초록 ▼

본 연구에서는 기상청의 단기 수치예보모델과 K-water의 수문모델을 연계한 원스톱물관리 시스템을 구축하고 강수량 및 유입량에 대한 예측결과를 평가하였다.
LDAPS 예측 결과는 36시간 누가 강수량이 50 ~ 200 ㎜ 구간에서는 과소추정하고 있지만 물관리에 활용성이 충분한 것으로 분석되었다. UM3.0과 RDAPS예측 결과는 전반적으로 과소 예측을 나타내는 것으로 분석되었다. 관측 누가 강수량에 따른 RMSE를 분석한 결과 관측 누가 강수량이 커짐에 따라 RMSE도 증가하는 경향을 나타내고 있다.LDAPS 강수량 예측 결과

본 연구에서는 기상청의 단기 수치예보모델과 K-water의 수문모델을 연계한 원스톱물관리 시스템을 구축하고 강수량 및 유입량에 대한 예측결과를 평가하였다.
LDAPS 예측 결과는 36시간 누가 강수량이 50 ~ 200 ㎜ 구간에서는 과소추정하고 있지만 물관리에 활용성이 충분한 것으로 분석되었다. UM3.0과 RDAPS예측 결과는 전반적으로 과소 예측을 나타내는 것으로 분석되었다. 관측 누가 강수량에 따른 RMSE를 분석한 결과 관측 누가 강수량이 커짐에 따라 RMSE도 증가하는 경향을 나타내고 있다.LDAPS 강수량 예측 결과를 K-DRUM 모델의 입력자료로 활용하여 각 예측 별로 유입량을 계산한 결과 관측 유입량이 많아질수록 산포도가 커지는 경향을 나타내었다. UM3.0과 RDAPS 예측을 활용한 경우 전반적으로 유입량을 과소 예측하는 것으로 나타났다.LDAPS 예측 강수량을 COSFIM 모델을 통해 군남홍수조절지에 적용한 결과 횡산수위국의 초기 인명대피를 위한 예경보 수위인 1 m 이상을 상회할 수도 있다는 결과를 사전에 관리자에게 제공할 수 있는 것으로 분석되었으며, 이는 군남홍수조절지의 운영관리에 있어서 매우 유용한 정보가 될 것으로 판단된다.
그리고 장기 물관리를 위해 GloSea5 모델의 예측 앙상블을 수문모델과 연계하여 장기적인 물관리의 의사결정을 지원할 수 있는 아키텍처를 설계하였다. 이는 GloSea5 모델에서 생산된 앙상블로부터 수문모델에서 필요한 기상요소를 추출해서 하천유량관리시스템의 수문모델과 연계하여 앙상블 유량예측 방법을 적용할 수 있도록 구성되었다.또한 용담시험유역에서 관측된 증발산량 및 토양수분량을 이용하여 K-DRUM 모형의 모의결과와 비교분석하였으며, 기후예측모델에서 제공되는 미래 기상요소 예측값을 이용하여 증발산량을 추정할 수 있도록 모델을 개선하였다. 그리고 기존 잠재증발산량에 0.7을 일괄 곱하여 유역 전반의 실제 증발산량을 추정하던 방식에서 토양함수상태를 고려하여 차등계산하는 방식으로 개선하였다.

Abstract ▼

Ⅳ. Research details and results
1. Analysis of meteorological-hydrologic model for one-stop water management
It is analyzed for meteorological model(LDAPS, UM3.0, RDAPS) and runoff model(COSFIM,K-DRUM) to manage short-term water. The UM is integrated numerical weather prediction model to have

Ⅳ. Research details and results
1. Analysis of meteorological-hydrologic model for one-stop water management
It is analyzed for meteorological model(LDAPS, UM3.0, RDAPS) and runoff model(COSFIM,K-DRUM) to manage short-term water. The UM is integrated numerical weather prediction model to have the ability to operate region of Korea, East Asia and global. A merit of UM(Unified Model) is to increase resolution for forecasting area of global or local. A local prediction model is able to produce a detail prediction data according to adopting one between lateral resolution(1.5 ㎞) and vertical resolution(70 layer to 40 ㎞) and improve simulation performance for high-resolution meteorological forecast phenomenon with high-resolution data preprocess, continuous data assimilation process and physical process.
