We are glad the reviewer found the manuscript to be of interest, and appreciate their insightful and attentive comments. In our response below, we have copied and labeled (RX-X) individual reviewer comments followed by our response in italics to each.

R1-1. First, I would suggest adding more references to the “two-way” coupling studies to smooth the logic flow. While I think this paper might be one of the first to conduct a two-way coupling of an agent-based model with a hydrologic model at the US scale, the concept itself is not brand new. There are several previous studies have already done this at the basin scale (some of the basins are fairly large such as the Mekong, Niger, and Colorado Basins). Citing these previous studies will provide readers with a better understanding of how this two-way coupling concept was developed. 

We thank the reviewer for noting this and agree that additional references on two-way coupled ABM-hydrologic modeling studies would provide important context. Towards this end, we plan to add the following references to a paragraph in the introduction that points to the evolution of two-way coupled ABM-hydrologic models, primarily involving ABM couplings with SWAT and MODFLOW models. These efforts indeed introduce and demonstrate the two-way coupling concept that serve as an inspiration for our coupling of a farm ABM with a large-scale hydrological model.

Giuliani, M., Li, Y., Castelletti, A. and Gandolfi, C., 2016. A coupled human‐natural systems analysis of irrigated agriculture under changing climate. Water Resources Research, 52(9), pp.6928-6947.

Hyun, J.Y., Huang, S.Y., Yang, Y.C.E., Tidwell, V. and Macknick, J., 2019. Using a coupled agent-based modeling approach to analyze the role of risk perception in water management decisions. Hydrology and Earth System Sciences, 23(5), pp.2261-2278.

Yang, Y.E., Son, K., Hung, F. and Tidwell, V., 2020. Impact of climate change on adaptive management decisions in the face of water scarcity. Journal of Hydrology, 588, p.125015.

Castilla-Rho, J.C., Rojas, R., Andersen, M.S., Holley, C. and Mariethoz, G., 2017. Social tipping points in global groundwater management. Nature Human Behaviour, 1(9), pp.640-649.

Khan, H.F., Yang, Y.C., Xie, H. and Ringler, C., 2017. A coupled modeling framework for sustainable watershed management in transboundary river basins. Hydrology and Earth System Sciences, 21(12), pp.6275-6288.

Reeves, H.W. and Zellner, M.L., 2010. Linking MODFLOW with an agent‐based land‐use model to support decision making. Groundwater, 48(5), pp.649-660.

Lin, C.Y., Yang, Y.E., Malek, K. and Adam, J.C., 2022. An investigation of coupled natural human systems using a two-way coupled agent-based modeling framework. Environmental Modelling & Software, 155, p.105451.

Yang, J., Yang, Y.E., Chang, J., Zhang, J. and Yao, J., 2019. Impact of dam development and climate change on hydroecological conditions and natural hazard risk in the Mekong River Basin. Journal of Hydrology, 579, p.124177.

Yang, J., Yang, Y.E., Khan, H.F., Xie, H., Ringler, C., Ogilvie, A., Seidou, O., Djibo, A.G., Van Weert, F. and Tharme, R., 2018. Quantifying the sustainability of water availability for the water‐food‐energy‐ecosystem nexus in the Niger River Basin. Earth's future, 6(9), pp.1292-1310.

Klassert, C., Yoon, J., Sigel, K., Klauer, B., Talozi, S., Lachaut, T., Selby, P., Knox, S., Avisse, N., Tilmant, A. and Harou, J.J., 2023. Unexpected growth of an illegal water market. Nature Sustainability, pp.1-12.

R1-2. Second, I do think more technical details should be provided in the methodology section. While the development and the connection of the two sub-models are quite clear, several technical details are only mentioned briefly. For example, the setting of groundwater supply, the real-world meaning of the two PMP coefficients, and the calculation of the adjusted perceived cost of production. To me, it is totally ok to make some assumptions to make the model development feasible, but these details should be provided and the readers can judge themselves. Two related comments regarding methodology. 1) I think the authors should provide the ODD+D document in the supplementary material which becomes a standard in any ABM study. This will allow other ABMers to quickly understand the ABM setting. 2) I think a summary table to show the necessary data (sources, years, resolution, etc.) can help readers better understand the scope and scale of this model.

