Homogenization of GNSS IWV time series and estimation of climatic trends

Water vapor plays a key role in the Earth's climate as a dominant greenhouse gas. It is also the most efficient actor of heat transfer from the surface to the atmosphere and from low to high latitudes which shapes the global atmospheric circulation and weather systems. Monitoring and understanding the spatial and temporal variability and changes of water vapor are thus of crucial importance.

This work aims at computing decadal trends of total column Integrated Water Vapour (IWV) from a global network of ground-based GNSS observations. Although GNSS observations are available with high accuracy in all weather conditions, it has been shown that, over long periods of time, changes in instrumentation, in station location and environment, and in processing methods can introduce spurious shifts in the IWV time series and bias trend estimates. Homogenization is a crucial step to detect and correct such non-climatic signals.

We have developed a relative homogenization method which involves three steps.

• Segmentation. First, change-points are detected from the difference series (GNSS – reference) with the help of the GNSSseg segmentation package (Quarello et al., 2022). The method uses a difference series in order to cancel out the common climatic variations. It also accounts for changes in the variance on fixed intervals (monthly) and a periodic bias (annual) due to representativeness differences between GNSS and the reference (in our case the ERA5 reanalysis). Because the change-points detected in the difference series could be either due to GNSS or to the reference (ERA5), the next step is the attribution.
• Attribution. Second, the detected change-points are attributed to either GNSS or to the reference (ERA5) using a statistical test based on linear regression and a predictive rule based on the Random Forest learning algorithm (Nguyen et al., 2023). This step requires additional neighbors stations (at least one).
• Correction. The last step is the correction. Here the initial GNSS series is corrected only for the shifts which are attributed to the GNSS in the second step.

We will present results of the homogenization procedure applied to a global network of GNSS stations and discuss the impact of homogenization on linear trend estimates for stations that have more than 20 years of observations.

Quarello et al., 2022, https://doi.org/10.3390/rs14143379

Nguyen et al., 2023, https://hal-obspm.ccsd.cnrs.fr/IGN-ENSG/hal-04014145v1