The authors present a consolidated version of MESMER, a fast climate model emulator based on a previously published statistical methodology. This is a well-written technical contribution that I enjoyed reading. I commend the effort of the team in making the code more accessible, with a focus on software engineering aspects that (as the authors also note) are often overlooked in research code. I also appreciate the effort of bringing the different flavors of MESMER together, which I think could prove useful in making the emulator more widely used by a larger community of researchers and/or stakeholders that are hungry for climate information. I have a few minor comments on the paper that could be worth addressing, and some food for thought for future releases of MESMER given the current state of the model.

Minor Comments to the authors:

1. L56: “land surface annual mean temperatures” – I’m assuming this is actually surface air temperature over land (tas in CMIP6 jargon), and not the actual land surface temperature. Worth rephrasing for clarity?

2. On a similar note (L343): is there a reason why you chose to exclude the ocean from your emulation? It seems that the main audience of the code is researchers, and I can think of a lot of research problems where the ocean response matters significantly (evaporation, hydrological cycle changes, low cloud feedbacks and boundary layer stability, SST pattern effect, dominant modes of internal variability [ENSO/NAO/…], ocean heat content and TOA radiation,....) . Even for general stakeholders, things like marine heat waves are becoming more prominent and relevant for impact assessments, and knowing the ocean response seems quite key. Maybe worth including (not necessarily in this paper, but in the future)? It would probably even simplify the code to some extent as no masking would be necessary.

3. Figure 4 caption and elsewhere: I just wanted to flag the possibility for any web link to be broken in the near future (few years timescale), whereas the paper is usually thought to be a much longer-term stand-alone contribution that will likely outlive the current web links (e.g. if you decide to change domain, not maintain mesmer anymore, or other possible unforeseen circumstances the links will be broken with no real possibility of updating them, since the paper after publishing is usually not updated anymore). Not necessarily suggesting to remove links (I see why they could be useful), but something to think about. I’m flagging this because I’ve recently read a few papers from the 2010s with links embedded in them that are now broken and not really useful anymore.

4. L383: can you be a little more precise on what you mean by “annual maximum temperature”? i.e., what time resolution is used to calculate the maximum (for example, is that the annual maximum monthly temperature, the annual maximum daily temperature or the annual maximum hourly temperature). Might be worth using CMIP6 jargon if it is defined in CMIP6 and it is not a postprocessed variable.

5. I tried to pip install the mesmer-emulator package and run the “Emulating near surface temperature on land with MESMER” tutorial, but it looks like the mesmer.example_data file was not included in the pip install (I believe along all the other *.py files in the mesmer/mesmer directory, e.g. anomaly.py, etc). I am using an Ubuntu machine and just followed the installation commands in the doc page, i.e. ran python -m pip install mesmer-emulator[complete] in a virtual environment. In my experience, a research software package becomes quite successful if the user experience is frictionless, meaning that it works out-of-the-box without much debugging to be done on the user side. Small details like this one might discourage users to keep digging into mesmer, which would undermine lots of great work that could be really useful to a lot of people. Just a suggestion to pay attention to these details and extensively test with different people, machines, …

6. On a similar note to #5, I wonder if it could be useful to develop an out-of-the-box version of mesmer (sort of like an old-fashioned executable, or maybe some high-level wrappers) that produces some basic output based on a few input parameters (e.g. N realizations of gridded tas for a specified experiment and model, where N, model, experiment are input parameters). In other words, I found the tutorials a bit too detailed (there is still a lot of processing, data handling, … ) for the casual user that just wants to try the model out.  Or perhaps there could be a range of tutorials, from a really simple one for the casual user, to the more processing/data handling ones that could be geared towards developers and people that want to dig more into mesmer. Not necessarily for this paper but it’s something that might be worth thinking about for the future if the goal is to make mesmer a widely used emulator in the community.