This manuscript outlines the development of a tool that can estimate the cost of groundwater pumping across the globe. The cost and availability of groundwater pumping is a timely and important area of research, particularly in the context of climate change and increased pressures on our water supplies. My expertise is in hydrogeology and water management, and as such I will be primarily commenting on those aspects.

This manuscript is very well written and brings aspects of hydrogeology and economics together in a clear manner. I do have several concerns with respect to the methodology, particularly the explanation and description of the drawdown assessments, and with the exclusion of any other aspect of the hydrologic cycle within the analysis.

• I don’t feel that this tool evaluates groundwater supply, as described by the title and implemented in the research. It seems to provide one static quantification of availability but doesn’t include any other aspect of the hydrologic cycle that effects the ever evolving groundwater volumes and availability across the globe. I think you could argue that this tool can evaluate changes in supply due to pumping, but the lack of connection with any other part of the hydrologic cycle makes the claim of evaluating groundwater supply very thin.
• Recharge needs to be mentioned WAY before the very end. There are various sources of estimated recharge rates across the globe. I understand that it would bring in a lot of uncertainty, but this is already wrought with uncertainty, I’m not sure how much it would change it. This is connected back to the exclusion of any other part of the hydrologic cycle.
• Lateral inflows are inherently part of Theis – it is assumed that there are infinite sources of water available laterally. Saying you aren’t including them is erroneous unless you have modified Theis to include boundary conditions of some sort.
• If the wells are pumped for 100 days (which may be very short for many parts of the globe), are they in recovery for the remaining part of the year? Is that simulated or do you just pause the groundwater levels after 100 days and start from there the next year? Both have obvious assumptions and limitations but it is not clear from the manuscript which approach is taken. I would hope that recovery is enabled through inclusion of modified Theis.
• Some discussion of the uncertainty in all of the datasets you use as inputs would be beneficial. Particularly because they are dependent on data that is now over a decade old. For example, how does Fan et al. (2013) capture different aquifer units? There are many instances where irrigators use deeper aquifer units that are overlain by shallow, unconfined units. In addition, how does Gleeson et al. (2014) capture this same issue? What about fractured rock aquifer which are prolific in many parts of the work and are very productive.

As a result of the concerns above, in addition to some more specific, yet related comments provided below, I suggest this manuscript be returned for major revisions.

Specific Comments

• Figure 3: Do you check to see if the well interference and Jacob correction result in a violation?
• Line 220: The accepted definition of saturated thickness is the depth from water table to a bottom confining unit, not to the depth of the well. The well can draw water from below as it follows pressure gradients. It is fine to keep this definition, but I would be clear that you are defining it much differently than the convention.
• Line 241-244: You are assuming that there are always adjacent wells? Or do you do this when the number of wells in the grid meet a certain criteria?
• Line 252-254: You should provide the main categories of aquifers that you use – I presume they are unconfined and confined? The reader should be provided this information without having to look through supplemental information. I would guess this category determines whether the correction is needed.
• Line 264: Here you mention an ‘off period’ – is this simulated as recovery (as per comment above)?
• Line 362: Some context for these two depths would strengthen this work – do they correlate with particular crops?
• Line 373-375: Some regions of the Ogallala are already well beyond these value, some having already depleted most of their resources. This treatment of the aquifer as one large unit is inconsistent with how it actually works. To account for this, just reword to say that on average, it was 30% depleted – if you want upper and lower bounds you can look at recent Kansas Geological Survey reports to see the ranges within Kansas – this would communicate to the reader that you understand that these units do not operate as one big bathtub.
• Section 3.2.1: I struggle with this whole section because nothing here is new or novel to the hydrogeology community. I understand that this manuscript is reaching a multidisciplinary audience but given the length of the manuscript I think it could be moved to supplemental information.
• Line 453: How does the Vavailable term change with time? Again, the problem with this is that you are completely removing GW from the hydrologic cycle. There is data and research that can support bringing it back in (e.g. inclusion of recharge), and I don’t feel that doing so is an unreasonable request.
• Figure 5: When is this representative of? Groundwater supply is not stationary and constant. Also interesting that the Great Lakes are not removed from this reporting, as with other inland lake regions - was there a reason for this?
• Section 6.1: Since you highlight the ability to work at a variety of scales (Figure 11), I would think that the first step could be to calibrate against smaller-scale depletion – for example the well documented depletion in the High Plains Aquifer that you discussed earlier.
• Section 6.3: As described in several previous comments, I think this is a bigger issue than this one paragraph insinuates. I don’t have additional comments beyond those given in sections above, but rather point to this as one place that can be extended to better capture the implications of the rest of the hydrologic cycle on this work.