A COSFIM(A Coordinated Operation System for Flood control In Multi-reservoirs) is model to determine appropriate size and time of discharge with analysis of flood volume flowing into dam during flood period. A model of flood analysis is composed of lumped model concerning short-term single storm event models for an ability of flood analysis and a prompt decision-making. K-DRUM(K-water hydrologic & hydraulic Distributed RUnoff Model) have a merit, which is able to calculate runoff of long term as well as short term, and is comprised of grid-based multi-layer model, estimation module of horizontal runoff and vertical infiltration in basin. The model is possible to trace hydraulically rainfall-runoff using method of kinematic water and to interpret accurately runoff applied by auto-calibration for initial soil condition. It is proper for automatic generation function of DB input data to form automation system.
2. Development of one-stop water resources operational system using data of a numerical weather prediction model
The one-stop water management pilot system is built by short-term numerical prediction models (LDAPS, UM3.0, RDAPS) of KMA and Hydrologic models (K-DRUM, COSFIM) of K-water used by field work. K-DRUM model works online(Receiving weather forecast data, operating K-DRUM model automatically and offering forecast and warning information to flood manager through forecast monitoring system.), and COSFIM model which needs experience of flood manager and realtime condition data works off line(manager operates COSFIM model with weather forecast data and connects with monitoring system to manage water resources).It is composed of a server to provide visualization of a product, a database server to store and acquire the hydrologic databases on parallel cluster and a system for extracting and converting data for the operation of the hydrologic model. A operation server of hydrologic model conduct a collection of DB hydrologic data and shot-term numerical prediction data. A visualization server provide the graphic representation of the water management and the result collection of a operation server.
The one-stop water management pilot system gathers and saves short-term weather forecast data and hydrological data, and operates all programs automatically by modules for each programs. The monitoring webpage provides all products of this system to offer convenience of users.
The user can check precipitation forecast of the Korean Peninsula including each dam basin with interpolation of precipitation on numerical prediction models(LDAPS, UM3.0, RDAPS). The graph of reservoir operation display precipitation, runoff, discharge, water level, observation time and analysis, is adjusted to brightness for user to grasp simply every prediction time. The user perform decision making for the reservoir operation with forecast of the present condition, precipitation, discharge, observation of water level, and rainfall-runoff modeling on a graph of the reservoir operation. A graph of runoff analysis show the basin average rainfall and runoff of each dam basin at the top and the bottom of the graph, the location including the degrees of latitude and longitude is indicated at the right side of the graph. Users can download the analysis table of product error from the web, and it can help users decision-making via prediction and observation produced by the one-stop water management service programme.
The adequate service environment is needed for effective use of graph and numerical data produced by the one-stop water management system. For this, the monitoring page of the one-stop water management is produced by web based user interface. It is operational on web based access environment, and the user in authority is available to access server with the area of internet connectivity. It is possible for users to use contents with menu and address bar via accessing page. It is consist of the UI on graph screen for same format of initial page, most graph and image page.
3. One-stop water resources operational system assessment
We used the values for evaluating the prediction result, which predicted cumulative value,observed cumulative value, the rate of predicted and observed cumulative values(%), predicted mean value, observed mean value, the deviation and deviation ratio of the predicted mean value and the observed mean value, standard deviation of predicted and observed value, standard deviation ratio of predicted and observed value(%). In addition, we use the index to be used mainly calculated and observed values in model for quantitative reliability assessment. It was used A dimensionless index of NSE (Nash-Sutcliffe Efficiency), statistics techniques of index errors PBIAS (Percent Bias), and the ratio of mean square error and standard deviation for observations RSR (RMSE-observations Standard deviation Ratio).
LDAPS, UM3.0 and RDAPS used as precipitation forecast results as input data to COSFIM and K-DRUM models performed on the result of the dam inflow forecasting. LDAPS prediction results found that underestimation between 50 ~ 180 mm precipitation and overestimation with below 50 mm precipitation cases. During cases with less than 200 mm precipitation were shown to be a very underrated. This is a very crucial issue in water management, and it is considered to further analyze. LDAPS observed error rate in proportion to the volume of rainfall predictions (VER) appeared significantly variability with less than 50mm. And the results of analyzing the observation of the accumulated precipitation showed a tendency to increase in accordance with the precipitation and RMSE at the same time. UM3.0 and RDAPS prediction results were analyzed as representing underestimated results.
