The manuscript “Spatio-temporal patterns and trends of streamflow in water-scarce Mediterranean basins” by Laia Estrada et al. proposes an intricate modelling exercise on a set of drainage basins of the Catalan River Basin district, with the main objectives of providing a tool useful for water management and assessing the spatio-temporal patterns and trends of stream flow during the first two decades of the 21st century.

The purposes and the extent of the exercise as well as the contribution of the stakeholders and the innovative design of the model calibration-validation are the main strengths of the manuscript. Nevertheless, there are severe methodological inadequacies that not only put into question model results but also would provide inadequate examples on how this kind of exercises may be correctly done.

Working hypotheses:

In the Discussion section (lines 397 and subsequent), the authors quote several publications that “report a correlation between afforestation in the headwaters and decreases in streamflow” and indicate that “during the first two decades of the 21st century this trend has continued, with forested area going from 46% of our study area in 2000 to 56% in 2018”.

Disregarding these clear alerts, the authors “did not account for land use changes, although we did include the effects of increased evaporation due to warming, which may be more relevant (Buendia et al., 2016).”

But Buendia et al (2016), in the Final Remarks section state: “Overall, results have indicated that increased forest areas are the major driver of reduced streamflows and the magnitude of peak floods.”

The working hypothesis that the increase of forest cover can be omitted as possible cause of temporal trends of streamflow should be stated in the methods section, and the reason claimed by the authors for this omission is untrue.

Given the results of several previous works, it is likely that land cover change is a more relevant driver of recent hydrological changes in this area than climate warming. Both the overall and spatial flow trends simulated by the model become highly doubtful and are not compared with actual ones.

Analysis of modelling results:

Contrary to its recurring attribution as a ‘physically based model’, SWAT is an empirical model without a sound physical basis. The core of SWAT is the Curve Number Model that is undoubtedly an empirical model.

This is not just a rhetorical question but is relevant to the interpretation of the modelling results. In a physically based model there might be some hope that the internal model variables (stores and fluxes) are acceptable if the simulated discharge is so (but Anderton et al., 2002). However, when a conceptual or empirical model is calibrated using streamflow data, “Model performances measure the correctness of estimates of hydrological variables generated by the model and not the structural adequacy of the model vis-à-vis the processes being modelled” (Klemes, 1986). In other words, the model not necessarily gives the “good answers for the good reasons” (Grayson et al., 1992; Beven, 2002; Kirchner, 2006), so model fluxes not directly used for calibration are highly suspect of being model artefacts.

Furthermore, the uncertainties associated to the model simulations (at least those used for calibration) must be analysed to provide the users with estimates of the risks in decision making (Grayson et al., 1992; Beven and Binley, 1992; Beven 2006; Herrera et al.,2022...).

Finally, In a sub-section section of the ‘Materials and Methods’ section named ‘Data analysis’, the authors included the calculation of many hydrological indicators, but contrarily to the title of the sub-section, this analysis was made (if I am not in error) not on the original ‘data’ but on internal (not calibrated) model results. Therefore there is no assessment on how these indicators represent the ones of the actual hydrological regimes. 

Overall manuscript assessment.

In spite of the valuable strengths stated above, the modelling exercise is based on the inadequate working hypothesis that warming is the main driver of hydrological trends in this area and manages several principal and internal model outputs as actual data without any assessment of the uncertainty associated with these simulations.

Recommendations.

Both the importance of the objectives and the magnitude of the modelling exercise deserve finding some feasible way to improve the soundness of the project.

The fact that the encroachment of forest cover in the studied catchments is a likely or very likely driver of the hydrological response involves a difficulty for the modelling exercise but an opportunity for water management. Indeed, if climate were the main driver of the hydrological response, management strategies for adaptation to the climate change would be limited. Conversely, if forest cover is the main driver, it can be managed to reduce the ‘green water’ consumption and increase the ‘blue water’ delivery (Falkenmark, 2000) as a climate change adaptation strategy.

Using SWAT for simulating the hydrological response to forest cover change is a cumbersome and risky task, taking into account the poor or very intricate examples available (Haas et al., 2022; Karki et al., 2023).

But the flow simulations made may be used to test the null hypothesis that the climatic forcing is sufficient to explain the observed flow records, analysing whether there are time increasing model residuals that could be attributed to the role of increasing forest cover extent or density. This exercise may be made in most of the gauging stations used, providing a map of the hydrological changes attributable to the encroachment of forest cover. The statistical significance of trends should be made following the recommendations issued by the IPCC (Mastrandrea et al., 2010).

Unfortunately, the hydrological indicators analysed in the manuscript may be obtained for the flow records at the gauging stations, but any comparison with the simulated ones is expected to give inconsistent results because it is not possible to determine if the differences are attributable to modelling errors or to the role of the hydrological role of forest encroachment.

Finally, the maps of figures 4 to 7 should be discarded because these results are highly suspect of being modelling artefacts because do not take into account the role of forest cover change and these are internal model outputs not calibrated and of unknown uncertainty.

References:

Anderton, S. P., Latron, J., White, S. M., Llorens, P., Gallart, F., Salvany, C., & O’Connell, P. E. (2002). Internal evaluation of a physically-based distributed model using data from a Mediterranean mountain catchment. Hydrology and Earth System Sciences, 6(1), 67-84.

Beven KJ, Binley AM. 1992. The future of distributed models: model calibration and uncertainty prediction. Hydrological Processes 6: 279–298.

Beven, K. (2002). Towards an alternative blueprint for a physically based digitally simulated hydrologic response modelling system.Hydrological processes, 16(2), 189-206.

Beven, K. (2006). A manifesto for the equifinality thesis. Journal of hydrology, 320(1-2), 18-36.