Niazi and colleagues present a distributed model to estimate the cost of groundwater extraction based on the available water volume. Many regions in the world rely on groundwater extraction for their water supply. Meanwhile, climate change and anthropogenic activity have pushed the groundwater balance out of statistical stationarity. Thus, the topic of estimating groundwater extraction cost under changing environmental conditions is timely and of interest to the readership of the journal.

The manuscript is well-written and easy to follow. The introduction gives a concise and clear overview of the model's place in the existing state of research. The methodology is clearly outlined and the results give a good impression of the model's capability.

My major concern is that the spatio-temporal dynamics of groundwater recharge and non-anthropogenic losses such as phreatic root water uptake are neglected, which—in my opinion—makes the model operate under rather constraining assumptions that are unsuitable for long term predictions.

I appreciate the parsimony of the presented model and am aware of the additional complexity that adding (eco)hydrological processes would bring. But I think it is crucial to address this issue. Perhaps the most feasible way to address this issue would be a model coupling (one- or two-directional) with a global hydrological model that could account for the seasonal and long term spatio-temporal dynamics of groundwater.

Regarding the cost model, I have some minor questions:

1. How do you account for inflation?

2. For global and long-term predictions, how do you account for variation in the cost of energy?

To address my major comment, I recommend major revision of the manuscript.

The well-written paper by Niazi et al is about a physics-based groundwater extraction and cost accounting model (Superwell) that allows the computation of GW supply-cost curves at a global 50km x 50km resolution. The paper fills the gap between the available purely physical and purely economic datasets related to GW. Physical datasets include remotely sensed products, in-situ sensor networks, proxy measurements, numerical model outputs, and other methods for storage and fluxes that do not consider the costs of pumping and the related human feedback. Many of these datasets do not have global coverage.  Economic datasets are usually lumped and do not consider the hydrogeological properties of the underlying aquifers or the long term dynamics of the resource. As a result, most agent-based models that aim to study water management issues in a basin do not incorporate well-hydraulics. Similarly, basin-scale integrated assessment models that can be used for planning or scenario generation have rudimentary GW components due to the missing coupling between the physical and economic components of GW pumping.

As someone interested in coupled human-water systems and socio-hydrological studies at multiple scales, the reviewer is pleased to receive this work as it would allow the incorporation of the GW cost curves as simple look-up tables without the need to deal with the extreme spatiotemporal complexity of groundwater flows that couple with equally complex water usages. Superwell is based on a simplified first-principles  model that requires physical parameters such as porosity, permeability, aquifer thickness etc. Some other required parameters such as transmissivity, drawdown depth, hydraulic conductivity, radius of the well’s influence, etc. have been derived from the basic parameters. Similarly, economic parameters such as capital and maintenance costs, energy costs etc have been used. Some more complex aspects such as pumping behavior, regulations, transportation costs, treatment costs have not been modeled or simplified. For example, it is assumed that each well pumps for 100 days / year and the remaining 265 days are assumed for recovery. The hydro-economic computations are similarly simple and easy to track.

For each grided cell, 6 scenarios have been generated by crossing 2 ponded depths with 3 intensities of volume depletion. As a non-expert, I am unable to assess whether these choices are appropriate or if more scenarios should have been generated.

Some specific comments/questions are below:

1. The simplicity of the model and its potential for immediate incorporation into socio-hydrological models seems to be the model’s biggest strength. For example, the authors have indicated the ease of incorporation in an ABM-based study for irrigation decisions in the US (ref: Yoon et al. 2024) by simple look-up tables.
2. The biggest concern is that the GW model does not have a recharge component. Also, the interactions with natural surface flows and irrigation is missing. This is acknowledged by the authors but it makes the non-expert whether the hydrology can be relied upon.
3. The method for model calibration / evaluation is a bit complicated to follow. It is totally appreciated that the model only provides a plausible range of future pumping rates (hence the value of the multiple scenarios). But why historical data in limited geographies and limited timespans can not be used for validation is not clear (If it is due to the unavailability of global data, perhaps a regional downscaled study should be planned). While the significance of expert-based evaluation is fully appreciated, the methodology referred by the authors (Gleeson et al 2021) suggests all three methods simultaneously, i.e. observation based-comparisons, expert evaluations and model-based evaluations and their interdependence. Furthermore, they advocate uncertainty quantification for a robust evaluation.
4. Some behavioral assumptions, e.g. the well-deepening feature of the model and the choice of 100 days for pumping seem arbitrary. Can the authors provide some strong justifications for these choices?
5. The various applications mentioned in the paper are very exciting. The application paper by authors in Nature Sustainability (Niazi 2024d) opens exciting directions for future work and further development of the model.
6. The compilation and availability of both input and output datasets and code are themselves a major contribution, for which the authors must be congratulated.

Overall, this is a well written paper and it is recommended for publication with minor revisions.