Remote Sensing-Assisted Physical Modelling of Complex Spatio-Temporal Nitrate Leaching Patterns from Silvopastoral Systems^†

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This article was submitted to the journal Remote Sensing, so it's expected that the key issues presented here will be the use of RS techniques in research, analysis of their accuracy, conclusions, and methodological recommendations. However, the application of remote sensing is a secondary issue here, based on general assumptions and simplifications, and lacking field verification. This could be considered a serious flaw in the article, undermining the validity of its publication in Remote Sensing.

The topic of remote sensing in the article comes down to the use of RGB images obtained from drones to determine the LAI index, which is needed in the Daisy model used in the research as input data for the prediction of interception and evapotranspiration (by determining the kcrop coefficient).

The relatively homogeneous structure of the studied vegetation (poplar monoculture and pasture) facilitates effective LAI prediction based on drone images. However, this prediction was made using a very simple methodology, and the article's authors do not verify this methodology in any way, assuming in advance that the methods selected based on the literature will be effective in a given case and provide accurate results.

Suppose the authors conducted ground-based studies during the drone flights (LAI measurement with dedicated instruments, subcanopy spherical images, subcanopy solar radiation intensity measurement). In that case, the results that can be used to calibrate the remotely sensed LAI measurement and assess the error of such measurements should be included in the analysis (by developing an appropriate methodology).

After appropriate corrections, the article can be resubmitted.

At the same time, it should be noted that the high value of detailed studies of nitrate concentration in soil and their analysis, supported by clear drawings, means that the article can be successfully redirected to journals dealing with the impact of agriculture on groundwater quality (e.g., Water, Sustainability). This is especially true as the conclusions from the study may contribute to modifying the management of free-range pig farming in order to reduce nitrogen leaching into groundwater.

Detailed comments:

Line 141

It is a pity that a camera capable of capturing images in the near-infrared (NIR) channel wasn't used. This would have significantly facilitated vegetation coverage analyses, for example, by enabling the calculation of remote sensing indices such as NDVI.

Line 152

 Can this visual assessment be classified as remote sensing, as the subsection title suggests, or was it conducted by another method? In other words, how was the low vegetation mapped?

Line 153

 Estimation of the area fraction of soil rooted either deep (> 10 cm) or shallow (< 10 cm) requires conducting systematic core sampling, followed by determining the sample root presence across the depth of the sample. Basing this assessment solely on visual inspection in the field is challenging to consider a scientific method that yields reliable results. Please clarify whether soil cores were collected and analyzed during the study.

Line 156

The research's imperfection is that the effects of poplar crown reduction were not measured, replacing them with an arbitrary simplification.

Line 157

For individual pixels representing canopy cover (CC), was the value determined as a binary or a percentage (quantitative)? Analysis of Figure 3 suggests that this was a binary assessment, which should be considered an oversimplification of the methodology for determining LAI values.

Line 158

 The calculation methodology recommended for preparing data for the Daisy Model, which calculates CC based on LAI according to the formula AC = 1 − exp(KI Lai ), where Ac = CC, was reversed here. It should also be remembered that k exhibits seasonal variability, which was not accounted for. Calculating LAI based on binary CC likely results in a significant reduction in the spatial resolution of LAI, which would likely be apparent if LAI maps were presented.

Line 227

It is recommended that kriging be compared with other popular interpolation methods, pointing out its actual advantages. The choice of the analysis radius is crucial for its accuracy, so it is worth presenting the analysis in variants with different radii. Selecting the optimal radius is often an iterative process. A 1-meter analysis radius with a 7-meter distance between measurement points seems significantly too small. It is recommended to choose a radius of at least 10 meters; this recommendation should be verified by analyzing variograms.

Lines 230-234

It is worth presenting the developed functions on graphs and specifying in more detail how they were subsequently used.

Line 509

Figure A1 contains only one aerial photo, which does not seem to be consistent with its description. The location of the trees is inconsistent with the description (tall trees are visible on the left, but according to the description, they should be on the right).

