ${\mathsf{\chi }}^2$ statistic is calculated as follows:

$${\mathsf{\chi }}^2=\sum _{i=1}^n{\displaystyle \frac{{\left(O_i-E_i\right)}^2}{E_i}}$$

(10)

Herein, $O_i$ represents the empirically observed frequency or magnitude corresponding to category $i$, while $E_i$ denotes the theoretically expected (simulated) frequency or magnitude for the identical categorical designation, and n signifies the total number of discrete categories or intervals under analytical consideration.

A reduced ${\mathsf{\chi }}^2$ value signifies a greater alignment between observed phenomena and simulated predictions, indicating a higher efficacy of the model, whereas an increased value denotes a significant discrepancy between the theoretical framework and empirical data. Within the scope of our research, these ${\mathsf{\chi }}^2$ values offer essential insights into the degree to which the spatial distribution of probabilistic rainfall, as derived from climate change scenarios, corresponds with actual observed rainfall patterns in the study area. This statistic can be compared against critical thresholds established from the Chi-square probability distribution to ascertain whether the observed deviations surpass those expected from random variation. The Chi-square analytical framework ultimately facilitates the differentiation between random probabilistic variations and systematic deficiencies in the model, thereby allowing for a quantitative evaluation of the reliability of climate models in replicating the spatial heterogeneity of extreme precipitation events. This understanding is crucial for the formulation of localized adaptation strategies.

3. Results

3.1. Interannual Comparison of Precipitation Quantiles

Prior to analyzing future scenario projections, this study sought to evaluate the statistical congruence between observational data and climate change scenario outputs. Precipitation quantiles were estimated at 615 meteorological stations across South Korea using progressively expanding datasets, beginning with records up to 2015, then sequentially incorporating additional annual data through 2024. Figure 3 presents these interannual precipitation quantile distributions as box plots, facilitating comparative analysis between OBS and projections from four climate models representing distinct Shared Socioeconomic Pathways (SSP126, SSP245, SSP370, and SSP585). This analytical framework was applied across multiple return periods (T = 10, 30, 50, and 100 years) to comprehensively evaluate precipitation characteristics and potential early manifestations of climate change signals.

The 10-year return period precipitation quantile distributions (Figure 3a) exhibit remarkable temporal stability throughout the analysis period, with median precipitation values consistently ranging between 180 and 200 mm and interquartile ranges (IQRs) spanning approximately 160–220 mm. Maximum rainfall values, appearing as statistical outliers, reached approximately 320 mm. All SSP scenario projections demonstrated substantial concordance with observational data, with no discernible systematic bias evident. For the 30-year return period (Figure 3b), precipitation magnitudes predictably increased, with median values predominantly occupying the range of 250–270 mm and interquartile ranges extending from approximately 220 mm to 290 mm. The temporal consistency of these distributions persisted throughout the decade, with no statistically significant trends detected in central tendencies. The 50-year return period rainfall distributions (Figure 3c) were characterized by median values ranging between 260 and 280 mm and interquartile ranges spanning approximately 240–310 mm. A notable development within this return period was the emergence of more pronounced extreme values during 2022–2024, where maximum values approached approximately 570 mm—representing a subtle increase compared to earlier years. The 100-year return period distributions (Figure 3d) exhibited the highest precipitation values, with median values consistently ranging between 310 and 330 mm and interquartile ranges spanning approximately 280–360 mm. This return period demonstrated the most pronounced extreme outliers, with maximum values reaching 650–700 mm during 2022–2024, a discernible increase compared to earlier years when extreme values typically remained below 650 mm.

Several significant patterns emerge from this analysis. Rainfall distributions demonstrated remarkable temporal stability across all return periods, suggesting that potential climate change impacts on precipitation regimes in South Korea may manifest over longer timescales than captured within this decadal dataset. Despite representing substantially different emissions pathways, all four SSP scenarios produced remarkably similar rainfall distributions, indicating that regional precipitation patterns during this timeframe may be governed more by larger climatic oscillations than by scenario-specific forcing mechanisms. While central tendencies remained consistent, the magnitude of extreme outliers showed a slight increasing trend in the later years (2022–2024), particularly for higher return periods. This pattern potentially represents an emerging signal of climate change impacts on precipitation extremes. The consistency between observed data and model projections indicates satisfactory model skill in reproducing precipitation patterns during this period, though significant divergence in rainfall patterns under different emissions trajectories may become more pronounced in subsequent decades.

