Characteristic extraction and control on severe, drought in Japan through timely and spacial pattern classification

本研究提出了适合于水文气象事件模式分类的目标函数。研究结果表明：1）引入ISODATA对气压、气温、海温空间分布的水文气象资料进行分类;利用模糊集理论、分形理论和图像处理理论，建立了特征模式与样本模式的相似性模型，通过适应度分析了上述事件与日本干旱形势的关系，并利用遗传算法优化模糊神经网络单元，对日本干旱季节的降水量进行了估计。此外，遗传算法将降水量测量站划分为具有相同气象特征的几个区域。（4）利用递归神经网络对旱情进行实时预测，以调度水库防御严重灾害。以1994年某水库为例，验证了所提出的方法的有效性。（5）利用多点模拟方法，结合空间和时间相关性，采用等效变分法估算了全球变暖引起的干旱灾害。（and.vi）提出了一种新的干旱事件发生概率估算方法，即利用模式识别的概念，不仅估算降水总量，而且估算降水特征模式。

In this Grant, we proposed the objective function suite to pattern classification on hydrological and meteorological events. The following results are obtained ;i) ISODATA was introduced to classify the hydrological and metetrological data of spacial distribution of air pressure, air temperature and sea surface temperature. Similarity between feature pattern and sample one is formulated with fuzzy set, fractal and image processing theories.ii) Relation between above events and drought situation in Japan was analyzed through fitness rate.iii) Precipitation of drought season in Japan was estimated with fuzzy-neural network whose units were optimized by genetic algorithm (GA). Moreover GA classified Gauging stations of precipitation into several zones with same meteorological characteristics.iv) For real-time prediction of drought situation was taken with recurrent typed-neural network to operate reservoir against severe disaster. One reservoir was applied for the data of 1994 to verify the proposed methodologies.v) Drought damage due to global warming was estimated with the equivalent variation method by considering spatial and timely correlation with multi-point simulation method and.vi) New approach for occurrence probability of drought event was proposed by evaluating not only total volume but also feature pattern of precipitation with pattern recognition concept.