Reason for less water supply shortages under climate change condition: evaluation of future rainfall data

Future climate data are frequently used in various studies related to global warming or climate change (Mishra & Herath 2015; Zahmatkesh et al. 2015; Kim et al. 2018; Roy et al. 2022; Yoosefdoost et al. 2022; Li & Burian 2023). These future climate data are those generated by considering future climate change scenarios. Generally used scenarios are those proposed in the evaluation report of the IPCC (Intergovernmental Panel on Climate Change). Various general circulation models (GCMs) or regional climate models (RCMs) are used for this data generation. As the spatial resolution of the generated data is rather coarse, spatial-temporal downscaling is additionally applied to generate finer-resolution data (Corte-Real et al. 1995; Rockel 2015; Dahm et al. 2016; Sharma et al. 2023). These finer-resolution data are generally used for the basin-scale hydrological analysis.

The future climate data focuses on the target year of 2100, and are based on a range of scenarios for projected changes in greenhouse gas concentrations (Collins & Senior 2002; Jevrejeva et al. 2010; Raftery et al. 2017). Thus, the generated future climate data covers almost a 100-year period from the present to 2100. The most frequently used future climate data are rainfall and temperature, which is also the same in the fields of hydrology and water resources (Bouraoui et al. 2002, 2004; Andersson et al. 2006; Merritt et al. 2006; Marshall & Randhir 2008; Zahabiyoun et al. 2013; Khaniya et al. 2020; Shaikh et al. 2022). For example, Kite et al. (1994) evaluated the possible change of runoff characteristics in the Mackenzie River basin, Canada. The future data used in the study were those generated by the CCC (Canada Climate Center) GCM II. Prudhomme & Davies (2009) evaluated the possible change of river flows in several UK basins and concluded that the flow rate would decrease in most basins. Sidiqi & Shrestha (2021) simulated hydrological changes in the Kabul River basin, Afghanistan using the future climate data generated under the RCP-4.5 and RCP-8.5 scenarios. This study also showed that available water resources will decrease in the future. In Korea, Kim et al. (2019) evaluated the future climate data generated by six GCM predictions and concluded that the southern region of Korea would become more vulnerable to drought.

However, there exist questions about whether the future climate is realistic or not (Wilby & Harris 2006; Kay et al. 2009; Najafi et al. 2011; Johnson et al. 2016; Kochendorfer et al. 2017). Most concerns are focused on future rainfall data. For example, Chandra et al. (2015) and Hosseinzadehtalaei et al. (2017) mentioned the possible uncertainty of the rainfall intensity–duration–frequency (IDF) curve derived by analyzing the future rainfall data. Jeong et al. (2004) argued that the change in rainy and dry periods is too big in the future rainfall data. Also, Kim & Joo (2015) pointed out that there are too many small rainfall events in the future rainfall data.

These issues have led to the correction of the future rainfall data (Refsgaard & Storm 1996; Kuczera & Mroczkowski 1998; Piani et al. 2010; Eum & Cannon 2017). For example, Berg et al. (2012) applied and compared several bias correction methods from simple scaling and additive corrections to more advanced histogram equalization corrections to the RCM data, and Gudmundsson et al. (2012) applied quantile mapping to correct the bias of future rainfall data. Watanabe et al. (2012) compared five bias correction methods including the delta method and quantile-based mapping, and evaluated the degrees of similarity to the observed data. Hwang & Ahn (2005) showed the possible use of the artificial neural network (ANN) in bias correction of future climate data, and Song et al. (2019) also tried to use the quantile mapping and random forest for the calibration of GCM data. However, it has not been confirmed yet that the corrected future rainfall data are fully acceptable.

This study focuses on the issue of future rainfall data. This study is concerned more with the long-term characteristics of the future rainfall data. The long-term rainfall characteristics are related to the occurrence and persistence of dry years, and this long-term analysis can be helpful for evaluating the possible change of multi-year drought (Mehta & Yadav 2022). On the other hand, the short-term rainfall characteristics include the annual mean number of rainy days, the mean rainfall duration, the mean rainfall intensity, the mean length of no-rain days (i.e., the mean duration between two rainy days), etc. With these derived short-term rainfall characteristics, any possible change of flood or short-term drought may be conjectured. Additionally, the future runoff is going to be simulated by the PRMS (Precipitation Runoff Modeling Simulation) model, which is then to be used as input to a dam reservoir. By applying a reservoir operation method, the possible change of dam storage and water supply can also be evaluated. With this evaluation result, the water supply system under future climate change conditions can be evaluated. This study considers sets of future climate data obtained from three different data providers: Global Water Bank (GWB). The Boryeong Dam, located in the western part of the Korean Peninsula, is considered as an example water supply system.

