Visible Near-Infrared Spectroscopy and Pedotransfer Function Well Predict Soil Sorption Coefficient of Glyphosate

First reviewer’s comments

Authors’ response

Maybe something should be added in the introduction with regards to remote sensing?

Thank you for this comment. A paragraph was added in the introduction where we introduced remote sensing and satellite spectral data (line no 106 – 118). We also added in conclusion with a perspective related to future use of RS satellite data (line no 578 – 579).

When soil samples have been collected? During one season? I suppose not if some "point" samples have been added. Does it have an effect on Kd analyses?

The samples were not collected during one season and the seasonality has no effect on Kd estimation.

The NFs are quite high (Table 3). How was the number of factors to be used with the PLS calibration decided

As the sorption coefficient is a complex soil property, relying on several other soil properties, it is not unusual to obtain somewhat higher NFs then if it was a model for example clay content. Additionally, we are working with data sets comprising different origins what ads to model complexity and increases the number of factors used.  This is best illustrated by comparing the combined model and Danish models both showing highest NF, and the Greenlandic set which is much more local and with much lower NFs. We have added this information in the manuscript to explain this to the reader as well.  Line no. 400 – 402.

We have added information on how the NF were selected in line no 231 – 232. We also replaced the term “no. of factors (NFs)” with “no. of latent variables (LVs)” in order not to confuse the reader.

Second reviewer’s comments

Authors’ response

I would like to suggest that instead of emphasizing the study in differences because of the two countries, you should base the study on the relation to soil types. This will add more depth and nuance to your research, as soil types can play a critical role in determining the outcomes of certain processes.

Thank you for this comment. We agree with the reviewer. This was also our aim. The two countries represent differences in soils- the geographic location dictates the differences in geologies thus differences in soils. To illustrate the variability and difference in parent material please see the below maps. The parent material of samples collected from Denmark (A) is more variable than that of Greenlandic samples (C). It consists of soils developed on Weichsel glacial deposits, mainly tills, Saale glacial deposits and the younger Weichsel out wash plain, marine, sand dune and saltmarsh deposits. The geology of Greenland presents a much more uniform characteristic with all sites representing soils developed on granodioritic gneiss and granites. As the parent material effects the composition of the entire soil matrix it effects soil spectra as well. We agree that the differences in soil types were not stressed out sufficiently, and we added more information and explanation on this in materials and methodology section (line no 138 – 152) for clarification.

Third reviewer’s comments

Authors’ response

The introduction of background is too simple. Many of the latest work has not been introduced, which is not enough to fully explain the significance of this study. In particular, the application of image processing technology in other fields should also be introduced. Therefore, the authors are suggested to add some literatures. e. g.,

Thank you for this comment. We are not sure which other works the reviewer is referring to. We aimed at introducing to the laboratory spectroscopy, as this is the focus of the paper. however, we did add a short section on RS, and image processing technique in line no 108 – 112 by following the suggested literature.

The introduction of the proposed method is too simple. The authors only describe the method through a large number of words. What is the schematic diagram involved? What are the specific formulas? Why do you choose these methods? These need to be added. It is suggested that the authors improve this part.

The schematic diagram (Figure 2) and specific formulas (equation 1 and 2) are added in method section. The reasonings for choosing these methods are already explained in line no 75 – 80.

The experimental part needs to be improved. Can you add more ablation experiments? Can you compare it with some similar latest methods?

We do not fully understand this comment. The reviewer is not detailed enough as to what ablation experiments and which latest methods he is referring to.

Our study was intended to develop a PLSR vis–NIRS model commonly used in lab spectroscopy as it provides easy to interpret results and usually performs well. As we are dealing with relatively complex data originating from different geographic regions, we are comparing the PLSR models with ANN model. For both modelling types, we have applied the exact same types of pre-processing techniques (including different types of scatter corrections and derivative) in order to enable a fair comparison.  The result of our study showed that the ANN results were not very different from the PLSR results. We agree however, that it could be interesting to conduct a study where fine tuning of ANN models is in focus, yet with a condition that there is more data available. A condition which is a requirement and fundamental for applying deep learning/machine learning techniques.

In addition, there are some grammatical errors in the article, which need further careful proofreading.

Proofreading done.

Can the proposed method be applied to other plants or scenes?

We do not understand the question. Lab vis–NIRS technique has been applied to all sorts of agricultural materials and samples, that includes plant material as well However, our paper is not considering plants at all and is purely focused on soils. We do not understand what is meant by scenes either.