The authors’ responses to reviewer concerns are appreciated, and the author should proceed in making the proposed corrections, with a few exceptions. In a number of key areas noted below, the authors can do more to strengthen the paper – primarily through (1) following reviewer suggestions further toward assessing additional, likely predictors from publically available data sources; and (2) further investigating and justifying the use of such a long-lag climate index (MEI). While it is well known that such indices provide predictability, such demonstrations have typically been at shorter lags (ie, values immediately prior to the prediction period). The authors must do more to defend the idea (through analysis) that predictability pathways exist, and that shorter lag indices (eg, the MEI itself, but in January of the same year as the prediction) cannot provide better skill. The current result is not well supported by existing literature on index-based hydroclimate prediction. Further notes on these points are given below.

Response to Rev1-Q2/Rev2-Q3 – It would be worth assessing the skill of the NMME average at least (which is available operationally) over the region to see whether it can be included in a statistical framework as a predictor – even if it is not used in hydrological modeling. Even though the assessment recognizes limitations in the ability of practitioners to use data such as NMME, it should not be hard for the authors to extract the NMME data and evaluate whether it would be worth developing as an additional predictor.

Response to Rev1-Q3/Rev2-Q3 -- Similarly, it would be worth following the reviewer’s suggestion to see whether modeled SWE anomalies from GLDAS (also publicly available) would be a similarly useful predictor. Even if they have the same correlation with streamflow, if it is uncorrelated with antecedent flow, it may add skill. The reasons for not using MODIS are fair.

Response to Rev1-Q12/Rev2-Q4 – The significance of MEI as a predictor at a lag of nearly a year to the predictand still needs further justification. The Marriotti text does not adequately characterize the processes and potential predictability at such lags explicitly. Although one possible route toward this predictability could be, as suggested, the MEI prediction of fall winter climate variability, hence winter snow accumulation, presumably this could be more directly captured using the actual variables (ie, winter snow peak, accumulated winter precip) than a long-lag teleconnection. Even the use of MEI at a shorter lag from the predictand would intuitively carry more information. This result suggests a spurious set of correlations without greater analysis of the underlying physical mechanisms. Could the paper establish whether the MEI is skillfully related to the proposed intermediate dynamics though additional analyses? Would using predictors such as winter precipitation anomalies provide similar predictability? Further analysis would result in a broader and more insightful study to benefit potential implementation.

While the authors do not include all additions requested by the reviewers, and the methodology and datasets used in the paper could be likely be extended fruitfully in various ways that were discussed during the review, the authors have nonetheless provided analysis explaining their decisions not to accommodate certain requests. They have also recognized the concerns associated with including a predictor (MEI from yr-1) that appears valuable in the cross-validated study, though lacking in dynamical justification. This is an ever-present challenge in statistical prediction, thus it is not out of place that such an result could appear, granted the authors recognize the cautions that apply. Given the operationally-oriented objectives of the paper, I believe that the methods and case study narratives may be of value to the community, thus I recommend publication.