Author Response

Comment 1: This article was submitted to the journal Remote Sensing, so it's expected that the key issues presented here will be the use of RS techniques in research, analysis of their accuracy, conclusions, and methodological recommendations. However, the application of remote sensing is a secondary issue here, based on general assumptions and simplifications, and lacking field verification. This could be considered a serious flaw in the article, undermining the validity of its publication in Remote Sensing.

Response 1: Thanks for the comment at start - we would like to bring an interesting study to the journal readership and beyond of how remote sensing supports physically-based modelling in forward mode (see Fig. R1). Also, see the reviews of Dlamini et al. (2023) and Kasampalis et al. (2018), which point in the direction of increasing characterization of crop parameters by remote sensing assimilated into process-based models. The retrieval does not need to be too complex in order to be scientifically valid. Forward support can potentially provide physically based models with continuous, near-real-time data assimilation which can significantly improve forecast accuracy by dynamically updating the model's initial conditions and constraining its simulations with observed reality. Leaf area index (LAI), i.e., canopy cover (CC) of the poplar trees have been estimated with a pragmatic near- machine learning method known as image thresholding. We elaborate more on this. Also, it is true that the values were not directly validated, and as the reviewer also identifies later, the poplar canopies were homogeneous, so LAI development over time is more important and somewhat easier. The retrieved values were well within the variability for poplar reported by other studies, and this is also emphasized now in the paper. We updated the introduction with these points to strengthen the added value of remote sensing and the purpose of our study.

Figure R1. Basic structure of physically-based models and remote sensing support

 

Comment 2: The topic of remote sensing in the article comes down to the use of RGB images obtained from drones to determine the LAI index, which is needed in the Daisy model used in the research as input data for the prediction of interception and evapotranspiration (by determining the kcrop coefficient).

Response 2: Yes, we were particularly interested in the attempt to see the potential of the RGB imagery obtained from UAV, processed with ML-like method such as GIMP-GNU which uses near-fuzzy classification of each pixel based on different intensities (first luminance or brightness for the greyscale image, then weighted average of the R-G-B bands. This is why it is not a “true” ML method as the weighted average of the color channels is fixed, predefined (e.g., 0.3*R + 0.59*G + 0.11*B), but it is still fuzzy. We find this very interesting as the method is quite relevant for homogeneous canopies. See for instance Parker (2020) on the importance of and the difficulties measuring LAI. We included also in the discussion point on this to stimulate the use of more simple image analysis tools as the physically-based model is already very complex.

Also, about LAI, it is a control variable in the Daisy model (at least in the permanent vegetation module) needed for estimation of evapotranspiration, which is a state variable. The crop (and the soil) coefficients are fixed parameters in the Daisy model.  We added a short sentence in the method section to make this clearer.

 

Comment 3: The relatively homogeneous structure of the studied vegetation (poplar monoculture and pasture) facilitates effective LAI prediction based on drone images. However, this prediction was made using a very simple methodology, and the article's authors do not verify this methodology in any way, assuming in advance that the methods selected based on the literature will be effective in a given case and provide accurate results.

Response 3: We find an advantage of our study to showcase a simple approach using near-ML image analysis method to estimate CC and LAI. This is not new, as we point in the introduction, but very few studies involved belowground processes. If the canopy is nonhomogeneous, e.g., mixed forest, it is nearly impossible to estimate LAI on species level (e.g., Marron et al., 2018), but a total leaf area or composite leaf area would be more of interest. It is true we do not have field-measured LAI to compare the estimates with (not the purpose of the study, also due to budget), but there are previous data published for comparison. Precise match of the retrieved with observed LAI is not of use as the measurements of LAI are associated with high errors, and also the models are not that sensitive to subtle LA variations, even less so for seasonal or annual water balances. However, this is interesting and relevant and now included in the revised version of the paper.

 

Comment 4: Suppose the authors conducted ground-based studies during the drone flights (LAI measurement with dedicated instruments, subcanopy spherical images, subcanopy solar radiation intensity measurement). In that case, the results that can be used to calibrate the remotely sensed LAI measurement and assess the error of such measurements should be included in the analysis (by developing an appropriate methodology).

