This manuscript confuses me slightly, as it is halway between a dataset description and an article analyzing trends in streamflow. I would rather this this document stripped of all of its trend descriptions and analysis and simply, and dryly, present the dataset. I do see value in the dataset itelf, but the accompanying manuscript needs a fair bit of work in my opinion.  Principly, the authors must go into much more detail about all the 'off the shelf' models they have used instead of relying on citations, and they must also qualify many of their statements and choices- the statistical tests here are not rigorous and the language is imprecise in communicating about the data. The manuscript is also oddly organized, with information relevant to methods appearing in multiple sections (see below). I recommend a major revision instead of a reject only because the dataset itself could be useful to archive. If the authors wish to do so, I beleive they must redo this accompanying document. My comments below are not exhaustive- there are many similar instances to what is below that must be found and elminated/changed.

L51- For a dataset description paper, I don't think you can rely on these citations. I recognize that this paper does not describe those models, but it is not sufficient to simply list 'principles, inputs, calibration, and validation' without proof of the skill of the model nor its methods.

Section 2.2- same comment as above. We need to understand how EROS has been calibrated, and its resulting skill.

Section 2.3- Why not use a globally consistent forcing? I worry that this dataset is fine, but that we can't repeat these methods/data in other basins as the forcing is unique to France. 

Figure 1- Is much too hard to read. The text size is too small. The resoltuion is quite high, but the figure is not legible without zooming in quite a lot. Please remake with readable font sizes

Figure 2- I am not sure what this adds? I would delete it.

3.1- this is not the place for skill scores of EROS. your model is about temperature, not discharge, so this information belongs in methods, not results.

L127- "pretty good" is unacceptable professional writing. The atuhors must at least quantify how good the preformance is using some standard metric (RMSE, NSE, KGE, etc.)

Figure 3 and L110- what is the justification for the 'locally weighted smoothing'? Are you using these lines to justify your trends? you must use a statistical test for these trends- either an MK or an PWMK or other citable and defensible test.

Caveats- this is an odd section to include, and these skill scores belong elswhere, in results.

The paper is interesting and present significant effort – very detailed dataset of historical and future stream temperature along Loire River basin. I would expect that it can be published after revision. The main concern regarding various parts of the manuscript are presented below.

1. Rate and heterogeneity of discharge change during the historical period considering size of the basin seem too strong to be attributed to climate change solely. Some additional information about human impact on the discharge may clarify it.
2. It is not always clear what lower and upper limit of the range mean – “Indeed, 3 %–83% stations (resp. 50 %–100 %) on small and medium (resp. large) rivers had a RMSE< 1°C across seasons (see their Figure S9, bottom panel)”. Is it variation across seasons, sizes?
3. Does the model considering potential landscape change due to climate change? Are there some estimates how significant this impact can potentially be?

This paper reconstructed historical, and projected future discharge and water temperature data with high spatial resolution for 52,278 reaches over the Loire River basin. This dataset will be useful for further researches in this area, and the method for the dataset generation might be able to be applied in more regions in the future. Overall, this manuscript is reasonably organized and I think this manuscript is acceptable for publication with minor revision.

Specific comments

For all figures, if you want to describe the subfigures, please numbering each subfigure. For example, using a), b), c). And then using the numbers/letters refer to subfigures, instead of using the words like “top”, “left” to locate them.

Line 65-69: Do you have validation period, if yes, please specify it.

Line 71: “(see http://www.drias-climat.fr/”, the bracket not closed.

Line 103: “the Nash-Sutcliffe efficiency of simulated daily Q is > 0.7 for Q, ln(Q), and √Q”, do you mean the NSEs of Q, ln(Q), and √Q are all >0.7? Since you already have the NSE of daily Q, why the NSE of ln(Q), and √Q still should be considered?

Line 105: What would be the possible reason for the underestimation of Q in winter and spring?