the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Sep 2025
Drought-induced land use and land cover change impacts on hydrology: Insights from the Harz mountains, Germany
Abstract. Drought conditions in Europe in 2018 and the following years led to significant land use and land cover (LULC) changes that affect the water balance. One of the regions that experienced strong changes are the Harz mountains in Germany, where drought conditions have led to tree mortality in its forested catchment areas. The aim of this study is to quantify these drought-induced LULC changes and analyze how they affect the water and sediment balance. To this end, remote sensing data and a deep learning model were used to derive annual LULC changes including dead tree areas between 2018 and 2023. These data were used to set up a SWAT+ model with dynamic LULC changes. The model was calibrated at the gauges of three headwater catchments of the Oker river. Kling-Gupta efficiencies indicate varying but at least satisfactory performance at all gauges during calibration (0.75 to 0.82) and validation period (0.63 to 0.79). Areas of dead trees were modeled as bare or sparsely vegetated (recent tree mortality) or as grassland (regrown areas). The LULC change analysis demonstrated strong performance (86–88 % overall accuracy) and revealed a decrease in coniferous trees by up to 46 % in one catchment (19 % and 25 % in the others). The hydrologic impacts were assessed by comparing a model run with LULC changes to a run without LULC changes. The results indicate a continuous decrease of evapotranspiration by up to -7.4 % in 2023 and a continuous increase of water yield by up to 11.3 % in 2023. The spatial assessment of modeled LULC change impacts shows strong increases in water yield and percolation and strong decreases in evapotranspiration associated with tree mortality. The increase in water yield can mainly be attributed to an increase in surface runoff. Changes in sediment yield indicate increased risk of soil erosion at areas associated with tree mortality. It was found that a dynamic model representation of tree mortality is necessary to account for these fast and strong changes and their impacts on hydrology. Moreover, the faster response of the catchment potentially increases the severity of flood events and the flood risk in downstream areas. The results underline that droughts significantly affect hydrology even after the end of the drought event. Therefore, afforestation with climate-resilient trees is needed to improve both flood and drought resilience in regions that suffer from drought-induced tree mortality.
Competing interests: Paul Wagner is a member of the editorial board of HESS. The authors declare that they have no conflict of interest.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(1975 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-3091', Santiago Valencia, 02 Nov 2025
-
AC1: 'Reply on RC1', Paul Wagner, 19 Feb 2026
Dear Santiago Valencia,
first of all, we would like to thank you for your constructive assessment of our manuscript. Please find our point-by-point replies below. Our comments are marked in italics.Major comments:
• The impacts of land use and land cover (LULC) change in the SWAT model are simulated based on the corresponding changes in the Leaf Area Index (LAI). Several studies have shown that inaccurate or unrealistic representations of LAI can significantly affect water balance simulations (e.g., Haas et al., 2022; Merk et al., 2024) and the assessment of LULC change impacts (e.g., Valencia et al., 2024; Zhang et al., 2020). Thus, I recommend showing the model’s performance in capturing LAI dynamics before and after tree mortality. Clarifying this aspect would also strengthen the interpretation of the reported changes in water yield resulting from LULC change. I am also wondering whether the authors considered changes in any soil parameter after tree mortality.
Reply: Thank you very much for the suggestion. We checked the LAI dynamics and will provide them in a revised version of the manuscript. We did not apply any changes to soil parameters.• Although the SWAT model was satisfactorily calibrated at the daily scale, I wonder whether a monthly scale might be more appropriate given the model’s limitations and the manuscript’s objectives:
o The watershed response to extreme precipitation events typically requires sub-daily analysis, as their duration is often shorter than one day, whereas the temporal resolution of SWAT outputs is daily. Moreover, the simulation of extreme events (e.g., floods) may also be influenced by the irregular spatial distribution of precipitation gauges across the study watersheds (Lines 402–415).
o The main analysis (Fig. 8) is conducted at a monthly scale, which provides a clearer representation of LULC change impacts on the watersheds' balance.
Therefore, I encourage authors to provide a more detailed explanation of the selected temporal scale as well as the associated limitations.