The forest type affects much on the evapotranspiration amount in the basin (NIFS 2016). In the Boryeong Dam basin, the mixed (mixed with needle leaf and deciduous; pale yellow-green in the figure) covers 36% of the basin area, other mixed (light green) is 23%, and pitch pine (blue-green) is 14%. As the percentage of needle leaves in the Boryeong Dam basin is rather high, it is assumed that the evapotranspiration loss could be significant. Finally, the soil map shows that the dominant soil is loam (86%), and sand (7%) and clay (7%) also exist. As the loam, the dominant soil in the Boryeong Dam basin represents a moderate infiltration rate, the overall runoff ratio in this basin may also be assumed moderate.

As can be seen in this figure, the daily maximum temperature in the study area fluctuates from 0 to 35 °C. The daily minimum temperature is about 10 °C lower than the daily maximum temperature. Seasonality in the temperature data is very strong, which is also the same in the rainfall data. Basically, the variation of the rainfall data seems much higher than that of the temperature data, and daily rainfall records of more than 150 mm are also found in 2010, 2012, and 2018. The maximum daily rainfall in 2018 was more than 200 mm.

Three sets of future climate data are available in Korea, and thus, all three datasets were considered in this study. The first dataset can be obtained from the Global Water Bank (GWB, http://gwb-ccaw.re.kr). These data were generated by considering several GCM simulations based on four RCP (Representative Concentration Pathway) scenarios. A total of five GCM simulations were considered in the data generation, which includes bcc-csm1-1-m (Beijing Climate Centre Climate System Model version 1.1, moderate resolution), CanESM2 (Canadian Earth System Model version 2), CMCC-CMS (The Centro Euro-Mediterraneo sui Cambiamenti Climatici Climate Model, well-resolved stratosphere), CNRM-CM5 (Centre National de Recherches Météorologiques Coupled Global Climate Model, version 5), and NorESM1-M (Norwegian Earth System Model, Version 1, intermediate resolution). Daily maximum temperature, daily minimum temperature, and daily rainfall data are available over the domain of interest around the Korean Peninsula. Additional information about the GWB data can be found in Bae et al. (2020).

The second dataset considered in this study can be obtained from the Water Resources Management Information System (WAMIS, http://www.wamis.go.kr). The future climate data in this system, including the daily rainfall, and daily highest and lowest temperatures are available only at rain gauge stations. These data were generated by applying the spatial disaggregation (SD) technique and the quantile delta mapping (QDM) technique to the simulations of HadGEM2-AO (Hadley Center Global Environment Model version 2-Atmosphere Ocean).

Finally, the third dataset can be obtained from the Korea Meteorological Administration (KMA, https://data.kma.go.kr). This future climate data is also based on the HadGEM2-AO. This GCM data with a spatial resolution of 135 km was then downscaled using the HadGEM3-RA (Hardley Centre Global Environmental Model version 3-Regional) to make the data with a spatial resolution of 12.5 km for the regional domain around the Korean Peninsula. Once again, a statistical downscaling technique was applied to produce the data with a spatial resolution of l km. This future climate data from KMA also includes the daily rainfall, and daily highest and lowest temperatures.

The future temperature data given in Figure 5(a) show that they can be quite different from each other depending on the data provider. For example, the maximum temperature data from GWB show no obvious increasing trend, and the highest temperature rarely reaches 40 °C. This trend is also the same for the minimum temperature. The RCP-4.5 data are not provided in this figure, but they also look quite similar to the RCP-8.5 data. No obvious difference cannot be found.

