Abstract
In making a mesh climatic data, neural network and multiple regression were normally used to estimate climatic data from topographical factors in areas where observed climatic data do not exist. In this study, accuracy of both methods was analyzed by leave-one-out cross validation method. As validation data, normal value of daily mean temperature of 91observation points around Kanto area in Japan was used as dependent variable. 14kinds of geographical factors such as altitude, latitude, longitude, etc. were used as independent variable. When all the data were used in obtain the equation for prediction, the neural network showed smaller errors than the multiple regression. In the case of leave-one-out cross validation, the neural network showed larger errors. When the network with larger repetition of training, the error increased. This result indicates that the estimation with neural network is not necessarily precise.
