Identification and statistical analysis of global water vapour trends based on satellite data

Abstract

Global water vapour total column amounts have been retrieved from spectral data provided by the Global Ozone Monitoring Experiment (GOME) flying on ERS-2, which was launched in April 1995, and the SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) onboard ENVISAT launched in March 2002. For this purpose the Air Mass Corrected Differential Optical Absorption Spectroscopy (AMC-DOAS) approach has been used. The combination of the data from both instruments provides a long-term global data set spanning more than 12 years with the potential of extension up to 2020 by GOME-2 data on MetOp. Using linear and non-linear methods from time series analysis and standard statistics the trends of water vapour columns and their errors have been calculated. In this study, factors affecting the trend such as the length of the time series, the variance of the noise and the autocorrelation of the noise are investigated. Special emphasis has been placed on the calculation of the statistical significance of the observed trends, which reveal significant local changes from -5 % per year to 5 % per year. These significant trends are distributed over the whole globe. Increasing trends have been calculated for Greenland, East Europe, Siberia and Oceania, whereas decreasing trends have been observed for the northwest USA, Central America, Amazonia, Central Africa and the Arabian Peninsular. The idea of the comprehensive trend and significance analysis is to get evidence for the truth of these observed changes. While the significance estimation is based on intrinsic properties such as the length of the data sets, the noise and the autocorrelation, an important aspect of assessing the probability that the real trends have been observed is a validation with independent data. Therefore an intercomparison of the global total column water vapour trends retrieved from GOME and SCIAMACHY with independent water vapour trends measured by radiosonde stations provided by the Deutsche Wetter Dienst DWD (German Weather Service) is presented. The validation has been performed in a statistical way on the basis of univariate time series. Information about the probability of agreement between the two independently observed trends, conditional on the respective data, is revealed. On the one hand a standard t-test is used to compare the trends and on the other hand a Bayesian model selection approach has been developed to derive the probability of agreement. The hypothesis of equal trends from satellite and radiosonde water vapour data is preferred in 85 % of compared pairs of trends. Substantial evidence for the hypothesis of agreeing trends is found in 26 % of analysed trends. However, also disagreement has been observed, where the main reason has been identified on the one hand as the different spatial resolutions of the instruments. This means, that the radiosonde measurements can resolve very localised events, which is not possible with the satellite instruments. On the other hand, in contrast to the in principle continuously available (on a monthly mean basis) GOME/SCIAMACHY data, missing data in the radiosonde time series lead to trend discrepancies. The identification and validation of water vapour trends is an important step for a better understanding of climate change, but water vapour is not the only contributing quantity. Beside water vapour, decisive parameters are temperature, clouds, precipitation, vegetation and many more. A promising framework for the investigation of a multivariate data set of environmental variables is given by the Markov chain analysis. As a first approach, the Markov chain analysis has been applied to a bivariate water vapour -- temperature data set, where the global near surface temperatures are provided by the Goddard Institute of Space Studies (GISS) and cover a time span from 1880 to 2005. The temperature data are retrieved from ground stations and are mainly based on the Global Historical Climatology Network (GHCN). In the framework of a Markov chain analysis, the bivariate set of data is reduced to a univariate sequence of symbols, which can be described as a discrete stochastic process, a Markov chain. This Markov chain represents the water vapour -- temperature interaction or state of a region. Several descriptors have been calculated, such as persistence, replacement of and entropy. This approach is new in environmental science. Exemplarily two climate systems, the Iberian Peninsular and a region at the islands of Hawaii in the central Pacific Ocean, are investigated. The Markov chain analysis is able to retrieve significant differences between the two climate systems in terms of the characteristic descriptors, which reflect properties such as climate stability, rate of changes and short term predictability.