On the other hand, the increasing trend is obvious and steeper in the WAMIS temperature data than in the GWB data. The highest data marks higher than 40 °C, which is also frequent after 2050. The trend of daily minimum temperature data is also the same. The lowest daily temperature around 2010 was near −20 °C, which becomes −10 °C around 2100. However, the RCP-4.5 temperature data (not provided in the figure) were found to be quite different from the RCP-8.5 data in the WAMIS dataset. Just a moderate temperature increase was found in this data, such as in the GWB data, and the highest daily temperature was also found lower than 40 °C. Finally, the KMA temperature data look very similar to the WAMIS temperature data, since both the WAMIS and KMA data were generated based on the same GCM, HadGEM2-AO.

The future rainfall data in Figure 5(b) also shows different characteristics among the datasets. First, the GWB data looks very stable without any abrupt change within the data period. There are several events with more than 100 mm of daily rainfall. Only a few events contain a daily rainfall of over 150 mm. The maximum daily rainfall during the generation period from 2009 to 2099 is just over 200 mm. These characteristics of the generated rainfall are also similar to the rainfall data based on the RCP-4.5 scenario (not provided in the figure). On the other hand, the WAMIS data are more extreme. The daily rainfall of 200 mm or higher is very frequent, and, around 2100, the highest daily rainfall marks more than 900 mm. The KMA data seems to be in between the GWB and WAMIS data. They also show a higher frequency of daily rainfall 200 mm or higher, but the highest values are still less than 400 mm. The KMA rainfall data based on the RCP-4.5 scenario (not provided in the figure) seem to be a bit milder than those under the RCP-8.5 scenario.

As can be seen in Figure 6, the generated rainfall data are very different from the observed. Only the WAMIS data are similar to the observed. However, this difference in the data is mainly caused by the large difference in the rainy days. Basically, the number of rainy days is more than the observed in the GWB data and much more in the KMA data. For instance, the number of rainy days in the GWB data is nearly double that in the observed rainfall data. The rainfall intensity is roughly halved, while the rainfall duration approximately doubles. Similarly, the no-rain duration also reduced by half.

If focusing on the future data (i.e., from 2022 to 2099), first, the GWB data show a mild increasing trend until around 2040–2050. Until then, the annual rainfall has been increased by about 100 mm. Proportional to the annual rainfall, its variation has also been increased. The variance of annual rainfall seems to be highest around 2050. In this period, the maximum annual rainfall reaches 1,800 mm and the minimum becomes just around 500 mm. After that, the annual rainfall seems to become stable without any obvious increasing or decreasing trend. This trend was found similar in both cases under RCP-4.5 and RCP-8.5 scenarios.

The increasing trend of the WAMIS data is clearer and steeper than that of the GWB data. It is much steeper in the case under the RCP-8.5 scenario. The mean value was just around 1,250 mm in 2010, but it becomes around 1,750 mm around 2100. That is, the annual rainfall is expected to increase steadily in the WAMIS data, and, at the end of the simulation period, the increased amount of annual rainfall will be about 500 mm. It is also noticeable that, under the RCP-8.5 scenario, the highest annual rainfall marks more than 3,000 mm around 2100.

The KMA data shows totally different behavior than the previous two data. Different from the GWB or WAMIS data, the KMA data do not show any increasing trend. However, the mean value of the KMA rainfall is found much higher (by about 500 mm) from the beginning of the simulation period. It is also interesting to note that the annual rainfall under the RCP-4.5 scenario looks more extreme than that under the RCP-8.5 scenario. That is, the annual rainfall under the RCP-4.5 scenario fluctuates from 1,500 to 2,500 mm, but that under the RCP-8.5 scenario from 1,500 to 2,250 mm. The highest value in the KMA data is also higher than 3,000 mm, under the RCP-8.5 scenario.

Table 2 compares the monthly rainfall observed and generated for the entire period. In the spring season (March to May), the generated monthly rainfalls look a bit higher than the observed, especially the generated rainfall in April seems quite higher. However, the summer rainfall from June to August is totally different from the observed. The generated rainfall in June is much higher than the observed, but the generated rainfall in July is smaller or similar to the observed. The generated rainfall in August is also a bit higher than the observed. This result indicates that the summer monsoon starts earlier in the generated rainfall (Cho et al. 1997; Ha et al. 2005). The generated rainfall in the fall season (September to November) also looks similar to the observed, but the generated rainfall in September is much higher than the observed. As a result, the summer season seems to be elongated with much more rainfall. Finally, in the winter (December to February), the generated rainfall is found to be much higher than the observed. As the mean temperature in winter is below zero, much more snow is expected in the future in Korea.

