Interactive comment on “Modelling the hydrologic response of a mesoscale Andean watershed to changes in land use patterns for environmental planning” by A. Stehr et al

The authors present an interesting case study about the hydrologic response of an Andean watershed to land use change. While the model application, as reported in the text, is not very innovative from the scientific point of view, the number of SWAT modelling case studies from the Andean region is rather low compared to other regions. So there might be an interested audience. The topic of the paper is suitable for HESS, but the scientific quality of the paper is not yet sufficient for publication. In its current state the paper should be rejected. Based on an assumed general interest on

(mm day-1) which is approximated to 0.2 R, and R is the retention parameter (mm day-1), which varies spatially due to changes in land cover, soil type and slope, and temporally due to antecedent soil moisture conditions: Equation 2where CN is the curve number for the day.Equation 3Reply 1 b).As already mentioned in reply 1 of referee # 1, we did not expand the SWAT database of land uses.In SWAT, land uses are classified according to the land cover database proposed by Neitsch et al. (2002).The land cover database includes the description of 97 different land uses.Each land use is identified with a code.In the particular study case, the observed land uses where compared with those described by Neitsch et al. (2002).It was found that all the observed land uses are present in the database by Neitsch et al. (2002).Thus, the observed landuses were codified following the standard database.
Reply 1 c).SWAT includes three different methods for computation of evaporation/evapotranspiration, namely: Penman-Monteith, Priestley-Taylor and Hargreaves.As there are no records of wind speed, relative humidity and solar radiation in the basin, but maximum and minimum temperatures are available from meteorological stations we used the Hargreaves method.The Hargreaves method has shown good results in different type of climates (Jensen et al. 1990;Allen et al., 1998, Antonioletti et al. 1998, Droogers et al., 2002, Saghravani et al. 2009).Moreover, Jensen et al. (1990) (in Saghravani et al., 2009) compared different 20 evapotranspiration methods against lysimeter data.Of all methods that required only air temperature, namely, the Hargreaves method showed the best results.
Comment 2. The land use change from 1979 to 1994 is rather dramatic compared to other regions.From 1979 to 1994 it is 15 years, and now you might even have data for the next 15 years (-2009)?Can you evaluate your scenarios against recent observation data?Which of the five scenarios is closest to reality so far?Reply 2. Unfortunately, the most recent available landuse data are those presented in C2016 the paper for the year 1994.Thus it is not possible to perform a hard evaluation of the predicted scenarios against recent observation data.We agree with the comment of referee #2 on the dramatic observed landuse change between years 1979-1994.That fact make the presented study case a very interesting one and is one of the main motivations for the publication of this paper.Currently, researchers of the Environmental Research Center EULA-Chile are working on the image processing in order to generate actual landuse maps from satellite imagery.In the near future we expect to be able to submit a new paper with those results and the evaluation of the predicted scenarios.
Comment 3. Scenarios 1 and 5 seem to have a logical background, while scenarios 2 to 4 seem to be more a kind of sensitivity analysis.These three are not realistic, are they?The way how you developed the scenarios, mainly scenario 5, should be documented in more detail.Please describe the regression model (e.g.formulae) and document your assumptions.The generation of the scenarios should be reproducible.
Reply 3. We agree with referee #2: Scenarios 2 and 4 do not have a logical background.However, they provide a good idea of the limits inbetween discharge, i.e. water availability in the watershed, could change following landuse patterns.As those scenarios are not realistic, we decide to eliminate them in the paper.Analysis of scenarios 1, 3 and 5 provide the relevant information as well.Scenario 5 was obtained using a set of prediction variables such as elevation, slope, distance from native forest, distance of forest plantations, distance from urban areas and size ownership.The documentation and explanation of the development of this scenario has been incorporated on the manuscript at "4 Generation of probable land use scenarios" and reads as follows: To quantify the relationship between land cover changes and its causal factors, the maps of 1979-1994 were sprawl and results related to a set of predictor variables (change and non change) that were selected based on current knowledge of landuse changes process in the Vergara watershed (Echeverría et al., 2006;Echeverría et al., 2007;Altamirano et al., 2007;Aguayo et al., 2009), Table 4 shows this variables.An C2017 appropriate binary response variable was constructed from the observed forest expansion pattern and a logistic regression model were used to predict the probability of land cover change depending on the various predictor variables (equation 1; Table R-5).Equation 4Table R-5.Results of the adjustment of the logistic regression for forest plantation sprawl (** = p < 0.01).
Comment 4. In table 8 you show results for the scenarios 1 to 4, but not 5 -why?Scenario 5 is very interesting.The caption of fig. 10 is incomplete: is f) scenario 5? Reply 4. We agree with referee #2.Table 8 has been completed, showing scenarios 1-5 for all the subbasins.The table is: Table 8.Percentage of change respect to the baseline scenario for mean annual, wet season (May -October) and dry season (November -April) flows.
Comment 5.The description of the SWAT model should be carefully revised.Interception is not computed with the CN method, you mention surface runoff twice, but I think you mean generation and concentration of surface runoff etc.
Reply 5. We agree with referee #2.Interception is not directly computed when the CN method is use.It is consider together with infiltration and surface storage in a term called initial abstraction (I) which is approximated to 0.2 of the retention parameter (R), accordingly to: Equation 5In the text, we corrected: Interception, surface runoff and infiltration were computed with the curve number method.The surface runoff is computed following the kinematic wave approach, using Manning's relation for estimation of the runoff speed.By Surface runoff was computed with the curve number method.Water is routed using the kinematic wave approach, using Manning's relation for estimation of the runoff speed.