Reply: SWAT+ runs on a daily time step. We agree that the model’s capability to analyze sub-daily processes is limited. However, we believe that general conclusions can be drawn as e.g. that the increase of surface runoff may increase flood risk in the downstream. Even though many of the analyses are carried out on the monthly scale, we believe that calibration should be carried out on the daily scale, if daily discharge measurements are available to allow for the best possible process representation. Even though monthly scale calibration might yield a higher performance, it would significantly reduce the data that can be used for calibration (i.e. from 365 to 12 values per year) introducing more uncertainty. We will include statements on the used temporal scale and the associated limitations in the revised version of the manuscript.• Lines 169-180: To my best understanding, the SWAT model assigns weather data to each sub-watershed using the gauge located closest to its centroid. Therefore, I do not fully understand why the authors applied an interpolation to obtain 1-km spatial resolution precipitation, temperature, and humidity data. Please clarify why this step was necessary and what advantages it provides.
Reply: Please note that SWAT+ handles weather data differently than SWAT. In the SWAT+ model each spatial object is assigned the weather stations that are closest to its centroid [https://swatplus.gitbook.io/io-docs/introduction-1/climate]. Reasoning for interpolating weather data was to improve the spatial reliability of the model, as our focus is on spatio-temporal changes of hydrologic processes. We will change ‘spatial representation’ to ‘spatial reliability of the model’ in line 169 to clarify this.• Although forest loss also affects snow accumulation and melt, thereby altering watershed response and water balance (Dickerson‐Lange et al., 2021; Varhola et al., 2010), this aspect is only briefly addressed in the manuscript, particularly in the discussion section. I recommend that the authors further describe how forest–snow interactions are simulated in the SWAT model and their overall effect on the reported results.
Reply: Thank you very much for the suggestion. We will look into this and include this aspect in the discussion.• Figures 8 and 9: The authors should present water balance changes in standardized units (e.g., %) rather than in millimeters (mm) to account for differences in watershed area and the extent of LULC change.
Reply: Thank you very much for the suggestion. We carefully looked into this, but decided to use absolute units, because relative changes could be misleading if the original value is small. Watershed area will not affect the assessment when using absolute values (mm = l/m²). Moreover, one aim of figure 8 was to show how the extent of LULC change in the different catchments affects the response, so that we would not standardize over the extent of LULC change.Minor comments:
• Line 2: I suggest a small change in the title for better flow: “Impacts of drought-induced land use and land cover changes on water balance: Insights from the Harz Mountains, Germany”
Reply: We will follow part of the suggestion and change the title to: ‘Impacts of drought-induced land use and land cover changes on hydrology: Insights from the Harz mountains, Germany’. We prefer to stay with the broader term ‘hydrology’ as we draw conclusions that go beyond the water balance.• Line 72: I suggest adding a paragraph in the introduction to describe how hydrological models such as SWAT are useful for simulating the impacts of LULC change on water balance.
Reply: Thank you for the suggestion. We will add a paragraph in this regard.• Figure 1: I encourage the authors to include the locations of meteorological gauges and to use a coordinated system in degrees (e.g., WGS 84) for clarity.
Reply: We will revise the figure in this regard. However, we would like to underline that we interpolated the weather data.• Line 148: Why did the authors use a different spatial resolution in the digital elevation model (DEM; 5 meters) and LULC maps (10 meters) to define the HRUs? How many HRU were computed?
Reply: The resolution is related to data availability and the needed accuracy. The DEM was available at 1 m resolution and was upscaled to 5 m as this yielded a reasonable representation of the stream network. For the LULC maps we used the highest possible resolution given the 10 meter resolution of Sentinel-2 data. This setup resulted in 4842 HRUs in the three headwater catchments for the LULC map 2023.• Lines 162-163: What is the percentage of missing values in weather data?
Reply: The percentage of missing values depends on the stations and the variable. We will revise the section and add the information.• Lines 169-180: What method was used to estimate evapotranspiration?
Reply: The Penman-Monteith method was used within the SWAT+ model to calculate potential evapotranspiration. We will include this information in the revised paper.• Table 1: I suggest naming the parameters with capital letters for consistency with SWAT documentation and previous studies.