These differences noticed in the generated rainfall can have both positive and negative effects in Korea. First, the rainfall increase in April can help the agricultural sector. It has an important meaning because the spring drought is very common in Korea. However, the rainfall decrease in October and November may be a disadvantage in securing the water resources. In Korea, the typhoon season ends in September. Prior to this, dam reservoirs in the country are required to maintain a restricted water level for flood control purposes. This restricted water level is much lower than the normal high water level. Additional rainfall in October or November helps secure water resources to fill the gap. However, regardless of these differences, both generated annual rainfalls look similar to the observed, except for the KMA rainfall data. The KMA rainfall data is too high to be realistic.

The difference between the dry and wet (rainy) seasons is clear in Korea. Generally, the dry season starts at the end of June and ends in the middle of September. Sufficient water should be secured during the wet season, particularly toward its end. This stockpiled water is then used throughout the year. If enough water is not secured, then the possibility of spring drought increases. On average, a moderate drought repeats every 2 years, and a more severe one repeats every 5–10 years. It is also possible to have little rainfall during the wet summer season. In this case, the drought situation becomes worse. This situation occurs about once every 10–30 years (Yoo 2006; Yoo et al. 2008).

The problem of severe drought and resulting water supply shortage is a long-term problem, which may not be easily observed in a simple analysis of hourly or daily rainfall characteristics. In general, monthly or annual data are analyzed to derive the long-term behavior of rainfall. It is also the same in the analysis of future climate data. In this part of the study, the long-term behavior of rainfall was derived by analyzing the annual rainfall data. The recurrence characteristics of the dry year (i.e., the year with smaller rainfall than a given threshold value) were investigated using the Poisson process. The Poisson process is widely used for modeling the occurrence of events such as flood, droughts, or rainfall itself (Yoo 2004).

Since the calculation of such probabilities in Equation (3) relies on only one parameter λ, it is of utmost importance to select the appropriate level of threshold so that the occurrence of dry years is independent. Yoo (2004) recommended using a threshold of mean – 0.5 standard deviation, or lower. The parameter λ is simply determined by the number of dry years divided by the total number of years.

This study analyzed the four different datasets. One is observed from 2009 to 2021, and the other three are from GWB, WAMIS, and KMA data. The generated data also covers the period of observed data, which makes the comparison of observed and generated data possible. So, in this study, the analysis of dry years was done twice; one for the old period from 2009 to 2021 and the other for the future period from 2022 to 2099. The threshold value for the application of the Poisson process was determined in the old period using the observed data, which was then applied to the future data. It was simply to apply the same condition to the future data and to conjecture the future condition of drought.

Using the parameter in Table 3, it is possible to calculate the probability of consecutive dry years, which can also be converted into the number of occurrences of consecutive dry years. Results are summarized in Table 4. The results in this table show that the generated data do not represent well the number of occurrences of consecutive dry years. In fact, the generated data have enough dry years, but these mostly last only for 1 year. For example, the frequency of just one-year dry year in the GWB data is too high (13 times under m – 0.5 s and RCP-4.5 scenario condition), but the frequency of two-year consecutive dry years is far smaller (just once under the same condition). The probability of three-year consecutive dry years is technically zero. This problem is slightly alleviated in the WAMIS data, but still, the frequency of just one-year dry year is high. This result is obviously opposite to the observed.

The results in this table also confirm that the future data underestimate the risk of consecutive dry years. For example, when considering m – 0.50 s as the threshold, the mean duration of dry years in the observed data was estimated to be 2.5 years. However, the mean duration was estimated to be just 1–2 years in the generated data. This problem becomes worse in the future data. The mean duration of dry years was estimated to be just 1–1.3 years depending on the generated data. When considering m – 0.75 s as the threshold, the results are also similar. In the generated data, the mean duration of dry years was just 1 year. Consecutive dry years rarely occur. If this generated data is true, then there will be no water shortage problem in the future. The current water supply system in Korea is capable of handling this type of one-year drought.

