Letter to Editor for Revision

Thank you for the consideration of this manuscript and providing feedback from the reviewers. We have reviewed all of the comments and incorporated their suggestions and feedback to various degrees. Below are some additional comments alongside these revisions to provide context on some of our edits.

First, while revising aspects of the paper, we discovered a minor coding error in our data. Fixing this issue did change some of our model results, but usually by a few percentage points and did not significantly impact the interpretation of the results.

Reviewer 1 has asked to see alternative formulations of other models we tried. This is also related to another question on research about other factors that impact growth of pathogens. We have added a new section 2.5; substantially revised section 3.1; and included a new model in section 3.1. We included a multivariate model that combines the hybrid model alongside common hydrometerological predictors, which we had earlier created but did not include. We've included discussion on the lack of significance for those predictors in section 3.2. The table describing the predictors used in the model is now on Table 2.

Reviewer 1 has asked for additional citations in the discussion section on models used in other research. We have reorganized the section to callback some of the literature and review of Material and Methods to place it in context of our findings.

Reviewer 1 has requested a more extensive and deeper comparison of this predictive modeling results to other research efforts across the globe. In our re-organization of the Discussion in section 3.2, we have clarified the distinction between "prior-day nowcast models"--which describes the modeling approach taken in other research--and the "hybrid model". We have revised the discussion section to help clarify this point since it was unclear in the initial submission. There is also a deeper discussion on the predictors in section 2.5.

Reviewer 1 asks an interesting question on why Chicago Park District does not look for pathogenic organisms, instead of just indicators. Unfortunately, the research team has not worked with Chicago Park District on this decision so we are unable to provide further insight.

Reviewer 1 asks whether we attempted to make a model to predict the 5 "outlier" beaches, which mostly have long breakwaters. When building the model, we referenced prior literature that discussed the effects of such breakwaters on water quality. We then confirmed the relationship between breakwater length and frequent water quality exceedances. Accordingly, we ultimately chose to assign rapid testing to those beaches to realize practical benefits, which we've elaborated on in the discussion in section 3.2.

All other comments and clarifications from Reviewer 1 have been incorporated.

Reviewer 2 requested that goodness-of-fit tests be included in the analysis. Typically, goodness-of-fit in machine learning is done through out-of-sample testing (such as this paper) and does not traditionally use tests like Chi-Square. We have incorporated Matthews Correlation Coefficient (MCC) to provide some more general interpretation that is akin to goodness-of-fit (e.g., R-squared). MCC does not have a threshold for significance like Chi-Square, but it does have characteristics like R-squared so researchers have a better sense of fit. We hope this is valuable and meets the intent behind the request.

All other comments and clarifications from Reviewer 2 have been incorporated.

We have included with our submission a version of the manuscript detailing how it differs from the original submission. Our paper is generated from code and the version history was tracked and managed using code versioning software, rather than word processing software. Therefore, we used a software tool in order to show changes.

We look forward to future correspondence.

Sincerely,