In the revised manuscript, we plan to add details on the treatment of groundwater supply in the framework, interpretation of the PMP coefficients, and the calculation of adjusted perceived cost of crop production. We likewise plan to incorporate a summary table on data sources that were used to initialize and parameterize the model. We will also add ODD+D documentation to the supplemental materials of the manuscript, which will provide further description of the technical details of the model in a standardized manner familiar to the agent-based modeling community.

R1-3. Third, I understand that this manuscript is testing a hypothetical experiment and I think it is fine for model development purposes. But I think the authors should still provide the calibration results and partially demonstrate that the model they developed (at least more or less) captures the overall trend and pattern of historical data (e.g., crop area and/or streamflow). Otherwise, it is very difficult to convince readers that the model is suitable for hypothetical experiments. One of the major criticisms of ABM is that these models are “toy models” that do not reflect reality. Since the authors use historical data to calibrate their PMP parameters, they should show the results as evidence that this is not just a toy model. Again, several two-way coupled ABM studies in recent years have already shown the calibration results to demonstrate the model's credibility. This is the most critical comment I have.    

We thank the reviewer for their comment regarding the importance of calibration. First, we note that the two individual sub-models have gone through their own independent calibration procedures (to a degree that is common in both their respective literatures). For the farmer sub-model, the data-driven positive mathematical programming approach and calibration is intrinsically embedded within the model development process. The identification of the PMP parameters during the first phase of model development is based on observed data, such that the PMP closely reproduces observed cropping areas for the historical period (we will provide test results in the supplemental materials of the revised manuscript to verify this). As such, the PMP can be viewed as a revealed preferences methodology for specifying farm behavior, which is a commonplace economic treatment of agent behavior but not without its limitations. With our revisions, we also plan to include sample local sensitivity tests (e.g., changes to crop prices, water availability, etc.) to demonstrate the reasonability of farm model response. For the VIC-MOSART-WM hydrologic modeling, attempts at calibration are detailed in various previous studies. The VIC simulation has been calibrated and is considered a benchmark for the United States Bureau of Reclamation (USBR, 2014), while the addition of the MOSART-WM component is evaluated on the basis of its performance in improving modeled river flow and reservoir storage outcomes to observations (Voisin et al., 2013; Hejazi et al., 2015).

The reviewer’s comment regarding calibration of the coupled model still remains, however (hereon we specifically focus on irrigated cropped areas as our calibration performance metric, given this is the novel outcome that our model treats dynamically). While we believe a calibration over the full historical period of interest is outside the scope of the effort for reasons we will elaborate below, we can nonetheless assess the reasonability and plausibility of model behavior by comparing irrigated crop area outcomes from our simulations with known cropping response during isolated periods of drought (over which we assume that non-hydrologic external factors are relatively stable). This can provide us with an indication of how our adaptive model is performing relative to the alternative (no cropping adaptation). Following the reviewer’s comment, we have conducted such a comparison for several states in the Western U.S. for the early 2000s period, during which much of the region was in drought. Over these isolated periods (over which we assume, albeit imperfectly, that non-hydrological conditions are relatively steady), our model generally reproduces changes in irrigated crop areas that are commensurate with observed changes (and in all but a couple of cases gets the direction of cropping change correct). This can be readily contrasted with a non-adaptive model, in which the irrigated crop area changes would all be zero based on the fundamental design of the model. We believe results such as these indicate that our approach offers a clear improvement over the non-adaptive cropping assumptions embedded in a traditional LHM as assessed against real-world data.