Response 4: Yes definitely, but verification of LAI was not objective of our work, which was instead focused on the modeling part, whereas remote sensing was seen as support for the Daisy model. We included in the discussion a suggestion to integrate LAI verification in future studies. However, bear in mind that the Daisy model water balance simulation is not that sensitive to subtle LAI variations, because it is the water availability and root water uptake which plays larger role.

 

Comment 5: After appropriate corrections, the article can be resubmitted.

Response 5: We addressed all comments and implemented appropriate corrections. We believe your comments have significantly improved the manuscript and its understanding.

 

Comment 6: At the same time, it should be noted that the high value of detailed studies of nitrate concentration in soil and their analysis, supported by clear drawings, means that the article can be successfully redirected to journals dealing with the impact of agriculture on groundwater quality (e.g., Water, Sustainability). This is especially true as the conclusions from the study may contribute to modifying the management of free-range pig farming in order to reduce nitrogen leaching into groundwater.

Response 6: We would like to keep our original intention targeting Remote Sensing and the special issue because our aim is to trigger even more articles on remote-sensing assisted physical modeling of belowground phenomena (also GHGs emissions). If this is possible with a relatively simple coupling based on vision analysis as in this paper, it will be of highly added value and frameworks can be improved and upgraded as more knowledge builds up. Otherwise, the paper can certainly be redirected, e.g., in SI “Advancing UAV-Based Remote Sensing: Innovations, Techniques and Applications” or SI “Remote Sensing Band Ratios for the Assessment of Water Quality”, or just a regular article in the journal. Let us know.

 

Detailed comments:

Comment 7: Line 141

It is a pity that a camera capable of capturing images in the near-infrared (NIR) channel wasn't used. This would have significantly facilitated vegetation coverage analyses, for example, by enabling the calculation of remote sensing indices such as NDVI.

Response 7: The camera DJI Mini 3 Pro does not capture images in the NIR, else we would have used it. We also added technical description of the camera as another reviewer asked similarly.

 

Comment 8: Line 152

Can this visual assessment be classified as remote sensing, as the subsection title suggests, or was it conducted by another method? In other words, how was the low vegetation mapped?

Response 8: The visual assessment of herbaceous coverage was not classified by remote sensing in the manuscript. Visual assessment is often used for low vegetation of herbaceous character otherwise impossible to measure. Visual assessment was conducted by experienced technical staff using a standardized scale. We added this in the manuscript. See also Manevski et al. (2019).

 

Comment 9: Line 153

Estimation of the area fraction of soil rooted either deep (> 10 cm) or shallow (< 10 cm) requires conducting systematic core sampling, followed by determining the sample root presence across the depth of the sample. Basing this assessment solely on visual inspection in the field is challenging to consider a scientific method that yields reliable results. Please clarify whether soil cores were collected and analyzed during the study.

Response 9: In line with the other reviewer comment, we deleted this sentence and refer only to the visual assessment of grass canopy cover in order to avoid confusion as these other assessments were conducted for other purpose. However, the technical staff has long experience in visual assessment of low-canopy herbaceous vegetation. We have the data provided on Fug. R2 as an example, showing the herbaceous vegetation cover across the paddocks and zones, where we can clearly see low values in the tree zone in the middle, and also low values over time as the grass-clover canopy diminishes.

Figure R2. Herbaceous vegetation cover in the study.

 

Comment 10: Line 156

The research's imperfection is that the effects of poplar crown reduction were not measured, replacing them with an arbitrary simplification.

Response 10: It is true that LAI was not measured and we extensively compared the image-derived values with previous studies under comparable pedo-climatic conditions. For short rotation coppice, LAI in the first year is typically between 0.5 and 2.0 (ref), whereas in well-established coppice it is typically 3-8, depends a bit on genotype too, i.e., LAImax increase has been reported from <2 in early seasons to 4–7.6 in later seasons, depending on the genotype (Broeckx et al., 2015). Upper extremes reported in the literature (including intensive trials) reach 9 m²·m⁻² in very productive coppice stands (Fuertes et al., 2022; Tripathi et al., 2018), and we can deem our trials as such due to the favourable genotype × site × management conditions.