3.2. Statistical Evaluation Using Bias, Correlation, $R^2$, and RMSE

This study conducts a comprehensive temporal analysis of climate model performance metrics by utilizing precipitation quantiles to evaluate the alignment between empirical observations and projections derived from four different Shared Socioeconomic Pathways (SSPs) in South Korea for the period spanning 2015 to 2024. Subsequently, Figure 4a illustrates the temporal progression of bias in the rainfall projections based on precipitation quantiles across the four SSP scenarios. All pathways exhibited an increasing trend during the initial years (2015–2017), followed by relative stabilization with minor fluctuations thereafter. Bias values initiated at relatively low levels in 2015 (ranging from 2.30 mm for SSP245 to 3.39 for SSP126) before progressively increasing through most of the study period. By 2022, bias peaked for most scenarios (9.09 mm for SSP126, 7.69 mm for SSP245, 7.44 mm for SSP370, and 8.60 mm for SSP585) before showing a slight decline in 2023–2024. Notably, SSP585 showed significant improvement in 2024, with bias decreasing to 5.91 mm, comparable to SSP245’s 5.88 mm. Throughout most of the study period, the intermediate-emissions scenarios (SSP245 and SSP370) maintained lower bias values than their more extreme counterparts, suggesting better representation of observed precipitation conditions.

Figure 4b illustrates the temporal evolution of correlation coefficients between observed precipitation data and model projections. These values demonstrated consistent improvement across all scenarios, increasing from approximately 0.92 in 2015 to 0.97 by 2024. This upward trend indicates progressive enhancement in the linear relationship between model outputs and empirical observations. SSP370 maintained marginally superior performance throughout most of the analysis period, though by 2024, all scenarios had effectively converged, with correlation values ranging narrowly between 0.96 and 0.97. The temporal evolution exhibited two distinct phases: rapid improvement from 2015 to 2018, followed by more gradual enhancement through 2024.

Figure 4c presents the R² evolution across the decade. R² values increased from approximately 0.84–0.85 in 2015 to 0.86–0.90 by 2024, indicating enhanced explanatory power of the models over time. SSP245 achieved the highest R² value by 2024 (0.90), closely followed by SSP585 (0.90) and SSP370 (0.89), while SSP126 showed more modest improvement (0.87). Notably, SSP585 demonstrated significant improvement in later years, overcoming its initially lower R² values during 2015–2017.

Figure 4d illustrates the RMSE reduction across all scenarios. RMSE values decreased from initial values of approximately 14.85–15.29 mm in 2015 to 10.63–12.59 mm by 2024, reflecting substantial improvements in projection accuracy. SSP245 achieved the lowest RMSE by 2024 (10.63 mm), marginally outperforming SSP585 (10.77 mm) and SSP370 (11.31 mm), while SSP126 maintained slightly higher error rates (12.59 mm). The reduction pattern showed three distinct phases: rapid improvement (2015–2017), stabilization (2018–2021), and renewed enhancement (2022–2024).

This comprehensive evaluation reveals that while all models improved over time, intermediate scenarios (particularly SSP245) demonstrated slightly better overall performance. Most notably, by 2024, performance metrics had substantially converged across all scenarios, suggesting that in the near-term (2015–2024), climate model performance is less dependent on specific emissions pathways than on general improvements in modeling capabilities and data assimilation techniques.

Table 1. The mean values of interannual outcomes (2015–2024) for statistical performance indicators derived from precipitation quantile estimates across various return periods and SSP scenarios.

Return Period (T-Years)	SSP126				SSP245				SSP370				SSP585
	Bias	Corr	R2	RMSE	Bias	Corr	R2	RMSE	Bias	Corr	R2	RMSE	Bias	Corr	R2	RMSE
10	6.48	0.94	0.86	13.83	5.57	0.94	0.87	13.41	5.36	0.95	0.87	13.08	6.40	0.94	0.86	13.87
30	11.31	0.94	0.82	20.47	11.33	0.94	0.82	20.39	10.63	0.94	0.84	19.64	11.50	0.94	0.82	20.80
50	14.46	0.93	0.80	24.78	15.08	0.94	0.79	24.89	13.94	0.94	0.81	23.97	14.58	0.93	0.79	25.07
100	20.02	0.92	0.75	32.36	21.21	0.93	0.75	32.39	19.44	0.93	0.76	31.57	19.88	0.92	0.75	32.42

