Current general circulation models are affected by biases in simulated clouds in the Southern Ocean (SO) which are despite some recent progress are still too large, resulting in biases in shortwave and longwave radiative transfer. These biases are also present in the HadGEM3 model, a UK Met Office model which has been a focus of development in the Deep South National Science Challenge with the aim to more accurately project future climate in New Zealand. The amount of cloud has more ice phase than expected. Satellite observations have been used extensively to study this problem, but the predominantly low level cloud in the SO cannot be reliably observed from space due to the prevalence of overlapping cloud and other limitations such as ground clutter in active instruments. We use observational data from a number of SO voyages to assess clouds in HadGEM3, and contrast it with the MERRA-2 reanalysis, which provides very different cloud phase and cloud occurrence results in the SO. We use ceilometer observations collected on voyages to assess cloud vertical distribution, mini micropulse lidar observations to assess cloud phase and radiosonde observations to assess tropospheric stability and humidity profiles. The ceilometer observations cannot be compared directly with the model due to attenuation of the lidar beam in thick cloud, and we use a ceilometer simulator developed for this project for a like-for-like comparison. We also apply this evaluation to a number of experimental HadGEM3 runs produced with different choices of cloud scheme parameters in order to fix the SO cloud bias.