This is a well written article that provides important insight into the complicated drivers of sporadic-E.  The developed model is the best to date at capturing the spatial and temporal variations, thereby providing a significant contribution to the field.  

Key Comments:

1) In the beginning of Section 3.2 Results of Reconstructing E-region Morphology, it would be helpful to provide more information on the COSMIC data used to create the model.  For example, what CDAAC file-types are you using and what is the time range of data used for analysis?

2) Throughout the paper, the data is binned and analyzed over a single variable to show trends.  For this binning, do you present the average value of the bin or some other parameter such as the maximum?  Also, when the data is displayed as a function of a single variable (such as Figure 5), do you average over all of the other parameters?  More information on the binning and averaging would be extremely helpful.  

Minor Comments:

-Line 78: Please define ECMWF here

-It would be helpful to add commas in the large numbers such as in 937592 in Line 124

-Line 126: We => we

-Line 131: It’s unclear what you mean by “The principle of RF is done”

-Lines 234-235: I suggest replacing “missing of Es” with “lack of Es”

-Line 245: Is this supposed to be a winter to summer meridional flow?

-Figure 3: It’s tough to see the variation from this color scheme because of the few high latitude patches with a large S4max…perhaps you could change the colorbar so the peak S4max color is lower, which would help to show more of the mid-latitude patterns.  

General Comments

This is a high quality manuscript offering a novel insight into how AI techniques can be used to support the prediction of Es layer morphology. The model is able to capture the Es climatology incredibly well, and therefore represents a significant step forwards in the field of Es layer modelling. A number of minor typographical corrections have been suggested to aid readability, but overall the article is well written and structured.

Specific Comments

Does the first paragraph in the results section (Lines 189-197) fit better in the methods section?

It is mentioned that the model offers “faster computation speed and a wider valid time scope” (Line 341) but the specific time scope is not detailed. In lines 357-358 it says the wider valid time scope spans approximately one full solar cycle. This would be useful to move to when it is first discussed in line 341. However it isn’t clear what the actual time range of the application and its predictability is – what calendar years is the model able to predict Es layers in? The time scope is wider than what? The SELF-ANN github page is referenced but the time scope isn’t clear from the readme there either.

Technical Corrections

• Line 1: Consider rephrasing to “Ionospheric sporadic E layers (Es) are intense plasma irregularities between 80 and 130 km in altitude, which are generally unpredictable”.
• Line 26: It may be useful to elaborate/ break this down further into the different formation mechanisms, splitting the references up.
• Line 88: “We has” -> “we have”
• Figure 3: Consider revising colour scheme to improve contrast between areas of high/low intensity. Do the boxes need to be shaded? Just a red outline may be clearer.
• Lines 284-285: Consider rephrasing to “due to the ions failing to converge vertically” or “since the ions fail to converge vertically”
• Line 297: “in this study, as depicted in” -> “in this study is depicted in”
• Line 321-324: Consider removing “trained on COSMIC satellite RO data” and putting foEs in brackets to aid readability.
• Line 325: Consider removing “square”.
• Line 392: “In sum” -> “In summary”