Modeling Monthly Nitrate Concentration in a Karst Spring with and without Discrete Conduit Flow
Round 1
Reviewer 1 Report
Two methods were compared for the karst spring discharge and nitrate transport at Silver Springshed. Generally it is interesting. However, due to the setting of the nitrate input, the simulated results regarding nitrate concentration is not that promising. As authors discussed maybe a finer resolution is needed and I also strongly suggest to consider a seasonal cycle of nitrate input if possible in this manuscript or in a future study. I have the following comments:
Major comments:
- CFPv2 is superior to EPM for discharge simulations. However, the nitrate concentration simulation shows much smaller difference between the two approach. Will the constant nitrate concentration in groundwater recharge lead to a systematic problem? Because normally the concentration is high after fertilizer application. Is the seasonal pattern tested?
Minor comments:
- Line 21-24: providing some numbers for the slight difference and also for the big improvement by using CFPv2 will be quantitatively clearer.
- Line 61: what is the 12-40% improvement? Monthly water balance, or performance metrics?
- Line 72: check the in-text citation style.
- Line 171: why leave out discharge in the period of 2017-2019 Feb?
- Line 191: why using 190.5m not an integer such as 200m?
- Fig 5a: maybe switch to m/s since the description in the text is m/s.
- Fig 5b: are these the major conduits that were used in the model. Are there any smaller conduits? If Any, will they affect the results?
- Line 274: The effect of denitrification should be justified. The variation of water table and oxygen concentration may largely influence the nitrogen concentration.
- Line 377: the number of point source proportional to population density should be given.
- Line 381: how to calibrate the steady-state model since the real system is barely under steady-state such that what are the “observations”?
- Line 380: why calibration method was used? What is the objective function?
- Line 471: is that monthly average?
Author Response
Response to Reviewer 1 Comments
Two methods were compared for the karst spring discharge and nitrate transport at Silver Springshed. Generally it is interesting. However, due to the setting of the nitrate input, the simulated results regarding nitrate concentration is not that promising. As authors discussed maybe a finer resolution is needed and I also strongly suggest to consider a seasonal cycle of nitrate input if possible in this manuscript or in a future study. I have the following comments:
Major comments:
CFPv2 is superior to EPM for discharge simulations. However, the nitrate concentration simulation shows much smaller difference between the two approach. Will the constant nitrate concentration in groundwater recharge lead to a systematic problem? Because normally the concentration is high after fertilizer application. Is the seasonal pattern tested?
Thank you for the comments. For agriculture land, input nitrate concentration could be high after fertilizer application. Different land use has different input nitrate concentration, therefore, input nitrate concentration was determined based on land use and population density in this study. Nitrate concentration in groundwater recharge could be high, so the the nitrate concentration in groundwater recharge was set to 4 mg/L for agricultural land and 0.12 mg/L for forested land (Phelps, 2004). For residential and transportation lands, the population density was used as the factor for estimating the spatial and temporal change of nitrate concentration in groundwater recharge. As population increases, the nitrate concentration from septic tanks and lawn fertilizer may also increase within a given residential area. Further, more impervious areas (i.e., transportation corridors) may be developed as a result of population growth, and more stormwater runoff from these areas are generated. These data were used to calculate nitrate concentration in groundwater recharge in this study. Since the detailed description has been illustrated in Gao et al., 2020, there is not too much descriptions put in the manuscript.
The seasonal pattern could affect the effects of conduit flow as well as the modeling results. For example, regional precipitation could affect spring discharge to some extent. Indeed the authors eager to analyze how precipitation influence groundwater flow in a karst aquifer and spring discharge in monthly scale. However, the monthly precipitation data for the study area could not be obtained. Therefore, the effect of precipitation on spring discharge was not analyzed in this paper. But this point has been considered as one of future research in lines 686-688 of the revised manuscript.
Lines 686-688:
“Although the general trend of nitrate in spring discharge is well simulated by the model in this study, the variations in spring discharge and nitrate (e.g., dilution) response to storm events in the model should be considered.”
Minor comments:
Point 1: Line 21-24: providing some numbers for the slight difference and also for the big improvement by using CFPv2 will be quantitatively clearer.
