Future Hydrological Drought Analysis Considering Agricultural Water Withdrawal Under SSP Scenarios

Hydrological drought is assessed through river flow, which depends on river runoff and water withdrawal. This study proposed a framework to project future hydrological droughts considering agricultural water withdrawal (AWW) for shared socioeconomic pathway (SSP) scenarios. The relationship between AWW and potential evapotranspiration (PET) was determined using a deep belief network (DBN) model and then applied to estimate future AWW using projections of the twelve global climate models (GCMs). 12 GCMs were bias-corrected using the quantile mapping method, climate variables were generated, and river flow was estimated using the soil and water assessment tool (SWAT) model. The standardized runoff index (SRI) was used to project the changes in hydrological drought characteristics. The results revealed a higher occurrence of severe droughts in the future. Droughts would be more frequent in the near future (2021–2060) than in the far future (2061–2100) and more severe when AWW is considered. Droughts would also be more severe for SSP5-8.5 than for SSP2-4.5. The study revealed that the increased PET due to rising temperatures is the primary cause of the increased drought frequency and severity. The AWW will accelerate the drought severities in the future in the Yeongsan River basin.


Introduction
Droughts are complex natural processes that begin with a lack of precipitation and lead to a lack of soil moisture due to decreased river flows, and they adversely affect plant and crop growth and human lives. Drought indices are generally used to assess drought conditions in an area and to examine the various characteristics of droughts, including their severities, which is the ratio of stream water use to accredited stream water permits. The model was applied to project future AWW levels from the PETs estimated by the twelve GCMs for two scenarios, . The selected GCMs are frequently used for climate studies in East Asia and provide good performance when applied to Korea (Kim et al. 2021a, b). Although the spatial resolutions of GCMs are all different, as shown in Table S1, the GCMs used can show good performances after spatial downscaling and bias corrections, as presented in Song et al. (2021). The soil and water assessment tool (SWAT) was used to simulate river runoffs, and the SRIs with and without AWW were compared for two future periods (e.g., 2021-2060 and 2061-2100). Finally, the uncertainties in future hydrological droughts were analyzed using the REA method, which has been commonly used in uncertainty analyses for various climate projections.

Procedure
The procedure used in this study is shown in Fig. 1. This study consists of a total of six steps. The first step was to conduct bias corrections of the simulations of the twelve GCMs using the quantile mapping method. The performances of the bias corrections were evaluated by using several statistical performance indices. The second step was to formulate the SWAT model for the Yeongsan River basin by using historical observed data and AWW. The SWAT parameters were calibrated using the SWAT-CUP (Calibration and Uncertainty Procedure; Abbaspour et al. 2007) based on the observed discharges at Geukrakgyo (bridge), as shown in Fig. S1. The third step was to estimate the future AWW outcomes using GCMs that were based on the derived relationship between PET and AWW for the observation period. The PETs were calculated using Thornthwaite's (1948) method, which has been commonly used due to its high performance (Song et al. 2022;Aschonitis et al. 2022), and the relationships were quantified using the DBN. The fourth step is to estimate the river flow for twelve GCMs and two SSP scenarios. The fifth step was to calculate the SRIs and analyze the future drought characteristics. The future period was divided into the near future (NF)  and far future (FF) (2061-2100). Step 1 Step 2 Twelve CMIP6 GCMs (bias corrected data) Step 3 Twelve CMIP6 GCMs (bias corrected data) Future climate and runoff analysis considering future AWW SSP5-8.5 SSP2-4.5 Step 4 Step Finally, uncertainty analyses were performed using REA for the runoff and drought indices with or without AWW. Several studies have estimated AWWs using meteorological factors, population, and agricultural production (He et al. 2021;Zhang et al. 2021). However, this study considered only the meteorological factors because the population and agricultural production showed significant variations for the past period.