Utilizing LDAPS as input data of the K-DRUM model in order to calculate observed accumulated precipitation, results showed relatively good relationship with less inflow. Based on statistical analysis, scatter plots were likely to wider with heavy precipitation cases. In case of large amount of inflow is likely to be a very important issue in this flood water management that is considered to require additional analysis. UM3.0 rainfall prediction showed underestimated results. In case of prediction of runoff, overall results showed the opposite pattern when accumulated inflow were relatively low. Further analysis of the impact that the K-DRUM model results with observed precipitation carried out in the past three days in early forecasting stage is needed. RDAPS prediction appeared to underestimate the inflow.
Using rainfall forecasts generated by the LDAPS, COSFIM model analysis was conducted targeting Hoengsan where the upstream water level stations of Gunnam flood control. Through LDAPS predicting precipitation to COSFIM model over Gunnam flood control water level station, Hoengsan example of how applying the results of 1 m or more alert level through a model that can provide the administrator in advance which is expected to be very useful in water sources management. In order to prevent the damage occurred by hydrological disasters that cannot be identified by only observations of rainfall stations, utilizing the LDAPS predictions can be significant factor in Gunnam flood control operations management.
4. Water management availability review of probability long-term forecast
By using probability information of the Korea Meteorological Administration long-term forecast estimate future inflow, and perform an analysis to suggest a long-term forecast methods that can be applied to the water management. Analysis performance process is largely composed of 4 steps ;’selecting the optimum probability distribution type of past precipitation’, ‘making cumulative distribution function of the future precipitation considering the probability long-term forecast’, ‘creating precipitation ensemble based on predicted cumulative distribution function’, ‘creating inflow rate ensemble based on predicted cumulative distribution function’. First, in ’selecting the optimum probability distribution type of past precipitation’, using monthly precipitation data from the past in the coverage area of Andong dam basin, we estimated the most appropriate probability distribution each month through calculating the parameter of various probability distribution and goodness of fit test. In‘making cumulative distribution function of the future precipitation considering the probability long-range forecast’, calculating the third quartile of the past probability distribution adopted in an appropriate distribution, setting a reference value of a probability forecast category(BN, NN, AN), and we make the distribution of future precipitation by combining probability long-term forecast. Forecast data used in the analysis was used to precipitation probability forecast data from June 2014 to February 2015 which released nearest forecast information from May 2014 to January 2015. Third, ‘creating precipitation ensemble based on predicted cumulative distribution function’, after converting the probability forecast value to cumulative distribution function to account for uncertainty of the long-term forecast, on the basis of the probability distribution of the forecasted precipitation generated a random number and was calculated by the ensemble of precipitation. Finally, by entering the resulting ensemble of rainfall on water budget model calculate predicted inflow rate, compare and analyze the actual observation value.
For precipitation, the deviation of the actual observation contrast predicted ensemble mean was greater than the past mean value in most month, RMSE also it showed a higher value. In case of inflow, error of the ensemble mean and historical average compared to the actual observed value is similar, but before the August, predicted ensemble average was less than the historical average, and after the August, showed a substantially greater tendency than the historical average. The result of the RMSE predicted that ensemble average was somewhat small, but, did not very different from the average of the past average. synthetically, when predicted ensemble average was compared to historical average, it was simulated the result closer to the observed values.
Error of calculated inflow rate through the research process is that period of data which changed long-term forecast of the probability method in Korea Meteorological Administration was only 9 month, Error that may occur due to different area of forecast data adopted in research, not Andong dam, Daegu-Gyeongbuk was excluded. Furthermore, the error of the actual flow rate is dependent on the accuracy of the climate model for calculating the probability long-term forecast and water budget model, but part on the accuracy of both the model itself was excluded. Therefore, deviation and RMSE of the analysis inflow is determined that could be used as a reference in the water management application of long-term forecast data, and this study use the probability of climate variables which provide probability long-term forecast for the first attempt in the country to water management, and is significant to present a quantitative applicability of the forecast.
5. Architecture design of ensemble based decision-making system
There are things to consider, depending on the planning and operations periods for dam operations. In order to make decisions appropriate to each situation, such as hydrological prediction as future expected inflow, water demand, reservoir water level of plan point, discharge for improving water quality by specific time, operation plans in preparation for the drought that lasts for a long period of time is needed.
