In 2015, South Korea suffered one of the worst droughts in recent years. Seoul and Gyeonggi and Gangwon provinces experienced severe drought conditions, receiving less than 43 percent of the annual precipitation average of the past 30 years. Additionally, the 2015 summer precipitation was less than half of the average. The lack of summer precipitation induced serious shortages in dam storages, which are important supplies for the dry season. K-water, a public company managing South Korea’s public water supply system, is fighting to secure public water supply and minimize potential damage that may occur before the subsequent wet season. 

To date, K-water has been putting in a lot of efforts into developing a sustainable water supply based on conservative perspective, one that expects a 20-year low-flow frequency dam inflow. However, recent droughts threaten current dam operation rules and water supply systems. In addition, climate change and climate variability are increasing drought risk in South Korea. Therefore, it is urgently necessary to incorporate long-term dam inflow forecasts into their decision-making for more flexible dam operation plans and drought management strategies.

The purpose of this study is to make a guideline to fully utilize the long-term dam inflow forecasts for the dry seasonal multi-purpose dam operation. To this end, we first conducted a literature review on best practices for climate information use among water managers. Next, we developed a method that produces probabilistic dam inflow forecasts based on ensemble streamflow prediction (ESP) and APEC Climate Center’s dynamic climate forecasts (i.e. Drip). In addition, we seek to improve the predictability of the future areal averaged precipitation over Soyang dam basin. Finally, we looked at the current dry seasonal dam operation of K-water and proposed a guideline that probabilistic dam inflow forecasts can be to support the dam operation’s decision-making process.

The following results were drawn from our analysis.

1) Based on the literature review, we concluded some insights on enhancing the use of long-term dam inflow forecasts for dam operation. First, the effective integration of seasonal forecast in dam operation decision requires long-term collaborative research between information providers and practitioners. Second, forecast providers are needed to enhance the mutual understanding of the intrinsic process of dam operation decision-making and the process of the dam inflow forecast. Third, long-term research-based relationships between forecast provides and dam operators must be adequately funded and supported to develop effective decision-support systems. Finally, governments should provide incentives for innovation that facilitate the use of long-term dam inflow forecast in decisions.

2) This study detected significant decreasing trends (95% confidence interval) in dry-seasonal runoff rates (=dam inflow / precipitation) in three dams basins (Soyang, Chungju, and Andong). Changes in potential evapotranspiration (PET) and precipitation indices were examined to investigate potential causes of decreasing runoff rates trends. However, there were no clear relations among changes in runoff rates, PET, and precipitation indices. Runoff rate reduction in the three dams may increase the risk of dam operational management and long-term water resource planning. Therefore, it will be necessary to perform a multilateral analysis to better understand decreasing runoff rates.

3) To develop reliable dry-seasonal dam inflow forecast models this study evaluated the performance of three different hydrological models (i.e. GR4J, SAC-SMA, and PRMS). Because GR4J and SAC-SMA do not have snow and PET simulation, this study added a simple temperature-based snowmelt module and Oudin’s potential evapotranspiration method to these two models. Micro genetic algorithm (-GA) was employed to optimize the parameters of each hydrological model. We used five statistics based on Nash-Sutcliffe efficiency as the objective function. The results showed that three hydrological models using the square root Nash-Sutcliffe efficiency can simulate dam inflow during both the dry season and the full year period.

4) In order to select the best hydrological model we used two performance matrices, correlation coefficient and the root mean square error. SAC-SMA and PRMS showed better performances than GR4J for simulating dry-seasonal dam inflow of Soyang dam basin. Finally SAC-SMA was selected for the dam inflow forecast model based on three criteria: calculation time, number of parameters, and simplicity of model structure. To determine the warm-up period of SAC-SMA, we tested the conversion period of simulated flows using 300 randomly generated states variables in SAC-SMA. In conclusion, a three year period was enough to avoid the initial condition problem of SAC-SMA model simulation.

5) This study developed probabilistic dam inflow forecast methods, ESP and dynamic reservoir inflow prediction (DRIP). To evaluate the performance of ESP and DRIP we employed ranked probability skill score (RPSS) and probability of detection (or hit rate). ESP showed good predictability with high RPSS (>0.5) for every month and every lead times (1-month, 2-month, and 3-month). DRIP also showed good performance in predicting dam inflow for every month except June.

6) To improve the predictability of June precipitation forecast, we tried to estimate the predictability of climate variables forecasts including SLP, PREC, T2M, T850, U850, V850, and Z500. A five-fold cross validation method was used to identify reliable predictability of climate variables for forecasting Soyang’ June precipitation. The highest predictive regions for June precipitation were located near the Gulf of Mexico. Therefore, developing statistical models between predictive regions and observed rainfall will be necessary to improve the predictability of Soyang dam June precipitation.

7) Finally, this study proposed a guideline to utilize dam inflow forecasts in decision-making on dam operation regarding droughts. We suggested several scenarios that guide how long-term forecasts can be used in dam operation decisions in terms of each response stage and according to dam storage levels.

The probabilistic dam inflow forecasting technique developed in this study can contribute to the dry-seasonal multi-purpose dam operation decision-making by supporting useable information of long-term forecasts and its reliability. In addition, probabilistic dam inflow forecast will be useful to establish K-water’s monthly and annual dam operation plans, manage efficiently water supply and demand, and maintain dam storage level against drought. 

1. Introduction​

    1.1 Research background and Motivation 

    1.2 Research objectives

    1.3 Research Approach 

  ​

2. Literature review for water management decision-making based on climate forecast​

    2.1 The utilization of climate information for water management decision-making  

    2.2 INFORM project in California, USA​

    2.3 Regional studies

    2.4 Discussion and summary

3. Analysis of changes in dry-seasonal dam inflow​

    3.1 Study area and methodology 

    3.2 Trend in dry-seasonal dam inflow​ 

    3.3 Changes in potential evapotranspiration and precipitation indicators

    3.4 Discussion and summary

4. Improving the skill of dam inflow forecast​

    4.1 Hydrologic modeling

    4.2 Probabilistic dam inflow prediction

 ​

    4.3 Comparison of performance of ESP and DRIP

5. Toward improving precipitation forecast skill in the basin scale​ ​

    5.1 Hydrologic modeling

    5.2 Probabilistic dam inflow prediction

6. Development of guidelines for supporting dry-seasonal dam operation based on probabilistic dam inflow forecast​​

    6.1 Review of current dam planning and operation practices

    6.2 Guideline of utilizing dam inflow forecast into dam operation practices

    6.3 Recommendations

7. Conclusion​​

    7.1 Key results

    7.2 Implications for practice

    7.3 Challenges and remaining issues