Study Area and Datasets
The study area consists of the Yeongsan River, which has an area of 3,371.4 km 2 . The average annual temperature and rainfall amounts are 14.0°C and 1,293 mm, respectively. The area consists of forest (45.4%), agriculture (35.5%), urban areas (7.3%), grasslands (5.2%), water bodies (3.4%), bare lands (1.8%), and wetlands (1.4%). The Yeongsan River Basin, which is located in the southwestern region of Korea, is known as a large-scale agricultural area, even after industrialization. The most common major crop is rice, although pears, onions, and garlic are also grown. The Yeongsan River often lacks sufficient river flows during the dry season. This study used climate data from the Korea Meteorological Administration, topographic data from the Environmental Geographic Information Service (EGIS), and hydrology data from the Water Resource Management Information System (WAMIS) and the Yeongsan River Flood Control Office. As shown in Fig. S1, the Yeongsan River Basin consists of 21 subbasins, which are divided based on the digital elevation model (DEM). The data from six meteorological stations near or in the study area and from one water-level station were used in this study.

GCMs and Shared Socioeconomic Pathway
GCMs include physical processes in the atmosphere, oceans, glaciers, and the surface, which make them useful for analyzing climate change and estimating future climates due to the increasing concentrations of greenhouse gases. GCMs provide climate information at a grid spacing distance of a minimum of 125 km and maximum of 400 km. The IPCC developed new climate scenarios for Assessment Report 6 (AR6), which consider social and economic factors together. New SSP scenarios were defined that considered various land use and greenhouse gas emission conditions. Therefore, compared to the existing RCP scenarios, the SSP scenarios are being upgraded or improved. This study used 12 CMIP6 GCMs, described in Table S1, for SSP2-4.5 and SSP5-8.5.

Quantile Mapping Method
Since GCMs are composed of data in the form of grids, differences in precipitation values and hydrological factors occur when compared to the actual observed points. Many methods have been used for bias corrections of GCM data (Song et al. 2020). Quantile mapping is an effective representative method to correct the difference between the GCM-simulated and observed values (Dosio and Paruolo 2011). In this study, spatial interpolation of the GCM grid data was performed using the inverse distance weighting method, and bias corrections were then applied using quantile mapping. As shown in Eq. (1), the quantile mapping method transforms a model variable ( P m ) by using the integral probability transformation function such that the newly constructed distribution becomes equal to the distribution of the observed variable ( P O ).
where P O is the observed precipitation, P m is the GCM precipitation, and h is the transformation function. Thus, the observed precipitation is calculated as the inverse function of the cumulative distribution function (CDF), as expressed in Eq. (2), where F m is the CDF of P m and F O −1 is the inverse function of the CDF of P 0 .

SWAT and SWAT-CUP
The SWAT can simulate runoff, sedimentation, and nutrients in watersheds due to alterations in land use, soil types, and land management conditions. The SWAT model was used in this study to estimate the flows under future climate change scenarios. There are two methods for parameter optimization of SWAT models: manual calibration and automatic calibration. The SWAT model provides a manual calibration function, but the calibration results may vary depending on the user's knowledge and skill level. SWAT-CUP was developed to perform automatic calibrations of SWAT parameters. Parametric optimizations of the SWAT model can simulate runoff from a wide range of large-scale watersheds by using observational data.
SWAT-CUP proposes various algorithms (including PSO, PARASOL, GLUE, MCMC, and SUFI-2), but the most frequently used method for daily and monthly analyses is SUFI-2 (Abbaspour et al. 2007), which has also been used in recent studies (Ahmed et al. 2022). SUFI-2 is an automated solution that makes it easy to carry out the calibration process under the time constraints that can be accomplished (Sloboda and Swayne 2011). Yang et al. (2008) found that SUFI-2 requires the smallest number of model runs to achieve good calibration and prediction uncertainty results. Therefore, this study used the SUFI-2 algorithm. McKee et al. (1993) selected the gamma distribution function (GDF) for fitting monthly precipitation data series and suggested that this procedure can be applied to other variables relevant to drought (e.g., streamflows or reservoir contents). The GDF was defined by Thom (1966) as follows:

