This is a continuation project. For my Rymdstyrelsen project ”MAPS” (dnr 2022-00149), we wanted to - train a CNN-type of system to detect a specific pattern in the Arctic sea ice; - [since this worked very well] apply this method both to observations and climate model output. Since the pattern we are interested in is a rare event, occurring in approximately 0.1% of the pixels of the individual image, we used a Unet architecture. The classification returned many false positive pixels, so we added a filtering step using a Gaussian Mixture Model. The GMM was the heaviest to run and the reason why we exceeded our quota on several occasions. The work on observations was published a few months ago in The Cryosphere. Starting this summer, we expanded the work to detect the same feature in climate model output, using all models available (18 with daily sea ice data, and a further 35 with monthly data) and applying to them our combined Unet+GMM method. We also used storage resources for ancillary data (atmosphere and ocean variables), to explain the causes and look at the consequences of our sea ice pattern. The GMM again had to be adjusted and tested, since models and observations need different thresholds, which consumed our resources in December and January. This work will be submitted before the end of the month; the complete draft is with the co-authors. We therefore need to continue having access to compute and storage resources, at least while the manuscript that is about to be submitted undergoes peer-review. I suspect that the reviewers will complain that the GMM still leaves out too many false-positives and that we need to fine-tune its threshold. We are also awaiting a recruitment decision that would provide a second PhD student to work on this project, who would adapt our method to detect that same pattern in summer, which is way harder.