General comments

Kenya is increasingly growing agricultural products for export. The water needed for commercial export farms competes directly with the water needed for agropastoral communities in Kenya. The authors did study this water competition. They developed a model that integrates a hydrological model with a decision model that simulates adaptation decisions by agropastoralists. After model calibration, they used the model to study the effect of upstream abstraction by commercial export farms on drought risk and its impact on agropastoral communities. The results show that these effects are relatively small compared to the effect of drought periods themselves.

The developed tool in this study is novel due to integrating a hydrological model with a human decision-making model. The upstream-downstream interactions and competition occur in many parts of the world and are therefore very relevant for the EGUsphere readers. The current increase in commercial export farms in different African countries makes the study very topical.

The setup of the study and the main results seem sound. However, the current manuscript raises a large number of questions and comments, which are listed below.

Major comments

Eq. 2: The water demand plays an important role in your study. I suppose that DRYP calculates the volumetric water content θ. If so, mention that here. It is not clear how extensive the irrigation is. Probably most agricultural fields are rainfed. Clarify how you determine the area that is irrigated.

Line 246: The sentence “Greenhouses are modelled as ‘closed systems’ and irrigation water is not added to the model while there is evaporation.” is unclear. Do you assume evaporation at the greenhouse locations?

Line 273: You mention that you list only the relative factors of the hydrology-related parameters in Table 2. Therefore all the factors are dimensionless. However, in this way the reader does not have any information on the actual values of the hydrology-related parameters. Therefore include in Table 2 a column in which you specify the reference values, such as Table S2 in the Supplement.

Line 295: In general de term “model validation” is reserved for the application of a model to different circumstances: a region or period for which the model was not calibrated. Am I right that the simulation results which you show in Figure 4 were derived after model calibration? In that case, you cannot call this “validation of the model”.

Figure 4: Milk production is one of the criteria you use to quantify the impact of water extraction. The simulated milk production does, despite the calibration, strongly deviate from the measured milk production. Especially in the dry periods, which are the focus of your study, the milk production is grossly underestimated by the model. Discuss how this may impact your results and conclusions.

Figure 6: What is the unit of soil moisture in the top graph? Earlier you used volumetric water content (-), which can never exceed 1.0.

Figure 6: This figure shows only the differences of scenarios 2 and 3 with the baseline scenario. In order to put these changes in perspective, I recommend to show also the time series in time, as depicted in Figure S6 of the Supplement.

Line 364: You discuss here where soil moisture is decreased. However, looking at the values in Figure 7, this decrease is always < 0.01. You should mention this here.

Lines 415-417: Here you discuss serious reductions (22 and 36%) of stream flows as a result of your study. However, you did not show these reductions in your paper. In Figure 7 you only mention absolute reductions (m³/d). You might add these % reductions in the Supplement as a function of time and refer to this information here.

Lines 482-483: “it should be noted that not all factors are included in this study and more factors may influence the adoption of drought measures”. Can you mention some of these factors?

Line 485: In line 59 you write: “The main goal of this paper is, therefore, to develop a coupled hydrological and agent-based model (ADOPT-AP) to investigate the influence of upstream large scale commercial export farms on downstream drought risk and adaptation by agropastoralists.” In this Conclusion section you discuss the influence of commercial export farms. However, how do you evaluate the performance of ADOPT-AP? Which parts of the model framework perform well and which parts need further development?

Lines 21-22 (abstract) + 489-490: You state: “The analysis shows that in the scenarios where these farms are replaced by forests or communities, drought conditions are alleviated by increasing soil moisture, streamflow, and groundwater tables.” Your results show that the simulated increases in soil moisture, stream flow and groundwater depth are very small and have a minor effect on crop production, milk production and distance to water. In my view, the current statements are too firm and should be put more in perspective.

Minor comments

Line 55: The phrase “minimum water availability” is unclear. Replace by “minimum river flows” (based on Lanari et al. (2018)).

Line 203: Equation 4 should be equation 2.

