Abstract
The aim of this study is to estimate distribution of meteorological factors in high resolution in order to thermally evaluate a land use in urban area. In this study, distributions of monthly mean average, maximum and minium air temperature were estimated by using 3 layered back propagation neural network system. The input data sets for this neural network system were (1) height above sea level, (2) normalized vegetation index, which was calculated by using LANDSAT TM data and (3) distance from seashore. The supervised data set for learning were meteorological data observed at 30 stations around the Osaka Bay. R^2 value and the sum total of error show a sufficient accuracy of this method with neural network system in comparison with the estimating method with multiple regression, which we have used before. The addition of distance from sea shore was recognized to improve an accuracy an estimation of air temperature both in sea side area and in inland area. After checking the over learning problem on the neural network system, distribution of monthly mean minimum air temperature around the Osaka Bay was drawn by this method.