Reply: Please note that we used the SWAT+ model. Parameter names have changed between SWAT and SWAT+. The naming convention for SWAT+ with lowercase letters is applied here.• Lines 194-195: Did you use the LULC of 2018 to run the model from 2010 to 2018?
Reply: Yes, we will make this clearer in the revised version.• I suggest moving to supplementary material the following figure and tables to reduce the extension of the manuscript: Table 3, Figure 5.
Reply: Thank you very much for the suggestions. We will move Table 3 to the supplementary material. However, we would prefer to keep Figure 5 in the main text, as it helps the readers to better understand the temporal component of tree mortality.• Lines 273-275: Moriasi et al. (2015) provide a widely used categorization for SWAT model performance.
Reply: Thank you very much for pointing that out. We are aware of the evaluation criteria by Moriasi et al. (2015) and chose not to apply them, because we feel that these generalized model performance classes might conceal important details. We therefore prefer to discuss the performance with regard to the statistics (including the KGE that is e.g. not considered in the suggested paper) and with regard to visual analysis of hydrograph and flow duration curves to provide a more comprehensive assessment of model performance.• Line 279: The authors use the term “hydrograph” to describe daily discharge time series. However, to my understanding, a hydrograph is typically used to describe a watershed’s response to rainfall events. Therefore, I suggest using a different term to avoid potential confusion.
Reply: We believe that the term ‘hydrograph’ is very well established in the way we used it, see e.g. McDonnell, J. J., and K. Beven (2014), Water Resour. Res., 50, 5342–5350, doi:10.1002/2013WR015141.• Figure 6: Authors should include the performance metrics shown in Table 3 here. I also encourage authors to show the observed and simulated time series for the validation period.
Reply: Thank you for the suggestions. As the metrics are shown in the table, we prefer not to repeat the metrics within a figure. Evaluation of models depends on statistical (table 3) and visual (figures 6 and 7) evaluation, which we prefer to keep separate. We will add a visual comparison for the validation period in the supplementary material.• Line 309-310: What does the vegetation period mean? Please clarify this sentence
Reply: The vegetation period is a synonym for growing season. As we used both terms, we will change ‘vegetation period’ to ‘growing season’ in the revised version.• In lines 317–318, the authors state that “While LULC changes led to increased peak flows, water yield is lower during the recession phase.” Could this be interpreted as an indicator of reduced watershed regulation capacity? See Meli et al. (2024)
Reply: Yes, indeed. We will include a statement on regulation capacity with reference to Meli et al. (2024) in the discussion.• Lines 365-370, lines 374-377, lines 419-424: These sections in the discussion look more like to introduction and methods.
Reply: Thank you for pointing this out. These are discussions of methodological details. We will revise these sections to provide more clarity.• Lines 430-431: The Authors state that “model results show that transpiration from plants and evaporation from interception in the canopy decreased, whereas soil evaporation increased”. However, I did not find this clearly supported in your findings. Furthermore, it is unclear how the authors derived the partitioning of evapotranspiration into transpiration and evaporation from the SWAT model outputs. Could the authors please clarify both aspects?
Reply: SWAT+ provides the partitioning of evapotranspiration as a standard output. The final year with the most pronounced LULC changes, shows a decrease of transpiration and evaporation from interception by 9% and an increase of soil evaporation by 61%. We will include these numbers in the revised version to support the statement.Citation: https://doi.org/10.5194/egusphere-2025-3091-AC1
-
AC1: 'Reply on RC1', Paul Wagner, 19 Feb 2026
-
RC2: 'Comment on egusphere-2025-3091', Anonymous Referee #2, 05 Dec 2025
The authors applied a classification model to identify land use and cover changes based on remote sensing products, for three mountainous catchments in the centre of Germany between 2018 and 2023. Then they applied a hydrological model to study the impact of these changes (focusing on tree mortality) on hydrologic regimes. The topic fits well in the scope of HESS and is of interest for the field. The paper is well structured, and generally well written.
