The authors study the changing streamflow dynamics of catchments in response to changing precipitation. Disentangling the influence of climatic/weather and catchment properties on the streamflow response is rather complicated. This is regularly found in the efforts to explain the variability of streamflow elasticity or flow duration curve properties between catchments. Both typically end up with a mixture of explanatory variables. The current manuscript is well written and technical solid – as far as I can tell. However, like reviewer 1, I have some questions regarding definitions and the robustness of the results, given those definitions. I also have some questions regarding correlation versus causation.

[1] As the other reviewer also states one relevant question is the definition of “streamflow sensitivity to precipitation”. Is the streamflow sensitivity to precipitation well defined when we use the ratio between annual streamflow and precipitation? The mathematical definition of sensitivity is the change in the output due to variations in the input of a system or model. Plots of Q versus P do not quite capture this definition because they only look at the response of the system to the input, without consideration of what state the system was in. The latter is capture by the idea of streamflow elasticity where one quantifies the change in streamflow due to the change in precipitation (or something else) from year to year. Elasticity is notoriously difficult to explain (or regionalize) while the Q-P relationship is often rather stable. Given that this type of analysis was part of Anderson et al. (2024, HESS, “Elasticity curves describe streamflow sensitivity to precipitation across the entire flow distribution”), I think it would be very good if the authors were to make the connection and discuss how their definition in this manuscript differs from previous work (incl. some of the authors) and what consequences the changing definition has.

[2] My second larger point is about the difference between correlation and causation. The authors work here, and many of the papers cited, use correlation to infer causation. While I fully agree with the type of analysis, I think that it would be good to at least discuss somewhat whether correlation can be used here to infer causation. This also includes the discussion use of some of the references. One example is in lines 487ff. where the authors state: “This decrease could be explained by reduced connectivity among bare patches (Urgeghe et al. 2010)”. The Urgeghe et al. study runs a model for a design storm and varies vegetation patches to show their role for runoff behaviour during the design storm. I find it quite a stretch to use this reference in support of long-term catchment water balance behaviour. The authors need to at least explain why they think this connection is valid. The second part of the sentence in lines 487ff. is “and increased soil evaporation due to an increase in solar radiation reaching the ground (Guardiola-Claramonte et al. 2011).” Isn’t the latter than coinciding with reduced transpiration? Is the reduction in transpiration not larger than the added soil evaporation (given the deeper capture of moisture through roots)?

This is just an example where I think where the authors could expand their discussion and argument. I just given an example, reflective of the wider discussion section. It would in general be good if the authors were a bit more explicit why the references cited are transferrable to their situation.

[3] Influence of length of about 30 time steps on the robustness of the stationarity test? It seems like a very short time span for such tests. And given the widely discussed limitations of using statistical significance for justification. This is not a criticism of the approach, but a question of how one can assure robustness of the results?

[4] Temporal connectivity of drought events? Is there relevance to the temporal sequence of drought periods for this analysis. Even though I appreciate that the short time series might make this difficult to study.

[5] How relevant are conclusions that show differences of 2-3%? This should be quite below the amount of uncertainty one would expect in precipitation and/or streamflow observations even in good circumstances.

I found this is an interesting article. I have a few comments detailed below. Minor revision is requested.

1. Title: I wonder whether the article title chosen by the authors is clear enough. I found it very general and therefore not really convincing on the original results it brings. For example should the annual scale of the analysis be mentionned.
2. Abstract: Are the 2%-3% evolutions significant given all the other uncertainties in data?
3. Introduction: The runoff-to-precipitation ratio was heavily analysed in studies based on the Budyko approach. I find it may be useful to more explicitly make a link with the studies which analysed the sensitivity/elasticity of this approach to various variables and discuss how the proposed study can be linked to these previous works (e.g. Xue et al., 2020, https://doi.org/10.1016/j.jhydrol.2020.125589).
4. Section 2.1: I liked the fact that a large data set was used in this study. However I missed a discussion on data quality and possible dependency of results to the type of data used. For example, satellite products are known to be subject to large biases, which are not uniform whatever the regions or conditions. Besides they often show non stationary behaviour over time due to changes in algorithms or data. How these uncertainties may impact results shown in this study? A more detailed description of data used on these aspects would be useful.
5. Section 2.1: I found that Table S1 would be better placed in the main text of the article. This table is important to understand the variety between data used, e.g. in terms of periods available. I was also wondering which quality checks were done on the data used and how gaps in series were processed and accounted for in the models. If all catchments were plotted on a Budyko-type plot, could some specific/outlier behaviours be detected?
6. Section 2.1: Could there be any influence of year-splitting on results, especially on the memory to drought conditions? The use of hydrological years make sense, but it will likely split drought events in two parts, i.e. straddling two years. I was also wondering if the hydrological year was determined catchment by catchment or if an homogeneity was sought between catchments in a same region or under similar climate type.
7. Section 2.1: Which exponent values were used in the Box-Cox transformation?
8. Section 4: I was not really convinced by several points in the discussion were the authors try to find explanations to the results found. These explanations remain hypotheses and should more clearly be presented as such.
9. Section 4: The catchment memory to past conditions is heavily dependent on geology. Could the authors find a link between their results and geological characteristics?