Solange M. Leite

THE AUTOREGRESSIVE MODEL OF CLIMATOLOGICAL TIME SERIES: AN APPLICATION TO THE LONGEST TIME SERIES IN PORTUGAL

The autoregressive model is extremely useful for the representation of many geophysical time series. The present paper deals, in the first place, with the analysis of sets of meteorological observations (temperature and precipitation) made sequentially in time, from 1856 to 1994, at equidistant time intervals Δt = 1 year. The best fitting is an autoregressive model of order 45. A thorough discussion of the temperature observation time series is presented, showing two main salient features: an approximately 0ċ6°C secular trend during the past century and a definite warming in the 1980s and early 1990s. The estimated values of temperature and precipitation are compared with the observed data series.

Long-term variability of the temperature time series recorded in Lisbon

As a case study for application in climate change studies, daily air temperature records in Lisbon are analysed by applying advanced statistical methodologies that take into account the dynamic nature of climate. A trend analysis based on two non-parametric tests (Spearman and Mann–Kendall) revealed the presence of statistically significant upward trends in the maximum temperatures, mainly during March. The minimum temperatures do not present significant trends, with the exception of March where a relatively weak positive trend is detected. A singular spectral analysis combined with a maximum entropy spectral analysis enables the detection of regularities in the annual mean time series of the maximum and minimum temperatures. A quasi-periodic oscillation with a peak period of about 50 years is superimposed in the linear trends. At the maximum temperature, a secondary oscillation with a peak period of nearly 20 years is also identified. No other regularities are isolated in these time series. The study is enhanced by applying an extreme value analysis to the extreme winter and summer temperatures. The generalized extreme value distribution family is shown to provide high-quality adjustments to the distributions, and a description of the temperatures related to different return periods and risks is given.