Summary

The submitted manuscript examines elevation dependent warming (EDW) over the global tropics and subtropics, using a combination of gridded analyses, reanalyses, and global climate model simulations. With these datasets, the authors characterize the magnitude of EDW, the relative roles of forced response and internal variability in EDW, and the roles of various forcings and feedbacks in shaping EDW. The authors do a great job of focusing in on fundamental aspects of EDW and addressing them with thoughtful, clear, concise, and theoretically sound analyses. The addressing of forced response vs. internal variability and the formal diagnosis of mechanisms in a forcing/feedback framework are novel and very valuable additions to the EDW literature. The prose is crisp and easy to follow. The figures nicely summarize the results. The authors leave some questions unanswered (e.g., factors governing the role of cloud feedbacks in EDW, the role of unresolved terrain) but those omissions are reasonable given the scope of the study and the limitations of the datasets analyzed. I only have some small points for the authors to consider before the manuscript should be ready for publication. 

General comments

1. I think it is OK that the authors choose to take a global perspective and focus on global datasets that facilitate their analyses. However, I think that the authors should at least briefly acknowledge/discuss what might be missed in such a framework. Regions of interest for EDW are often in complex terrain that is not sampled well by observation networks used in historical (re)analyses. Furthermore, the terrain features associated with EDW may be poorly resolved in GCMs, which can bias the representation of key processes like accumulation of mountain snow cover (relevant to the snow albedo feedback) and the orographic clouds (relevant to cloud feedbacks). How much do the authors think these factors might affect their results? 
2. In several places, the authors mention that the surface albedo feedback is expected to be "stronger at colder temperatures" (or something similar). I don't think this is quite right. The snow albedo feedback, which dominates the surface albedo feedback over land, is not necessarily stronger at cold temperatures. It is often strongest when near zero deg. C, since small temperature changes can have big effects on melting and rain vs. snow. I suggest the authors tweak their wording to reflect this.

Specific comments

• Introduction: Though it is a few years old, the Pepin et al. (2015) EDW review paper still seems to be the best overall review on the topic. Though they site this eventually, I think the authors should cite it more prominently in the general intro as well. 
• L 28–29: might rephrase this to focus on regions where snow/ice is plentiful, but near freezing point of water (very high/cold locations may be insensitive to the SAF, due to persistently below-freezing condition in current and perturbed  climate states). See general comment #2.
• L 35–36: Some studies have attempted to quantify and compare the role of different physical mechanisms in producing EDW, though perhaps in a less global and/or comprehensive manner as this study. Those contributions should be more clearly acknowledged here. 
• L 50–51: Two potential issues with this dataset should at least be acknowledged:  1. This relatively coarse resolution is insufficient to resolve many important topographic features 2. Trends for many high-elevation grid cells may have large uncertainties, due to a lack of station locations in mountainous regions.  
• L 50: "observational side...": I suggest adjusting this wording when writing about ERA5, since it is not purely observation-based. 
• L 53–55: What is the grid spacing for the CESM1-LE runs?
• L 57: To what extent might the relatively coarse resolution of the models used (as compared to relevant terrain features) bias the simulated EDW?
• L 93–95: See general comment #2
• Section 4: This section provides a nice perspective that is often missing from the EDW literature
• Section 5.3: This analysis is great. It concisely breaks down the relevant physical processes without falling into some of the analysis pitfalls of many previous studies that take a more statistical approach. 
• L 259–260: See general comment #2
• L 273–276: One key aspect that is worth mentioning is the strong influence that mountains have on modulating clouds. The role of the terrain itself in modulating the cloud cover response could be an important part of the story (e.g., https://doi.org/10.1175/JCLI-D-21-0379.1). Also, some of these orographic effects on clouds may not be well resolved in the relatively coarse resolution simulations considered here. 
• Section 6: Given the authors' focus on the tropics/subtropics, it would be nice to include a few sentences of discussion to place their results more specifically in the context of previous EDW work that focuses on that part of the world. 

The authors use the forcing-feedback framework and a feedback analysis based on radiative kernels to examine the mechanisms causing amplified warming at higher elevations in low latitudes. The science is strong, and the manuscript is clear. I have two major and a few minor comments:

I see some similarity between the elevation-dependent warming described here and the overall tendency for amplified warming in the upper Tropical troposphere compared to the surface. The latter is caused by the fact that moist deep convection keeps the Tropical lapse rate close to the moist adiabat, which is steeper in warmer climates. This is somewhat akin to the finding that MSE convergence drives EDW, and both are connected to the increase in water vapour in a warming climate following Clausius-Clapeyron.

The authors derive forcing as a residual – could you use the CO2 kernel instead, possibly with more idealized runs? Deriving forcing as a residual means there is no residual error to check the quality of the decomposition, which is unfortunate.

Minor comments:

l7: For people who do not know the topic, the sign of EDW (larger warming at high elevation) is first mentioned halfway through the abstract, and it is not quite clear if this is a new finding or corresponds to what was known before.

Fig. 1: Why use standard seasons when the data spans the equator? Wouldn’t local summer/winter be more consistent?

If you insist of changing the axis scaling within a Figure, please make a clear visual mark of that.

Fig. 2 What are the gray lines?

Footnote p 15 – could you include this in the main text?

l. 316 – see comment above, CC-relation plays a role in MSE gradients