The authors have developed a global geomorphic model of fluvial floodplains, notably at 90 m resolution. The authors made use of global elevation, flow direction, and drainage area models along with the global HydroBASINS boundaries for their analysis.

The methods presented here closely follow the methods of Nardi et al., 2019 (i.e., GFPlain250) which uses height above nearest drainage (HAND) with a floodplain hydraulic geometry (FHG) thresholding scheme. FHG suggests that potential inundation depth can be represented as a function of a river’s upstream draining area. Here, the authors argue that FHG parameters optimized for each basin, as opposed to global values, will better represent the spatial heterogeneity of global basins and ultimately be of benefit to floodplain delineation. 

The authors propose an iterative process with starting values based on previous knowledge that converges on suitable parameter values for each global basin. The authors found that Parameter b in the FHG model loosely corelates with a basin’s aridity index. Finally, the authors use two global hydrodynamic inundation maps (JRC and GAR) and another geomorphic floodplain model (GFPlain250) as reference data for comparison.

The authors have a logical claim; basins across the world are heterogenous and locally optimized FHG parameters could produce better models of floodplains when compared to global parameters. Of note, Nardi et al., justified the use of global coefficients by finding reasonable measure-of-fit values with varying b parameters. However, they also supported the notion that regional values for the scaling law parameterization could be further refined to capture local climatic variations.

SHIFT data and code were easily accessible. In North America, SHIFT aligns well with GFPlain with some noticeable differences. Specifically, in North America, SHIFT tends to estimate a narrower floodplain in comparison to GFPlain.  Both products have notable examples of areas identified as floodplains that are omitted by the other. 

I have several comments I would encourage the authors to consider.

• 170 - I’m unclear on why 34 ‘major’ river basins were selected for further analysis. The authors rely on the results in these 34 basins as evidence throughout their paper. Please explain the selection of these basins, what is significant about them, and what is the justification for analyzing them independently. 
• 233 – This could use more explanation. Why did the authors choose to define river as a function of UPA versus using the delineated river network in MERIT Hydro? Why select 1000 km2? The authors touch on this at the end of the paper.
• 350 – What method was used to resample to 1-km? 
• 352 – I see many permanent water bodies in the final SHIFT product. (e.g., the North American Great Lakes).
• 450 - Use of overall accuracy overly rewards correctly classifying the 94.5% (author’s estimates) of the world’s land area that is not a floodplain. I would be more persuaded by overall accuracy if the authors were to limit their accuracy analysis to some reasonable distance from your river network (e.g., 1km, 10km). 
• 454 – I would think to prove “the effectiveness of our parameter estimation scheme in capturing information from the reference maps”, I would need to see this same accuracy measurements but with global values used (e.g., the Ndari et. al., values: b= 0.3, a = 0.01) and the deltas.
• 472 – “Superiority” is an overstatement. Agreement does not equate to superiority.
• 560 – I’m not sure I would call FHG correlation with hydroclimatic conditions ‘reasonable’. There is a loose correlation. Earlier the authors described it as ‘statistically significant but weak’.  That is a more apt description.
• 561 – I’m not convinced this loose correlation proves effectiveness of the methods.
• 566 – I’m not convinced of “superior consistency”. Sometimes SHIFT is part of the highest agreement pair in a basin and sometimes it is not (Fig 7). The authors mention “superior consistencies” in the abstract as well. I’m not sure how to interpret that phrase.
• Fig 7 – It looks like GFPlain has higher agreement with GAR and SHIFT has higher agreement with JRC. Any explanations as to why this is?
• Why include JRC & GAR and SHIFT & GFPlain combinations in the choropleth map? I’m less interested in where the two hydrodynamic models (JRC & GAR) or the two geomorphic models (GFPlain & SHIFT) agree and I’m more interested in where SHIFT outperforms or underperforms against GFPlain. That is, where does GFPlain better align with hydrodynamic models and where does SHIFT better align with hydrodynamic models?
• Fig 7 - The color combinations for SHIFT + GAR and JRC + GFPLAIN are indistinguishable.