The subject is interesting, and the analysis has a lot of potential to shed light on impacts of droughts on hydrologic regimes in forested and mountainous catchments. The case study is also very interesting, with one of the basins that experienced 46% of forest loss. It is clear that the authors have put a lot of work in the modelling approaches which are at the core of the analyses presented in the manuscript. We believe that the publication is potentially interesting but need a lot of refinement and additional analysis for the conclusions to be robust. Our main comments are:
- The main issue we have with the paper, is that conclusions rely on the results of a deterministic hydrological model application. The model is SWAT, a well-established model, which has been calibrated on overall discharge from the period with LULCC (2018-2023). Model performance is acceptable, but not great (0.65 daily NSE, model underestimation of discharge in all catchments, particularly over summer, problems to capture the peaks). The authors state that this performance alone proves that the model can capture LULCC and that we can draw conclusions from it. In our opinion, to prove that the model can capture LULCC you should show that the inclusion of LULCC (comparing the model with and without) improves overall model results and bias, particularly over time. This should include a sensitivity analysis over the main choices of parameters and model setup, to show that the results are robust.
- We have a few general questions on the training and performance of the deep-learning classification model:
- The data used to train the model should be explained better. If we understood well, ESA and ELC10 do not include dead trees, which have been included in the training sample manually from satellite images for 2 years. The use of multiple products raises questions on their consistency, which might cause detection of LULCC.
- The model has been trained over the 3 catchments using 2 annual values (2020-2021). Then, it has been used to identify LULCC over 3 more years (2019,2022, 2023). Deep learning models require the highest amount of data among ML methods. This raises the question: what is the added value of training a DL model over a database which is objectively small, if you could manually identify LULC/dead trees also for the other 3 years and probably have higher accuracy?
- Authors highlight “strong performance” of the classification method (86-88%). However, for critical classes for LULCC, as grassland and dead trees, the performance is much lower (around 30% of false positive, and 30% of false negative). What’s the impact of this uncertainty on your results?
- The authors draw conclusions on sediment transport and erosion risk from a part of the model which is completely uncalibrated and not verified with any observation.
- If we understand well, the impact of LULC changes on hydrologic extremes (floods) is supported by a figure of a single flood event (Fig.9). A more comprehensive analysis should be carried out but, since the model has problems to describe floods, we suggest drawing conclusions of flood response changes directly from discharge observations.
Other comments (by Line number or chapter name)
Introduction: the literature review on land use and cover change and impacts on hydrology is a bit minimalistic and could be developed more as the core of the papers focuses on that subject.
Study area description is missing important information as geology, climate description, soil description, forest management, monitoring network (stream and rainfall gauges). The rain gauges should be indicated on the figure.
L104: How did you do the manual collection for the classification of dead trees? Was it manual classification based on image analysis? For what dataset and what period was this done? How are they merged with the other datasets?
L104: Is forest management information available to clearly identify where trees were cut/removed and use that as ground truth data?
L120: Please explain how the pixels are classified (highest probability? uncertainty?). For the land use and cover maps used for training the land model, the description is missing some information on their reliability (how good are they to identify land use classes of interest?), their availability (why just 2020 and 2021?).
L 145: hydrological model: SWAT classification is driving the results. More information is needed on SWAT structure, and how SWAT represents these classes, possibly with the main equations driving the change (e.g. ET, infiltration), and if they are verified in regions similar to the study area.
L155: Is the representation of dead trees as barren/sparsely vegetated appropriate? Isn’t this typical of poor soil conditions, while areas under dead trees in this area are quickly cover in lower vegetation? This seems like a key choice for model setup.
L169: You mentioned high uncertainty in the precipitation data. As this is impacting the water balance that you are assessing in your study, why not integrating radar QPE? Table 1: In the calibration, there are a lot of values of very sensitive parameters (CN, evaporation) at the boundary of the plausible interval. This is not a good sign; can you address this?
Table 3: Consider presenting the performances of the model without LULC changes, and the evolution of models performance in time
Figure 7: The message of this figure is not very clear
Figure 8: A chart with the evolution over time of the different LULC classes would help understand how and when the models should start to deviate from each other (Figure 8).
Figure 10: There is a lot of information, and it’s not always easy to distinguish colour variations.
Discussion: The discussion is quite descriptive of the results and data limitations. Data limitations are mostly focusing on precipitation estimates, while the uncertainty of model setup and calibration is not argued. In our opinion the discussion would benefit from more specific objectives and tests (e.g. how much does the impact change in wet and dry years? Across seasons? What’s the difference between the 3 catchments? Does it match your expectations?)