presents the statistical measurements averaged by the recurrence period for each year. The data indicates a consistent decline in all statistical measurements as the recurrence period lengthens. Specifically, the bias exhibits an approximate threefold increase, rising from a range of 5.36–6.48 mm for a 10-year period to a range of 19.44–21.21 mm for a 100-year period. Concurrently, the R² shows a significant decrease, dropping from 0.86–0.87 to 0.75–0.76. Furthermore, the RMSE more than doubles, increasing from approximately 13.08 mm to around 32.42 mm, thereby quantitatively illustrating the growing uncertainty associated with forecasting extreme weather events.

Among the four scenarios analyzed, SSP370 consistently exhibits superior performance across all recurrence periods, characterized by the lowest bias, the highest correlation coefficient, the highest R² value, and the lowest RMSE. SSP245 demonstrates comparable performance; however, SSP126 and SSP585 tend to exhibit higher error rates. These findings indicate that predictions of extreme precipitation events are associated with significantly greater uncertainty compared to those of more frequent weather events. This has critical implications for infrastructure design and flood management strategies. It is essential for practitioners to consider this heightened uncertainty when utilizing climate projections for long-term recurrence events.

3.3. Chi-Square Test for Consistency Analysis

Figure 5 is represented to demonstrate the simple computation of the χ² statistics as summarized in

Table 2. The chi-square (χ²) test statistics and p-values to evaluate precipitation quantiles across various return periods and SSP scenarios.

	Return Period (T-Years)
Scenarios	10		30		50		100
	${\mathsf{\chi }}^2$	p-Value	${\mathsf{\chi }}^2$	p-Value	${\mathsf{\chi }}^2$	p-Value	${\mathsf{\chi }}^2$	p-Value
SSP126	23.6455	0.0105	29.8894	0.0010	33.9740	0.0004	39.7561	0.0000
SSP245	19.7886	0.0644	31.3528	0.0020	37.2585	0.0004	44.0228	0.0001
SSP370	17.0244	0.0901	26.3772	0.0035	31.0016	0.0013	35.8341	0.0001
SSP585	24.4455	0.0138	32.6602	0.0011	39.7236	0.0007	40.9854	0.0001

. Figure 5 shows the interannual histograms of the precipitation quantiles (T = 10 years) at all 615 observation sites for each year from 2015 to 2024, together with the mean histogram obtained by averaging those ten years. For each year and each SSP scenario, a χ² test was applied to compare the spatial distribution of observed and modeled quantile values across the 615 stations. The resulting ten annual χ² values were then averaged to produce the summary χ² statistic reported in Table 2.

Table 2 presents the mean chi-square test statistics and corresponding p-values for precipitation quantiles across four SSP scenarios (SSP126, SSP245, SSP370, SSP585) and four return periods (10, 30, 50, and 100 years). These statistics provide a formal assessment of the goodness-of-fit between model-projected and observed precipitation distributions.

The analysis reveals a systematic pattern of increasing chi-square values with longer return periods across all scenarios, with χ² values rising from a range of 17.02–24.45 for 10-year return periods to 35.83–44.02 for 100-year return periods. This pattern indicates progressively larger discrepancies between modeled and observed precipitation distributions as event rarity increases. Correspondingly, p-values decrease substantially from the 10-year to 100-year return periods across all scenarios, reflecting stronger statistical evidence for rejecting distributional equivalence between modeled and observed extreme precipitation patterns for rarer events.

Among the four scenarios, SSP370 consistently demonstrates superior performance across all return periods, exhibiting the lowest chi-square values (17.02 for 10-year, 26.38 for 30-year, 31.00 for 50-year, and 35.83 for 100-year return periods) and, consequently, the highest p-values. Notably, for the 10-year return period, SSP370 produces a non-significant result (χ² = 17.02, p = 0.0901), indicating that its distributional properties cannot be statistically distinguished from observations at the conventional α = 0.05 significance level. SSP245 similarly shows a marginally non-significant result for the 10-year return period (χ² = 19.79, p = 0.0644), while both SSP126 and SSP585 yield significant deviations even at this shortest return period.