 

Comment 11: Line 157

For individual pixels representing canopy cover (CC), was the value determined as a binary or a percentage (quantitative)? Analysis of Figure 3 suggests that this was a binary assessment, which should be considered an oversimplification of the methodology for determining LAI values.

Response 11: Vision analysis based on histograms, i.e., image thresholding was the main method, we have explained this in the revised manuscript. The main result of a CC is always a binary image, despite the fuzzy classification of each pixel based on different intensities and the RGB thresholds. See also reply to comments 1-2.

 

Comment 12: Line 158

The calculation methodology recommended for preparing data for the Daisy Model, which calculates CC based on LAI according to the formula AC = 1 − exp(KI Lai ), where Ac = CC, was reversed here. It should also be remembered that k exhibits seasonal variability, which was not accounted for. Calculating LAI based on binary CC likely results in a significant reduction in the spatial resolution of LAI, which would likely be apparent if LAI maps were presented.

Response 12: We estimated CC, i.e., LAI for each pixel of the UAV imagery and LAI maps were indeed generated; however, they depicted nearly-single-value, i.e., notable variation was detected only between the two main treatments pruned (P) or tall (T) as also shown in Figure 4 in the manuscript, and we thus decided to depict only the zones per treatment and run the Daisy model at this level. I included a short clarification on this.

 

Comment 13: Line 227

It is recommended that kriging be compared with other popular interpolation methods, pointing out its actual advantages. The choice of the analysis radius is crucial for its accuracy, so it is worth presenting the analysis in variants with different radii. Selecting the optimal radius is often an iterative process. A 1-meter analysis radius with a 7-meter distance between measurement points seems significantly too small. It is recommended to choose a radius of at least 10 meters; this recommendation should be verified by analyzing variograms.

Response 13: Thanks for this comment. It was out of the objective of the study to analyze the effect of different interpolation methods or different radii- actually in ordinary kriging with the krige function in the gstat package of R, there is  no direct "radius" parameter but the function uses the variogram (sill, range, nugget) to define the spatial relationships between data points, i.e., as you say the "radius" of influence. On your suggestion that a 1-meter analysis radius with a 7-meter distance between measurement points, actually, it very much depends on the sampling schemes and the variable sampled- for soil nitrate. Not many studies reporting soil nitrate, but the few report semivariogram range values in the 5–50 range (e.g., Piotrowska, 2011; Van Meirvenne et al., 2003), but values of 150-200 m are also seen for larger or more homogeneous areas (e.g., clayey soils which tend to retain water and nutrients) or coarser sampling schemes (e.g., Ghidey & Alberts, 1999). We included this in the revised manuscript without emphasizing too much as this is not in focus.

 

Comment 14: Lines 230-234

It is worth presenting the developed functions on graphs and specifying in more detail how they were subsequently used.

Response 14: We present the models in a new figure. This is a well developed method for estimating nitrate leaching (Børgesen et al., 2001; Manevski et al., 2015) originating from Lord and Shepherd (1993).

 

Comment 15: Line 509

Figure A1 contains only one aerial photo, which does not seem to be consistent with its description. The location of the trees is inconsistent with the description (tall trees are visible on the left, but according to the description, they should be on the right).

Response 15: This is correct, the figure was aimed to show the poplars pruning treatment. We have replaced it with another image and might find even better one.

 

References:

Børgesen, C. D., Djurhuus, J., & Kyllingsbaek, A. (2001). Estimating the effect of legislation on nitrogen leaching by upscaling field simulations. Ecological Modelling, 136(1), 31–48.

Broeckx, L., Vanbeveren, S., Verlinden, M., & Ceulemans, R. (2015). First vs. second rotation of a poplar short rotation coppice: leaf area development, light interception and radiation use efficiency [First vs. second rotation of a poplar short rotation coppice: leaf area development, light interception and radiation use efficiency] [Research Articles]. iForest - Biogeosciences and Forestry, 8(5), 565–573. https://doi.org/10.3832/ifor1457-008

Dlamini, L., Crespo, O., van Dam, J., & Kooistra, L. (2023). A Global Systematic Review of Improving Crop Model Estimations by Assimilating Remote Sensing Data: Implications for Small-Scale Agricultural Systems. Remote Sensing, 15(16), 4066.