L390: You talk about drought impact: can you include a characterization of the years analysed? Which ones are dry? How much? Do they correlate with LULCC?
Conclusions: A bit short. Lacks specificity (e.g. “…decrease in evapotranspiration and increase in surface runoff, percolation, erosion risk were found”), relation with the literature, limitations, future challenges.
Citation: https://doi.org/10.5194/egusphere-2025-3091-RC2 -
AC2: 'Reply on RC2', Paul Wagner, 19 Feb 2026
Dear Reviewer,
first of all, we would like to thank you for your constructive assessment of our manuscript and your suggestions on how to improve it. Please find our point-by-point replies below. Our comments are marked in italics.The subject is interesting, and the analysis has a lot of potential to shed light on impacts of droughts on hydrologic regimes in forested and mountainous catchments. The case study is also very interesting, with one of the basins that experienced 46% of forest loss. It is clear that the authors have put a lot of work in the modelling approaches which are at the core of the analyses presented in the manuscript. We believe that the publication is potentially interesting but need a lot of refinement and additional analysis for the conclusions to be robust. Our main comments are:
1. The main issue we have with the paper, is that conclusions rely on the results of a deterministic hydrological model application. The model is SWAT, a well-established model, which has been calibrated on overall discharge from the period with LULCC (2018-2023). Model performance is acceptable, but not great (0.65 daily NSE, model underestimation of discharge in all catchments, particularly over summer, problems to capture the peaks). The authors state that this performance alone proves that the model can capture LULCC and that we can draw conclusions from it. In our opinion, to prove that the model can capture LULCC you should show that the inclusion of LULCC (comparing the model with and without) improves overall model results and bias, particularly over time. This should include a sensitivity analysis over the main choices of parameters and model setup, to show that the results are robust.
Reply: Thank you very much for the valuable suggestion. We agree that a comparison of the model with and without LULC change will be beneficial. However, we also argue that dynamic representations of LULC change should be used in a situation of such strong LULC change. If static representations are used, the effects of LULC change may be compensated through model calibration.
We therefore suggest the following to address the reviewer’s concerns:
a) We will add a paragraph reviewing the literature that shows that SWAT in general can represent LULC change. In particular, SWAT can also represent dynamic LULCC see e.g. our comparison study in which we particularly target this topic: Wagner, P.D., Bhallamudi, M.S., Narasimhan, B., Kumar, S., Fohrer, N., Fiener, P., 2019. Comparing the effects of dynamic versus static representations of land use change in hydrologic impact assessments. Environmental Modelling & Software, 122, 103987.
b) We will evaluate the performance of a model with a static LULC setup over time and compare it to the model with a dynamic LULC setup.With regard to the comment on model performance we would like to underline that model performance is thoroughly evaluate in the results section. We agree that peak flows are underestimated. However, underestimation is primarily found in the calibration period, whereas the validation period shows an overestimation in one catchment and a very good PBIAS (<2%) in two catchments. Moreover, we did not assess a systematic underestimation in summer. Performances differ for the different gauges, which we attribute to (i) the precipitation input as well as to (ii) trade-offs associated with the derivation of one set of parameters for all gauges. These points are discussed in detail in the discussion section.
2. We have a few general questions on the training and performance of the deep-learning classification model:
1. The data used to train the model should be explained better. If we understood well, ESA and ELC10 do not include dead trees, which have been included in the training sample manually from satellite images for 2 years. The use of multiple products raises questions on their consistency, which might cause detection of LULCC.
Reply: Thank you very much for the comment. To represent all relevant LULC classes, including dead trees, it was necessary to use multiple reference sources, as no single existing ground truth dataset provides comprehensive coverage of all classes of interest. In particular, the absence of a dead tree class in available LULC products required manual delineation. To minimize uncertainty, all products followed the same spatial resolution, class definition, and temporal alignment.
We will revise the data set description in this regard to provide more clarity. In addition, we will discuss the reliance on multiple reference data sets in the discussion section.2. The model has been trained over the 3 catchments using 2 annual values (2020-2021). Then, it has been used to identify LULCC over 3 more years (2019,2022, 2023). Deep learning models require the highest amount of data among ML methods. This raises the question: what is the added value of training a DL model over a database which is objectively small, if you could manually identify LULC/dead trees also for the other 3 years and probably have higher accuracy?