General: The authors argue that locally optimized FHG parameters better represent the climatic heterogeneity of the world’s basins than using global parameters. I would be more persuaded by a direct comparison of the two methods. That can be accomplished either by comparing SHIFT to the results of the author’s methods but with global FHG parameters (e.g., Fig 6 using b = 0.3, a = 0.01 globally) or a direct comparison of SHIFT and GFPlain to reference data (e.g., Fig 7 without the JRC & GAR and SHIFT & GFPlain250 combinations)

Essentially, the question is: Do locally optimized FHG parameters meaningfully improve the delineation of floodplains over global parameters and is there a spatial pattern of where those improvements occur? Any answer to those questions would be useful information for the community.

This manuscript tackles the floodplain mapping at the global scale. The authors use a methodology to estimate floodplains using a geomorphic approach (integrating heterogeneity), based on past studies such as Nardi et al. (2019). This methodology involves applying a geomorphic descriptor such as HAND (Height Above the Nearest Drainage) and globally optimizing its parameters to delimit floodplains, resulting in a global map with a resolution of 250m. In this study, the authors take a further step by optimizing the parameters of the same geomorphic descriptor (HAND) for more than 200 basins (delimitation at level 3 with respect to HydroBASINS). They consider heterogeneity in the production of a new map on a global scale, which is provided with resolutions of approximately 90m and 1km.

For the calibration and validation of the applied methodology, they relied on 500-year return period maps (JRC, GAR flood maps) and on the 250m Nardi resolution map (GFPlain250m), which presented a general precision greater than 0.85. Additionally, they facilitate access to the results through the following links: available at https://zenodo.org/records/10440609 and the main code at https://github.com/Mostaaaaa/SHIFT_floodplain.

The study is of interest and may be worthy to be published, but some effort should be made to better emphasize the impact of the study. In the following, you will find my comments.

Major comments

• The scaling of hydraulic depth is investigated at the global level, obtaining a very scattered graph. Data seem to be better aligned for larger basins, but some additional effort should be spent to explain the variability observed in other river basins. Climate cannot be the only variable controlling the scaling exponent. Other factors such as rainfall, river morphology, or land use could also impact the result.
• In Section 4.1, The authors discuss the uncertainty associated to the parameter b. In this section, results are not clear or do not display a clear pattern. It is also surprising that the results obtained over the larger basins still have a large uncertainty even if the regression function works better.
• Results should be better described. For instance, it would be valuable to have floodplain patterns obtained from SHIFT with the river network layer and one image showing the differences between SHIFT and a reference map. Additionally, it would be good to enlarge the images in Figure 4.

Minor edits:

1. Line 120-123: ‘overestimated floodplains in arid or semi-arid area as reported by existing assessments of geomorphic floodplains’ (Dhote et al., 2023; Lindersson et al., 2021). In these references, only Lindersson et al. refers to arid areas and their difficulties. While Dhote et al. only highlights the overestimation and underestimation of the descriptors HAND and TWI respectively, but does not talk about the relationship with arid areas.
2. Line 314-315: ‘This iterative process stops either when every data point fits within all moving windows, or if the procedure fails to converge towards a stable solution’. It could explain what is meant by a stable solution, for a better understanding.
3. Line 343: change ‘as the as the’ for ‘as the’.
4. Line 385: only the range of values obtained for the coefficient 'a', For what reason is not presented a graph as in fig.3 of parameter 'b'? If it is possible to provide the values of both parameters, so that this method can be studied at smaller scales focusing future studies in a single basin or a single region and its sub-basins.

It would be ideal to base the importance given to the parameter 'b' on the parameter 'a'.

1. 3b: include a legend.
2. 3a: change 'estimatio' for 'estimation' in the description.