Response 1: Thank you for the comments. The statistics indicating the slight difference and also for the big improvement by using CFPv2 have been added in lines 20-22 and 25-26 of the revised manuscript:
Lines 19-28:
“Moreover, simulated hydraulic heads in poorly karstified areas from the two models both show slight differences with the observations (the head RMSE values of calibration/validation for CFPv2 and MODFLOW models are 0.16 m/0.25 m and 0.26 m/0.17 m, respectively), indicating the inclusion of conduits may not affect the simulation considerably, and the lower the proportion of karstic area, the much slight affects brought from the inclusion of conduits in the model. For highly karstified areas, the CFPv2+CMT3D model could provide more accurate results (head RMSE of calibration/validation for CFPv2 and MODFLOW are 0.22 m/0.06 and 0.52 m/0.47 m, respectively), showing the coupled continuum pipe-flow framework may be more appropriate for applying to highly and maturely karstified areas where the variations in behavior of flow and contaminant transport be more affected by turbulent flow regime.”
Point 2: Line 61: what is the 12-40% improvement? Monthly water balance, or performance metrics?
Response 2: Thank you for the comments. The 12-40% improvement mentioned in the manuscript is model performance. The statement has been modified to be much clearer in line 62 of the revised manuscript:
Line 59-63:
“MODFLOW-Conduit Flow Process (CFP) was developed by USGS to simulate turbulent flow in a discrete network of pipes coupled with the traditional groundwater flow equation [14] and model performance has been evaluated by [15] for a spring in Florida, in which the model performance (match between observed and simulated discharge) improved by 12-40% compared to the results from MODFLOW.”
Point 3: Line 72: check the in-text citation style.
Response 3: Thank you for the comments. The citation style has been updated in lines 77 of the revised manuscript:
Lines 73-77:
“The Conduit Modular 3-Dimensional Transport (CMT3D) model is developed based on the MT3DMS source code, which adds the ability to simulate solute transport in conduits to MT3DMS so that it can be integrated with CFP or CFPv2 flow model to simulate flow and solute transport in matrix and conduits of karst aquifers [20,21].”
Point 4: Line 171: why leave out discharge in the period of 2017-2019 Feb?
Response 4: Thank you for the comment. The data for nitrate concentration are difficult to be obtained and are not available to some extent. In this study the data for nitrate concentration from Dec 2014 to Dec 2016 were available and they were obtained from USGS website or St. Johns River Management District website. The data from 2017 to2019 are available. The authors searched website and literatures, but the data in that period could not be found out. The data from Mar 2019-Mar 2020 were obtained from field experiment. The sample which was conducted by the corresponding author. To make the modeling step be reasonable, the authors used data from Dec 2014 to Dec 2015, and Jan 2016 to Dec 2016 to calibrate and validate the transport model, because these periods are within the periods for calibration and validation of the hydraulic model.
The data from Mar 2019-Mar 2020 were obtained from field experiment. The sample which was conducted by the corresponding author. Because the transport model has been calibrated and validated, these data were employed to investigate the change in nitrate concentration in spring discharge. The description has been updated in lines 179-181 of the revised manuscript:
“The nitrate concentration data from Mar 2019 to Mar 2020 were obtained from field experiment, which were employed to investigate the change in nitrate concentration in spring discharge.”
Point 5: Line 191: why using 190.5m not an integer such as 200m?
Response 5: Thank you for the comment. The model was developed from an EPM model (NSFEG model as mentioned in the manuscript). In the NFSEG model, the unit is US unit. The authors refined the grid and improved the model with the grid space as 190.5m.
Point 6: Fig 5a: maybe switch to m/s since the description in the text is m/s.
Response 6: Thank you for the comment. The unit in Figure 5a has been swiched to m/s and the figure has been updated in the revised manuscript:
Figure 5. (a) Hydraulic conductivity values set in layer 3 of the EPM model [31]; (b) The delineated conduits in the UFA and the distribution of sinkhole locations in the springshed.
Point 7: Fig 5b: are these the major conduits that were used in the model. Are there any smaller conduits? If Any, will they affect the results?
Response 7: Thank you for the comment. The conduits shown in Figure 5b are the conduits used in the model.
Conduit structure and location are difficult to be determined, so it is always a challenge for karst groundwater modeling. In this study the authors determined the conduit locations and structure based on sinkhole locations and tracer test documented in reports (FDEP 2015). The results for conduit structure and locations are reasonable since this method is also used by previous studies (Xu et al., 2015). There should be smaller conduits. However, those smaller conduit are not well developed and connected, so those kind of conduits may not pose big influence on results.
Point 8: Line 274: The effect of denitrification should be justified. The variation of water table and oxygen concentration may largely influence the nitrogen concentration.
