Continuous long-term sound sources are recorded at hydroacoustic station H03S, a threeelement hydrophone array south of Robinson Crusoe Island between 2014 April 23 and 2017 August 20. The origin of the signal between 3 and 20 Hz is investigated by using cross-correlation, array processing using plane wave beamforming and spectral analysis. Onebit normalization is successfully applied as a cross-correlation pre-processing step in order to suppress undesired earthquake events in the data. Traveltimes retrieved from averaged cross-correlations do not yield a coherent array direction of arrival. Averaged envelopes of the cross-correlations, however, indicate a coherent signal approaching H03S from a southsouthwest direction. Beamforming indicates two dominant backazimuth directions: 172 •-224 • (Antarctica) and 242 • (Monowai Volcanic Seamount). This continuous source field creates possibilities to investigate the applicability of acoustic thermometry at hydrophones H03 S1-S2. Cross-correlation and array processing indicate significant directional variation in local propagation, most likely related to the steep slope in the bathymetry near H03S. In addition, it is demonstrated that the ambient noise field is not sufficiently equipartitioned. It is shown that this causes a large error in the estimated temperature, primarily due to the short receiver spacing. These large errors have not been addressed in previous studies on deep-ocean acoustic thermometry. Hence, it is shown that acoustic thermometry does not perform well on small arrays such as H03S. The power spectral density yields a strong broadband signal in January-March, most likely related to iceberg noise. A narrow banded signal around 17 Hz between April and September corresponds to whale calls. The best-beam sound pressure levels towards Antarctica are compared to ERA5 climatologies for sea ice cover and normalized stress into the ocean, supporting the hypothesis of iceberg noise.