The paper represents a substantive body of work and is of interests to HESS readers. The reviewers were positive about the study and made constructive comments. The authors responded mostly soundly in their replies to the reviewer comments, as such that I believe the paper will significantly improve after a major revision and be publishable.

I have a couple of major comments. The first one is on evaluation. I agree with the authors that a period of 10 years is too short for independent evaluation of seasonal forecasts. This problem needs to be addressed for drawing meaningful conclusions from the results. An alternative to the methodology is to use a same period for predictor selection and model training. Yes there may be over-fitting in doing so, but any over-fitting will be punished in independent evaluation. Using such a methodology, a longer period of records can be left out for independent evaluation.

My second comment is related to the handling of correlated predictors. The authors should be aware of and make mention of alternative approaches. For example, One could build multiple models, each of which consists of only one or two predictors. Forecasts from the multiple models could be merged by using for example Bayesian model averaging. If needed, feel free to contact me for references on such an approach.

Minor comment. I agree with the second reviewer that the use of the word exemplary is not appropriate. The use of "solid" results and other subjective descriptors should also be avoided as much as possible.

The authors have done a thorough revision in response to the reviews. The study makes a useful contribution to statistical seasonal climate forecasting.

I have made edits throughout the document to improve the reading. I have also made a number of comments. These are annotated as sticky notes in the pdf manuscript.

I have two main comments for the authors to address:

(1) The running three month forecast needs to be clearly defined. If the final forecast (after merging forecasts for the same month but at three successive lead times), the sum needs to be divided by 3 (equations on page 11). If the final forecast is a coherent forecast for the total of the next three months, these equations are incorrect. The descriptions and references in the text also need to be tightened up. I have made some suggestions in my sticky notes.

(2) A major drawback of the work is that the forecasts are deterministic. Seasonal forecasts are subject to large uncertainty. The usefulness the forecasts are greatly diminished when the forecasts are deterministic. The authors should discuss this limitation in the paper.

The authors have addressed my two major concerns. The notations for the equations on page 11 are now clearly explained, and I am now fine with these equations. The authors also added some discussion on a drawback of the work, that is the forecasts are only deterministic.

The authors have made minor edits, which have improved the manuscript.

Well done to the authors for responding to the reviews very professionally.