Line 206: Change “θ is the water content (–) and θ_fc is the water content at field capacity (–)” to “θ is the volumetric water content (–) and θ_fc is the volumetric water content at field capacity (–)”

Line 268: Table 3 should be Table 2.

Line 272: Do you mean Table 2 with “Table 5”?

Lines 366, 381 and 387: Check figure references.

Line 388: produciton should be production

Line 397: resuling should be resulting

This is an interesting paper, which presents both a tricky setting with loads of water-related interactions and a fascinating methodology to study the setting. Having said that, I am afraid that to me the paper is less convincing than it should be on both aspects. My main reason for this assessment is that the paper focuses rather strongly on the "what" (model choices, data inputs) and tends to ignore the "why" (are certain choices allowed, what are consequences of choices?). I cannot determine whether model results are not a direct result from the modelling setup. Let me rephrase that: I am convinced that the model results are the direct result of the modelling setup, but cannot easily determine to what extent the agency of model agents is "free" enough to escape modelling setups. As far as I can judge, the ABM setup is quite deterministic. I do not have problems as such, as some of the ABMs I have worked on have deterministic procedures too. I do have problems with this text because the choices made are not clearly explained. I will not present detailed comments on all parts of the text, as what needs to be done first in my opinion is a clearer discussion on the model setup. Below, I share some feedback on selected parts of the text to illustrate my assessment.

Both in the abstract and the introduction, the lack of studies on the topic is connected to the ABM setup with "therefore". Why "therefore"? I can imagine very useful studies to be developed on the topic without ABM. In fact, the strong hydrological base of the ABM setup suggests that agency may be less required to study flow-related feedbacks. What does agency add?

The text moves between "drivers", "behavior" and "actions" when describing aspects that I think are very close if not the same. As soon as "actions" are "driven", where is the choice? To me, "behavior" reflects a longer-term pattern, whereas "actions" are shorter term. As soon as hydrological effects are produced by "actions", can they be an external "driver"?

The ODD protocol and other materials in the supplement are obviously more detailed compared to the text on model setup. This is to be expected, but I think some elements from the supplements need to be clearer in the main text.

• Household characteristics: does the framing on the best fit mean all households start the same?
• The same for self-efficacy and how household properties like gender influence decisions
• The PMT theory is from 1983. I would like to read more about why such an old one can still be used.
• How can a grid cell represent multiple household as one agropast agent when apparently the density can differ?
• My main concern is the time step. Decisions are annual according to the text. The supplement mentions that annual decisions on crops and livestock are at other moments, right? This would mean that decisions are not annual, or should not be assuming that decisions on crops and livestock should be related. After all, there is only so much one can do in the time given to an agent. With the hydromodel running on shorter time steps, why can agents not act more often? or does the hydromodel actually represent many decisions of human agents, but keeps them hidden and unrelated to the "real" decisions?

I find figure 4 not easy to understand, as the middle and lower panel do not seem to have a legend. This remark stand symbol for almost all of the figures, which i find not easy to understand. I do know how hard making useful figures can be...

The Discussion is extremely general and is not going into much detail on model results in relation to the setup or in relation to other literature. Paragraph 5.1 does not seem to need ABM anyway to make its points. 5.2 is a nice overview of some literature, but does not reflect back on the ABM proposed in the text beyond claims that this model is "valuable" in "capturing" and "understanding" things. I would agree on the "capturing", even when I would still need to know more about why model choices make sense. I take issue with the "understanding", as copying processes is not necessarily the same as finding out why these processes happen.

I will be totally clear: when reading the conclusion, my first response was a simple "wow". It was not an enthusiastic "wow, though, as I find the conclusion frustratingly brief and shallow. Apart from the numbers mentioned (why only those?), I could have written the same conclusion without any deeper analysis of the setting. Again, to be totally clear: this conclusion was almost enough reason for me to suggest a "reject" for this text. If this is all one should take from the study, it can be easily missed in the academic world. There must be more to share. As already said, I propose that sharing much more on reasons for model choices and on implications of these choices for results would be required. Only with these aspects clarified could one think about consequences of both for understanding and follow-up actions (research and policy).