연구보고서
- 저자
- 양유빈 박사
- 작성일
- 2016.01.23
- 조회
- 269
- 요약
- 목차
In South Korea multi-purpose dam serve important roles in contributing approximately 58 percent of annual water supply. Recent Korean dam operators are facing challenges in using traditional dam operation methods based on historical dam inflow data because of occurring more severe spring droughts in a changing climate. In 2014 longer-lasting drought from spring to early summer broke records in the lowest dam storage levels in several multi-purpose dams, inducing a restriction of water supply in some areas. To reduce the negative climate change and variability impacts on dam operation one of best options is that dam operators utilize long-range climate prediction information to have enough longer time to prepare for potential threat such as spring droughts.
This study developed two dry seasonal dam inflow prediction methods based on APEC Climate Center (APCC)’s seasonal climate prediction and teleconneciton between global climatic indices and historical dam inflow. To do this, we first evaluated the predictability of dry seasonal dam inflow forecast according to 10 climate models, 3 lead-times (i.e. 1-month, 2-month, and 3-month), five hydrological models with three different parameters’ set, and three domain areas that are used to consider the effect of using different grid number of climate prediction (i.e. 1-point, 4-point, and 9-point). Second, teleconnection between observed dam inflow and climatic indices obtained from the National Oceanic and Atmospheric Administration (NOAA) Climate Prediction Center (CPC) was analyzed to develop statistical seasonal streamflow prediction models. Finally, we proposed the hybrid approach that uses the combination of dynamical and statistical prediction information based on their predictability to improve the reliability of dry seasonal dam inflow forecasts. In addition, this study suggested to employ risk matrix to efficiently utilize uncertain seasonal dam inflow prediction into managing real dam operation and developing long-term dam management strategies.
The following results were drawn from our analysis.
1) Dam inflow forecast using APCC’s climate prediction (i.e. the dynamical approach) showed a significant predictability in every months with three different lead-times (0.05 confidence level). Especially, the predictability in January, March, and April with 1-month lead time showed best performance. However, the predictability were generally reduced as lead-time increases.
2) Using multi-model ensemble (MME) in the dynamical approach generally showed the better predictability compared to using the products by an individual model. MME showed the best performances or similar with the predictability of the best individual model when most of individual models showed positive correlation with observed value. However, the predictability of MME showed very low when some prediction showed positive correlation but the others had negative correlation.
3) The predictability of dam inflow was dependent on evaluation criteria (i.e. Pearson vs. Spearman rank correlation coefficients). Especially, the performance of using individual model product showed big difference by the selection of evaluation criteria. MME showed relatively less sensitive to the selection of criteria than using a individual model.
4) The performance of the dynamical approach was also highly affected by a combination of a climate model and hydrologic model selection. However, the predictability according to the prediction domain showed relatively small differences compared to the choice of climate model or hydrologic model.
5) Our results showed that the uncertainty in the dynamical seasonal prediction was mainly attributed to climate model prediction and hydrologic modeling. In the hydrologic modeling, the uncertainty caused by the different model structures was larger than those by model parameters. This finding indicates that the dynamical approach will be needed to seek optimal combination of climate models and runoff models to improve the predictability of seasonal dam inflow at a certain region.
6) Statistical approach using teleconnection between global climatic indices and dam inflow observation showed significant predictability in forecasting a dry seasonal dam inflow at the different lead-times (0.05 confidence level). In particular, March, April, and May showed a significant high predictability in the Soyang dam.
7) The verification of the statistical approach showed overall good performance for 2006-2011 which period was not used in developing the statistical models. This results indicated that statistical model has potential to apply for a real prediction, although further verification process using a longer observation will be needed to identify the availability of the statistical approach.
8) This study developed the hybrid approach for seasonal dam inflow forecast based on the dynamical and statistical prediction information. The statistical prediction can cover the wider range of lead times up to 12 months compared to those of dynamical approach (up to 3 months). For example, if a dam operator wants to predict March dam inflow in February, the dam operator can have the three dynamical forecasts (i.e. 1-month lead in February, 2-month lead in January, and 3-month lead in December) and eight statistical forecasts (i.e. 3-month lead to 12-month lead except 10-month & 11-month due to no significant forecast). Using 11 prediction the dam operator can understand mean March inflow and its uncertainty.
9) Finally, this study proposed a risk matrix to utilize seasonal dam inflow forecast in making decision on dam operation. This approach use a two-dimensional risk matrix that is consisted of 3 potential available water capacity (PAWC) and 3 likelihood ranges (i.e. low, medium, and high). PAWC is estimated by the function of current available dam storage, expected dam inflow (20yr low flow), and scheduled water supplies. The likelihood is defined in terms of probability of possible dam inflow based on dynamical and statistical prediction. The benefit of using the risk matrix is that it can easily identify the risk and allow for specified implementation of mitigation measures.