Hi, just heard the great talk by @liellnima at the ML4ESM workshop, and was super excited to see that xMIP is used! I wanted to reach out and see if there is any way I can help integrate xMIP better with ClimateSet. I particular I noted this in the [paper](https://arxiv.org/pdf/2311.03721) ``` The structure processor was implemented with xmip (Busecke and Abernathey, 2020) and follows their guidelines. Note that it is significantly slower than the CDO-implemented modules. ``` I am very curious to learn what causes the performance difference. Most of the corrections of xmip are applied lazily and should be quite cheap! If you have a minimal example that we could work through, it would be great to post an issue on [the xmip issue tracker](https://github.com/jbusecke/xMIP/issues). A small comment on the citation: It would be tremendous if you could update the citation to the actual software package [doi:10.5281/zenodo.7519179](https://zenodo.org/badge/latestdoi/215606850) in the published paper. That would help me a lot in tracking useage and justify working on this software. Again many thanks for using xMIP and please feel free to reach out with any questions/comments.