Falkenmark, M. (2000). No Freshwater Security Without Major Shift in Thinking: Ten-year Message from the Stockholm Water Symposia. Stockholm International Water Institute. ISBN 91-973359-5-9. https://dlc.dlib.indiana.edu/dlc/bitstream/handle/10535/5149/PB-Report_No_freshwater_security_without_major_shift_in_thinking.pdf?sequence=1&isAllowed=y

Grayson, R. B., Moore, I. D., & McMahon, T. A. (1992). Physically based hydrologic modeling: 2. Is the concept realistic?. Water resources research, 28(10), 2659-2666.

Haas, H., Reaver, N. G., Karki, R., Kalin, L., Srivastava, P., Kaplan, D. A., & Gonzalez-Benecke, C. (2022). Improving the representation of forests in hydrological models. Science of The Total Environment, 812, 151425.

Herrera, P. A., Marazuela, M. A., & Hofmann, T. (2022). Parameter estimation and uncertainty analysis in hydrological modeling. Wiley Interdisciplinary Reviews: Water, 9(1), e1569.

Karki, R., Qi, J., Gonzalez-Benecke, C. A., Zhang, X., Martin, T. A., & Arnold, J. G. (2023). SWAT-3PG: Improving forest growth simulation with a process-based forest model in SWAT. Environmental Modelling & Software, 164, 105705.

Kirchner, J. W. (2006), Getting the right answers for the right reasons: Linking measurements, analyses, and models to advance the science of hydrology, Water Resour. Res., 42, W03S04.

Klemeš, V. (1986). Operational testing of hydrological simulation models. Hydrological sciences journal, 31(1), 13-24.

Mastrandrea, M.D., C.B. Field, T.F. Stocker, O. Edenhofer, K.L. Ebi, D.J. Frame, H. Held, E. Kriegler, K.J. Mach, P.R. Matschoss, G.-K. Plattner, G.W. Yohe, and F.W. Zwiers, 2010: Guidance Note for Lead Authors of the IPCC Fifth Assessment Report on Consistent Treatment of Uncertainties. Intergovernmental Panel on Climate Change (IPCC). Available at <http://www.ipcc.ch>.

The spatio-temporal analysis of streamflow patterns and trends over the last 20 years in Spanish Catalonia proposed in the article is, in my opinion, useful and interesting.

A large number of measurements are included in the article, and a major modeling effort is made to generate flow series that are continuous in space and time from data that are mostly discontinuous (which is always more or less the case everywhere). The analysis of patterns and trends involves numerous indicators of different flow characteristics (magnitude, duration, frequency). This makes for interesting and original results.

The authors state 3 objectives for their study:

1. Develop a useful modeling tool for water management
2. Propose a new calibration strategy that overcomes conventional approaches
3. characterize spatio-temporal patterns and trends of streamflow.

In my opinion, the demonstration made in the article for the first 2 objectives is not completely satisfactory.

On the first objective, the authors mention in section 2 "Co-development with end-users".

The "end-users" are not clearly defined (what type of structure do they belong to? how many are there? how were they chosen? did some refuse to participate? what was their level of appropriation of hydrological sciences and modeling?)

The authors indicate that they aim to help end-users understand how the model works, through training sessions. No details are given on the number of these sessions, their content or the end-users' prerequisites. No feedback is offered or analyzed on this appropriation phase (were there any evaluations following the training sessions? how did the end-users progress? what mastery levels were reached?).

The authors insist on "the inclusion of valuable expert knowledge on actual management practices". Is this to be understood as co-development of the model? Or rather as consultation to parameterize the water use rules of the reservoirs (in the same way as soil experts would have been consulted to parameterize soil properties in the model)?

In my opinion, the full description of the methods and the retrospective analysis are insufficient for this issue to be presented as an objective of the article. It would seem more appropriate to treat this point as a step in the parametrization of the model, based on expert data from the field. The authors' view of their collaboration with stakeholders could be discussed in the article, but as it stands, I don't consider that there is a clear demonstration of co-construction (what would have been the results of the modelling without this consultation on water use rules?).

With regard to the second objective, the proposed calibration/validation technique is interesting, but raises a number of unresolved questions.

In my opinion, the authors give this technique an exaggerated benefit in relation to the results shown in the article. Does this technique really provide better results than a traditional calibration/validation technique? It's quite possible, but it's not demonstrated in the article. In my opinion, it would require an article of its own to demonstrate this. This technique could simply be presented in the "materials and methods" section. Its positive aspects and limitations could be discussed. But positioning it as an objective of the article seems too strong, as do the claims that it best captures the spatio-temporal variability of hydrological processes in the study area.

Another point concerns the fact that only one land use is considered over the entire period, even though it may have varied, as the authors indicate. Would it have been more appropriate to calibrate the model on the flows of the period when this land use was in place? rather than calibrating on random periods?

On the third objective, the trend analyses are really interesting and raise several questions as to their interpretation.

Trends and patterns are based on model simulations. Calibration/validation performance is uneven between periods and between basins. I think it would be useful to associate a level of confidence with the indicators produced, depending on the quality of the modeling. This would allow us to temper the conclusions regarding patterns and trends.

Several causes are cited for interpreting flow trends: precipitation, rising temperatures (which should lead to an increase in evapotranspiration) and changes in land use.

Be careful, however, as the evolution of these causes in relation to flow changes is not quantified: l.392-393 the authors mention an absence of trend in annual rainfall, which is not quantified by a test. In addition, there may be trends in rainfall at other time steps and key periods in the year that influence river intermittency.

l.407-410: this summary is very probably true, but the article does not deal with forecasting future flows.

In conclusion, it seems to me that the objectives of the article should be reformulated to focus on the 3rd objective. The other two are, in my opinion, features of the methodology and should be presented and discussed as such.