Separating the impact of urbanization (urban effect temperature trends) from observational data at urban monitoring sites is an important problem in the detection of global warming impacts. A statistical method using principal component analysis of the temperature data in an area was developed to separate the urban effect temperature trend from the observed temperature record at urban stations, including those without nearby rural reference stations. After the mathematical description of the method, an example of the analysis for model data on which some trends are superimposed (given) at any stations was shown for an explanation of the method. This example not only clarified the performance (impact, weight) of the superimposed trend in the trend for each principal component score (Z-score) but also showed how all the superimposed trends were separated from the model data by this method. This method was applied to monthly mean temperature data for the past 73 years (1920-1992) from 51 meteorological stations in Japan, and the urban effect temperature trend at each station was separated from the observed temperature trend. As a result of the analysis, it was clarified that in Japan the maximum urban effect temperature trends at stations with a population of over 100 thousand in 1993 were 1.0-2.5°C/100 years, which were almost the same as those in the United States and China. The urban effect temperature trends in big cities are larger in the cold season than in the warm season, being maximum in winter or autumn and minimum in summer. After removing the urban effect temperature trend, the annual mean temperature trends at stations in Japan over the period analyzed ranged from 0.5°C/100 years in northern Japan to 1.1°C/100 years in western Japan with the areal mean for Japan of 0.8°C/100 years. The positive trends are most notable in winter and spring (December to May), and show areal mean values of 1.0-1.6°C/100 years with the maximum in March. On the other hand, they are not marked in summer and autumn. In particular, in the northern part of Japan, negative trends appear from July to November. The area with negative trends below -1.0°C/100 years extends over most of the northeastern half of Japan in July, but does not reach the southern part of Japan, which has a slight positive trend. As a result of these regional differences in temperature trends, the north-south temperature gradient increases in summer and autumn with the maximum trend difference of 2.0°C/100 years in July and October.
View full abstract