In this study, the Precipitation Runoff Modeling System (PRMS) model was used to analyze the rainfall-runoff processes in the Boryeong Dam basin. The PRMS model is a long-term runoff model developed by the United States Geological Survey (USGS) (Leavesley et al. 1983; Markstrom et al. 2015). The PRMS model can simulate various hydrological processes such as interception, infiltration, effective rainfall, direct runoff, interflow, and baseflow with given input data and model parameters. The input data consists of meteorological and ground data. Meteorological data include precipitation, maximum and minimum temperatures, solar radiation, and evaporation. Ground data include digital elevation models (DEMs), land cover maps, forest type maps, and soil maps. Ground data are used to determine the parameters representing surface characteristics.

In the PRMS model, the hydrological process consists of five steps. Firstly, precipitation reaches the basin surface, some of which is blocked and evaporated by vegetation. Secondly, the precipitation that reaches the ground surface is stored on the surface by the impervious layer, which can either be evaporated or drained. The remaining precipitation infiltrates the soil and enters the soil reservoir. Thirdly, water stored in the soil reservoir either evaporates or becomes runoff, while the rest enters subsurface or groundwater reservoirs. Fourthly, some of the water stored in the subsurface reservoir becomes the interflow (or intermediate runoff), and the remaining water enters the groundwater reservoir. Lastly, groundwater runoff occurs from the groundwater reservoir. The surface runoff, subsurface runoff, and groundwater runoff are combined to form the total runoff in the basin.

The parameters of the PRMS model are categorized into two types: easy-to-estimate parameters (EEPs) and difficult-to-estimate parameters (DEPs). EEPs are associated with basic input data, interception, surface runoff, and water behavior in the soil. These are determined by the characteristics of the target basin, such as climate, topography, soil, and land use. EEPs can be easily determined by compiling numerical data in GIS. DEPs are those related to evapotranspiration, surface runoff, the behavior of water in the soil, and the behavior of groundwater. DEPs are difficult to estimate precisely with given data. So, when estimating the initial values of those DEPs, the proposed values in the PRMS model or the estimates from previous literature can be used. Alternatively, parameters from other basins with similar characteristics to the target basin can be transferred and used.

As the area of the Boryeong Dam basin is small, the sub-basin division was not performed. Estimation of the EEPs was done first and then the DEPs. The hydrological survey report on the Guem River basin (MOCT 2006) was also considered in the parameter estimation, which includes a chapter on the parameters of the PRMS model. The data collected in the year 2017 were used for the parameter estimation, and the data collected in the years 2018 and 2019 were used for the validation.

The EEPs of the Boryeong Dam basin were estimated in GIS using numerical information like DEM, land cover map, forest type map, and soil map. First of all, the basin area, average elevation, latitude, longitude, and slope can be obtained using the DEM. The land cover map is used to calculate the impervious area of the basin. The forest type map is used for determining the parameters related to interception. Finally, the soil map is used to calculate parameters related to soil.

For determining the DEPs, a three-step process is carried out, which includes the calibration for total runoff volume, calibration for peak runoff, and calibration for the recession limb. First, for the calibration of total runoff, an evapotranspiration-related parameter was used. In case the observed and simulated total runoff volumes are similar, this step is not necessary. After adjusting the total runoff volume, the calibration for peak runoff was carried out. The parameters related to surface runoff and soil internal conditions were adjusted to control the peak runoff. Finally, calibration for the recession limb was performed by adjusting the groundwater-layer parameter.

Table 6 compares the basic statistics of the observed and simulated runoff results from 2017 to 2019. In 2017, the parameter estimation period, the simulated value was 2.6 m³/s, which was very close to the observed average runoff of 2.7 m³/s. The difference in the total runoff volume was also less than 10%. Based on the Ministry of Construction and Transportation (MOCT 2006), less than a 10% difference in the total runoff volume is assumed to be an appropriate simulation result. The RMSE between the simulated and observed runoff was also very small at just 2.9 m³/s.

The result in the validation period of 2018–2019 was also similar to 2017. The simulated mean runoff was 3.86 m³/s, which was also very close to the observed value of 3.60 m³/s. The simulated runoff peak was 177.7 m³/s, which is also similar to the observed value of 186.9 m³/s. The total runoff volumes were also similar to each other, whose difference was less than 10%. However, the RMSE was rather high at 7.3 m³/s, which seemed to be related to the higher variation of rainfall and runoff in these 2 years. Due to that, the standard deviation of runoff in these 2 years was much higher.

