Streamflow forecasting plays a critical part in water resources management. In this paper, streamflow forecasting models are developed for Great Britain. The usefulness of the models are demonstrated through the Hydrological Outlook Portal (https://ukho.ceh.ac.uk/).

There are three comments on the paper.

First of all, the paper is more like a report, rather than a research article. It is mainly due to that the methods presented along with the results. For example, in “4.2 Performance of the forecast distribution”, the equation of CRPS is illustrated and then the detailed results are presented. It is pointed out that for research articles, methods and results are mostly presented in different sections. In this way, all the methods are illustrated in the same section, so that people can better understand the framework of the proposed methods.

Secondly, the rainfall forecasts that drive the hydrological models can be improved. In “4.1 Performance of the ensemble mean”, the performance of raw rainfall forecasts is examined. It is not surprising to see that the performance of raw forecasts is not satisfactory. It is noted that peer studies have developed post-processing methods to exploit the skill of raw rainfall forecasts. The authors are suggested to considered forecast post-processing.

Thirdly, there exist some forecasting systems in Europe. Is it possible to conduct some comparisons?

References:

Alfieri, Lorenzo, Peter Burek, Emanuel Dutra, Blazej Krzeminski, David Muraro, Jutta Thielen, and Florian Pappenberger. "GloFAS–global ensemble streamflow forecasting and flood early warning." Hydrology and Earth System Sciences 17, no. 3 (2013): 1161-1175.

Li, Wentao, Quan J. Wang, and Qingyun Duan. "A variable-correlation model to characterize asymmetric dependence for postprocessing short-term precipitation forecasts." Monthly Weather Review 148, no. 1 (2020): 241-257.

Zhao, T., Bennett, J.C., Wang, Q.J., Schepen, A., Wood, A.W., Robertson, D.E. and Ramos, M.H., 2017. How suitable is quantile mapping for postprocessing GCM precipitation forecasts?. Journal of Climate, 30(9), pp.3185-3196.

Wang, Quan J., Yawen Shao, Yong Song, Andrew Schepen, David E. Robertson, Dongryeol Ryu, and Florian Pappenberger. "An evaluation of ECMWF SEAS5 seasonal climate forecasts for Australia using a new forecast calibration algorithm." Environmental Modelling & Software 122 (2019): 104550.

Huang, Zeqing, and Tongtiegang Zhao. "pyNMME: A python toolkit to retrieve, calibrate and verify seasonal precipitation forecasts." Environmental Modelling & Software 166 (2023): 105732.

Review of HESS Manuscript

“Improved seasonal hydrological forecasting for Great Britain”

Please find attached my review of the manuscript.

1. Scope

The scope of the article is inside the scope of HESS.

1. General evaluation

The authors present a method to improve seasonal hydrological forecasting in Great Britain, in which they conditioned the numerical meteorological forecast with historical weather data and used this forecast as input to a hydrological model to produce seasonal forecasts.

In general, I found the paper difficult to follow. In my opinion, it lacks a consistent story telling. Also, I think in most cases the findings are not conclusive, and the language is not in line with scientific discussion (see examples below).  Moreover, the correlations presented in Figure 2 are quite weak, compromising the efficiency of the method. Consequently, I believe major revisions are necessary before moving on with the review process.

1. Specific comments

Line 33: You should define sub-seasonal to seasonal timescales (weeks, months…). Also, the temporal resolution of the forecast (daily, sub-daily, etc.)

Line 40: It would be beneficial to give examples of low- and no-regret actions.

Line 57: What do you mean reasonably well?

Line 64. I would suggest removing “Section 2 describes our hydrological forecasting scheme in detail.” as it is stated below, where you give the structure of the paper.

Line 65-80: This is part of the methods and not the introduction. I would suggest moving this section.

Line 71-75:  I do not understand what you are explaining. Can you elaborate further?  Why 3 subsamples, where does the 63 come from? If the pseudo-observations come from historical records, how are you including physically plausible extreme rainfall events that have not previously been observed?

Line 75-80: This also requires further explanation. What are you matching? The large-scale atmospheric circulation patterns with what?

In general, I think what lacks in the introduction is the importance / impact that this seasonal forecast can have.  A small paragraph justifying why they are used, what type of decisions can be taken, what are the limitations.  

Line 110: How do you pass from daily total rainfall to 15-minute inputs that the G2G model needs (mentioned in line 107)?

Line 127: Which is the original scheme?

Line 135: What is MSLP (this has not been introduced before)

Line 137: What criteria do you use to select analogue years? How did you match them?

Line 139: NAO was not introduced before

Line 140: What is the “than” referring to here?

Line 143: Why are you referring to a method that will be explained in a publication that has not been published yet?  This is not how references work. You need to either explain it here or wait for the other paper to have at least a preprint and refer to that.

Line 147: What is “any” referring to here? Also, by using a proxy and then using that proxy to calculate a quantity, you are not avoiding biases, you are just changing the model you are using.

Line 148: Less good =  worse.

 Line 150:  I do not agree with that; observational means will vary depending on the resolution. Hourly, daily and monthly precipitation means are different.

Line 153: Three sub-periods are 3 months?

Line 158: Why are the extremes better sampled?

Line 155-162: I believe this needs to be explained better, right now the writing is confusing and there are a lot of ideas not properly explained. How are you sampling not-independent samples? Are you quantifying the covariances you talked about?

Line 172: I believe the references for section 2.1 and 2.2 are mixed.

Section 3.

I have multiple questions about this section, especially the correlation values that you presented in Figure 2. If I understand correctly you used samples of the hindcast period to evaluate the rainfall forecast model (so comparing the model results against past observed data). Moreover, on the top right of figure 2 you present the mean correlation.

The values you have there are extremely low. Saying that the “correlations are not particularly strong” is a huge overstatement. A correlation of 0 means no correlation at all, and most of your cases are values around 0.  Also, you have negative correlations. If you are calculating the correlations between the forecasted and the observed variable (so same variable), and the correlation is negative, it is a clear sign that the model is not working at all. Moreover, how did you get that a threshold of 0.33 to define statistical significance correlation?  

I think the evidence in this section suggest that the forecast model used to predict precipitation does not work. Is there any reference of how other models perform? How do this compare, for example, with the seasonal forecast of ECMWF? I would suggest that if your model is not working you take another model.

Line 232: I agree that taking another model is as the “ground truth” give you the advantage to validate in ungauged locations, but it might be worth to also compare against observed values where you have them and present both metrics. In GB there is the CAMELS-GB dataset that have discharge locations in over 600 stations, which you can compare against.

Line 235: What do you mean reasonably well? Also, if you use a dataset as CAMELS-GB there is a label that says what basin are relative unaffected by human impact, so the argument that comparing against observed flows would be unfair is not an actual argument.

Line 257: What do you mean by “the performance is good overall”? In a scientific article you should indicate the metrics also in the text, otherwise is completely arbitrary.  

Line 260: How did you define the statistically significant threshold? How does it compare to other models? Do you have any reference hydrological model? Why do you have negative correlations for JJA in figure 3?

Your use of language is quite general and not in line with a scientific article. Expressions as:

• “are statistically significant over much of the country in many months”
• “forecast ensemble means perform less well in northern England, where hydrological memory is not particularly long, nor rainfall forecasts especially skilful. ”

are subjective and do not quantify anything.

Figure 3. The name of the figures should be descriptive by its own, not a reference to other figures.

Section 4.2.  I think here also comparing your results against observed data would be useful. It would give you an idea of how your model performs against real data.