This data set contains daily files of vertically integrated water vapour (precipitable water, prw) derived from combined measurements of the HATPRO and MiRAC-P microwave radiometers (see Walbröl et al. 2022, https://doi.org/10.1038/s41597-022-01504-1) operated aboard the Polarstern during cruise PS122 (MOSAiC expedition). The data cover the time range from 22 October 2019 until 02 October 2020 with a temporal resolution of about 1 second. Both microwave radiometers were mounted on top of the OCEANET-ATMOSPHERE container (Macke et al. 2010, doi:10.1051/epjconf/201009006; Engelmann et al. 2021, doi:10.5194/acp-21-13397-2021). HATPRO is a Humidity and Temperature Profiler (Rose et al. 2005, doi:10.1016/j.atmosres.2004.12.005) that measures atmospheric radiation emitted from water vapour along a rotational water vapour aborption line in the K-band (7 channels from 22.24 to 31.4 GHz) and from oxygen along the absorption complex in the V-band (7 channels from 51.26 to 58 GHz). The HATPRO measurements are also sensitive to radiation emitted from hydrometeors (cloud liquid water). MiRAC-P is a Low Humidity Profiler (LHUMPRO-243-340, Mech et al. 2019, doi:10.5194/amt-12-5019-2019), also known as the Microwave Radiometer for Arctic Clouds - Passive, that measures atmospheric radiation at higher frequencies along a strong water vapour absorption line in the G-band (6 channels from 183.31 +/- 0.6 to 183.31 +/- 7.5 GHz), and in the window channels (no distinct absorption lines) 243 and 340 GHz. We developed a Neural Network retrieval to derive improved integrated water vapour estimates through the combination of zenith HATPRO and MiRAC-P observations compared to single instrument retrievals (see Walbröl et al. 2024; doi:10.5194/egusphere-2024-1301). Error estimates in terms of root mean squared error are included in each file as a variable (prw_rmse). Quality flags of each instrument (flag_h for HATPRO and flag_m for MiRAC-P) are also included in each file and can be used to filter data with bad quality. The data has good quality when the flag values (flag_h and flag_m) are 0 or nan (depends on the importing routine). Additionally, data where flag_h is 32 can also be used with care.