Standardized Runoff Index
where and are shape and scale parameters, respectively, and Γ is an ordinary GDF. These and parameters can be determined by following the equation shown in supplementary S1 and S2.
The SRI (Shukla and Wood 2008) was developed to assess hydrological droughts while considering runoff data based on the SPI methodology. The SRI is an index that can be calculated similarly to the SPI by using only runoff data. Similar to the SPI, the SRI can calculate drought indices at daily or monthly scales by using observed and predicted runoff data. A hydrological drought defined by the SRI is extreme if this value is less than -2, severe if it is less than -1.5, moderate if it is less than -1, and mild if it is less than 0. An SRI greater than 0 indicates no drought. (1)

0.7 Reliability Ensemble Average
The REA, which was developed by Giorgi and Mearns (2003), can evaluate the similarities between GCM outputs and observed data for historical periods and can consider the differences in GCM projections for future periods. Therefore, it quantifies the reliability by considering both historical and projected data, as shown in Eq. (4).
where w i is the weight of the GCM simulation and w B,i is an error term between the GCM simulations and observations in the historical period. In addition, w D,i represents the difference between the GCM projection and historical period. In this process, the initial value of w D,i is calculated as where GCM is the difference between the averages of a GCM for the historical period and observed data; is the difference between the maximum and minimum values of the moving average; and m and n are the weights for the error and difference terms, respectively.

Bias Correction for GCMs
The performances of the bias-corrected precipitation and temperatures for twelve GCMs are shown in Fig. S2 and Tables S2 and S3. The results show improvements in the precipitation and temperature results of the GCMs after the bias corrections. The coefficients of determination (R 2 ) of the GCMs increased from 0.04 to 0.99 for precipitation. The root mean square error (RMSE) for the GCM precipitation decreased from 16.29 to 1.93. The standard deviation (SD) of the bias-corrected GCMs also becomes closer to the SD of the observed precipitation. In the case of temperature, the R 2 of the GCMs increased from 0.88 to 1.00, and the RMSE decreased from 4.64 to 0.05. The SDs of the bias-corrected GCMs become close to the observed SDs. For the RMSEs, the MPI-ESM1-2-LR precipitation exhibited the best performance of 15.33, and IPSL-CM6A-LR exhibited the worst performance of 17.95 before the corrections. After the corrections, MRI-ESM2-0 exhibited the best performance of 0.5, and INM-CM4-8 exhibited the worst performance of 5.16. For the SDs, IPSL-CM6A-LR exhibited the best performance of 12.29, and KIOST-ESM exhibited the worst performance of 6.99 before the corrections. After the corrections, all GCMs exhibited SDs that were close to the observed SDs. For the R 2 values, all GCMs exhibited low performances, e.g., 0−0.1 before the corrections, but the performances exceeded 0.9 after the corrections.
For the temperatures, MIROC6 exhibited the lowest RMSE of 3.87, and KIOST-ESM exhibited the highest RMSE of 6.95 before corrections. After the corrections, all GCMs exhibited RMSE values of 0.06 or less. For the SDs, IPSL-CM6A-LR exhibited the best performance of 11.58, and KIOST-ESM exhibited the worst performance of 5.15 before the corrections. After the corrections, all GCMs exhibited SDs greater than 9.40 or near the observed SDs. All GCMs before the corrections exhibited R 2 values of approximately 0.75 before the bias corrections, while after these corrections, they exhibited R 2 values of 0.9 or greater. The daily GCM data performances in this study area were best for MRI-ESM2-0 for precipitation and ACCESS ESM 1-5 for temperature.