Reply: We trained the models over larger areas in and around the Harz Mountains, rather than limiting the training to the three catchments alone. Furthermore, the objective of using a DL approach was not to maximize year-specific classification accuracy through extensive manual interpretation, but rather to develop a scalable and transferable framework capable of producing spatially consistent land-use/land-cover (LULC) maps across multiple years and catchments.
Manual identification of LULC classes and dead trees for the years 2018, 2019, 2022, and 2023 is prohibitively expensive in terms of time and resources. We therefore used available annotations and labels (e.g., ESA products and tree species maps) as a starting point and complemented them with targeted manual annotations for dead trees.
3. Authors highlight “strong performance” of the classification method (86-88%). However, for critical classes for LULCC, as grassland and dead trees, the performance is much lower (around 30% of false positive, and 30% of false negative). What’s the impact of this uncertainty on your results?
Reply: The strong performance mentioned in the manuscript is related to the overall accuracy achieved by the DL model, which demonstrated the robustness of the proposed methodology in general terms for LULC classification. Going deeper on specific classes, dead trees had a lower performance than other classes, due to the scarcity of training samples. Given the overall accuracy of the LULC classification discharge predictions should be less effected as discharge integrates spatio-temporal changes in water fluxes. However, the uncertainty in the input maps obviously translates in the spatial impacts shown in the maps.3. The authors draw conclusions on sediment transport and erosion risk from a part of the model which is completely uncalibrated and not verified with any observation.
Reply: We agree with the reviewers that results regarding sediment yield should be interpreted carefully as the model was not calibrated for sediment transport. This is because sediment data were not available. We do not argue that the overall magnitude of changes is captured precisely. However, as we use the same model that relies on the same methods for calculating sediment yield, we argue that relative changes can be used as an indicator of erosion risk and therefore only expressed the changes in risk classes not to create a false impression of accuracy. We believe that erosion risk assessment complements the hydrological analysis. Moreover, we also show that model results on sediment changes are plausible given the correlation with surface runoff.1. If we understand well, the impact of LULC changes on hydrologic extremes (floods) is supported by a figure of a single flood event (Fig.9). A more comprehensive analysis should be carried out but, since the model has problems to describe floods, we suggest drawing conclusions of flood response changes directly from discharge observations.
Reply: Thank you, we gladly clarify that this is an exemplary figure to show the impacts of LULC changes on the daily time step. We chose this example for two reasons: 1) it shows the impact of LULC changes on peak flows and during the recession phase and 2) it explains the decrease in water yield in March 2022 that can be observed in Figure 8 on the monthly scale – which is against the general behavior of increasing water yield. We agree that the model’s capability to analyze floods is limited. We therefore only draw general conclusions as e.g. that the extremes are exacerbated as peak flows are increased and low flows are decreased.Other comments (by Line number or chapter name)
Introduction: the literature review on land use and cover change and impacts on hydrology is a bit minimalistic and could be developed more as the core of the papers focuses on that subject.
Reply: Thank you for the suggestion. We will revise the section in this regard.Study area description is missing important information as geology, climate description, soil description, forest management, monitoring network (stream and rainfall gauges). The rain gauges should be indicated on the figure.
Reply: Thank you for the suggestions. We will revise the section in this regard.L104: How did you do the manual collection for the classification of dead trees? Was it manual classification based on image analysis? For what dataset and what period was this done? How are they merged with the other datasets?
Reply: The dead tree class was generated through manual interpretation of satellite imagery. Specifically, dead trees were visually identified using high resolution Planetscope imagery (with 3 m spatial resolution) for the years 2020 and 2021. The samples were then resampled to 10 m spatial resolution and merged with the other dataset. We will include this information in the revised manuscript.L104: Is forest management information available to clearly identify where trees were cut/removed and use that as ground truth data?