Beyond the calibration of individual model components and the plausibility evaluation described above, a full time-series calibration of irrigated cropped areas using the coupled model presents several challenges. While our study focuses on and endogenizes the influence of water availability on irrigated cropping decisions, there are many factors outside of this that influence irrigated cropping patterns. Other factors such as crop prices, crop production costs, areas equipped with irrigation, the existence and capacity of dams/reservoirs, subsides on crop production, and inter-basin water transfers are among these external influences. To conduct a coherent historical calibration, we would need to account for all of these time-varying factors alongside our dynamic treatment of hydrologically-driven water availability, one that is extensive over CONUS and conducted for a model time period of reasonable length to capture adequate variability of the coupled system (in our view, at least ~50 years), a major and likely unprecedented undertaking which we consider outside the scope of the current analysis. Similarly, a comprehensive calibration of our model would require a reliable record of observed Surface Water/Groundwater irrigated cropped areas over the historical period of calibration, for which we are unaware of an existing dataset. For these reasons, we have taken the approach to strongly emphasize that the effort is a hypothetical analysis (rather than a historical reconstruction) in our initial manuscript, as the reviewer notes.

We intend to include the full results of our new plausibility evaluation in the supplemental materials of a revised manuscript. We also intend to include further details regarding limitations of our analysis given the calibration gaps noted by the reviewer.

Reclamation, U.S., 2014. Downscaled CMIP3 and CMIP5 climate and hydrology projections: Release of hydrology projections, comparison with preceding information, and summary of user needs. Denver, CO: US Department of the Interior, Bureau of Reclamation, Technical Services Center.

Voisin, N., Li, H., Ward, D., Huang, M., Wigmosta, M. and Leung, L.R., 2013. On an improved sub-regional water resources management representation for integration into earth system models. Hydrology and Earth System Sciences, 17(9), pp.3605-3622.

Hejazi, M.I., Voisin, N., Liu, L., Bramer, L.M., Fortin, D.C., Hathaway, J.E., Huang, M., Kyle, P., Leung, L.R., Li, H.Y. and Liu, Y., 2015. 21st century United States emissions mitigation could increase water stress more than the climate change it is mitigating. Proceedings of the National Academy of Sciences, 112(34), pp.10635-10640.

I have some minor comments below to help the authors to improve the readability. 

R1-4. Line 85: Does this mean groundwater is an "infinite underground reservoir?" It is ok for this assumption but need to make it clear to the readers.

Yes, groundwater is treated as an infinite reservoir over the modeled time horizon, however production capacity and costs are fixed to the historical period (i.e., a farmer agent cannot simply produce groundwater free of cost or volume limitations in the face of surface water shortage). We will clarify this in the revised manuscript.

R1-5. Line 97: This is interesting. Is there a reason why this 50 km threshold? How sensitive is this threshold?

The 50 km threshold represents a reasonable estimate of a distance cutoff for most diversions. In the initial implementation of this threshold-based approach, Biemans (2011) performed a sensitivity analysis on this threshold value, concluding that an increase to 100 km would increase demand by 4 percent while a decrease in the buffer to 25 km would decrease demand by 18 percent. The selection of 50 km is also chosen due to computational tradeoffs. As the buffer increases, additional agents/cells have access to any given reservoir, increasing the computational requirement for the reservoir water allocation algorithm.

Biemans, H., Haddeland, I., Kabat, P., Ludwig, F., Hutjes, R. W. A., Heinke, J., von Bloh, W., and Gerten, D. (2011), Impact of reservoirs on river discharge and irrigation water supply during the 20th century, Water Resour. Res., 47, W03509, doi:10.1029/2009WR008929.n 

R1-6. Line 103: I assume the hydrological proxy during the calibration period is like a long-term average?

Yes, the hydrological proxy during the calibration period can be viewed as a long-term average of a hydrologic state (either a surface water reservoir level or a runoff value). Agents look to this hydrologic state to formulate their expectation of water availability. In simulation mode, the dynamically simulated hydrological proxy results in changes to agents’ expectation of water availability as the model advances in time.

R1-7. Line 173: I assume “Data Sources and Processing” mean Section 2.3? Because there is no sub-section with this title.

Yes, we will correct this in the revised manuscript.

R1-8. Line 188: Is there a specific reason why this period: 2010-2013 is used?