SWAT Formulation
After considering the AWWs in the Yeongsan River basin, the SWAT parameters were optimized by using SWAT-CUP. The SWAT parameters that are related to groundwater levels, HRU, watersheds, soil, and channel routing affect the runoff process. The NSE was used as an objective function in SUFI2 for optimizing the SWAT parameters. Several studies recommend 500 to 1500 repetitions to optimize the SWAT parameters when using SUFI2. This study used 1000 repetitions.
The optimized values of the SWAT parameters are shown in Table S4. Figure S3 shows the calibration and validation results at Geukrakgyo station for 2012-2021. The results showed that R 2 increased from 0.75 to 0.92, and NSE increased from 0.53 to 0.84. The RMSE, Pbias, and RSR values decreased from 34.64 to 19.86, from -12.00 to -1.90, and from 0.69 to 0.39, respectively. For the validation period, the results were R 2 =0.90, NSE= 0.80, RMSE= 26.13, Pbias= -3.00, and RSR= 0.45. As a result, the SWAT model was calibrated well for the river runoff simulations.

Estimation of Future AWW
The parameter settings for the deep learning model were determined for the AWWs, as shown in Table S5. The monthly PET values for 2011-2019 begin to increase in April and reach their maximum values in July and August, accounting for 55% (310.4 mm) of the annual total PET (787.8 mm) (Fig. S4a). The AWWs reflect the annual cycles of agricultural activity; these values begin to increase in May and reach their maximum values in September (Fig. S4b). The annual average AWW was 26.7%, and the maximum was 63.2% in June. The greatest AWW reflects the extreme drought that occurred in 2015.
The RMSE, NSE and R 2 values of the model in this study were found to be 0.14, 0.87, and 0.97, respectively, confirming its good performance. The DBN model was then used to estimate the future AWWs from the projected PETs that were estimated for the SSP scenarios, as shown in Fig. 2. The future AWWs show the annual cycles of Korean agricultural activity from May to September. For SSP 2-4.5, the AWWs in NF during the farming season increased by an average of 2.6% compared to the reference period. The largest increase was 4.2% in July. Similarly, the AWWs in FF showed increases with an average of 3.2% compared to the past. The largest increase was 4.8% in July. During the nonfarming season, the AWWs were projected to increase by an average of 2.6% in NF and 2.8% in FF. The future AWWs were projected to decrease only in April on average by 4.6% in NF and 3.8% in FF.
For SSP 5-8.5, the AWWs in NF during the farming season increased by an average of 3.6% compared to those in the past. The largest increase was 5.1% in July. NorESM2-LM projected the highest increase in AWW, with an average of 4.5%, and KIOST-ESM projected the lowest increase of 1.8%. The AWWs in FF were projected to increase by an average of 4.4% compared with the reference period. The largest increase was 5.9% in July. Similar to NF, NorESM2-LM exhibited the largest increase in AWWs, with an average of 5.3%, and KIOST-ESM exhibited the smallest increase, with an average of 3.2%. During the nonfarming season, the AWWs were projected to increase by 2.7% in NF and 3.1% in FF. The AWWs were also projected to decrease in April, similar to SSP2-4.5.
The future AWWs were estimated to be higher for SSP5-8.5 than for SSP2-4.5 and larger in the FF than in the NF. In addition, these values were projected to increase more in the farming season than in the nonfarming season. The largest increase was projected in the summer month of July. The increased AWWs mainly resulted from the global warminginduced increases in the average monthly temperatures.

Estimation of Climate Variables
The seasonal future precipitation and temperature changes are shown in Fig. S5. The future precipitation amounts in NF were projected to increase by an average of 12.7% compared to the past for SSP2-4.5. The precipitation levels in spring and winter were projected to increase the most, by 93.5% and 67.3%, respectively, while the levels were projected to decrease in summer and fall by 12.9% and 17.8%, respectively. The future temperatures in NF exhibited an average increase of 1 °C compared to the past. The highest increase was projected in winter by 2.2 °C.
The future precipitation levels in FF were projected to increase by an average of 17.6%. These followed the same trend as NF. The results exhibit the greatest increases of 102.9% in spring and 79.1% in winter and largest decreases of 9.4% in summer and 16.5% in fall. The annual average temperatures were projected to increase by 2.2 °C in FF, with the greatest increase of 3.3 °C in winter.

