Review of “Bottom mixed layer derivation and spatial variability over the 

central and eastern abyssal Pacific Ocean” 

The authors use a series of historical and novel abyssal observations in the Eastern Pacific and report on the spatial variability of the bottom boundary layer. This is achieved through testing a series of methods to identify the bottom boundary layer. Following this they use a machine learning technique to identify the contributions to that variability from the bottom topography (depth, slope and roughness) and internal tide dissipation. To the best of my knowledge, this manuscript would be the first publication applying this machine learning approach to the bottom boundary layer. I believe that, following some additional work to strengthen the paper as described below, it would be worthy of publication in Ocean Science.  

It is not clear to me why the authors made the choice to infer the salinity using the GMM. If I understood line 112 correctly there is a CTD on the bottom lander system so I expect they have measured salinity? If there is something I missed here, I think it should be made clearer in the text. At the very least, I would like to see some of derived salinity profiles or T-S diagrams with a comparison with what salinity data is available (either from the cruises or from climatology) to give confidence. 

It is useful somewhere to identify that there is potentially an important difference between a mixed layer defined by hydrography and an active mixing layer where turbulence is acting over the layer. This is very important when inferring the importance of the boundary layer in e.g. water mass transformation. The surface mixed layer community have started to address this nuance and we should too. 

The machine learning technique (as I understand it) is simply identifying where the bottom mixed layer height has the same spatial structure as the potential contributions. This leaves me a little concerned about how to interpret the results given the apparent overlap between some of the inputs. For example, the two main components identified (the bottom depth and the total dissipation) have very similar distributions in Figure 7. I think that the uncertainties and limitations of this method need to be discussed more in the text and acknowledged in the interpretation of the results. 

A thought that you might want to bring into your discussion (optional), if the machine learning is identifying the total dissipation as an important control (subject to my concerns above) does that imply that the hydrographic definition of the boundary layer is likely similar to the active mixing layer for the bottom boundary layer for this region. It also implies strong local control of the hydrographic boundary layer thickness rather than control by relatively far-field topography (e.g. set by the sill depth between basins). This could be an important result. 

Line 129 – 130 > It is not clear to me what this sentence means 

Line 133 – 136 > These sentences seem repetitive 

Line 145 > Define what the difference is relative to (seafloor / deepest part of profile / etc) 

Line 195 > Remove remained 

Line 222 > repetition of datasets 

LIne 301 > I didn’t understand why the importance of dissipation was enhanced here. If I understood correctly both the dissipation and bottom depth are acting in the same direction to reduce the BML so either could be driving the change?