In spite of the lower accuracy of the NWM-derived ZTD when compared to the ZTD estimated from geodetic techniques such as GNSS, the comparison with a reanalysis model such as ERA Interim allows the detection of possible biases or drifts in the GNSS solutions. The comparison of the three types of GNSS-derived ZTD (U.Porto, IGS and EPN) with ERA Interim shows that both IGS and U.Porto solutions are stable with a uniform accuracy for the whole period, with a standard deviation of the differences with respect to ERA Interim of 12 mm. The EPN solutions evidence several discontinuities over the analysed period, the most relevant occurred in 4 November 2006, which introduces a bias of about several millimetres in most stations. Therefore, IGS online products are suitable for application in studies such as sea level change where the long-term stability of the products is an important issue. For the same purpose, the EPN products must be corrected before November 2006, for example, by removing the mean difference with respect to a stable source such as ERA Interim. In situ measurements of surface atmospheric pressure analysed for a set of 63 GNSS stations reveal that a number of stations have discontinuous time series, often several periods with invalid measurements and calibration problems (biases, drifts and jumps), requiring careful editing. This highlights the need for an adequate maintenance of the instrumentation, in order to improve the utility of these data sets. Although very important as ground truth information for validation purposes, these measurements are not suitable for an operational use in applications such as the GPD (GNSS-derived Path Delay) algorithm for computation of the wet path delay effect in coastal altimetry. Using only valid in situ observations at barometric sites free of calibration problems, results show that the global grids of sea level pressure provided by ECMWF operational model, either at 0.25 or 0.125 spacing, or the ERA Interim reanalysis product at 1.5 1.5, allow the estimation of the hydrostatic component of the tropospheric delay with an accuracy of 1–3 mm at a global scale, being better than 2 mm in 90% of the analysed sites for the operational model and in 87% of the sites for ERA Interim. This performance of ERA Interim is remarkable, considering the relatively low resolution of this product when compared with the finer grids used of the operational model.