The authors build a framework that links direct and indirect effects of ENSO on global soybean production with a focus on North and South America. The authors use linear structural causal models (SEM) to quantify the impacts of spring and summer soil moisture as well as extreme heat on soybean yield anomalies in the main growing regions. In addition, ENSO variability is linked to extratropical SST patterns, local weather and soybean yields in Brazil, Argentina and the US.

General comments:

The paper is very well written and clearly structured. Findings are compared with results of similar studies and put into context. Using SEM models to link ENSO, SST and soil moisture, as well as soil moisture, heat and yields is an innovative approach that builds on previous research which only focused on parts of the causation chain. Limitations and potential future research are well described. The title, however, is misleading as the majority of the analysis refers to correlations between ENSO, SST, soil moisture and soybean yields in each of the three breadbaskets. Spatial correlations between the three production areas via ENSO conditions are only mentioned in the Discussion and not explicitly analysed. Thus, I suggest a change of title or additional analysis of spatial correlations, e.g. of soil moisture conditions or crop yields between the regions.

Specific comments:

The authors state that crops are particularly vulnerable during the reproductive time and identify relevant months for South America and months for North America. I suggest  specifying that this refers to soybeans as soil moisture sensitive growing periods differ between crops and regions. There are also differing definitions of sensitive growing periods. In addition to the papers that the authors cite, USDA has published crop calendars with slightly differing moisture sensitive growing periods: https://ipad.fas.usda.gov/ogamaps/cropmapsandcalendars.aspx

Regarding the selection of counties, using rainfed production and harvested areas makes sense. It would be useful to provide an estimate of the total share of global production that the study covers in the end. The authors state that the US, Argentina and Brazil account for 80% of global soybean production. After excluding a few relevant US states, is the share of global production in your study still significant?

I have trouble to understand why soil moisture drives (assuming causality) extreme heat. I would assume the driver of soil moisture is heat. Please provide more information on the underlying land-atmosphere coupling you are referring to in the methodology paragraph (It is mentioned in the discussion later. I suggest referring to it already earlier in the manuscript).

There is an important difference between correlation and causation, as I am sure the authors are well aware of. I also assume that the SEM methodology considers this. It would be helpful if the authors elaborated on this further, especially regarding the example soil moisture -> heat or heat-> soil moisture.

Hamed and co-authors present an analysis framework to link crop yield anomalies to crop growing conditions and subsequently the underlying climate drivers in a causal chain of analysis. They build on past work to demonstrate that their method is relevant in the case of multiple crop yield shocks to soybeans in North and South America. The authors present a well written and well motivated study with easy to interpret graphs. I thank the authors for the time and care that has gone in to the manuscript. I generally think the manuscript sound and I have only three minor comments and suggestions for the authors to consider.

Specific comments:

1. For central Brazil, the relative soybean growing seasons have changed over time with the increase in Safrinha cycle cropping. If you are using a static harvested area map and crop calendar to weight the climate anomalies and produce a regionally aggregated weather time series to relate to the regional crop yield time series, the change in dominance from traditional crop cycles to a safrinha soy-maize crop cycle may introduce error. Your approach is a reasonable enough as it is, but this limitation may be worth mentioning in the context of the smaller variance explained by climate variables in central Brazil as compared to SESA. South Brazil does not produce much Safrinha cycle soy-maize crop rotations, so the analysis in South Brazil would not be strongly affected by this.
2. The soybean growing season in the US (May-Oct) intersects typical ENSO development (~Jul) and decay (~Mar) such that one could develop reasonable hypotheses that the intersection of the soybean season with either a developing ENSO event (Jul-Oct) or the lagged effect of a decaying ENSO event (Apr - Jun) might affect the soybean growing season. Can your causal framework distinguish between these two different cases, and if so what do the conclusions say about whether we should be considering developing ENSO events, decaying ENSO events, or both when evaluating the effect of ENSO on summer-grown crops in the US? It would be helpful to clarify this, especially because the past literature you cite (e.g. Anderson et al. 2017a, 2017b, 2018) outlines the effects of ENSO primarily as developing events, although Jong et al. (2020) highlight the importance of antecedent SST anomalies in the west pacific for US summertime heat during La Niñas (https://journals.ametsoc.org/view/journals/clim/33/14/jcliD190701.xml).
3. Clarify what is meant by “persistent” La Niñas. Do you mean multi-year, or La Niña events that persist into AMJ?