We select the 2010-2013 period due to both data availability as well as historic drought conditions:  1) the Cropland Data Layer, a critical input for a our data workflow, is only available starting in 2008, 2) the start of the 2010s were a period in which historic drought over the United States was relatively low as a baseline, and 3) the USDA Farm and Ranch Irrigation Survey is only available in 2013. We combine these data sources together (CDL starting in 2010) and consider them a historic representation of 2010-2013 conditions. We will add this explanation/justification to a revised manuscript.

R1-9. Line 233: I assume farm-level optimization is agent level in Section 2.2?

Correct, we will clarify to indicate “agent-level” in our revisions.

R1-10. Line 246: I think the calculation is Water Demand = crop area * irrigation requirement. In the ABM, when you adjust water demand, did you change both crop area and irrigation requirement? Or do you only change the crop area?

We only change crop area. The irrigation requirement (e.g., the depth of irrigation water required per unit land area of crop planted) is assumed static over our model run, i.e., we do not consider the impacts of climate change on the irrigation requirement for our analysis. We will clarify this in our revisions.

R1-11. Line 252 and 265: Are these assumed to be the same throughout the simulation period? It is fine if that is the case, but should be clarified. 

Yes, all unit prices and costs are assumed to be the same throughout the simulation period. We will clarify this in the revised manuscript.

R1-12. Line 256: Do you mean Table S1?

Yes, we will make this correction.

R1-13. Line 286: Is this a typo for PerceivedCost?

Yes, we will make this correction.

R1-14. Line 292: Can you provide an equation?

Yes, the equation (included below) will be added to the manuscript:

PerceivedCostAdj = (Yield * Price) – Profit, when Profit >= .10 * Yield * Price,

PerceivedCostAdj = (.90 * Yield * Price), when Profit < .10 * Yield * Price

R1-15. Line 312: I think you might need to mention what is VIC first. Also, is irrigation water availability = streamflow in each grid?

Noted. Considering the reviewer’s comment below as well (R1-16), we will merge sub-section 2.3.6 into 2.4 after the introduction of the VIC-MOSART-WM descriptions in the following sub-sections, since 2.3.6 pertains to both the agent model and the hydrologic model. “Irrigation water availability” refers to both local streamflow in a coincident grid cell to the farmer, as well as water made available via allocation from a surface water reservoir. We will also clarify this in the revised manuscript.

R1-16. Line 322 (Section 2.4): Don't really think you need one page of these backgrounds since these are already published. So maybe merge this section with 2.3.6 to smooth the logic flow? Otherwise, a sudden mention of simulated irrigation water availability is a bit logical jump.  

Noted. We will merge this section with 2.4 and also shorten as well, given this material is published elsewhere as the reviewer correctly notes.

R1-17. Line 371: I think an equation to show how you calculate water shortage is still necessary.

Yes, we will include an equation showing our calculation of water shortage as below:

WaterShortage = WaterDemand – LocalWaterSupply - ReservoirWaterSupply

R1-18. Line 393: I think some text in the caption of Fig 2 should be moved (or copied) to the main text like how you calculate the blue and orange bars and dots. 

We will plan to include this information (calculation of the colored bars and dots) in the main text as well, while also keeping it in the caption so readers have access to the information in both locations.

R1-19. Line 394: Is it a typo for Fig2b?

Yes, we will correct the typo in the revised manuscript.

R1-20. Line 426 and 444: I think you need to at least show the Eastern US results in the supplementary material because you do emphasize in your abstract and introduction that this is a CONUS study. But currently, there are no results showing this scale.   

Thank you for calling attention to this. We will include complete CONUS results (i.e., a full CONUS map of Figure 2b) in a new section of the Supplemental Materials.

R1-21. Line 435: It is a bit unclear how you calculate Fig 3e-h? Is it a long-term average? Counting every model year?

The nature of the classification is provided in lines 435-441, copied below. The classification is conducted looking over every year of the model run. For example, an  agent is classified as “crop expansion/contraction” if the stated criteria is satisfied, considering every model year of the 60-year model period. We will clarify the nature of the classification in our revisions. We will also include a brief explanation in the caption of Fig 3 for clarity.

