| Abstrakt: |
Recent years, flood disasters have become rampant due to climate change. For the risk mitigation, building up a hydrological run-off model which can be used accurately and quickly helps communities prepare effective disaster prevention measures. Tank Model is a hydrological run-off model proposed in Japan. On one hand, its calculation is relatively simple. On the other hand, many unknown parameters have to be identified. Thus, the more random number it generates at a time, the more it takes time for calibration. This research examines the optimal method of identifying the unknown parameters by Monte Carlo method, considering the improvement of efficiency for practical use. Montel Carlo method can generate the massive number of random samples and help us select the best combinations of unknown parameters. Among the generated random numbers, the optimal parameters are searched which can fit to the actual measured values with the minimal number of samples. This research examines how many random numbers need to be generated at minimum to obtain the optimal parameters. The number (N) of random samples were seven kinds of 100, 1,000, 10,000, 100,000, 1,000,000, 10,000,000 and 100,000,000. Every time random samples generated by Montel Carlo method, the maximum and minimum numbers of each unknown parameter among the five best combinations were applied for the retrieval range of the next simulation. In this way, the simulation was repeated ten times for each kind of sample number. As the result, the watershed with simple land-use could obtain the optimal parameter with fewer samples (N = 1,000) than the watershed with complex land-use type (N = 1,000,000). With these sample numbers, the prediction accuracy for both watersheds were high. It is considered that the complex land-use watershed had lower accuracy rate because the water runoff was influenced by the underdrainage facilities. [ABSTRACT FROM AUTHOR] |
