1. We thank the reviewer very much for this general comment and gladly provide reasoning for the used methodology. 

We agree with the reviewer that a calibration using one static land use map for streamflow was applied. We chose to use the map of 2019 as it provides the most detailed land use classes and is closer to the two periods for which water quality data was available. Moreover, land use is not as static as it seems as crop rotations have been implemented. With regards to water quality, the SWAT model is calibrated for water quality data in 2009-2011 with the land use map of 2010 and validated for measured data in 2018-2019 with the land use map of 2019.

While we agree that a dynamic implementation of land use changes would be necessary to represent dynamic changes, in this study the model is used to compare the averaged output from three static model runs to the associated changes in land use. We therefore believe that a calibration using a static land use is suitable. We however evaluated the performance of the model with regard to streamflow for the other two land use maps and showed that the parameters derived during calibration are also suitable for these land use maps (using land use map 1987, NSE: 0.75 - 0.80, KGE: 0.82 - 0.88, PBIAS: -1.1 - 12.2; using land use map 2010, NSE: 0.76 - 0.81, KGE: 0.82 - 0.88, PBIAS: -2.5 - 10.3). 

2. We thank the reviewer for this helpful comment. We will provide additional information as requested to improve the description.

3. We will clarify how the model has been calibrated. We used an auto-calibration approach based on 8000 model runs for streamflow and 5000 model runs for sediment, TP, and TN. Parameter sets for these runs have been generated using Latin Hypercube Sampling method. A multi-metric approach was used for streamflow including KGE and RSR as objective functions. For sediment, TP, and TN, NSE was used as the objective function. We will clarify this in section 2.3.3. 

4. The stars are used to indicate the extreme outliers. We will add a legend for explaining all notations (e.g., circles, stars) in Figure 6.

5. There is not one baseline for all comparisons in Figure 7. The figure shows differences between the model runs (each is performed between 1990 and 2019) with a land use map of the first stated year and the second stated year, i.e. the first stated year in each column may be referred to as baseline (e.g. 1st column 1987-2010, baseline 1987; 2nd column 2010-2019, baseline 2010; 3rd column 1987-2019: baseline 1987). We will add more details in section 2.3.4 to clarify this point. 

Response to general comments: 

1) We thank the reviewer for this very constructive comment. We agree that we can broaden the introduction to highlight the suitability of SWAT for this study. Specifically, we will shorten the first and second paragraph about the importance of LULC and add a new paragraph, which will introduce the importance and suitability of using SWAT to assess the impacts of land use changes on hydrology and water quality. We will also highlight the high capability of the enhanced SWAT version (with two active groundwater aquifers) of modeling groundwater process and associated water quality components in lowland catchments. 

2) Thank you very much. We believe that it is important to show that the model is capable of accurately representing water quality and quantity. We therefore include an evaluation on a daily time step. We will however re-evaluate which parts are most important for our study and to make the manuscript more concise, we would agree to move part of the results of model calibration and validation (section 3.1) to the supplementary materials. 

3) We use the land use map of 2019 (detailed land use classes) for the SWAT calibration and validation for streamflow. The other land use data have been used for land use scenario model runs. 

Following the reviewer’s comment, we will remove the descriptions and instead provide a table to summarize which data has been incorporated into SWAT. 

4) Thank you very much. There are a few findings related to lowland characteristics in the dynamics of water quality loads (in Figure 7). E.g., due to land use change between 1987 and 2019, sediment, TP, and TN loads decreased least in the central part of the catchment with a lower topography (lowland). This may be related to the more intense exchange between groundwater and surface water and a higher contribution of nutrients from groundwater to stream in lowlands. We would like to follow the reviewer’s suggestion and add a few statements regarding this to better highlight findings related to lowland characteristics.

5) We will consider these comments and improve abstract and conclusion.

6) We thank the reviewer for this constructive comment! We will add comparisons to other watersheds in the discussion part. 

Response to other specific comments:

1) Thank you. We will delete some details and shorten the abstract properly.

2) Thank you for pointing this out. In L41 it should indeed be land use changes. We will work on and improve the first paragraph of the introduction.

3) We will rephrase it. Thank you!

4) We will provide a paragraph for the SWAT model in the introduction to improve this.

5) As stated in L163 only the land use map of 2019 was used for model setup, model calibration and validation. We will rework section 2.3.3 to make this very clear. 

6) Thank you very much for this helpful comment. The daily sediment load (in L107) is calculated based on daily concentration of total suspended sediment concentration (in L181) and measured daily streamflow. We will provide this information somewhere applicable. 

7) Actually, we calibrated TP, TN, and sediment in the same years (2009-2011) as we stated in L196. But specifically, we calibrated TP and TN during the period of 08/08/2009-10/08/2011, and sediment during the period of 30/10/2009-07/08/2011, which differs slightly due to the availability of measurements. 

