Reliable seasonal hydrologic forecasts are essential for managing water resources, especially drought detection and mitigation, reservoir operation, energy planning for hydropower plants, water supply management, and many other related activities. Recent advances in dynamical seasonal climate forecasts using coupled atmosphere-ocean-land General Circulation Models (CGCMs) can provide reliable information to predict hydrological conditions at longer lead-times (up to 6 months). In this context, dynamic-model-based seasonal hydrologic prediction can play an important role in transferring advanced scientific knowledge from the climate research community to end-users in society, such as water resource managers and decision-makers.

This study developed a Multi-Model Ensemble (MME)-based seasonal hydrologic forecasting technique which generates forecasts with up to 3-months lead-time based on APEC Climate Center (APCC) MME climate forecasts for eight dam basins in South Korea. This seasonal hydrologic forecasting technique employs five monthly water balance models, namely ABCD, Guo, VUB, WBM, and Xiong. Applying the monthly water balance models has certain advantages: simple model structures, easy implementation, reliable performance for monthly runoff simulation, and minimal requirements in terms of climate forcing data. In particular, inputs to the water balance models, such as monthly temperature and precipitation can be obtained directly from the APCC MME climate forecasts.

This study added a simple temperature-based snow accumulation and snow melt module and the Hamon’s potential evapotranspiration calculation module into the original water balance models to consider snow dynamics and water loss from evapotranspiration. The Shuffle Complex Evolution (SCE) algorithm developed by the University of Arizona (SCE-UA) was used to calibrate the parameters of the water balance models. Five water balance models were calibrated with three distinct objective functions (for a total 15 models) to account for parameter uncertainty in hydrologic forecasts.

Bayesian model averaging (BMA) and a simple model averaging (SMA) were employed to generate more skillful forecasts by weighting the individual forecasts from the fifteen water balance models. BMA is a statistical scheme for probabilistic MME prediction that is more skillful and reliable than the original ensemble members produced by several competing water balance models alone. The BMA scheme was applied to the transformed hydrologic ensemble members to obtain a single set of BMA weights for each basin using all the data points from the reforecast period (hindcast period).

This study used Spearman rank correlations of the streamflow forecasts and observations as the forecast skill score. For quantitative verification of the performance of the seasonal hydrologic forecasts in detecting above-normal flow and below-normal flow for the dry (November through May) and wet (June through October) seasons, the frequency bias index (FBI), false alarm ratio (FBI), probability of detection (POD), and Critical Success Index were computed, based on a 2×2 contingency table.

This study attempted to answer the following research questions using the MME-based seasonal hydrologic forecasting technique; (1) Do APCC MME seasonal climate forecasts have sufficient skill to provide useful hydrologic forecasts over South Korea?; (2) Does the predictability of seasonal hydrologic prediction vary depending on location, lead time, and time of year?; and (3) How can the most skillful seasonal hydrologic predictions be made, given the climate predictions? Furthermore, is it better to develop MME for seasonal hydrologic prediction using Bayesian model averaging (BMA) or simple model averaging (SMA)? Can the BMA approach give more useful forecast information than SMA?

The following conclusions can be drawn from our analysis.

1. The seasonal hydrologic forecasts based on 1-month lead APCC MME climate forecasts can provide useful dam inflow forecast skill for certain months and basins.

2. 2-month lead and 3-month lead hydrologic forecasts also demonstrate significant forecast skill but only for a few months and basins. However, there was no forecast skill for June and September for all basins. 2. The performance of the seasonal hydrologic forecasts varied depending on the dam basin, lead time, and month. In particular, the 1-month lead hydrologic forecast skill in February through May was higher than that of the other months. Furthermore, model averaging techniques, such as BMA and SMA, did not have much of an influence on the hydrologic forecast skill.

3. The skill scores of the seasonal hydrologic forecasting system during below-normal streamflow conditions demonstrated better performance than those during above-normal conditions, indicating that seasonal hydrologic forecasts using APCC climate forecasts may be more useful for drought management than flood management at monthly scales.

Our results showed that an MME-based seasonal hydrologic forecast using APCC MME seasonal forecasts has potential for dam inflow forecasting in eight dam basins. Therefore, the seasonal hydrologic forecasts developed in this study could contribute to alleviating potential economic losses related to water management by providing reliable forecasts and early warning at an extended lead-time over South Korea.

1. 서론

2. 수자원계절예측 프레임워크

3. 연구 자료 및 방법

3.1 연구대상유역

3.2 월 강수 및 기온 관측자료

3.3 APCC MME seasonal forecast

3.4 월물수지 모형(Monthly water balance model)

3.5 물수지 모형의 매개변수 검보정

3.6 다중모형앙상블 (multimodel ensemble) 기법

3.7 예측성 검증

4. 연구내용

4.1 월물수지모형의 정확도 평가

4.2 BMA를 이용한 모형별 가중치 산정

4.3 유출량 모의에서의 BMA와 SMA 비교

4.4 수자원계절예측의 예측성 분석

5. 주요 결과 및 향후과제