Evaluating Urban Sewage Sludge Distribution on Agricultural Land Using Interpolation and Machine Learning Techniques
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis study employed a conventional manure spreader to evaluate the distribution of urban sewage sludge on agricultural land.
1. I do not prefer the abbreviation sewage sludge (SS); in my opinion it is better to write both words in full.
2. The image included in Figure 2. The improved manure spreader used in trials. A very general picture of the spreader, please update this image with a picture showing the location of the sensors used and the improvements made to the machine according to this study.
3. Line 118 (The sludge collected in each tray was analyzed using ANOVA to determine if there were significant differences) What are the confidence levels for the ANOVA analysis?
4. In section 2.5, Machine Learning Classification on field data, you did not explain how the machine learning algorithms were used and what the inputs and outputs of the hyperparameters for each model used are. This part needs more explanation and detail.
5. How were the performance metrics calculated for the models?
Author Response
REVIEWER 1 ANSWERS 1. I do not prefer the abbreviation sewage sludge (SS); in my opinion it is better to write both words in full. We thank you for your suggestion and acknowledge that the choice of SS as an abbreviation is not the best! We apologize for the unfortunate misunderstanding. We have replaced all SS with SWS, as suggested by the second reviewer. This highlights the SeWage part of the object. 2. The image included in Figure 2. The improved manure spreader used in trials. A very general picture of the spreader, please update this image with a picture showing the location of the sensors used and the improvements made to the machine according to this study. Thank you for bringing this to our attention. We have replaced the image with a new, more detailed one. 3. Line 118 (The sludge collected in each tray was analyzed using ANOVA to determine if there were significant differences) What are the confidence levels for the ANOVA analysis? We apologise for not stating this. We have added that we considered differences with a p-value < 0.05 to be significant. 4. In section 2.5, Machine Learning Classification on field data, you did not explain how the machine learning algorithms were used and what the inputs and outputs of the hyperparameters for each model used are. This part needs more explanation and detail. Thank you for your suggestion. We have updated Section 2.5 of the manuscript to include a detailed description of how the machine learning algorithms were used, specifying the inputs and outputs for each model, as well as the process of optimising the hyperparameters. In particular, we described: - The input variables for each algorithm (e.g. interpolated values, spatial coordinates). - The outputs generated (distribution zone classifications, performance metrics such as AUC and MCC). - The tuning process of the hyperparameters (e.g. selection of the number of neighbours for the kNN and configuration of the neural network). 5. How were the performance metrics calculated for the models? We apologize for not stating this. We have updated the manuscript in section 2.5 to clarify how the performance metrics were calculated: 1. Metrics used: The performance metrics, including AUC, CA, F1 score, precision, recall and MCC, were calculated using the test set data, separated from the training data in a 70/30 ratio. 2. Calculation method: - AUC was calculated using the R package pROC, based on the ROC curves generated for each model. - CA, precision, recall and F1 score were derived from the confusion matrices. - MCC was calculated to assess the correlation between predictions and observations, which is particularly useful with unbalanced data sets. 3. Iterations for robustness: The models were iterated 100 times, with average results reported to reduce variability due to randomisation in the division of the data.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe topic is relevant, but I have to highlight the following:
Abstract
- It is mainly an introduction and shows only a brief instead of the methods and evidence-based existing results. I suggest to improve it.
Introduction
I have to suggest using different abbreviations for sewage sludge. I prefer to use SWS (highlight the "sewage"), the original one is not the best. I have concerns about the use of sewage sludge in PA. Firstly after the pandemic, the case of the not properly handled SWS is problematic. Secondly, the nutrient content of the SWS is regionally related so the nutrient content is not standardizable. But this is my concern, and this kind of usage is optionable.
Materials and methods
The 2.1. 2.2 are nice. The 2.3 is extremely brief (a few sentences of the ANOVA, reasons, and ideas are missing). 2.4 is clear, it is nice to see the QGIS usage. 2.5 the use of R language is also welcome. Lines 166-174 are nice, but additional information is suggested. What about the NNA, MT, MP, and ME ? What about the specific settings of the keras? What about the seeds? So I suggest adding the most relevant information, I see this is not IT specific journal, but this information is mandatory. Data preprocessing steps (176-177) also brief, the type of normalization, main settings of the caret packages are necessary. The 70/30 rate is nice. The briefly mentioned performance metrics are nice, but at least a short why-we-use-it description is suggested.
Results
Table 1, figure 3 more detailed additional interpretations are suggested, we can see the numbers, but what are these values representing (for the readers)? Additional information the Tukey's test (how can we interpret it) also suggested. Figure 4 interpretation is nice.
232- confusion ?matrices? not matrix? I suggest adding few sentences about figure 5. There is a near-full-page-size figure but a detailed interpretation is suggested.
Discussion
As the methodology is so brief, the discussion is too long. I suggest removing the duplication (make it shorter, and more clear), and mentioning the value of the results is missing for me. 309 the mention is Kriging is nice, but we have to read it in the methods subchapter. Maybe some of these lines would be better to put in the introduction/methodology chapters. The performance metrics are mentioned in several lines (302, 329, etc.).