For longer return periods (30, 50, and 100 years), all scenarios produce statistically significant deviations from observed distributions (p < 0.01), though SSP370 consistently maintains the most favorable fit statistics. The worst-performing scenarios vary by return period, with SSP585 showing the highest chi-square values for 10-year and 30-year return periods, while SSP245 demonstrates the poorest fit for the 100-year return period (χ² = 44.02, p = 0.0001).

These findings complement earlier performance metrics by providing formal statistical evidence of the increasing challenges in accurately reproducing extreme precipitation distributions with longer return periods. The results further support the observation that intermediate emissions scenarios (particularly SSP370) demonstrate superior skill in representing observed precipitation patterns compared to either lower (SSP126) or higher (SSP585) emissions pathways during this study period. This pattern suggests that internal model dynamics and initialization may be more influential in determining near-term projection accuracy than the specific characteristics of emissions scenarios.

3.4. Future Precipitation Quantiles

Figure 6 illustrates the temporal progression of the precipitation quantiles (T = 2 years) across all 615 sites in South Korea, spanning the years 2015 to 2100, under four distinct emission scenarios: SSP126, SSP245, SSP370, and SSP585. The 2-year threshold was chosen because it strikes an ideal balance between event frequency and extremity: it occurs often enough in both observations and model simulations to produce statistically robust estimates yet remains sufficiently rare to reflect meaningful changes in severe rainfall. By contrast, longer return periods yield much fewer exceedances, leading to wider confidence bounds and poorer trend detectability. In the figure, observed quantiles are plotted in black (solid line for the median), while each scenario’s median projection is surrounded by shaded bands that indicate the full range of results across sites, thereby illustrating both the central tendency of future extremes and how projection uncertainty widens over time.

The analysis results, organized by scenario, indicate markedly different trajectories and levels of uncertainty. The SSP126 scenario (green) demonstrates the slowest and most consistent increase in precipitation quantiles projected until 2100, characterized by a gradual rise that represents the narrowest increase among all scenarios. In contrast, the SSP245 scenario (blue) exhibits a relatively steeper trend, accompanied by a broader range of uncertainty compared to SSP126, reflecting increased variability due to heightened emissions. The SSP370 scenario (purple) reveals a somewhat different pattern, with its median trajectory indicating a more modest increase than SSP245 by 2100, despite a steeper trajectory during mid-century periods. Notably, SSP370 displays a significantly wider uncertainty range than SSP245, particularly in the latter half of the century, suggesting greater variability in potential outcomes despite similar median projections. This expanded uncertainty range highlights the increased sensitivity of precipitation patterns to the medium-high emissions pathway, even when central tendencies appear comparable to intermediate scenarios. Finally, the high-emission scenario SSP585 (red) shows the steepest increase and the broadest distribution of outcomes, maintaining a distinct separation from other scenarios throughout the latter half of the century.

The SSP370 scenario is pivotal as it provides a balance between moderate emissions and policy adaptability. It anticipates that precipitation quantiles will reach approximately 228 mm by the year 2100, serving as a midpoint between the modest increases projected by SSP126 and the substantial rises indicated by SSP585. Nevertheless, the increasing uncertainty associated with SSP370 post-mid-century underscores the heightened variability linked to intermediate emissions, thereby necessitating the implementation of adaptive strategies and highlighting the critical importance of emissions reductions to mitigate associated risks. Overall, projections indicate that South Korea will face an escalation in extreme precipitation events throughout the 21st century across all emission scenarios. The low-emission pathway represented by SSP126 suggests smaller and more predictable changes, which may indicate potential for stabilization through rigorous mitigation efforts. Conversely, SSP585 predicts more significant increases and considerable uncertainty, particularly after 2070, which complicates water management and adaptation strategies. SSP370 thus underscores the necessity for flexible approaches to address the evolving climate risks while reinforcing the urgency of emissions reduction to limit future impacts.

3.5. Spatial Distribution of Future Changes

Figure 7 shows the spatial distribution of percent changes in precipitation quantiles corresponding to the 100-year recurrence interval under the SSP scenarios across South Korea. This analysis is based on data from 615 observation stations (indicated by black dots), which ensure comprehensive national coverage and facilitate a detailed assessment of regional hydrological responses to future climate change. The findings reveal considerable spatial variability in precipitation changes, indicating that such differences are likely to result in systematic alterations to regional runoff patterns.

