Time series analysis and identification of trends in a Mediterranean urban area

Abstract

During the last decade there has been considerable interest in the stochastic analysis of temperature trends. This attention usually focuses upon global and hemispheric time series. Analyses of global temperatures are of obvious importance but it is equally of interest to analyse questions of warming at a more ‘micro’ level. The objective of the paper is to illustrate that the appropriate estimation of a local record trend component necessitates the use of flexible methods. A long time series of temperature data from a Mediterranean urban area is analysed. The data comprise a series on monthly mean surface temperatures for Valencia (Spain), a Mediterranean city in terms of its urban structure and climatology. Trends are extracted from the temperature data available from 1938, using both annual and seasonal averages. Attention is focused on applying non-parametric methods which allow the analysis of non stationary trends in comparison to other methods. A local polynomial regression fitting and wavelet thresholding indicate that there is an upward trend in the annual series. The bandwidth in the local regression model selected by cross-validation follows short-term cycles rather than tracing out the long-term component. Therefore to extract the long-term trend a higher smoothing parameter is selected. The trend visualization plots show that the increase in the mean temperature per year is non-stationary. A change-point detection test indicates a trend change in the series around the year 1968. There is compelling evidence of a trend increase after this year, which is probably due to increase urbanization. The analysis of monthly observations is also discussed. Seasonal differences in trend patterns are more clearly observed when the monthly data are averaged across seasons. There have been upward shifts in the trends of spring and summer temperatures.