Response 8: Thank you for the comment. Chemical reaction was not considered in this study, and the reason has been specifically stated in lines 352-360 of the revised manuscript:
“Nitrate reduction is negligible in the UFA and other aquifers due to low ammonium and dissolved organic carbon (DOC) concentrations; and significant nitrate reduction is unexpected when there is low carbon availability [38,39,40,41,42,43]. Also, flow velocity is high in conduit especially around the Silver Springs where karst features are maturely developed, so the residence time for the solute could be short. Moreover, conduit transport dispersion and chemical reaction were not considered in the simulation of contaminant transport in the Woodvill Karst Plain which is also located underneath the UFA, due to the limited residence time [22]. Therefore, nitrate is set as the non-reactive solute in this study, and conduit transport dispersion and chemical reaction are not considered.”
Point 9: Line 377: the number of point source proportional to population density should be given.
Response 9: Thank you for the comment. The relationship between point source and population density has been added in lines 377-382 of the revised manuscript:
“In this study, the nitrate concentration in groundwater recharge (, [M/L3]) were assumed to be linearly proportional to population density ([pop./L2]):
C(input)=k*Pd-B (6)
The slope k [M/L3] and the background concentration B [M/L3] were estimated during calibration of the transport model.”
Point 10: Line 381: how to calibrate the steady-state model since the real system is barely under steady-state such that what are the “observations”?
Response 10: Thank you for the comment. Hydraulic conductivity and conduit parameters were modified during calibration of the steady-state model. These parameters do not change with time even though the real system is not steady state. This is the reason why these parameters can be calibrated by steady state model. The observations used for calibration of the steady state model are mean hydraulic heads and spring discharge.
Point 11: Line 380: why calibration method was used? What is the objective function?
Response 11: Thank you for the comment. Trial-and-error method was used for calibration in this study. The statement has been added in line 394 of the revised manuscript. The objective of the calibration is to make the simulations and observations be close and the RMSE values between simulation and observation be small enough.
Line 394: “Trial-and-error method was used for calibration in this study.”
Point 12: Line 471: is that monthly average?
Response 12: Thank you for the comment. Yes that is monthly average.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The work presented seems extremely interesting, both in the theoretical-methodological and in the applied point of vue.
The work is more focused on the physical aspects of the models (criteria, applicability, validation) than on understanding and explaining the spatial dynamics (both vertically and horizontally) of the aquifers. For this interpretation, it would be very useful to take into account the precipitation values in the aquifer recharge processes.
Although this analysis of the physical-mathematical aspects of the models is not my specialty, the application to this concrete and difficult-to-analyze territory, due to the superposition of karst and non-karst aquifers, seems to work well.
That is why I am of the opinion that the work should be accepted.
Comments for author File: Comments.pdf
Author Response
Response to Reviewer 2 Comments
General comments:
The work presented seems extremely interesting, both in the theoretical-methodological and in the applied point of vue.
The work is more focused on the physical aspects of the models (criteria, applicability, validation) than on understanding and explaining the spatial dynamics (both vertically and horizontally) of the aquifers. For this interpretation, it would be very useful to take into account the precipitation values in the aquifer recharge processes.
Although this analysis of the physical-mathematical aspects of the models is not my specialty, the application to this concrete and difficult-to-analyze territory, due to the superposition of karst and non-karst aquifers, seems to work well.
That is why I am of the opinion that the work should be accepted.
Dear reviewer 2:
Thank you for the comments which could greatly improve the quality of the manuscript. We have revised our manuscript according to your suggestions and comments. Our responses and corresponding revisions are listed below.
Specific comments:
Point 1: Lines 141-143: In this first presentation of the aquifer, it would be important to:
- characterize the types of limestone that form the aquifer;
- give an idea of the dimensions and modes of hydrological functioning of the karst cavities;
- give an idea of the type of land use practiced on the surface.
Response 1: Thank you for the comments.
- The type of limestone that forms the aquifer is Ocala Limestone which is described in line 119 of the manuscript.
- The idea of dimensions and modes of hydrological functioning of the karst cavities have been updated in lines 145-152 of the revised manuscript:
Lines 145-152:
“The subsurface drainage system is continuing to evolve today, as evidenced in part by frequent occurrence of sinkholes [27]. There are numerous closed sinkhole depressions with permeable bottoms in the springshed (Figure 1), which are varied in dimensions and most prevalent in the mature karst terrane west of the Ocklawaha River, facilitating the recharge activity and correspondingly nitrate transport to subsurface. The holes result from collapse and flow of the sandy and clayey cover into limestone solution cavities. If a sinkhole suddenly develops near a point of groundwater discharge, such as a well or spring, temporary turbidity in the discharging water may result [27].”
- The idea of the type of land use practiced on the surface has been added in lines 138-141 of the revised manuscript:
Lines 138-141:
“The land use and land cover (LULC) in the springshed is comprised of agriculture, residential, forest, barren land (i.e., sparse vegetation and fallow ground), rangeland, wetlands and streams, and lakes (data obtained from https://data-floridaswater.opendata.arcgis.com/).”