Fuertes, A., Sixto, H., González, I., Pérez-Cruzado, C., Cañellas, I., Rodríguez-Soalleiro, R., & Oliveira, N. (2022). Time-course foliar dynamics of poplar short rotation plantations under Mediterranean conditions. Responses to different water scenarios. Biomass and Bioenergy, 159, 106391. https://doi.org/https://doi.org/10.1016/j.biombioe.2022.106391

Ghidey, F., & Alberts, E. E. (1999). Temporal and Spatial Patterns of Nitrate in a Claypan Soil. Journal of Environmental Quality, 28(2), 584–594. https://doi.org/https://doi.org/10.2134/jeq1999.00472425002800020024x

Kasampalis, D. A., Alexandridis, T. K., Deva, C., Challinor, A., Moshou, D., & Zalidis, G. (2018). Contribution of Remote Sensing on Crop Models: A Review. Journal of Imaging, 4(4), 52.

Lord, E. I., & Shepherd, M. A. (1993). Developments in the use of porous ceramic cups for measuring nitrate leaching. Journal of Soil Science, 44(3), 435–449. https://doi.org/10.1111/j.1365-2389.1993.tb00466.x

Manevski, K., Børgesen, C. D., Andersen, M. N., & Kristensen, I. S. (2015). Reduced nitrogen leaching by intercropping maize with red fescue on sandy soils in North Europe: a combined field and modeling study. Plant and Soil, 388(1-2), 67–85. https://doi.org/https://doi.org/10.1007/s11104-014-2311-6

Manevski, K., Jakobsen, M., Kongsted, A. G., Georgiadis, P., Labouriau, R., Hermansen, J. E., & Jørgensen, U. (2019). Effect of poplar trees on nitrogen and water balance in outdoor pig production – A case study in Denmark. Science of the Total Environment, 646, 1448–1458. https://doi.org/https://doi.org/10.1016/j.scitotenv.2018.07.376

Marron, N., Priault, P., Gana, C., Gérant, D., & Epron, D. (2018). Prevalence of interspecific competition in a mixed poplar/black locust plantation under adverse climate conditions. Annals of Forest Science, 75(1), 23. https://doi.org/10.1007/s13595-018-0704-z

Parker, G. G. (2020). Tamm review: Leaf Area Index (LAI) is both a determinant and a consequence of important processes in vegetation canopies. Forest Ecology and Management, 477, 118496. https://doi.org/https://doi.org/10.1016/j.foreco.2020.118496

Piotrowska, A. (2011). Spatial Variability of Total and Mineral Nitrogen Content and Activities of the N-Cycle Enzymes in a Luvisol Topsoil [journal article]. Polish Journal of Environmental Studies, 20(6), 1565–1573.

Tripathi, A. M., Pohanková, E., Fischer, M., Orság, M., Trnka, M., Klem, K., & Marek, M. V. (2018). The Evaluation of Radiation Use Efficiency and Leaf Area Index Development for the Estimation of Biomass Accumulation in Short Rotation Poplar and Annual Field Crops. Forests, 9(4), 168.

Van Meirvenne, M., Maes, K., & Hofman, G. (2003). Three-dimensional variability of soil nitrate-nitrogen in an agricultural field. Biology and Fertility of Soils, 37(3), 147–153.

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

First of all, I would like to congratulate the authors on this comprehensive and well-structured study addressing nitrate leaching dynamics in silvopastoral systems.
The integration of UAV imagery, geostatistical analyses, and a process-based model (Daisy) represents an innovative framework for quantifying spatio-temporal nutrient fluxes in complex agroecosystems.
I found the work highly valuable; however, there are several issues in the manuscript that should be addressed before publication, as outlined below:

 

GENERAL COMMENTS

My main concerns arise from the processing of the UAV imagery used in the study.
First, the paper does not specify the flight characteristics, which would ideally be summarized in a table including essential details for orthorectified image generation, such as flight extent, flight type (longitudinal, grid, etc.), overlap between photographic captures, flight altitude, velocity, duration, or camera angles.
Without this information, it is difficult to fully assess the applied methodology, and I have serious doubts regarding both the geometric and radiometric calibration of the images.