8) Yes, we agree to move Table 2 to the supplementary material.

9) Yes, we will include more details in this regard. 

10) Here we wanted to state respective changes between 1987 and 2010, between 2010 and 2019, and between 198 and 2019. We will follow the reviewer’s advice and clarify it.

11) “Wold criterion” is a common terminology in PLSR method proposed by Svante Wold (Wold et al., 2001). It is often referred as a criterion for assessing the variable importance for projection (VIP) statistic. “by word’s criterion” means “according to word’s criterion”. We will change “by word’s criterion” to “according to word’s criterion”.

12) Thank you! In fact, we did not evaluate the performance of the model for the monthly data in this case, because there are only 24 (calibration) or 12 (validation) monthly water quality observations, which is not a reliable basis for model evaluation. However, it can be expected that the model performance is better on a monthly scale, as over- and underestimations tend to balance out. This statement is in agreement with our modeling experience and the literature. We will provide references to support our statement.

13) We will modify the color scheme in Fig.3.

14) “agricultural grasses” indicates grasses grown on fields by farmers for feeding livestock. 

15) We will add a legend for the box plots. 

16) We chose to show absolute changes, as these absolute changes are used in the PLSR approach. We believe that it is better to keep the response variables (changes of water quality and quantity) and explanatory variables (changes of land use indicators) consistently in absolute values. Moreover, the absolute change allows the reader to better compare changes for different variables (e.g. if evapotranspiration is increased by a certain absolute value, water yield is probably decreased by the same absolute value). We would therefore prefer to keep showing absolute changes.

17) Thank you for spotting this. The term may be confusing. We suggest to change it to “changes of the rural landscape”, as about 75% variations in sediment, TP, TN are explained by the changes of this rural landscape, which means the change of rural land use plays a key role in influencing water quality variations.

18) Yes, we will denote the meaning of “AIa”, “CONTIGAWa” right before them, respectively.

19) We agree with the reviewer, and will remove the sentence in L448-449.

20) We agree and will not highlight the methodology here.

• Thank you very much for reviewing this manuscript and providing constructive comments. We should highlight that the model accurately simulated daily long-term streamflow at the available three gauges in the catchment. We are therefore confident that the model is capable of representing spatio-temporal water fluxes in the catchment. Moreover, we agree with the reviewer that long-term measurements of water quality on a daily time step are rather scarce. However, for the study area observations for three time periods in three decades were available, which is a comparatively good data basis. We will gladly address the points raised by the reviewer to improve the manuscript and provide further details on the used methods.

Introduction

• Thank you very much for the constructive comment. We will revise and improve parts of the introduction to better highlight the scientific challenge and frame the scientific question of this study. Particularly, we will rewrite or improve the sentences as suggested by the reviewer.
• Thank you for his very helpful suggestion. We will describe the key information on L86-89 with more details, and meanwhile briefly summarize what previous scientific studies found using PLSR for land use change and hydrology.
• We will add some more recent references.

Terminology

• This is indeed a key term and we will provide a description of “landscape metric” and examples of explaining its meaning where it first appears (Line 63).
• Thank you very much. We will add an explanation of the term “patch”.
• Thank you for spotting this. This is a typo. We will take care of this and also check the rest of the manuscript carefully.

PLSR

• Thank you very much. We will work on this section. Specifically, we will explain more about how land use changes are employed as predictors and how modeled dynamics in water quality and quantity are incorporated as response variables for PLSR. In addition, we will provide some necessary details on calibration, cross-validation and performance assessment of PLSR in the supplementary material. 
• Thank you. Actually, it is correct that we had only three available land use maps (1987, 2010, and 2019). We analyzed three time intervals, by analyzing the change in land use from 1987 to 2010, the change in land use from 2010 to 2019, and the land use change between 1987 and 2019. We will make that clearer in the revision.
• Yes, we can provide an example to explain the method of 50 random repetitions on 10 equal segments.
• Yes, the sentence will be rephrased to “According to Wold’s assessment criterion, the predictor …”. 
• Thank you. We will define this difference on L 64-65 properly. Landscape metric is a quantitative indicator that describes specific spatial characteristics (i.e., spatial fragmentation or structure) of land patches, e.g., the connectedness degree or shape complexity of the land patches of one class or the entire landscape. A composition metric normally refers to the abundance of patches belonging to one class within the landscape, without considering the spatial character, placement, or location of patches, e.g., areal percent of a land use class. 
• We will provide a brief description.
• Yes, we will rephrase and improve the sentence.
• Thank you very much. We agree that this will help and we will try to work out a sketch to visualize the meaning.