Reply: Forest management or harvesting records were not available for the study area at the spatial and temporal resolution required for this analysis. However, based on visual analysis of time-series satellite imagery, areas where trees were cut or removed could be identified within the study area.L120: Please explain how the pixels are classified (highest probability? uncertainty?). For the land use and cover maps used for training the land model, the description is missing some information on their reliability (how good are they to identify land use classes of interest?), their availability (why just 2020 and 2021?).
Reply: Pixels were classified based on the highest predicted probability. We will clarify this in the revised version.
We will also add more information about all reference datasets used in the study. The reported overall accuracies of the ESA maps will be added to the supplementary data. For example, class-wise accuracies showed that in ESA 2021, user’s and producer’s accuracies were high for tree cover (80%, 92%), cropland (81%, 79%), and water bodies (90%, 86%), while grassland (72%, 67%) and built-up areas (66%, 73%) exhibited moderate accuracies.
Regarding their availability, ESA LULC maps are only available for the years 2020 and 2021. Although other LULC products exist (e.g., CORINE), they have a coarser spatial resolution and were therefore not suitable for this analysis. We also selected ESA WorldCover maps (2020/2021) over ELC-10 (2018) for model training to take advantage of their larger training sample sizes.L 145: hydrological model: SWAT classification is driving the results. More information is needed on SWAT structure, and how SWAT represents these classes, possibly with the main equations driving the change (e.g. ET, infiltration), and if they are verified in regions similar to the study area.
Reply: Thank you very much for the suggestion. Within the SWAT+ model the Penman-Monteith method was used to calculate potential evapotranspiration and the curve number method was used to estimate surface runoff. Both methods are commonly applied in Europe. We will include this information in the revised paper and will refer the reader to the SWAT+ documentation for further insights.L155: Is the representation of dead trees as barren/sparsely vegetated appropriate? Isn’t this typical of poor soil conditions, while areas under dead trees in this area are quickly cover in lower vegetation? This seems like a key choice for model setup.
Reply: We chose this class as these areas usually do not show a lot of vegetation. If they show more vegetation (e.g. regrowth), they are usually classified as grassland.L169: You mentioned high uncertainty in the precipitation data. As this is impacting the water balance that you are assessing in your study, why not integrating radar QPE?
Reply: Thank you very much for the suggestion. We actually tested different precipitation products when setting up the model. Among these were also radar QPE based RADOLAN data. The RADOLAN data led to strong overestimation and was therefore not used. Given the analysis we carried out in the region, we believe that the uncertainty stems from the spatial heterogeneity of precipitation in the region and an insufficient density of rain gauges to measure spatial patterns. Therefore, we used the mean annual precipitation pattern as a covariate for interpolation to better account for spatial heterogeneity in the precipitation input.Table 1: In the calibration, there are a lot of values of very sensitive parameters (CN, evaporation) at the boundary of the plausible interval. This is not a good sign; can you address this?
Reply: We adjusted parameter ranges to optimum ranges within the physically plausible parameter ranges. Optimum ranges were narrowed down with the help of preliminary model runs. Due to the narrower ranges, it is more likely that parameters are closer to the boundaries. Moreover, even if an optimum existed outside of the ranges, we would not want to go beyond physically plausible ranges.Table 3: Consider presenting the performances of the model without LULC changes, and the evolution of models performance in time
Reply: Thank you very much for the suggestion. We will evaluate the performance of a model without LULC changes over time and compare it to the model with the dynamic LULC setup.Figure 7: The message of this figure is not very clear
Reply: Flow duration curves are commonly used to show model performance for high, middle and low flow sections. We will add a statement on our model evaluation approach in the methods section.Figure 8: A chart with the evolution over time of the different LULC classes would help understand how and when the models should start to deviate from each other (Figure 8).
Reply: Thank you very much for the suggestion. We feel that depicting five classes for three catchments might be too much – given that the figure is already big. In addition, LULC change is on an annual scale, whereas the figure depicts monthly impacts. We will however look into this and evaluate if this information can help to explain the depicted impacts.Figure 10: There is a lot of information, and it’s not always easy to distinguish colour variations.
Reply: We will revise the figure to improve readability.Discussion: The discussion is quite descriptive of the results and data limitations. Data limitations are mostly focusing on precipitation estimates, while the uncertainty of model setup and calibration is not argued. In our opinion the discussion would benefit from more specific objectives and tests (e.g. how much does the impact change in wet and dry years? Across seasons? What’s the difference between the 3 catchments? Does it match your expectations?)