Point 2: Lines 151-152: In this case, these points should be on the map in fig.1, which could be slightly enlarged...
Response 2: Thank you for the comment. The onsite sewage tanks are widely used by residential in Marion County, so the density of the tanks is big. These points almost constitute an area within Marion County. The locations of the sewage tanks have been re-stated in the revised manuscript in lines 159-162 of the revised manuscript:
Lines 159-162:
“(e.g., wastewater leakage from onsite sewage tanks and municipal waste treatment systems as well as the nitrate-contained storm runoff with accumulated nutrient loading from impervious roads, which are mostly distributed in Marion County, especially in the western part of the Silver Springs) (Figure 1)”
Point 3: Lines 180-184: It would be good to point out some bibliography to present these models...
Response 3: Thank you for the comment. The references have been cited in this section in lines 190 and 192 of the revised manuscript. Also, in the following sections, more details and references are described and cited to make the modeling steps be clear.
Lines 190-192:
“The flow model was developed using CFPv2 model based on an equivalent porous media (EPM) model [16]. Simulation of nitrate concentration in Silver Springs was accomplished by integrating the CFPv2 model with the CMT3D model [20].”
[16] Reimann, T.; Liedl, R.; Birk, S.; Bauer, S. MODFLOW-CFPV2. 2013a, Dresden Technical University, Dresden, GE.
[20] Spiessl, S.M. Development and Evaluation of a Reactive Hybrid Transport Model (RUMT3D). Ph.D Dissertation. Hydrogeologic, Georg-August-University Gottingen 2004.
Point 4: Lines 191: It seems ok to me. However, isn't the dimension of this grid too big considering that it is a karst aquifer?
Response 4: Thank you for the comment. The CFPv2 model employed in this study was developed from an EPM model. The degree of karstification is not much intense in the study area. In the model although 190.5m by 190.5 m grid was used, the karstification areas were represented by CFP conduit nodes and pipes. Therefore, conduit flow could be simulated accurately.
Point 5: Line 240: “In” the EPM model.
Response 5: Thank you for the comment. The word “In” has been revised to “in” in line 249 of the revised manuscript:
Line 249: “According to [31], high hydraulic conductivity cells were used for representing karst features in the UFA in the EPM model (Figure 5(a)).”
Point 6: Figure 8: What is the relationship with the values of regional precipitation. Can these values be included in the graph?
Response 6: Thank you for the comment. Regional precipitation could affect spring discharge to some extent. Indeed the authors eager to analyze how precipitation influence groundwater flow in a karst aquifer and spring discharge in monthly scale. However, the monthly precipitation data for the study area could not be obtained. Therefore, the effect of precipitation on spring discharge was not analyzed in this paper. But this point has been considered as one of future research in lines 686-688 of the revised manuscript. Also, the corresponding author analyzed the effect of mean annual precipitation on spring discharge in the same study area, which included in Gao et al. (2020).
Lines 686-688:
“Although the general trend of nitrate in spring discharge is well simulated by the model in this study, the variations in spring discharge and nitrate (e.g., dilution) response to storm events in the model should be considered.”
Gao, Y.; Libera, D.; Kibler, K.; Wang, D.; Chang, N.B. Evaluating the performance of BAM-based blanket filter on nitrate reduction in a karst spring. Journal of Hydrology 2020, 591, 125491. Doi: 10.1016/j.jhydrol.2020.125491.
Point 7: Figure 9: Here, the relationship of the discharge values and the concentration of nitrates with the values of precipitation seems equally important...
Response 7: Thank you for the comment. In this study, the monthly precipitation data for the study area could not be obtained, so the effects of precipitation on spring discharge and nitrate concentration were not analyzed in this paper. But this point has been considered as one of future research in lines 686-688 of the revised manuscript. Also, the corresponding author analyzed the effect of mean annual precipitation on spring discharge in the same study area, which included in Gao et al. (2020).
Lines 686-688:
“Although the general trend of nitrate in spring discharge is well simulated by the model in this study, the variations in spring discharge and nitrate (e.g., dilution) response to storm events in the model should be considered.”
Gao, Y.; Libera, D.; Kibler, K.; Wang, D.; Chang, N.B. Evaluating the performance of BAM-based blanket filter on nitrate reduction in a karst spring. Journal of Hydrology 2020, 591, 125491. Doi: 10.1016/j.jhydrol.2020.125491.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Thanks for the responses. For a future study, I would suggest to use an automatic calibration scheme, if possible, to minimize the subjective tunning of model parameters. If some seasonal scenarios can be made, I think you may also find something interesting.