Regarding geometric calibration, the study presents methodological limitations that reduce the reliability of the results. There is no evidence of the use of ground control points (GCPs) or of a properly planned photogrammetric flight with sufficient longitudinal and transversal overlap to allow accurate orthorectification. Based on the information provided, it also seems that no geometric rectification was applied, but rather that images were simply captured by attempting to reproduce approximately the same camera position across dates. If that is the case, the lack of an orthorectified image would introduce the following sources of error:

i) Image distortions: In the absence of camera parameters or optical corrections, the images are likely affected by various distortions, most notably barrel distortion, especially when using a low-cost commercial drone whose sensor and lens are not calibrated for photogrammetric purposes. Therefore, geometric correction using ground control points is essential to mitigate these deformations.

ii) Spatial misalignment: Taking only a single image per flight introduces translations and rotations between dates, since it is practically impossible to capture images from exactly the same position. This results in each capture containing slightly different objects.

iii) Imprecision in exterior orientation: In a multitemporal analysis, precise georeferencing of orthomosaics is critical, as even minimal shifts between dates can cause spatial misalignments that directly affect the comparison of results.

Moreover, the manuscript does not indicate whether any radiometric correction was performed. Without radiometric normalization, differences in solar illumination, cloud cover, time of day, or camera angle may lead to variations in brightness and color that do not reflect real surface changes. Consequently, the HSV transformations applied are likely influenced by a lack of radiometric standardization. Values may vary not due to vegetation changes but due to differences in exposure, white balance, or sensor temperature.
Additionally, in uncalibrated RGB cameras, the nonlinear response of each channel introduces radiometric biases that prevent reliable comparisons across areas or acquisition dates.

In this context, were any in situ canopy cover measurements performed to correlate them with UAV-derived estimates? Such validation would help confirm the correspondence between field and remote-sensing data. Otherwise, was any comparison conducted between the measured nitrate concentrations and those simulated by the Daisy model? This would help assess the reliability of the LAI used as model input.

 

DETAILED COMMENTS

Materials and Methods
Line 153: I would suggest removing this line.
Line 160: Were only three of the nine flights used?

Figure 3: What explains the increase in green area in the center of the study site on the date 2023-10-10?

Author Response

Comment 1: First of all, I would like to congratulate the authors on this comprehensive and well-structured study addressing nitrate leaching dynamics in silvopastoral systems.
The integration of UAV imagery, geostatistical analyses, and a process-based model (Daisy) represents an innovative framework for quantifying spatio-temporal nutrient fluxes in complex agroecosystems.
I found the work highly valuable; however, there are several issues in the manuscript that should be addressed before publication, as outlined below:

Response 1: We would like to thank the reviewer for the positive and important comments, we tried to accommodate all.

 

GENERAL COMMENTS

Comment 2: My main concerns arise from the processing of the UAV imagery used in the study.
First, the paper does not specify the flight characteristics, which would ideally be summarized in a table including essential details for orthorectified image generation, such as flight extent, flight type (longitudinal, grid, etc.), overlap between photographic captures, flight altitude, velocity, duration, or camera angles.
Without this information, it is difficult to fully assess the applied methodology, and I have serious doubts regarding both the geometric and radiometric calibration of the images.

Response 2: We have included the flight characteristics in textural form as it is not necessary to add a table. We also refer to the technical specifications of the camera elaborated in the user manual available online, which is also referenced (SZ DJI Technology Co., 2023). Also, thanks for pointing that the DJI Mini 3 Pro RGB camera does not provide in-depth geometric and radiometric calibration but there are factory-calibrated and auto-calibrated parameters, in addition to several user-specifications to ensure reliable image acquisition details. We have included this in the paper to also made readers aware about the image acquisition system. However, we believe this has no implications on the results since the canopy cover (CC) is treated independently (per day) and also the canopy was very homogeneous.