First, the relative difference between precipitation quantiles for the reference year 2024 and projected precipitation quantiles for the year 2100 was calculated as the change ratio across all 615 observation sites. To represent the spatial distribution of these calculated site-specific change ratios, spatial interpolation was performed using the PyKrige library in Python (3.13.3 version), applying a spherical variogram model with a grid resolution of 0.02° through the Ordinary Kriging technique. During this process, additional Python libraries were utilized, including Cartopy for visualization, NumPy and SciPy for numerical computations, and Shapely for boundary delineation operations.

Spatial analysis indicates that precipitation increases across South Korea under all SSP scenarios; however, the magnitude and spatial distribution of these changes exhibit substantial regional variability closely associated with emissions trajectories. A persistent spatial pattern is observed in which the northeastern and eastern regions experience markedly greater precipitation increases compared to the southwestern and southern coastal areas. This east–west gradient in change becomes more pronounced with higher-emission scenarios, implying significant regional differences in sensitivity to climate forcing.

Under SSP126, projected changes are relatively modest, with most regions experiencing increases between 5% and 15%. The spatial distribution of these increases is relatively uniform, particularly in the northeastern and central regions, suggesting that strong emissions mitigation may lead to more moderate and evenly distributed hydrological impacts across the peninsula. For SSP245, pronounced spatial heterogeneity emerges, with precipitation increases in the range of 10% to 20%. The central and eastern inland areas exhibit relatively high change rates, with the most pronounced increase observed along the eastern coastline near 36° N. In contrast, western regions record relatively modest changes, resulting in a clear east–west contrast in projected precipitation increases. The SSP370 scenario shows distinct areas of increased precipitation concentrated in central and select coastal regions, with overall change rates similar to SSP245 but presenting a somewhat less intense pattern. SSP585 exhibits the most significant precipitation changes, characterized by widespread increases exceeding 20% across much of the peninsula, with the most pronounced increases concentrated in the central inland area.

These findings indicate that precipitation patterns in Korea are projected to undergo spatially heterogeneous changes throughout the 21st century, with both the magnitude and distribution of changes being strongly dependent on emissions trajectories. This underscores the necessity for region-specific adaptation strategies, particularly in areas expected to experience the greatest precipitation increases. Moreover, the pronounced differences between low- and high-emission scenarios emphasize the critical role of mitigation efforts in limiting both the magnitude and spatial variability of future hydrological changes across the Korean Peninsula.

4. Discussion

This research provides a comprehensive evaluation of changes in precipitation quantiles throughout South Korea under different SSP scenarios. Our findings reveal that the SSP370 scenario demonstrates the closest alignment with currently observed extreme precipitation patterns among all analyzed scenarios, despite exhibiting some notable inconsistencies with observational data. The similarity in precipitation quantiles between current observations in South Korea and the SSP370 scenario—which represents a medium-high emissions pathway marked by sustained fossil fuel reliance and minimal climate policy implementation—bears significant implications. This finding suggests that the nation’s precipitation patterns may already be transitioning in a manner consistent with considerable future warming, thereby emphasizing the critical need for more robust climate mitigation strategies.

The implications of these findings are substantial. Should precipitation patterns persist along the SSP370 trajectory, South Korea is anticipated to face considerable increases in flood-related damage to infrastructure, agricultural systems, and human settlements. The potential economic repercussions could be severe, with a recent study indicating that flood damages may escalate by as much as 25% under the RCP4.5 scenario [15]. Despite ongoing investments in climate resilience initiatives—including improved drainage systems, flood-control dams, and early-warning networks—the anticipated severity of extreme events under SSP370, and more critically under SSP585, may surpass the capacity of current protective measures. Therefore, resilience planning must explicitly incorporate analyses based on more severe scenarios to test and upgrade infrastructure capacity, insurance schemes, and emergency response protocols. Given that current observations align with the SSP370 pathway, a shift towards a more moderate SSP245 trajectory would necessitate highly aggressive carbon reduction strategies. In the absence of such interventions, there is growing concern that actual emissions and their corresponding climate impacts may approximate those projected under the SSP585 scenario, which represents the most adverse outcome outlined in this study.

4.1. Model-Observation Divergence and Its Implications

A critical finding of this study is the substantial discrepancy between observed precipitation trends and climate model projections. This divergence warrants extensive discussion given its implications for climate adaptation planning. Our analysis reveals that actual precipitation quantiles in South Korea have remained relatively stable or experienced slight declines during the recent period (2015–2024). This contrasts sharply with the consistent upward trends predicted by all climate models across SSP scenarios, raising fundamental questions about the reliability of model projections for extreme precipitation events.