Hydrological prediction as future expected inflow is affected by precipitation,evapotranspiration-related factors offered through long-term weather forecasts and also expected to cause significant element of uncertainty at the same time. Through the conjunction of hydrological model results and long-term forecasts by the Meteorological numerical model GloSea5 model for future hydrological forecast, in this study, it was looking for ways that can help in water management decision-making process.
GloSea5 model was to test operations in September 2013 as a fifth version of the ensemble prediction system, in June 2014 being applied to the probability forecast operations.1 month 28 ensemble, 3 months, and generates a probability of long-term forecast using 42 ensemble. In this study, GloSea5 model based ensemble has been designed in conjunction with hydrological models(SSARR, K-DRUM) an architecture that can support decision-making for long-term water management.
Long-term runoff model is usually done daily and take into account the losses due to evapotranspiration, besides precipitation, maximum and minimum temperatures, wind speed, dew point temperature, radiant energy (in the case of observed dat, hour of daylight) is required. It is also necessary to consider the boundary surface elevation data of the temperature and wind speed. If these meteorological elements extracted from the boundary surface elevation, existing Historical Rainfall Scenarios ensemble flow prediction methods can be utilized is replaced by the ensemble produced in GloSea5 model. Long-term runoff model is usually carried out daily. Due to consideration of losses of evapotranspiration, temperature(maximum and minimum), wind speed, dew point temperature, radiant energy and elevation are additionally required. Therefore, the system was built to consider ensemble streamflow prediction from GloSea5 in conjunction with hydrological models.
The ensemble based flow forecasting is basically to utilize meteorological information as input data, because the flow has to be simulated through the hydrological models, it can be highly dependent on the accuracy of meteorological information and hydrologic models. Methods of improving the accuracy of the ensemble based flow forecasting are to increase middle and long term meteorological forecasting accuracies and hydrological model’s accuracies. There is a method that the weight to the probability of occurrence with respect to the long-term weather forecasts, and the probability is given as a method of using the correction technique and using statistical methods as a way to improve the weather forecast accuracies. Methods to improve the hydrological model simulation accuracy are a lot of ways to apply correction using simple statistical methods, such as weather forecasts.
6. Application of K-DRUM model to Yongdam-dam basin and improvement of the model
The results of estimating evapotranspiration using flux tower observed data from January 2013 to December in Yondam-dam test watershed, estimated evapotranspiration was gradually increased in January but did not actively, it was drastically increased in May to August while rapidly decreased in November and it did not occur in December. Monthly averaged evapotranspiration was 1.56 ㎜/day as maximum values from May to September, averaged value was 0.15 ㎜/day from January to April and it was 0.41 ㎜/day from September to December. The values was significant seasonal effects. Daily evapotranspiration variation was 0~3.96 ㎜/day and total daily averaged evapotranspiration was estimated to 0.81 ㎜/day. Soil moisture content was divided into surface layer(10 cm, 20 cm) and deep layer(40 cm or more), soil moisture values in deep layer(40 cm or more) had highly increased when preceding rainfall occurred. Soil moisture content appeared differently depending on soil characteristics and rainfall intensity, but generally, soil moisture values was 20~30 % in period of non precipitation, the values increased to 40% in precipitation period.
Predicted temperatures (daily maximum and minimum), wind speed, dew point temperature used to assess evapotranspiration by climate prediction model. And short-term weather prediction model hourly data are available from the research. Taking into account the soil moisture condition was improved in such a way that a differential calculation then multiplying 0.7 to the existing batch of evapotranspiration. Using the observed evapotranspiration in Yongdam test basin, flow was simulated considering evapotranspiration in order to verify the accuracy of the K-DRUM model. Total observed flows was 7,151 ㎥/s and simulated flows was 8,257 ㎥/s, simulated one as compared to observed one was estimated at 115 %. As annual rainfall values was 1303. 5 ㎜, and areal rainfall values applied to basin area(930 ㎢) was 14,030 ㎥/s, Observed flow was 51% compared to total areal rainfall and Simulated one was about 59%. In other words, basin runoff ratio is similar to the domestic basin characteristics about 51%.
As a result, simulated flows by K-DRUM model considering evapotranspiration and snow fall and melting is well fitted to the observations. simulated annual actual evapotranspiration values was 449 ㎜ as compared to simulated total potential one was estimated at 43 %. soil moisture content was underestimated as compared to observed one from start time of simulation up to 150days, and after that simulated one was similar to observed one. It is assumed to be due to not considering the previous amount of snowfall because simulation was performed since January 1 as start point. Since the summer, however, there appears a pattern similar to the observed value, and it is expected to reproducible for simulation of the soil moisture content.