“…while Figure 3e-h assigns farms to crop adaptation categories based on the amount of crop adaptation simulated over the model period. For the latter, agents are assigned to one of four categories depending upon the level of crop adaptation activity: 1) “crop expansion/contraction” if the ratio of an agent’s annual minimum surface-water irrigated crop area is less than 80 percent of the annual maximum surface-water irrigated crop area, 2) “ crop switching” if the predominant crop’s share of the total crop makeup for any given agent (measured in terms of crop area) changes by at least 5 percent between any two years of the model run (which do not need to be consecutive), 3) “both” if the agent satisfies both criteria 1 and 2 above, and 4) “none” if the agent satisfies none of these criteria.“

R1-22. Line 470: I don't see the results of agricultural profits. Are you showing any figures or tables?

We have not included figures/tables of agricultural profits to keep the manuscript at a reasonable length and to maintain focus on the primary outcome of our model (changes in irrigated crop area and ensuing water shortage). We also note a nuance – our model outputs expected agricultural profit rather than actual agricultural profit, as impacts of water shortage on crop yields and ensuing profitability are not calculated by the model. However, we believe expected agricultural profit results are still useful and insightful, and will plan to include additional figures showing expected agricultural profit results in the Supplementary Materials.

R1-23. Line 545: I think there is another limitation worth mentioning which is social norm effects. Farmers' behaviors are heavily affected by their social networks (neighbors, friends, etc.). There are ABM out there showing this already and can be considered in the future. 

We thank the reviewer for calling attention to this important social phenomenon and agree this is a limitation of the current approach (as well as an opportunity for improvement). We do additionally note that the large-scale nature of the current effort presents an interesting conceptual challenge for representation of social norms. A single representative farm in our case can represent hundreds of neighboring farms in reality. Implicitly, this aggregation assumes that all farms accounted for by a representative farm follow the same decision model and set of norms.

We will call attention to this limitation and challenge in the revised manuscript, and also provide reference to key previous farm ABM-hydrologic studies with explicit treatment of social norms noted below:

Hu, Q., Zillig, L.M.P., Lynne, G.D., Tomkins, A.J., Waltman, W.J., Hayes, M.J., Hubbard, K.G., Artikov, I., Hoffman, S.J. and Wilhite, D.A., 2006. Understanding farmers’ forecast use from their beliefs, values, social norms, and perceived obstacles. Journal of applied meteorology and climatology, 45(9), pp.1190-1201.

Lin, C.Y., Yang, Y.E., Malek, K. and Adam, J.C., 2022. An investigation of coupled natural human systems using a two-way coupled agent-based modeling framework. Environmental Modelling & Software, 155, p.105451.

We are glad the reviewer found the manuscript to be interesting and are thankful for their insightful comments. In our response, we have copied and labeled (RX-X) individual reviewer comments followed by our response in italics to each.

R2-1. My main comment is regarding the use of surface water only. Most irrigation in the USA is from groundwater. How are the results impacted by this assumption? Groundwater pumping can be considered an adaptation strategy when surface water runs out.

The reviewer rightfully notes the importance of groundwater for crop irrigation in the United States. We firstly clarify that groundwater is accounted for in our formulation of farmer agent behavior; however, the hydrologic state of the groundwater system (i.e., groundwater levels, wellfield capacities, and groundwater quality) is assumed to remain static over the course of the model run. In the farmer agent formulation, agents can access groundwater for crop irrigation, but this production comes at a cost to the farmer (per volumetric unit of groundwater produced) and is limited by a capacity of groundwater production (that is assumed to be the groundwater production for irrigation estimated via observations during the calibration period). The latter is perhaps the most significant limitation of groundwater treatment in our effort. In our revisions, we will clarify the treatment of groundwater in both the farmer agent formulation as well as on the hydrologic side of the model (see also response to Reviewer #1s R1-2 comment).

