A major challenge in cryoseismology is that signals of interest are often buried within the high noise level emitted by a variety of environmental processes. Particular DistributedAcoustic Sensing (DAS) data often suffers from low signal-to-noise ratios (SNR) poten-tially resulting in a multitude of undetected events of interest, which further remain un-analyzed. To record seismicity, we deployed a DAS system on Rhône Glacier, Switzer-land, using a 9 km long fiber-optic cable that covered the entire glacier, from its accu-mulation to its ablation zone. The highly active and dynamic cryospheric environment,in combination with poor coupling, resulted in DAS data characterized by a low SNR.Our objective is to develop and evaluate a method to effectively denoise this cryoseis-mological DAS dataset, while comparing our approach to state-of-the-art filtering anddenoising methods. We propose the J-invariantcryo denoiser, specifically trained on cryo-seismological data and capable of separating incoherent environmental noise from tem-porally and spatially coherent signals of interest, based on a self-supervised J-invariantU-Net autoencoder. The method enhances inter-channel coherence, improves waveformsimilarity with co-located seismometers, and increases SNR. The comparison of differ-ent methods shows that our approach obtains the highest gain in SNR and highest sim-ilarity with co-located seismometers, while suffering from denoising artifacts in rare cases.The proposed denoiser has the potential to enhance the detection capabilities of eventsof interest in cryoseismological DAS data, hence to improve the understanding of pro-cesses within Alpine glaciers.