C2018
Comment 6.The deep aquifer in SWAT should be handled with care.You can remove a lot of water from your watershed when you set a high value for the rchrg_dp parameter -as you did (0.5 -1 as reported, that means up to 100% of the percolating water!).Does this correspond with the local hydrogeologic situation?When removing so much water (violating the continuity equation for your watershed), it should be documented where the water is transferred and why it is.Reading such parameter values for rchrg_dp and a relatively high bias of your model results gives cause for serious concern.
Reply 6.We agree with referee #2.The parameter rchrg_dp bounds indicate the minimum and maximum value that this parameter can take, and certainly 1 is a very high value.There is a typographic error in the next.In fact, the value of the lower limit of rchrg_dp parameter is 0.05 instead of 0.5.In the presented simulations rchrg_dp parameter ranged between 0.05 and 0.1.
Comment 7. The results of the sensitivity analysis (Section 6.1) are not discussed.How do they compare with the results of others?Did you observe a specific behavior of the model, which can be related to the local situation?Is there a recommendation to use different parameter ranges for PARASOL in your region?Table R-6.Sensitive parameters obtained in the referred studies According to table 6 we observed a similar behaviour of SWAT for applications in the basins located in south-central Chile, i.e.Vergara and subbasins, to that reported for watershed located in the northern hemisphere by the aforementioned researchers.