In the current discussion section (lines 545-550), we note implications of our treatment of groundwater on our study conclusion. Namely we note that the “ability of farms to increase groundwater extraction in response to surface water shortage, as well as changes in the availability and cost of groundwater (e.g., due to depletion of groundwater in stressed aquifers) are not currently represented in our modeling framework. These potential responses may either mute the simulated water shortage changes (e.g., instances in which farmers increase groundwater pumping to accommodate surface water shortage) or heighten them (e.g., instances in which increasing groundwater depletion results in even higher shortages and ensuing adaptive responses).” It is difficult to hypothesize a consistent implication given these considerations. In some cases, farms may indeed compensate surface water shortage by simply pumping more groundwater where physically available and economically viable, though this would still impact their agricultural profitability. In other cases, groundwater production for irrigation may already be at its limits in terms of production capacity, cost, and/or water quality and further dwindling as aquifers are increasingly stressed. In these cases, declines in groundwater accessibility may interact with surface water availability in unexpected ways. We will elaborate upon this in our manuscript revisions.

Given these considerations, we also note ongoing work that is outside the scope of the current manuscript, but is focused on the addition of a CONUS-scale groundwater response simulator into the MOSART-WM-ABM framework introduced in this manuscript. This improvement, which is still in its early stages, will allow us to address the groundwater dynamics that the reviewer calls attention to. We will note this ongoing effort as a future research direction in our revised manuscript.

R2-2. Line 30: Please quantify how much water is consumed and withdrawn by irrigation in the world and the USA.

We agree that this information would provide helpful context for the manuscript, and will include estimates in the introduction of a revised manuscript. While global and national irrigation estimates are prone to substantial uncertainties (Puy, et al. 2022), a recent review (McDermid et al., 2023) attempts to synthesize the state of knowledge regarding quantitative estimates of consumption and withdrawal for irrigation, which we will draw upon. We will also highlight some model-specific studies that attempt quantifications of specific facets of irrigation water withdrawal and consumption, such as those that the reviewer notes in comment R2-3 below.

Puy, A., Sheikholeslami, R., Gupta, H.V., Hall, J.W., Lankford, B., Lo Piano, S., Meier, J., Pappenberger, F., Porporato, A., Vico, G. and Saltelli, A., 2022. The delusive accuracy of global irrigation water withdrawal estimates. Nature communications, 13(1), p.3183.

McDermid, S., Nocco, M., Lawston-Parker, P., Keune, J., Pokhrel, Y., Jain, M., Jägermeyr, J., Brocca, L., Massari, C., Jones, A.D. and Vahmani, P., 2023. Irrigation in the Earth system. Nature Reviews Earth & Environment, pp.1-19.

R2-3. Line 32-33: There is extensive work on irrigation expansion done by Rosa and colleagues. Elliot et al., 2014 did not quantify irrigation expansion, but impacts of climate change on current irrigation.

https://www.science.org/doi/full/10.1126/sciadv.aaz6031

https://iopscience.iop.org/article/10.1088/1748-9326/aadeef/meta

https://www.pnas.org/doi/abs/10.1073/pnas.2017796117

https://iopscience.iop.org/article/10.1088/1748-9326/ac7408/meta

We thank the reviewer for their clarification of Elliot et al., 2014 and agree with their assessment. Upon revisiting the manuscript, we realized that our sentence was phrased in a confusing manner that could be prone to misinterpretation. We will rephrase the sentence to clarify that Elliot et al., 2014 evaluates the impacts of climate change on current irrigation.

We also thank the reviewer for pointing our way to the insightful studies by Rosa and colleagues that attempt to quantify potential irrigation intensification and expansion. In a revised manuscript, we plan to include reference to the four publications the reviewer notes, which can improve the context, motivation, and relevance of our efforts. We will also make note of these studies as they pertain to future directions of research in the Discussion of the revised manuscript (e.g., the PNAS paper and opportunities to endogenize irrigation expansion changes supported by reservoir storage increases in future model versions).

R2-4. Line 35: Recent literature that quantified the contribution of dam-based water storage on irrigation: https://www.pnas.org/doi/abs/10.1073/pnas.2214291119

Thank you for calling attention to this important study. While we are under the impression that  the study does not particularly fit into the list of references included in line 35, which are all studies associated with large-scale hydrologic models of a distinct lineage, we nonetheless think the findings of the study are important to highlight. We will include a separate sentence indicating its relevance to our work in our revisions.