The radiosounding network in the Arctic, despite being sparse, is a crucial part of the atmospheric observing system for weather prediction and reanalysis. The spatial coverage of the network was evaluated using a numerical weather prediction model, comparing radiosonde observations from Arctic land stations and expeditions in the central Arctic Ocean with operational analyses and background fields (12-hr forecasts) from European Centre for Medium-Range Weather Forecasts for January 2016 to September 2018. The results show that the impact of radiosonde observations on analyses has large geographical variation. In data-sparse areas, such as the central Arctic Ocean, high-quality radiosonde observations substantially improve the analyses, while satellite observations are not able to compensate for the large spatial gap in the radiosounding network. In areas where the network is reasonably dense, the quality of background field is more related to how radiosonde observations are utilized in the assimilation and to the quality of those observations.