Forecast of Meteorological Data Utilizing State-Space Model Utilizing Metric-Multi Dimensional Scaling

Abstract

Based on Bayesian statistics using the expected a posterior (EAP) estimation, we forecast time evolution of meteorological data, such as the temperature, at a target point via information on a set of time-series of the temperatures at sampling points selected by the metric multi-dimensional scaling (metric-MDS), without using information on that of the target point. Using numerical calculations with respect to the climate statistics in Kanto district, we clarify that the metric-MDS can select a set of sampling points whose data are similar to that of the target. Then, we clarify that the EAP estimation succeeds in predicting time evolution on the temperature in Maebashi using the set of time-series of the temperatures at the selected sampling points around Maebashi. Also, we find that the EAP estimation predicts the time evolution of the temperature more accurately than the conventional autoregressive model.