On the Properties of Zenith Total Delay Time Series from Reprocessed GPS Solutions

Global Positioning System observations from stations in regional and global networks have proven to sense the conditions of the atmosphere, especially the water vapour content of the troposphere. Zenith Total Delay (ZTD) derived during the processing of GPS data is a measure of the total atmospheric delay along the signal path between satellite and receiver antennas and arises mostly from the hydrostatic and wet parts of the atmosphere. Having taken surface pressure and temperature into account, ZTD can be converted into an estimate of the Integrated Water Vapour (IWV) content of the atmosphere, which when derived from homogenously reprocessed GPS observations, is emerging as an important parameter in the monitoring of climate change. Especially, the long-term trend and variations in IWV together with their associated uncertainties are of high interest as atmospheric water vapour is the dominant greenhouse gas. To date the trend estimates and their uncertainties are widely determined with assumption that the stochastic properties of the time series follow a random, ie. white noise, process. However, if ZTD and IWV are directly linked to climate processes, one would expect that the underlying noise process has similar character to that found in other climate parameters, which have been modelled by means of an autoregressive process. If this proves to be true, the trend estimates and their uncertainties in ZTD and IWV may have been underestimated up to this day of an order of magnitude.
In this research, we examine the properties of both deterministic and stochastic parameters of the ZTDs that were estimated by the consortium of the British Isles continuous GNSS Facility (BIGF) and the University of Luxembourg TIGA Analysis Centres (BLT) for GPS data collected by a global tracking network of more than 700 stations (repro2 solution). The analysis has been started with the homogenisation of the ZTD time series, which is an important task to provide homogeneity over the long-term. Here we used all previously reported discontinuities for a single station along with those added after manually inspecting the time series. This procedure did lead to a total number of 2505 discontinuities for this data set. Next, all significant oscillations were identified with spectral analysis and thereafter modelled with a Least-Squares Method. The residuals were subjected to noise analysis with different stochastic models. The results showed that an autoregressive model of fourth order combined with a white noise process is the optimal model for the ZTD time series. Finally, we provide an optimum evaluation of the ZTD trends and their uncertainties for selected climate zones, which were established according to the Köppen-Geiger climate classification.