Techniques of Real-Time Flood Forecasting and Low Flow Prediction for Reservoir Operation

本研究取得的主要成果如下：1.与政府部门合作开展了水库调度实际情况问卷调查。通过对集水面积超过50 km2的201座水库资料的总结，指出提前3小时的实时预报对洪水预报至关重要，约有一半的水库采用径流模型进行洪水预报，但实际运用中很少使用过滤技术。为了检验长期和短期径流模型(LST-II)和Tank模型在实时洪水预报中的性能，将这两种径流模型应用于黑木坝流域18年(1979-1996年)的观测资料。结果表明，两种径流模型均能较好地模拟长期径流和洪水径流的观测过程线。将扩展卡尔曼滤波、统计线性化和反演法等滤波技术应用于黑木坝流域提前期为1小时和/或3小时的实时洪水预报。对流域19个洪水的预报结果表明，采用滤波技术预报洪水的精度较高，以预报流量的相对误差衡量的精度基本相同。利用黑木坝流域的洪水资料，比较了几种降雨预报方法在实时洪水预报中的性能。结果表明，假设未来降雨量与当前强度不变的最简单的方法效果最好，应用AR(自回归)方法需要设定系数的限值。

Main results obtained in this study are as follows :1. A questionnaire on actual circumstances of reservoir operation was carried out in cooperation with government agencies. It is outlined by summarizing the information on 201 reservoirs with catchment over 50km^2 that real-time forecasting up to 3 hours ahead is of critical importance for flood flow, and that runoff models are used for flood forecasting in about half of the reservoirs, but the filtering techniques are rarely used in actual reservoir operation.2. In order to examine the performance of Long-and Short-Term Runoff model (LST-II) and Tank model used in real-time flood forecasting, both runoff models were applied to the data of 18 years (1979-1996) observed in the Kuroki Dam basin. It is shown that observed hydrographs both in long-term runoff and flood runoff were simulated with good accuracy by both runoff models.3. Such filtering techniques as extended Kalman filter, statistical linearization techniques and back calculation method were applied to real-time flood forecasting with lead time of 1 and/or 3 hours in the Kuroki Dam basin. The results of 19 floods in the basin demonstrate that more accurate flood prediction is attained by adopting the filtering techniques, and that the accuracy evaluated by relative error of predicted discharge by each technique is almost same.4. The performance of several methods for rainfall prediction in real-time flood forecasting was compared by using flood data in the Kuroki Dam basin. It is shown that the simplest method, in which future rainfall are assumed to be constant with the present intensity, gives best results, and that setting the limitations of values of coefficients is necessary for applysing the AR (auto-regressive) method.