목차 Contents

• 표지 ... 1
• 제출문 ... 2
• 보고서 요약서 ... 3
• 요약문 ... 4
• SUMMARY ... 12
• CONTENTS ... 22
• 목차 ... 25
• 표목차 ... 27
• 그림목차 ... 29
• 제1장 연구개발과제의 개요 ... 36
• 제1절 연구배경 및 목적 ... 36
• 제2절 연구내용 및 범위 ... 39
• 1. (단기 물관리) 강우-유출모델기반 원스톱 물관리 시스템 기술개발 ... 39
• 2. (장기 물관리) 장기 물관리 의사결정 지원시스템 아키텍처 개발 ... 40
• 제2장 국내외 기술개발 현황 ... 41
• 제1절 기상예측 정보의 수자원 분야 활용 현황 ... 41
• 제2절 기상-수문 모델 연계 기술 사례조사 및 분석 ... 45
• 1. 유럽의 EFAS(European Flood Awareness System) ... 45
• 2. 일본의 GFAS(Global Flood Alert System) ... 48
• 3. 미국 Puyallup강 유역의 유출분석 시스템 ... 49
• 4. 영국 Thames강 유역의 유출분석 시스템 ... 54
• 5. 기후변화 평가 및 예측분야 기상-수문모델 연계 활용 사례 ... 55
• 제3절 확률장기예보 정보를 활용한 물관리 서비스 분석 ... 58
• 1. 국외 확률장기예보 현황 ... 58
• 2. 장기예보의 물관리 활용 현황 ... 62
• 3. 앙상블 유량예측에 대한 국내･외 사례 분석 ... 66
• 제3장 연구개발수행 내용 및 결과 ... 71
• 제1절 기상-수문모델 원스톱 연계를 위한 모델 특성 분석 ... 71
• 1. 단기 기상예측모델 ... 71
• 2. 강우-유출모델 ... 73
• 제2절 기상 수치예보모델 자료를 활용한 원스톱 물관리 시스템 구축 ... 76
• 1. 기상-수문모델 원스톱 연계를 위한 시스템 설계 ... 76
• 2. 수치예보모델 자료변환 ... 79
• 3. 원스톱 물관리 시스템의 수행을 통한 산출물 생성 ... 92
• 4. 원스톱 물관리 시스템의 모니터링 프로그램 ... 114
• 제3절 원스톱 물관리 시스템 평가 ... 121
• 1. 대상 댐유역 특성 분석 ... 121
• 2. 예측결과 평가 방법 ... 129
• 3. 댐 유역 강수량 예측결과 평가 ... 131
• 4. 댐 유입량 예측결과 평가 ... 153
• 제4절 확률장기예보의 물관리 활용성 검토 ... 180
• 1. 물관리를 위한 확률장기예보 필요성 및 적용 방안 ... 180
• 2. 월별 댐 유입량 생성을 위한 물수지 모델 ... 182
• 3. 확률장기예보 물관리 적용 및 결과분석 ... 184
• 4. 확률장기예보 물관리 적용성 검토 결론 및 제언 ... 195
• 제5절 앙상블 기반의 의사결정 지원시스템 아키텍처 설계 ... 196
• 1. GloSea5 모델 ... 196
• 2. 장기유출모델 ... 199
• 3. GloSea5 앙상블 기반의 의사결정 지원 아키텍처 설계 ... 201
• 4. 앙상블 유량예측 정확성 향상 방안 ... 212
• 제6절 K-DRUM 모델의 용담댐 시험유역 적용 및 모델 개선 ... 217
• 1. 용담댐 시험유역 증발산량 및 토양수분량 관측 ... 217
• 2. K-DRUM 모델 적용을 통한 유입량, 증발산량, 토양수분량 분석 및 모델 개선 ... 219
• 제4장 목표달성도 및 관련분야에의 기여도 ... 226
• 제5장 연구개발결과의 활용계획 ... 231
• 제6장 연구개발과정에서 수집한 해외과학기술정보 ... 232
• 제7장 연구시설 · 장비 현황 ... 233
• 제8장 참고문헌 ... 234
• 끝페이지 ... 242