C2019
The use of the parameter ranges showed in reply #1 of referee #2 is strongly recommended.
Table 4 in the manuscript will be replaced by the following table; in order to avoid redundance of information provided in reply # 1 of referee #2.It shows the ranking of the 4 most sensitive parameters in Tijeral, Rehue, Renaico, Mininco and Malleco.
Table 7. Ranking of the 4 most sensitive parameters in Tijeral, Rehue, Renaico, Mininco and Malleco Comment 7. In section 6.2 you discuss the calibration results for 2000-2002.You report that the model "satisfactorily reproduced the order of magnitude of the observed discharges".This does not sound convincing.Model and data uncertainty should be analyzed in more detail.For example: you report that the model subestimates (underestimates) peak discharge.I am not surprised when I see the availability of rainfall data: only one station within the mountain area and that station has a lot of missing values even within the calibration period.Which method did you apply to interpolate (?) rainfall data?
Reply 7. We are surprised that our statement "the model satisfactorily reproduced the order of magnitude of the observed discharges" does not sound convincing to the referee #2.Also, it is rare that one expect an underestimation of discharges when data availability is scarce; why not an overprediction?The presented simulations of the Vergara watershed with SWAT calculate complex non-linear hydrological process in a semidistributed manner solving a mathematical problem without analytical solution.Thus, it is reasonable to expect some differences between observed and calculated values.Based on our experience with hydrological modeling, there is no clear tendency to under or overpredict discharges depending on the data availability.Influence of hydrologic regime of the watershed, as well as the soil type, land uses might control model response.
Rainfall data where assigned to the different subbasins using the methodology incor-C2020 porated in ArcSWAT, i.e. rainfall data use to calculate runoff are obtained from the precipitation station centroid that is closest to the subbasin under consideration.Additionally, to model snow accumulation and melt each sub-basin generated in SWAT can be divided into 10 elevation bands in order to incorporate temperature and precipitation variations with respect to altitude (Hartman et al., 1999).For each sub-basin, different lapse rates for precipitation plaps (mm H2O/km) temperature tlaps ( • C/km) can be defined, which are then used to account for the differences in precipitation and temperature between these elevation bands.Reply 8.As referee #2 enunciates the sentence, it seems to make no sense.The sentence "changes tendency in time" in 6.2 and 6.3 read: "The model satisfactorily reproduced the order of magnitude of the observed discharges, and their changes tendency in time."Observed discharges exhibit a seasonal dynamics with a mean value that is about an order of magnitude higher in winter than in summer, with significant winter highwater events of about 8 days.
Comment 9. P. 3082, l. 20: "calibration period" -isn't it validation period here?Reply 9. We agree with referee #2.There is a typographic error in the manuscript.It read: calibration period Might read: validation period Comment 10.Table 2: what do the numbers mean?I'm not sure if I understood the table.It needs some explanation in the capture, also provide units (ha?) Reply 10.A landuse transition matrix shows the changes of landuse for different years.C2021  1997 and 1982, respectively.This stations could be left out in the paper.Nevertheless we consider that it is important to present all the data, because it illustrates a real situation of a basin with scarce hydrometeorological data, which is a characteristic feature in the region and represents a challenging topic in hydrological modeling: development, application, and/or modification/adaption of modeling tools for the correct estimation of the water balance components in basins with scarce data availability.In the text, we modified Table 3 in order to explicitly show the period used in the computations for each subbasin.8).This needs to be discussed in detail.
Technical Corrections: Language is understandable, but grammar and spelling need a revision to meet publication standards.Asking a native speaker is recommended.
Figure R-1 shows the subbasins and the location of the corresponding meteorological station.

Figure R- 1 .
Figure R-1.Subbasins and the location of the corresponding meteorological station Comment 8.In 6.2 and 6.3 you mention "changes tendency in time" -what do you mean with that?Are you able to say anything about trends in runoff, based on the short calibration/validation periods?

Table 2
present values in[ha].The total of each column and row indicates the area of the land covers for the years1979 and 1994, respectively.The values of the columns indicate the land cover changes occurred between 1979 and 1994 (e.g.30428 ha of native forest for a total of 133096 ha were converted to forest plantations).The values of the diagonal indicate the area that remained the same during the period (e.g.92 533 ha of native forest from a total of 133096 ha were maintained during the period 1979-1994).Comment 11.Only three of the five gauging stations have been operated in 1977, so are the data presented in table 3 really related to the period 1977-2002?Why do you include the two stations Rehue and Renaico in your study, could they even be left out?Reply 11.In fact the gauging station Rehue and Renaico have been operated since

Table 3 .
Mean monthly discharges [m3/s] at the different control points in the Vergara basin.Comment 12. P. 3083, l. 2: From your model results, you assume that the model can be applied to analyze the impact of land use changes on the hydrologic response.

Table 5 :
In my opinion the model bias is rather high.The percentage of change caused by the different land use scenarios is within a similar magnitude (table

Table R -
4. Parameters involved in the computation of surface runoff in the Vergara watershed.Table R-4.Parameters involved in the computation of surface runoff in the Vergara watershed. C2027

Table R -
4. Parameters involved in the computation of surface runoff in the Vergara watershed.(continued)

Table 8 .
Percentage of change respect to the baseline scenario for mean annual, wet season (May -October) and dry season (November -April) flows.

Table R -6.
Sensitive parameters obtained in the referred studies Fig. 6.Table R-6.Sensitive parameters obtained in the referred studies C2031