River Flow Projections
The future runoff changes were calculated by season for twelve GCMs, as shown in Fig. S6. For SSP2-4.5, the annual average flows in NF were projected to increase by 37.2%. The flows in spring would increase the most by 213.3% and decrease the most in fall by 13.9%. INM-CM5-0 exhibited the largest increase with an average annual flow of 57.7%, while CanESM5 exhibited the smallest increase of 20.1%. The average annual flows in FF exhibited increases of 45.1% compared to the reference period. The flow variations for each season were similar to those in NF. These variations were projected to increase by 238.5% in spring, 24.8% in summer, and 100.3% in winter and to decrease by 11.5% in fall. NorESM2-LM projected the largest increase in annual average flows of 70.7%, and CanESM5 projected the lowest increase of 33.3%.
The results for SSP5-8.5 were similar to those for SSP2-4.5. However, the ranges of the increases and decreases were larger. The annual average flows in NF was projected to increase by 32.1%, with increases of 200.1% in spring, 16.1% in summer and 69.9% in winter and a decrease of 17.1% in fall. IPSL-CM6A-LR exhibited the largest increase of 50.7%, and CanESM5 exhibited the smallest increase of 10.9%. The average annual flows in FF showed an average increase of 50.8%. IPSL-CM6A-LR exhibited the largest increase of 76.9%, while ACCESS ESM 1-5 exhibited the smallest increase of 18.9%. In NF, the increase for SSP2-4.5 was larger than that for SSP5-8.5, while the opposite was true in FF.

Estimation of Future Runoff Considering AWW
The results of runoff changes when considering future AWWs for different SSPs by season are shown in Figs. 3 and S7. The annual average flows exhibited decreases with increases in AWW. For SSP2-4.5, the flows in NF exhibited an average decrease of 4.0%. The decreases were the highest in summer (6.7%). KIOST-ESM exhibited the largest decrease in average runoff of 4.7%, with a decrease of 9.5% in summer, while IPSL-CM6A-LR exhibited the smallest decrease of 3.0%. In FF, the runoff was projected to decrease by an average of 4.3%, with the largest decrease in summer of 7.1%. On the other hand, MIROC6 exhibited the largest decrease of 5.2%, and KIOST-ESM exhibited the largest decrease of 9.6% in summer.
The annual average runoffs for SSP5-8.5 were projected to decrease, which were similar to SSP2-4.5. SSP5-8.5 also exhibited significant decreases in summer flows. During NF, the runoff was projected to decrease by 3.9% on average, with the largest decrease of 6.5% in summer. ACCESS ESM 1-5 and KIOST-ESM exhibited the largest decreases of 4.9% and 8.5%, respectively, in summer. In FF, the runoffs were projected to decrease by 5.2%, with a decrease of 8.0% in summer. CanESM5 exhibited the largest decrease of 13.9%, with a decrease of 18.0% in summer.

