[G07-4] GGOS observations: GNSS and co-locations

[G07-4-04] Performance of various homogenization tools on a synthetic benchmark dataset of GPS and ERA-interim IWV differences

^○Anna Klos¹, Roeland Van Malderen², Eric Pottiaux³, Olivier Bock⁴, Janusz Bogusz¹, Barbara Chimani⁵, Michal Elias⁶, Marta Gruszczynska¹, Jose Guijarro⁷, Selma Zengin Kazanci⁸, Tong Ning⁹ (1.Military University of Technology, Warsaw, Poland, 2.Royal Meteorological Institute of Belgium (RMI), Brussels, Belgium, 3.Royal Observatory of Belgium (ROB), Brussels, Belgium, 4.IGN LAREG, University Paris Diderot, Sorbonne Paris, France, 5.Central Institute for Meteorology and Geodynamics, Austria, 6.Research Institute of Geodesy, Topography and Cartography, Czech Republic, 7.AEMET (Spanish Meteorological Agency), Spain, 8.Karadeniz Technical University, Turkey, 9.Lantmäteriet, Sweden)

Within the COST Action ES1206 “Advanced Global Navigation Satellite Systems tropospheric products for monitoring severe weather events and climate" (GNSS4SWEC), a sub-Working Group (WG) on “Data Homogenization" has been set up. The aim of this group is to homogenize a worldwide Integrated Water Vapour (IWV) dataset retrieved from Global Positioning System (GPS) observations, the IGS repro 1 tropospheric product (1995-2010), by correcting for (artificial) break points due to e.g. instrumental changes. As at most stations, the ERA-interim IWV field output correlates well with the IWV retrieved by GPS, the former one is used as reference and the IWV differences between both sets are considered. The characterization of these IWV differences provided us typical trend values, seasonal oscillations and noise models, to build a synthetic benchmark IWV dataset of differences. We simulated these IWV differences over a period of 16.4 years with two different noise types: white, as well as the combination of white and autoregressive process. Then, we simulated offsets, trends and seasonal oscillations, as characterized from the real IWV differences. We created three variants of synthetic datasets referred to as: EASY, LESS-COMPLICATED and FULLY-COMPLICATED, depending on the level of complexity. All synthetic datasets were then subjected to homogenization: various detection methods (e.g. HOMOP, CLIMATOL, PMTred, STARS and non-parametric rank tests) were blindly employed to deliver the epochs of simulated offsets. In this presentation, we show the detection scores for each synthetic dataset type and each detection method We analyze the sensitivity of each detection method w.r.t. the complexity of the synthetic datasets. We also present the trend differences between before and after homogenization by the different approaches.