 

Comment 3: Regarding geometric calibration, the study presents methodological limitations that reduce the reliability of the results. There is no evidence of the use of ground control points (GCPs) or of a properly planned photogrammetric flight with sufficient longitudinal and transversal overlap to allow accurate orthorectification. Based on the information provided, it also seems that no geometric rectification was applied, but rather that images were simply captured by attempting to reproduce approximately the same camera position across dates. If that is the case, the lack of an orthorectified image would introduce the following sources of error:

i) Image distortions: In the absence of camera parameters or optical corrections, the images are likely affected by various distortions, most notably barrel distortion, especially when using a low-cost commercial drone whose sensor and lens are not calibrated for photogrammetric purposes. Therefore, geometric correction using ground control points is essential to mitigate these deformations.

Response 3: DJI Mini 3 Pro has vision system/depth-vision calibration, but only for the navigation/ obstacle-avoidance cameras, not to the main 1/1.3″ camera used for imaging as in our study. However, the system has gimbal calibration (geometric stability, physical mount for the camera) which was calibrated via the DJI Fly app (settings for “Gimbal Calibration”). This calibration uses internal scripts or software to align the camera gimbal properly and ensures that the camera remains level and stable, which indirectly improves geometric consistency (i.e., images do not suffer from horizon tilt or axis misalignment).

 

Comment 4: ii) Spatial misalignment: Taking only a single image per flight introduces translations and rotations between dates, since it is practically impossible to capture images from exactly the same position. This results in each capture containing slightly different objects.

Response 4: The UAV was flown at low altitude, 43 m, which greatly reduces the impact of Spatial misalignment (and camera exterior-orientation imprecision) on the final model.

 

Comment 5: iii) Imprecision in exterior orientation: In a multitemporal analysis, precise georeferencing of orthomosaics is critical, as even minimal shifts between dates can cause spatial misalignments that directly affect the comparison of results.

Response 5: Please see the reply to previous comment 19. Also, we did not compare between dates but estimated CC for each specific day, i.e., CC data were independent.

 

Comment 6: Moreover, the manuscript does not indicate whether any radiometric correction was performed. Without radiometric normalization, differences in solar illumination, cloud cover, time of day, or camera angle may lead to variations in brightness and color that do not reflect real surface changes. Consequently, the HSV transformations applied are likely influenced by a lack of radiometric standardization. Values may vary not due to vegetation changes but due to differences in exposure, white balance, or sensor temperature.

Response 6: The Mini 3 Pro offers manual control over ISO and shutter speed, which allows an operator to fix exposure settings. It also provides a histogram, “peaking level”, and an over-exposure warning in the DJI Fly app, which helps monitor exposure. However, you are current, it does not provide in-depth radiometric calibration as for multispectral cameras, such as flat-fielding, vignetting correction, spectral response calibration. We therefore relied on HSV analysis. This is included now in the manuscript, so readers are aware.

Comment 7: Additionally, in uncalibrated RGB cameras, the nonlinear response of each channel introduces radiometric biases that prevent reliable comparisons across areas or acquisition dates.

Response 7: True, that is why we used the HSV thresholding. H and S channels are relatively insensitive to the radiometry, however V involves image brightness, which affects intensities under different lighting conditions and can potentially the object of interest or capture unnecessary background. We believe the high contract between the canopy and the background and the highly homogeneous canopies introduced insignificant error. However, we included this comment in the paper as readers need to be aware about this uncertainty in future studies.

 

Comment 8: In this context, were any in situ canopy cover measurements performed to correlate them with UAV-derived estimates? Such validation would help confirm the correspondence between field and remote-sensing data. Otherwise, was any comparison conducted between the measured nitrate concentrations and those simulated by the Daisy model? This would help assess the reliability of the LAI used as model input.

Response 8: Please see reply to comment 3. We agree with your comment, however, in situ measurements of poplar LAI were not budgeted in this project, also due to the pruning treatment and the related uncertainty. Therefore we conducted an extensive literature review of studies under similar conditions to ensure the retrieved values are in observed range.