Availability of water resources depends on the dam basin area, dam reservoir size, and reservoir operation method. For example, it is advantageous to have a larger dam basin area to collect the water. However, if the dam reservoir is small, most dam capacity may be allocated for flood control. Smaller dams are thus general in small dam basins, and large dams are general in large dam basins. The mismatch between the dam basin area and dam reservoir size makes it difficult to secure more water resources. This problem can be alleviated a little by effectively operating the dam reservoirs. The Boryeong Dam basin is a small dam in a small dam basin. The dam reservoir is also small, which is known to be very vulnerable to drought (Jeong et al. 2016).

Five methods are typically mentioned as reservoir operational method (ROM). Those include Auto (automatic) ROM, SRC (spillway rule curve) ROM, Rigid ROM, Technical ROM, and SRD (scheduled release discharge) ROM (KICT 1996). Each dam should select one method by considering various factors like climate, dam reservoir, and dam basin. First, Auto ROM is a simple method. Until the target water level is reached, no discharge is allowed. After that, the discharge is determined by the spillway discharge rating curve. SRC ROM involves the gate control to mitigate the peak flow of design flood. The so-called spillway rule curve is used to control the flood volume. SRC ROM is known to be effective when the dam inflow is similar to the design flood. On the other hand, if the inflow is very different from the design flood, the downstream flood risk can become bigger. Rigid ROM is the most popular method in Korea. The method is also called the constant-rate and constant-magnitude method. That is, the discharge is determined by multiplying the constant rate by the inflow until the peak inflow, and after that, the peak discharge (i.e., the peak inflow multiplied by the constant rate) is discharged. Technical ROM is known to utilize the flood control volume most effectively if the predictive inflow hydrograph is available. That is, the discharge can be optimally determined by considering the predicted inflow hydrograph. Finally, SRD ROM fully uses the past experience of dam operation. Even though this method is very primitive, it can also be effective under the condition that the past experience contains various extreme conditions.

The Boryeong Dam also adopts Rigid ROM for its reservoir operation. To apply Rigid ROM, it is first necessary to determine the constant rate (R). This constant rate is determined to satisfy the condition that the flood control volume is filled by some portion of the dam inflow hydrograph. That is, the constant rate is determined under this condition as the ratio of discharge out of the dam inflow. The trial and error method is generally applied to determine the constant rate, which is 0.586 for the Boryeong Dam (K-water 2018).

The dam inflow data was simulated by the PRMS model along with the parameters determined in the previous chapter ‘Model parameters and runoff simulation’. Daily simulation was done with the input of daily rainfall, and daily maximum and minimum temperatures. Both the observed and generated data based on the RCP-4.5 and RCP-8.5 scenarios were considered. Especially the rainfall data used as input here were those corrected ones in the previous chapter ‘Overall trends of future rainfall data’. The inflow data were then applied to the Rigid ROM of the Boryeong Dam and derived the daily dam reservoir storage and the daily amount of water supply.

As the reservoir operation was done with the same initial condition observed on 1 January 2009, the behavior of the dam reservoir storage at the beginning of the simulation was all similar. However, in the simulation results with the observed rainfall and temperature data, even though the simulation period was short, the dam water level reached a low water level. The planned water supply could not be satisfied during that period. On the other hand, the simulation results with the generated rainfall and temperature data based on RCP scenarios were totally different. Under the RCP-4.5 scenario, even though the generated data period was much longer than the observed data period, there was no case that the water shortage problem occurred. Only in the case of considering the WAMIS data, the dam water level approached the low water level just once. A similar result was also found under the RCP-8.5 scenario. There was no water shortage problem in the GWB data, and only once was observed in the WAMIS data. During the entire simulation period, the dam water level was found to have remained very high. In fact, this situation is a very promising one, if it is true. There will be no water shortage problem in the future. Dam reservoirs will always be full of available water.