Reply: Thank you very much for the suggestions. We will use the suggestions to revise the discussion.L390: You talk about drought impact: can you include a characterization of the years analysed? Which ones are dry? How much? Do they correlate with LULCC?
Reply: Thank you very much for the suggestion. We will look into this.Conclusions: A bit short. Lacks specificity (e.g. “…decrease in evapotranspiration and increase in surface runoff, percolation, erosion risk were found”), relation with the literature, limitations, future challenges.
Reply: We will enhance the conclusion and add more specific outcomes.Citation: https://doi.org/10.5194/egusphere-2025-3091-AC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 2,039 | 244 | 39 | 2,322 | 27 | 26 |
- HTML: 2,039
- PDF: 244
- XML: 39
- Total: 2,322
- BibTeX: 27
- EndNote: 26
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Wagner et al. present a novel analysis of the impacts of drought-induced land use and land cover (LULC) changes, particularly tree mortality, on water balance using the SWAT+ model. This is particularly relevant as droughts are becoming more frequent under climate change, and most of the studies focus on the impacts of LULC changes associated with human activities (e.g., agriculture, deforestation, logging). Their results contribute to our understanding of how drought-induced tree mortality alters watershed water yield. Overall, I found the paper to be well-written and organized, and suitable for publication in the HESS journal. However, I have some comments that should be addressed before consideration for publication, as described below.
Major comments:
Therefore, I encourage authors to provide a more detailed explanation of the selected temporal scale as well as the associated limitations.
Minor comments:
References
Dickerson‐Lange, S. E., Vano, J. A., Gersonde, R., and Lundquist, J. D.: Ranking Forest Effects on Snow Storage: A Decision Tool for Forest Management, Water Resources Research, 57, e2020WR027926, https://doi.org/10.1029/2020WR027926, 2021.
Haas, H., Reaver, N. G. F., Karki, R., Kalin, L., Srivastava, P., Kaplan, D. A., and Gonzalez-Benecke, C.: Improving the representation of forests in hydrological models, Science of The Total Environment, 812, 151425, https://doi.org/10.1016/j.scitotenv.2021.151425, 2022.
Meli, P., Ellison, D., Ferraz, S. F. D. B., Filoso, S., and Brancalion, P. H. S.: On the unique value of forests for water: Hydrologic impacts of forest disturbances, conversion, and restoration, Global Change Biology, 30, e17162, https://doi.org/10.1111/gcb.17162, 2024.
Merk, F., Schaffhauser, T., Anwar, F., Tuo, Y., Cohard, J.-M., and Disse, M.: The significance of the leaf area index for evapotranspiration estimation in SWAT-T for characteristic land cover types of West Africa, Hydrol. Earth Syst. Sci., 28, 5511–5539, https://doi.org/10.5194/hess-28-5511-2024, 2024.
Moriasi, D. N., Gitau, M. W., Pai, N., and Daggupati, P.: Hydrologic and Water Quality Models: Performance Measures and Evaluation Criteria, Trans. ASABE, 58, 1763–1785, https://doi.org/10.13031/trans.58.10715, 2015.
Valencia, S., Villegas, J. C., Hoyos, N., Duque-Villegas, M., and Salazar, J. F.: Streamflow response to land use/land cover change in the tropical Andes using multiple SWAT model variants, Journal of Hydrology: Regional Studies, 54, 101888, https://doi.org/10.1016/j.ejrh.2024.101888, 2024.
Varhola, A., Coops, N. C., Weiler, M., and Moore, R. D.: Forest canopy effects on snow accumulation and ablation: An integrative review of empirical results, Journal of Hydrology, 392, 219–233, https://doi.org/10.1016/j.jhydrol.2010.08.009, 2010.
Zhang, H., Wang, B., Liu, D. L., Zhang, M., Leslie, L. M., and Yu, Q.: Using an improved SWAT model to simulate hydrological responses to land use change: A case study of a catchment in tropical Australia, Journal of Hydrology, 585, 124822, https://doi.org/10.1016/j.jhydrol.2020.124822, 2020.