This divergence represents a fundamental challenge to current climate modeling approaches and has been increasingly recognized in recent literature. Kim et al. [36] documented similar discrepancies in extreme precipitation trends across the Asian domain. They attributed this phenomenon to complex interactions between regional circulation patterns and global climate forcing that are inadequately captured in current GCMs. Their findings suggest that the observed stability in extreme precipitation may reflect compensating effects between enhanced moisture availability and changes in atmospheric circulation patterns, particularly in monsoon-dominated regions like Korea.

The increased bias observed in Figure 4a is particularly concerning as it may contaminate forecast signals. Since 2015, the climate model ensemble has incorporated anthropogenic forcing, resulting in amplified projected trends across various scenarios. Consequently, even after applying bias correction, the ensemble mean of the 23 GCMs continues to display a systematic upward trend. This intensification occurs because bias correction is typically calibrated with historical data and then applied to future projections without dynamic feedback mechanisms, potentially failing to account for evolving model biases under substantial external forcings such as increased greenhouse gas concentrations.

This model-observation divergence has profound implications that extend beyond academic interest. Chen and Sun [37] demonstrated that such discrepancies in precipitation extremes can lead to systematic biases in impact assessments. These biases potentially result in either over- or under-estimation of future risks depending on the region and season. In the context of South Korea, the stable or declining trends in observed extreme precipitation during 2015–2024 suggest that climate models may be overestimating the rate of intensification of extreme events in the near term.

This finding challenges the conventional assumption that warming directly translates to proportional increases in extreme precipitation intensity, particularly at regional scales where local meteorological factors play crucial roles. The observable lack of upward trends in actual precipitation data (Figure 6) while models consistently project increasing trends raises serious questions about model credibility for regional precipitation forecasting. If observed trends continue to deviate from model projections, current adaptation planning based solely on model outputs may be inadequately calibrated to actual regional climate evolution. This suggests the need for adaptive management approaches that can accommodate both the possibility of accelerated changes (as projected by models) and the current observed stability. Furthermore, the discrepancy highlights the importance of continuous monitoring and model validation, as reliance on projections alone may lead to maladaptive responses to climate change.

4.2. Study Limitations and Future Research Directions

Notwithstanding the significant findings of this study, it underscores critical limitations inherent in contemporary climate modeling methodologies. Employing the ensemble average of 23 General Circulation Models (GCMs) as a multimodel strategy may result in inaccuracies stemming from individual models that exhibit poor correlations with local and regional observations, despite bias correction implementation. Some models within the ensemble may inadequately capture regional dynamics, potentially compromising the overall reliability of the multimodel mean. The persistent divergence between observed and modeled precipitation trends may reflect fundamental limitations in how current GCMs represent regional-scale processes. These include complex interactions between large-scale circulation patterns, local topography, and land-surface feedbacks that are particularly important in monsoon-influenced regions [36,37].

The generally low χ² test results across various scenarios further substantiate the inadequacies of these models in accurately representing the spatial distribution of extreme precipitation occurrences. These statistical indicators reveal significant challenges in capturing the spatial heterogeneity of extreme precipitation events in South Korea’s complex topography. The poor performance in reproducing observed spatial patterns suggests that current GCMs struggle with the fine-scale processes that govern regional precipitation extremes. As noted by Hawkins and Sutton [38], the uncertainty associated with GCMs often exerts considerable influence on rainfall projections, particularly at regional scales that are pertinent to adaptation planning.

The methodological approaches utilized in this research warrant further scrutiny. Bias correction was performed using the comprehensive observational dataset from 1961 to 2024 as a fixed reference period, rather than employing interannual bias correction methods. This approach enhances stability and consistency, as variable reference datasets may introduce unnecessary fluctuations and undermine reliability. Recent literature underscores the efficacy of fixed-period quantile mapping. Turco et al. [39] assert that fixed-period quantile mapping techniques are among the most effective methods for rectifying distributional biases in precipitation data, particularly when applied to long-term climate projections. Their findings indicate that the use of fixed reference periods enhances the reliability of extreme precipitation indices, particularly in areas characterized by substantial interannual variability. However, the limitations of bias correction under changing climate conditions must be acknowledged, particularly when dealing with non-stationary climate systems.

