The variability in climate change is a crucial element in the hydrologic cycle. Slight discrepancies in climate can alter variations in the hydrologic processes of the hydrologic cycle [1
]. The effects of climate change are diverse, and they vary locally and internationally with their intensity and duration. Challenged with this realism of varying climate, law makers in expanded diverse institutions are progressively searching for quantitative descriptions of climate forecasting. Thus, they require projections of regional and climate changes that will influence humans, economies and ecosystems [2
]. Hence, general circulation models (GCMs) are the main mechanism for forecasting changes to future climate. Due to the intensity and severity of hydrological occurrences, it is globally recognized that the variations in climate can reshape the geographical and secular dissemination of water resources, thus causing extreme events such as droughts and floods [3
]. Therefore, the effects of climate variation on hydrological occurrences has been extensively studied [6
]. The hydrological influence of a changing climate on hydrology is usually analyzed using various climatological models with climate change events obtained from GCMs forced with emission scenarios. However, these results have been rarely used in management of water resources because of the existence of uncertainties in both future climate change projections from GCMs and assessments of climate variation effects on hydrology.
This study used 27 GCM data to quantify model uncertainty in three future periods to assess annual precipitation in the Cheongmicheon watershed in South Korea. Established from a physical theory, GCMs replicate observed characteristics of the recent climate, which are important mechanisms to predict future climate involving temperature and extreme precipitation for uncertainty [8
]. Describing and quantifying uncertainty in climate variation predictions is crucial not only for the sole aim of observation and acknowledgement but also for key perspectives to climate adaptation. The authors [9
] pointed out that uncertainty occurs in climate models predictions as a result of the uncertainty in predicting anthropogenic forcing (that is, the emissions scenarios or scenario uncertainty) and intermodal variations in physical parameterization process due to random differences and dependence on fundamental conditions. Hence, the precariousness on various factors should be scrutinized in a quantitative assessment.
Some studies on hydrologic impacts due to climate change have concluded that the choice of GCMs has a bigger impact on the hydrologic output compared to the choice of emission pathway [11
]. Moreover, the structural component of the hydrologic models is a vital part in the projected changes. Thus, the methodological context for climate change impacts on hydrology has a critical point that affects the projected outcome of future climate change as well as the adaptation or mitigation strategies that arise based on the information provided. Consequently, it is significant to evaluate the potential cause of the unreliability of the effects of climate variation studies and the outcome to a variety of impacts that result from the present state of science.
Moreover, some interdisciplinary studies, such as hydrology and climate hazard assessment, cannot meet the conditions of users who need to apply changes to extreme precipitation, and in order to close this loop, downscaling methods have been used and widely applied to various studies. The authors in [8
] also applied Bias-Correction/Spatial Disaggregation (BCSD) on extreme climate estimation over the north-eastern United States under three future scenarios. They indicated that downscaling performs differently for the three aspects of the eleven extreme indices, generally reproducing the character of temperature extremes better than precipitation extremes. Since statistical downscaling is usually versatile with less calculation, it can efficiently remove errors in historical simulated values. Similarly, data of this study was downscaled using two Bias-Correction/Spatial Disaggregation (BSCD) methods, namely, Simple Quantile Mapping (SQM) and Spatial Disaggregation with Quantile Delta Mapping (SDQDM) to preserve the long-term temporal trends in climate. This will provide useful insights that will be of interest to a range of decision makers as well as water managers in South Korea involved in the impacts of climate change hydrology and in the Cheongmicheon watershed.
To this end, this study employed a quantitative procedure considering the performance of the models and models averaging, known as Reliability Ensemble Averaging (REA). The REA is used to identify the uncertainty range and reliability of climate variation forecasting of 27 different GCMs of Coupled Model Intercomparison Project Phase 5 (CMIP5) for precipitation and temperature. This study used the term of ensemble referring to simulations of different individual GCMs and not to different attainments with identical model. Here, climate projections for all the GCMs under the Representation Concentration Pathway (RCP) 4.5 scenario were analyzed. The authors [13
] applied the REA method and took into account two reliability indicators that include model performance and model convergence. The former transcribes historical climate while the later acknowledges the best estimate of climate projection. Thus, to determine the model performance based on weights of Root Mean Square Error (RMSE) in quantifying uncertainty we follow the idea of [14
]. The authors used performance-based ensemble averaging technique on Regional Climate Models (RCMs) over South Korea by applying weights based on the inverse of the bias, RMSE and temporal correlation coefficient wherein they found out that the weightings are reduced for low model performance. Furthermore, in some preceding studies, performance-based ensemble methods by RMSEs have been found to significantly improve the ensemble averaging results [15
A number of studies have selected appropriate GCMs based on their performances in replicating past weather conditions. Nonetheless, these models have the constraint that the performances in a past period cannot assure the consistent performance in a later time [17
]. The concept of Multiple Model Ensembles (MMEs) to consider the uncertainty of climate change projection has been popularly applied in the hydrological impact analysis of climate change [17
]. Therefore, MMEs have been popularly used to apprehend feasible climate variation prediction by several models.