 

DETAILED COMMENTS

Materials and Methods
Comment 9: Line 153: I would suggest removing this line.

Response 9: Done, also see reply to comment 9.

Comment 10: Line 160: Were only three of the nine flights used?

Response 10: All 10 (not nine) flights were used, as written in the text.

 

Comment 11: Figure 3: What explains the increase in green area in the center of the study site on the date 2023-10-10?

Response 11: It is just a patch of not-well harvested canopy that remained greener for prolonged time before the late autumn defoliated everything.

 

References:

SZ DJI Technology Co., L. (2023). DJI Mini 3 Pro User Manual v1.6 2023.03.

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for your explanations. I see great potential for the future in the presented research. I think increasing the accuracy of LAI measurements using UAVs could improve the quality of water balance modeling (especially in studies where root density in soil layers was not measured, so root uptake is based on assumptions). Each model is only as accurate as the input values ​​provided. The more values ​​adopted from the ranges given in the literature, the less knowledge we gain about the actual functioning of a given area. Examining the spatial variability of hydrochemical processes at the field/animal enclosure scale, i.e., at a very detailed scale, is an ambitious goal. To achieve this, we need to provide accurate and precise values ​​of parameters describing the variation in chemical conditions – which was undoubtedly achieved – as well as parameters enabling hydrological modeling, in this case LAI, because soil conditions were rather uniform. The latter condition can certainly be met with the help of UAV-assisted remote sensing. Currently, there are methods available that allow for limiting assumptions and simplifications in favor of almost direct measurements, which, however, require calibration and aim to obtain continuous data (not discrete data).

comments:

Figure 5:
How was the LAI for 2022 determined based on UAV imagery, since according to Figure 2 there were no drone images or other forms of vegetation cover registration this year?

Line 537:
In the previous review, when I wrote "radius," I meant the "range" parameter set when generating the variogram.) What specific range value was used in the analysis shown in Figure 8 (from the given range of 5-50m)? Please don't make readers guess.

 

Author Response

Comment 1: Thank you for your explanations. I see great potential for the future in the presented research. I think increasing the accuracy of LAI measurements using UAVs could improve the quality of water balance modeling (especially in studies where root density in soil layers was not measured, so root uptake is based on assumptions). Each model is only as accurate as the input values ​​provided. The more values ​​adopted from the ranges given in the literature, the less knowledge we gain about the actual functioning of a given area. Examining the spatial variability of hydrochemical processes at the field/animal enclosure scale, i.e., at a very detailed scale, is an ambitious goal. To achieve this, we need to provide accurate and precise values ​​of parameters describing the variation in chemical conditions – which was undoubtedly achieved – as well as parameters enabling hydrological modeling, in this case LAI, because soil conditions were rather uniform. The latter condition can certainly be met with the help of UAV-assisted remote sensing. Currently, there are methods available that allow for limiting assumptions and simplifications in favor of almost direct measurements, which, however, require calibration and aim to obtain continuous data (not discrete data).

Response 1: Thank you very much- we have integrated also some of your reflections in the paper future work. Again, we were more interested in the annual and seasonal water balance, hence absolute accuracy was of secondary (thought not lesser in thought) importance.

Comment 2: Figure 5: How was the LAI for 2022 determined based on UAV imagery, since according to Figure 2 there were no drone images or other forms of vegetation cover registration this year?

Response 2: This explanation is included in the manuscript (see in the last paragraph in section 2.2), however, now also included in the image caption so it is easier for the readers to follow.

Comment 3: Line 537: In the previous review, when I wrote "radius," I meant the "range" parameter set when generating the variogram. What specific range value was used in the analysis shown in Figure 8 (from the given range of 5-50 m)? Please don't make readers guess.

Response 3: Good comment - and yes, I understood you from start, that is why the range of the semivariograms (basically also of the variograms) are included from other studies in the revised version. I did not report the range of our study because the kriging was conducted on polled data (see Figure 1, so from statistical perspective, it is "unfair" to report range for originally high sampling resolution, which was downgraded due to economic reasons. However, now included.

Author Response File: Author Response.pdf