EIGEN-TEMPERATURE MODEL FOR THE ANNUAL AIR TEMPERATURE MOVEMENT EVALUATION AND FORECAST

Abstract

Climate variability and its changes are issues of broader global concern. This study addresses the annual air temperature movement evaluation and forecasting based on principal component analysis (PCA). An Eigen-temperature model for describing the annual air temperature movement by employing PCA is introduced. Subspace for evaluation is generated by selecting principal orthogonal eigenvectors of covariance matrix of temperature data. The principal eigenvectors are called "Eigen-temperatures", since they are eigenvectors and each temperature movement is described by them. Each temperature movement is projected onto the subspace of eigenspace, and described by a linear combination of the Eigen-temperatures. Then, a forecast method for the temperature movement by employing the Eigen-temperatures is proposed. Forecast is implemented with polynomial curve fitting algorithm to estimate subsequent representation weights for the subsequent temperature movement with respect to the "Eigen-temperatures" generated by its previous temperature movements. The proposed Eigen-temperature model is applied to evaluation and forecasting for annual temperature movement at Tongchuan observation station of China from 1962 to 1971 and from 1994 to 2002. Experimental results agreeing well with actual observation values show workability of the proposed. Result analysis indicates its effectiveness that the proposed Eigen-temperature model is outperforming the classical AR model and the BP-ANN on the forecast tasks.