Cross-validation of GPS tomography models and methodological improvements using CORS network

Abstract. Using data from the Continuously Operating Reference Stations (CORS), recorded in March 2010 during severe weather in the Victoria State, in southern Australia, sensitivity and statistical results of GPS tomography retrievals (water vapour density and wet refractivity) from 5 models have been tested and verified – considering independent observations from radiosonde and radio occultation profiles. The impact of initial conditions, associated with different time-convergence of tomography inversion, can reduce the normalised RMS of the tomography solution with respect to radiosonde estimates by a multiple (up to more than 3). Thereby it is illustrated that the quality of the apriori data in combination with iterative processing is critical, independently of the choice of the tomography model. However, the use of data stacking and pseudo-slant observations can significantly improve the quality of the retrievals, due to a better geometrical distribution and a better coverage of mid- and low-tropospheric parts. Besides, the impact of the uncertainty of GPS observations has been investigated, showing the interest of using several sets of data input to evaluate tomography retrievals in comparison to independent external measurements, and to estimate simultaneously the quality of NWP outputs. Finally, a comparison of our multi-model tomography with numerical weather prediction from ACCESS-A model shows the relevant use of tomography retrieval to improve the understanding of such severe weather conditions, especially about the initiation of the deep convection.