However, this promising result ironically indicates the problem of future climate data. As analyzed in the previous chapter, the future climate data rarely contain a multi-year dry situation. The future drought is mostly short less than or equal to 1 year. This long-term characteristic of future rainfall data results in the simulation result that there will be no water shortage problem. As mentioned earlier, the current water supply system in Korea can overcome the one-year drought. Only the Han River Basin can overcome the two-year drought. Based on historical data analysis, this two-year drought occurred once every 10–30 years (KICT 2002; Yoo 2006; Yoo et al. 2008; Kim et al. 2011).

This study evaluated the appropriateness of GCM-based future climate data available in Korea. These data are those from GWB, WAMIS, and KMA. The evaluation of future climate data was mostly focused on the rainfall data, which was done from three different aspects. First, the basic characteristics of rainfall events such as the number of rainy days, rainfall intensity, rainfall duration, and no-rain duration were evaluated along with those observed. Second, the long-term behavior of annual rainfall data was evaluated with a focus on severe drought. The occurrence and persistence characteristics of dry years were evaluated by applying the Poisson process. Finally, the runoff was simulated for the study dam basin using the PRMS (Precipitation Runoff Modeling Simulation) model, and the dam water level was simulated by considering the adopted reservoir operation method. The simulated result was then used to evaluate the possible problem of water supply in the future. This study considered the Boryeong Dam basin in Korea as a study basin. Summarizing the results is as follows.

First, the generated future rainfall data showed rather smooth long-term increasing behavior from the observed rainfall data, except for the KMA data. The KMA rainfall data is about 50% higher than the observed, but the GWB data is just about 10% and the WAMIS data is about 20%. For both the GWB and WAMIS data, even though a little increasing trend was found in the future rainfall data, the range of the annual maximum daily rainfall was smaller than the observed. Also, the future rainfall data were found to have too many but small rainy days. The number of rainy days and the rainfall duration were about twice as high as the observed data. As a result, the rainfall intensity and the no-rain duration were about a half.

Second, the future rainfall data showed that the frequency of dry years was similar to the observed. However, in the future data, the occurrence of one-year dry years was found much higher, but far less occurrence of multi-year dry years was observed. It was very rare to have consecutive two-year or three-year dry years. Simply, the occurrence of very severe drought seemed to be significantly decreased.

Third, the future climate data showed that there would be no water supply problem in the future. In the observed data, the water level of the Boryeong Dam fluctuated greatly from the high water level to the low water level. However, in the application of the future climate data, the water level rarely fell below the 50% line between the high water level and the low water level. Almost every year, the water level touched the high water level to keep the dam reservoir nearly full. Only the WAMIS data under the RCP-8.5 scenario produced one water shortage case during the future simulation from 2022 to 2099.

The above summarized result must be very promising. However, it seems unacceptable as future climate change proposes higher flood risk and more difficulty in securing enough water resources. Many previous studies in Korea also showed that the drought in the future can be severe (Kwak et al. 2015; Ahn et al. 2016; Sung et al. 2018; Won & Kim 2020). For example, Sung et al. (2018) mentioned that the frequency of drought in the future may not be increased, but it can be severe with higher severity and longer duration. Won & Kim (2020) evaluated the relative effect of rainfall and temperature on drought in the future and concluded that the future drought could be severe mainly due to the temperature rise.

The result of this study that there would be no severe water supply shortages for the next 80 years until 2100 was mainly from the future rainfall data without any long-term high fluctuations. This result, on the other hand, clearly indicates the problems of the future rainfall data. Simply, the GCMs may not properly produce future rainfall data, especially if they lack the ability to generate the long-term characteristics of rainfall, such as the multi-year dry years. It is thus disappointing that these future rainfall data may not be used to evaluate the water supply system in the future, at least in the Boryeong Dam basin.

As this study was focused on the Boryeong Dam basin, the results derived in this study may not be generalized without any future applications to other basins. It is also possible to derive somewhat different results in other basins. However, the key issue is, if the future climate data reproduce the multi-year droughts. The authors of this study will also evaluate the other future climate data, as well as those for other basins.

This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (No. NRF-2021R1A5A1032433).

All relevant data are available from an online repository or repositories. (GWB: http://gwb-ccaw.re.kr/fcst/fcstFutrData.do. WAMIS: http://www.wamis.go.kr/wkc/wkc_download.do. KMA: http://www.climate.go.kr/home/CCS/contents_2021/32_2_user_analysis.php.)

The authors declare there is no conflict.