Comparative Analysis of Historical and Future Droughts
The annual average numbers of drought days with different durations during the historical period are shown in Fig. S8. The annual average numbers of drought days differ by duration for the reference period. Severe droughts (SRI < -1.5) increase at a high rate with increasing duration. The periods of moderate drought decreased from 62.1 days for 270-day duration droughts to 58.0 days for 90-day duration droughts, while those for severe drought increased from 13.0 days to 28.3 days. We also found that more severe droughts occurred as the durations increased.
The future annual average numbers of drought days and future severe drought ratios with and without AWW for SSP2-4.5 and SSP5-8.5 are shown in Fig. 4. For SSP2-4.5, the number of drought days decreased from 60.7 days to 58.5 days on average as the durations increased from 90 to 210 days. IPSL-CM6A-LR exhibited the largest decrease of 8.9 days, while MPI-ESM1-2-LR exhibited the smallest decrease of 2.2 days. On the other hand, several GCMs exhibited increases in drought days. INM-CM4-8 exhibited the largest increase of 10.4 days. In addition, the rate of severe drought occurrences decreases as the duration increases. INM-CM4-8 exhibited the largest decrease of 6.7%, while INM-CM5-0 The characteristics of drought occurrences were different when considering AWW. The numbers of drought days were projected to decrease by 2.8 days to 2.2 days, while the severe droughts decreased by 0.7% to 0.5%. This means that the drought periods increased as the durations increased when AWW was considered. The rates of occurrence for severe drought also increased.
For SSP5-8.5, the average numbers of drought days decreased from 62.5 to 57.3 days as the durations increased from 90 to 210 days. The drought periods were projected to decrease by all GCMs. NorESM2-MM exhibited the largest decrease of 8.8 days. The occurrences of severe drought decreased from an average of 39.8% to 37.4%. INM-CM4-8 exhibited the largest decrease of 14.3%, while MRI-ESM2-0 exhibited the largest increase of 7.3%. When AWW was considered, the numbers of drought days decreased from 5.2 to 4.8 days, and the severe drought rate decreased from 2.5% to 2.3%. This means that the drought periods increased as the duration increased, and the rates of occurrence for severe drought also increased when AWW was considered. For SSP 5-8.5, the numbers of drought days decreased for all GCMs when AWW was considered. The greatest decrease was projected by ACCESS ESM 1-5 from 90 to 2.8 days.
All GCMs projected increases in severe drought periods compared to the historical period. Although the drought periods would decrease in the future, the severe drought frequencies may increase. For all durations, future droughts would have higher rates of occurrence for severe droughts. FAWW mostly occurs during the farming season. In addition, the seasons in which droughts occur are generally winter and spring, and when FAWW is considered, the differences between the maximum and minimum values in winter and spring river flow are reduced.

Comparative Analysis of Drought According to Two Future Periods
The future annual average numbers of drought days and severe drought ratios in SSP2-4.5 and SSP5-8.5 are shown in Figs. S9 and S10, respectively. For SSP2-4.5, droughts were projected to occur more frequently in NF than in FF. However, the severe drought frequencies were projected to be higher in NF than in FF. The differences became more significant as the durations increased. The numbers of drought days in NF were projected to decrease in FF by an average of 8.5 days without AWW and by 8.6 days with AWW. CanESM5 exhibited the largest decrease in drought periods (33.5 days) from NF to FF without AWW. In contrast, IPSL-CM6A-LR projected the largest increase of 25.0 days. The drought periods using INM-CM5-0 exhibited an increase of 1.4 days with AWW, but these decreased by 1.7 days using MIROC6. Severe droughts were projected to decrease by an average of 0.5%. The decrease was projected to be 0.8% when AWW was considered. MPI-ESM1-2-HR exhibited the largest decrease in severe drought periods by 12.3%, and INM-CM5-0 exhibitedthe largest increase by 8.4%. For the differences in drought occurrences with and without AWW, MIROC6 exhibited the highest increase of 1.4%, and INM-CM5-0 exhibited the greatest decrease of 2.0%.
For SSP5-8.5, the drought days were projected to decrease by an average of 19.1 days. The decreases with AWW were projected to be as great as 21.1 days. All GCMs exhibited decreases in drought periods. MIROC6 projected the highest decrease of 30.3 days. The differences between the cases with and without AWW exhibited the largest increase for MIROC6 (0.9 days) and largest decrease for CanESM5 (17.8 days). The severe drought frequencies were projected to decrease by an average of 3.9% without AWW and by 4.0% with AWW. Without AWW, INM-CM5-0 exhibited the greatest decrease of 31.0%, and MRI-ESM2-0 exhibitedthe greatest increase of 12.5%. With AWW, MIROC6 exhibited the greatest increase of 3.1%, while CanESM5 exhibited the greatest decrease of 12.1%.
The future drought characteristics were different for different GCMs. Most GCMs projected that the droughts in NF would be more frequent and more severe than in FF. The reason is shown in Fig. S11. The temperatures in FF increase more than in NF and have values of 14.8 to 16.0 °C for SSP2-4.5 and 15.2 to 17.5 °C for SSP5-8.5. Thus, the PETs and AWWs increase, but in a relative sense, the precipitation levels in FF increase much more than in NF, and are projected to be 1558.0 to 1634.8 mm for SSP2-4.5 and 1523.4 to 1688.4 mm for SSP5-8.5. The monthly average flow rates in the two future periods are shown in Tables S6-S9. The drought characteristics were also different according to AWW considerations. Considering AWW, a decrease in drought days was projected from NF to FF for SSP2-4.5 but an increase from NF to FF was projected for SSP5-8.5. SSP2-4.5 projected a decrease in the severe drought rate from NF to FF, but SSP5-8.5 projected an increase.