The observed divergence between model projections and recent observations raises important questions about the temporal evolution of climate change impacts. While models consistently project increasing extreme precipitation under all warming scenarios, the observed stability during 2015–2024 may represent a temporary hiatus in the intensification process, potentially followed by accelerated changes in subsequent decades. This phenomenon aligns with findings from Alexander [40], who highlighted systematic differences between observed and modeled precipitation extremes across multiple regions globally. Alternatively, it may indicate that regional climate responses are more complex than currently represented in global models, with local and regional factors exerting stronger influence on extreme precipitation patterns than previously anticipated [36,37]. This uncertainty underscores the importance of maintaining robust observational networks and developing improved understanding of regional climate dynamics.

For future research, we emphasize the need to address the spatial scale limitations of GCMs for a relatively small territory like South Korea. Regional Climate Models (RCMs) with higher spatial resolution could potentially overcome some of the reproduction limitations identified in this study. An MME approach using RCM outputs would likely provide more reliable projections of extreme precipitation patterns and their spatial distribution. Furthermore, integrating non-stationary frequency analysis methods could better account for the evolving nature of precipitation extremes under climate change.

Given the significant model-observation divergence identified in this study, future research should prioritize the development of hybrid approaches that combine observational trend analysis with physically-based projections, as suggested by Min et al. [41] in their analysis of extreme weather events in Korea. Additional research is also needed to explore the drivers behind the divergence between observed trends and model projections, potentially incorporating atmospheric circulation patterns and local topographical influences that may not be adequately represented in current models. Enhanced model validation and regional calibration are essential to improve the reliability of future climate projections for extreme precipitation events.

5. Conclusions

This study effectively achieved its primary aim of identifying the Shared Socioeconomic Pathway (SSP) scenario that most accurately reflects the current patterns of extreme precipitation in South Korea, while also assessing the implications for future climate adaptation strategies. The central hypothesis positing that the observed precipitation quantiles in Korea correspond to a more aggressive emissions scenario than previously anticipated has been substantiated through thorough analysis.

In addressing the first objective of determining the alignment of scenarios with current observations, the SSP370 scenario exhibited the closest correlation with the observed precipitation quantiles for the period of 2015–2024. This outcome supports the hypothesis that Korea’s precipitation patterns are already shifting towards a medium-high emissions trajectory, characterized by ongoing reliance on fossil fuels and limited implementation of climate policies. The superior performance of SSP370 relative to more moderate scenarios (SSP126, SSP245) suggests that existing climate mitigation efforts may be inadequate to alter the trajectory of extreme precipitation events.

With respect to the second objective of quantifying future precipitation changes, all SSP scenarios forecast an increase in precipitation quantiles through 2100, albeit with significantly divergent trajectories. The SSP126 scenario indicates gradual increases, while SSP245 and SSP370 present intermediate and variable trends. In contrast, SSP585 predicts steep increases accompanied by high uncertainty. The alignment of current observations with SSP370 implies that, in the absence of aggressive carbon reduction measures, South Korea may encounter precipitation extremes approaching those projected under the most severe SSP585 scenario.

Regarding the third objective of characterizing changes in spatial distribution, this study anticipates considerable regional variations in precipitation quantile increases across South Korea. The SSP370 scenario, which most accurately reflects current conditions, forecasts increases of 10–20% in 100-year return period precipitation, exhibiting distinct regional patterns. This spatial heterogeneity carries significant implications for infrastructure design and flood management strategies, necessitating regionally tailored adaptation approaches.

The validation of our scientific hypothesis has substantial implications for climate adaptation in South Korea. The alignment of current precipitation patterns with the SSP370 pathway indicates that the nation is already experiencing climate impacts associated with considerable future warming. This finding underscores the urgent need to reassess existing flood prevention standards, infrastructure capacity, and emergency response protocols to accommodate more severe conditions than previously anticipated.

These results provide critical scientific evidence for policymakers to formulate more robust climate adaptation strategies. The identification of SSP370 as the most representative scenario should inform long-term infrastructure planning, water resource management, and disaster risk reduction policies. Future research should prioritize the development of higher-resolution regional climate models and the integration of these findings into comprehensive socioeconomic risk assessments to facilitate evidence-based adaptation decision-making in South Korea.