Abstract
A neural network was used to estimate climatic data from topographical factors in areas where observed climatic data do not exist. The neural network with a back propagation algorithm was used as a method of nonlinear multiple regression which has advantages over conventional linear regression, especially when the relation between climatic data and topographical factors is nonlinear.
As an example, four meteorological data (normal value of daily mean temperature, daily maximum temperature, daily minimum temperature and daily mean wind velocity) were estimated for 74 observation points around Shizuoka Prefecture from input data of 13 geological factors such as altitude, latitude, longitude, slope angles etc. The network consists of 3 layers which have 13 units in the input layer, 4 units in the output layer and 26 units in an intermediate layer. A linear multiple regression analysis was also used to compare the characteristics of the methods.
In all cases, the neural network showed smaller errors than the multiple regression. In the case of wind velocity where the data was rather scattered, the neural network showed its advantage clearly. Although the use of the neural network is accompanied by theoretical ambiguity, the result shows precise estimation is possible by neural network.