Furthermore, this study evaluated future drought severities for three general future periods to quantify uncertainty for all GCMs. Drought happens when there is little or no occurrence of rainfall over a long time and is most times referred to as meteorological drought and when this phenomenon keeps occurring, it generates agricultural, hydrological and later socio-economic drought [19
]. Thus, it is significant to evaluate drought severities at different intervals, intermittently over the year and then grasp drought impacts on numerous elements of the water cycle. Researches have been carried out across the world in modern days to evaluate distinct droughts at intervals of 1-, 3-, 6-, 12- and 24-months using Standard Precipitation Index (SPI), Streamflow Drought Index (SDI) and Standard Precipitation Evapotranspiration Index (SPEI) in the Loess region [21
]. The authors pointed out feeble trends with SPEI compared to SPI. The researchers [22
] analyzed the future drought of the Han River Basin using the RCP 8.5 scenario and showed that drought frequency will increase in that location, while [23
] projected the future drought in Korea using the RCP 8.5 scenario and found projected increases in both drought duration and severity. Therefore, this study aims to project future SPEI, SDI and SPI using RCP 4.5 scenario to evaluate drought severity. Since South Korea has been experiencing extreme droughts since 2014, this study investigated extreme future droughts under climate variation events.
A methodology for modeling GCM uncertainty as a result of the influence of climate variations on hydrology in the Cheongmecheon watershed is presented in this paper. The REA technique that is used in this study is vital for decision markers based on adaptation measures for climate change impact on hydrology with a significant curtailment in uncertainty due to performance-based ensemble averaging parameter. This curtailment in unreliability range, in contrast to other model performance and ensemble mean used in this work, proposed that REA is a feasible approach to determine future projections of precipitation in the watershed by reducing the contribution of poorly performing GCMs. The main ideology fundamental to REA method is to reduce the contributions of simulations that perform poorly in simulating current climate and future projections. Therefore, this study only extracted the most vital information from the multiple models simulated. The results indicate that model performance variability is seen as a point of uncertainty in the prediction of climate variation scenarios.
The suggested approach has a limitation of not taking into account the uncertainty due to model convergence from the REA. However, it should be acclaimed that weights given to the GCMs employing REA are based on model performance criterion. It is vital to note that the quality of the results presented in this study is based on the modeled performance criterion and weights based on the RMSE. Hence, the REA averaged relied on the quality of the observational data set, was used to determine the model bias. Therefore, our analysis does not consider model convergence criteria. The uncertainty range calculated using the REA method shows similarities across models but is intensified towards the 21st century.
Furthermore, as researchers are faced with often desperate climate change predictions from various GCM, it is essential to elaborate with the uncertainty around future projection of climate variation. It was also argued that, using various ensembles model information based on applied weights from the model performance criterion, it can also represent a vital feature of uncertainty lurking in climate change projection that should be discovered. Thus, the REA procedure gives a simple and versatile scheme to carry out such evaluation.
In addition, this study evaluates drought features using several GCMs for many intervals under RCP 4.5. The concept of weighting based on RMSE is recommended to lower uncertainty of climate predictions. The major outcomes of the present study can be shown as an incitement of the rainfall simulations using weights for multiple climate models. Projections of drought indexes show less uncertainty with SDI compared to SPEI and SPI. More significantly, the worst prediction occurred in 3-month duration of each index than long-term duration indicating that shorter rainy times could adversely determine water resources, with a broad effect for local human societies and ecosystems. The influence of these rainfall variations at the watershed level is necessary in order to develop plans in long-term water supply and demand, and thus to achieve sustainable management of water resources. Therefore, in future studies, especially those related to model spread, ensemble forecasting using more valid models and an evaluation of decreasing uncertainties is significant for a better comprehension of future variations.