1. Introduction The nature is composed of infinite process, and each process is surely deterministic (out of mention for the micro process on the level of quantum physics, which is under the uncertainty principle of Heisenberg, 1927), but affected by uncountable number of factors. What we are trying to do with modeling is to find the most dominating factors on a process and to simplify the process with an understandable structure which is composed of those several effective and observable factors. For any natural phenomenon that we are trying to forecast, if the spatiotemporal boundary or initial condition were exactly known, and if the model exactly simulated the process, then the computed phase path would provide an exact forecast. But, unfortunately, neither assumption is valid based on current technology or knowledge. One should bear in mind that there are always initial error in a model at the beginning of simulation and there are always additional error during a simulation generated by the imperfect model structure. To estimate the effect of those errors on the forecast results, it is necessary to supplement such deterministic forecasts with detailed information by estimates of forecast reliability. By this reason, the stochastic concept has been included in forecasting, and ensemble simulation has been used as a good tool for carrying those stochastic concepts in a computer simulation. Recent trends of flood forecast are away from the conventional deterministic forecasts of hydrographs toward offering probabilistic forecasts, which include its prediction uncertainty. Deterministic flood forecast specifies a point estimate of the predicted values, such as precipitation and river stages/discharges. On the other hand, a stochastic forecast specifies a certain probability distribution function of the predicted values. The predictive probability in a probabilistic forecast is a numerical measure of the certitude degree about the intensity of a flood event, based on all meteorological or hydrological information utilized in the forecasting process (R. Krzysztofowicz, 2001). Flood Forecasting System Using Weather Radar and a Distributed Hydrologic Model Sunmin KIM *, Yasuto TACHIKAWA, Kaoru TAKARA * Graduate School of Urban and Environmental Engineering, Kyoto University Synopsis A real-time flood forecast system is proposed with stochastic radar rainfall forecasts and recursive measurement update in a distributed hydrologic model. In the first part of the system, a radar image extrapolation model gives deterministic rainfall predictions and error fields are simulated to offer probable variation on the deterministic predictions. The error field simulation uses a random field generation method based on an analyzed error structure of the current time rainfall prediction. Then, the probable rainfall fields with generated error fields are given to a distributed hydrologic model to achieve an ensemble runoff prediction. In the second part of the system, the distributed model is coupled with the Kalman filter to utilize online hydrologic information by several techniques including Monte Carlo simulation scheme. Keywords: flood forecasting, weather radar, distributed hydrologic model 京都大学防災研究所年報 第49号B 平成18年4月 Annuals of Disas. Prev. Res. Inst., Kyoto Univ., No. 49 B, 2006