This research focuses on the seasonal prediction of precipitation in Thailand. Thailand faces difficulty of dam operation because of its climate and geography characteristics. Here, seasonal prediction of precipitation is effective for accurate dam operation, and urgently needed to prevent floods in summer and droughts in winter. Seasonal scale predictive skill, however, is still limited due to complex variability of atmospheric environment. It requires to consider mutual relationship between atmosphere, ocean and land. This research tried to contribute to understanding of this mechanism by using machine learning method. Nine different climate indices are investigated in order to extract features of those relationship and express the global mutual interactions in modeling. A couple of statistical models, Genetic programming model(GP) and Multi linear regression model(MLR) for comparison, are applied for prediction. GP has advantage of its learning process compared to other machine learning methods in that GP can avoid overfitting even with limited number of input data. As a result, GP succeeded to predict large fluctuations such as floods or droughts in some years, while MLR underestimated these fluctuations. On the other hand, both models produced large errors in other predicted years. GP still has to be improved for practical use for dam operation.
各種検索
サポート
ヘルプ