Extreme Drought Index
The minimum values of the drought index for SSP2-4.5 and SSP5-8.5 are shown in Tables 1 and 2, respectively. The results for the GCMs are all different. For SSP2-4.5, the minimum drought index was projected in FF (INM-CM5-0 of duration 270 days -3.48) rather than NF (CanESM5 of duration 180 days -3.28). The drought indices with and without AWW showed greater average decreases of 0.06 in NF than the values of 0.05 in FF for the 90-day duration. KIOST-ESM projected the largest differences in the minimum 90-day duration drought index with and without AWW as 0.55 in NF and 0.43 in FF.
Similarly, SSP5-8.5 exhibited decreases in the differences in the minimum drought index of 0.07 in NF and 0.03 in FF for the 90-day duration. KIOST-ESM projected large differences in the minimum drought index of 0.64 in NF and 0.48 in FF for the 90-day duration.
The minimum drought index was analyzed differently based on the GCMs. On average, droughts were projected to be more severe in FF and more severe with AWW. The differences in the minimum drought index between NF and FF was also larger with AWW.

Uncertainty in Future Drought
The REA results for 12 GCMs for the two SSPs by future period are shown in Figs. 5 and S12. The results showed that the reliabilities were higher when considering AWW. In particular, this difference was analyzed more clearly in FF than in NF. For SSP2-4.5, IPSL-CM6A-LR exhibited the highest REA of 0.34 with AWW and 0.32 without AWW. For SSP5-8.5, NorESM2-LM exhibited the highest REA of 0.22 with AWW and 0.19 without AWW. This means that the reliabilities increased when AWW was considered when calculating future river flows. This indicates the necessity of considering AWW for drought analyses. Figure 6 shows the uncertainties for the future drought index by duration days that were obtained using 12 GCMs. The results exhibited increases in reliability with increases in drought duration. The greatest REA occurred for the 180-day duration. IPSL-CM6A-LR showed the greatest reliabilities, which were close to 1 for all durations. Ultimately, the reliabilities of the drought index were higher when AWW was considered.

Conclusion
This study proposed a new approach to identify the future hydrological drought status by considering future AWWs. The proposed method was applied in the Yeongsan River Basin. The future AWWs exhibited increases at higher rates for SSP5-8.5 than for SSP2-4.5. The average annual precipitation amounts were projected to increase over time but were more variable for SSP5-5.8 than for SSP2-4.5. The annual average temperatures were also projected to increase, but at higher rates for SSP5-8.5 than for SSP2-4.5. As a result, the annual average flows were projected to increase. However, the annual average flows exhibited decreases with AWW due to the increases in AWW. The decreases in average seasonal flows due to AWW consideration occurred most in summer, which caused changes in the SRI pattern in the basin.
The drought characteristics were projected to vary for different GCMs. The results showed that drought periods in the future might decrease compared to the historical period, but the severe drought periods would increase. In other words, future droughts are expected to be more severe than those during the present. Droughts would be more frequent and severe in NF than in FF. Future droughts showed different characteristics with AWW, and showed increased severities with increases in AWW. The results indicate that the increased PETs under the SSPs is the major cause of the increasing drought frequencies and severities. The reliabilities for future droughts were higher with AWW. In addition, the drought reliabilities in FF were higher than those in NF. The results confirm the requirement of considering AWWs for future hydrological drought analyses to increase the reliability of drought projections.
Hydrological drought analyses are important for water resource planning and management in various sectors, including agriculture and industry. In the future, more grounded drought analyses can be conducted by considering the water balance of the region and more newly released GCMs of CMIP6.