The paper proposes a new methodology to obtain the best model evaluation, based on error metrics. It uses 38 different metrics, and through a clustering algorithm, it reduces the relevant metrics to less than half and then computes a composite metric to summarise the result. However, the paper does not demonstrate that this methodology allows for a ranking of the different models to build an informed ensemble mean. Is the Bergen Metric reasonably different from model to model or does it provide very similar results?  

Although it uses 38 different metrics, they all assume that the best model would be the one that has the best temporal synchronicity with the observations. While this is mostly true for the EURO-CORDEX evaluation simulations since they were forced by ERA-Interim, this is not the case for the historical runs. The latter were forced by GCMs and are only supposed to represent the historical climate. The introduction is also missing other types of performance metrics, which should also be added. E.g. metrics that evaluate the performance of the pdf or cdf are missing. Expand paragraphs 118 -121.

There are several model performance studies of the EURO-CORDEX domain, either for the entire domain or, specific regions. The different metrics employed should be referenced.  

The code provided does not allow for the reproduction of the manuscript’s results. The input data for the clustering code should be provided as well as the code to compute the final Bergen Metric.

A diagram with the complete methodology would be useful to better understand the different sections of the work.

The supplementary text S1 and S2 although relevant to the paper, are not referenced in the main text. Please add the corresponding references.

Lines 153 – 162 The Euclidean distance framework is not clearly explained. Please rephrase

Lines 186-189 Which version of E-Obs was used? There is already an E-Obs grid on 0.1^o resolution, why wasn’t this grid used? Add the justifying text to the manuscript.

Lines 198-206 belong in the introduction.

Line 209: Please provide the equations for each metric and the relevant references.

Table 1: the order of the last columns should be changed so that it matches the text.

I enjoyed reading the ms.

However, I have several concerns listed below:

L65: “38 different metrics” the reader can be curious where are they or how they are obtained from Eq7? Explain in more detail on section 3.3.

Eq 4-5-6-7: listing all four forms of the BM is confusing only Eq7 is enough for the reader.

Eq7: none of the components are innovative or directly related to distribution. They are all bias-sensitive components and may have overlapping (redundant) information.

In a recent HESS paper, spatial patterns of GCM/RCMs are used to the ranking.

https://doi.org/10.5194/hess-23-4803-2019

Applying cluster analysis doesn’t add much to the novelty of the metric. Each component must add something to the Bergen Metric.

One of the components in Eq7 should be histogram match (overlap %) for better discrimination power of the metric. Also the metric should be insensitive to the unit-differences since observation and simulation may have different units.

GCM to GCM comparison can be smooth and not unit issue; however, observed AET from MODIS is watt/m2 whereas hydrologic model flux simulations are in mm/day. Then ,Bergen metric cannot be applied to other hydroclimatologic problems.

EOF, SSIM, FSS, SPEM and SPAEF metrics should be covered in the literature review.