Conclusion
I suggest highlighting the evidence-based results, instead of general sentences. I suggest rethinking the abbreviation of SS.
Author Response
REVIEWER 2 ANSWERS Abstract It is mainly an introduction and shows only a brief instead of the methods and evidence-based existing results. I suggest to improve it. Thank you for your suggestion. We have added references in the abstract to the methods used and the main results obtained, within the size limits specified by the journal. Introduction I have to suggest using different abbreviations for sewage sludge. I prefer to use SWS (highlight the "sewage"), the original one is not the best. I have concerns about the use of sewage sludge in PA. Firstly after the pandemic, the case of the not properly handled SWS is problematic. Secondly, the nutrient content of the SWS is regionally related so the nutrient content is not standardizable. But this is my concern, and this kind of usage is optionable. We thank you for your suggestion and acknowledge that the choice of SS as an abbreviation is not the best! We apologise for the unfortunate misunderstanding. We have replaced all SS with SWS, as suggested. We added a critical note to the discussion by highlighting the variability in the nutrient properties of sludge. We have carried out this experiment to contribute to the development of technologies to regulate distribution so that precision farming strategies can be applied even with these simpler machines. However, the next steps will rightly have to take into account a preliminary or on-the-spot verification of the characteristics of the sludge to be spread. Materials and methods The 2.1. 2.2 are nice. The 2.3 is extremely brief (a few sentences of the ANOVA, reasons, and ideas are missing). 2.4 is clear, it is nice to see the QGIS usage. 2.5 the use of R language is also welcome. We appreciate your recommendation. We have expanded section 2.3 to provide a more complete description of the ANOVA analysis used, including an explanation of why ANOVA was used, i.e. to identify significant differences in mud distribution along field axes and between trials. 166-174 are nice, but additional information is suggested. What about the NNA, MT, MP, and ME ? What about the specific settings of the keras? What about the seeds? So I suggest adding the most relevant information, I see this is not IT specific journal, but this information is mandatory. Data preprocessing steps (176-177) also brief, the type of normalization, main settings of the caret packages are necessary. The 70/30 rate is nice. The briefly mentioned performance metrics are nice, but at least a short why-we-use-it description is suggested. Thank you very much for the information which we have followed. We have updated the manuscript with additional information on Neural network analysis (NNA): We described the design of the neural network, including the number of layers, neurons per layer, activation functions used (ReLU for hidden layers, sigmoid for output) and the use of dropouts to avoid overfitting. Metrics (MT, MP, ME): Details were given on the metrics monitored during training (MAE, MSE, accuracy) and their interpretation in the results. Keras settings: We specified that the Adam optimiser was used with an initial learning rate of 0.001. The batch size was 32 and callbacks such as EarlyStopping were used to avoid overfitting. Seeds: To ensure reproducibility, we set the seeds in R to use the Keras package, number 42. This setting was used for all experiments. Results Table 1, figure 3 more detailed additional interpretations are suggested, we can see the numbers, but what are these values representing (for the readers)? We very much appreciated your suggestion. We have expanded the description of Table 1 and Figure 3 in the manuscript to provide a more detailed interpretation of the observed values and patterns. In particular: Table 1: We have explained the significance of the ANOVA results, highlighting that the significant variability along the longitudinal axis (p < 0.001) reflects mechanical constraints on the distribution of sludge. Figure 3: We described how visual patterns highlight higher concentrations in the centre of the field, with practical implications for calibration and optimisation of distribution operations. Additional information the Tukey's test (how can we interpret it) also suggested. Figure 4 interpretation is nice. The comments were very helpful. So we extended the description of the Tukey test results to give a clearer and more detailed interpretation. In particular Description of Tukey's test: We introduced the test as a post-hoc method to compare means and characterise significant differences along the longitudinal axis. Interpretation: The results show three distinct zones of concentration, with significantly higher values in the central zone than in the periphery. We related these observations to the mechanical limitations of the manifold. Operational implications: We discussed how the results can guide practical interventions, including the calibration of the distributor or the use of advanced technologies to optimise distribution. 232- confusion ?matrices? not matrix? We apologise for the error. Thank you for reporting this. We have corrected 'confusion matrix' to 'confusion matrices' to reflect the reference to multiple models. I suggest adding few sentences about figure 5. There is a near-full-page-size figure but a detailed interpretation is suggested. Your suggestion was much appreciated by the authors. So we immediately proceeded to add details to Figure 5: We described the meaning of the confusion matrices, discussed the observed results and linked these observations to the general performance of the models (NN, RF, kNN). We also provided practical implications based on the identified error patterns. Discussion As the methodology is so brief, the discussion is too long. I suggest removing the duplication (make it shorter, and more clear), and mentioning the value of the results is missing for me. Thank you for your recommendation. We have shortened the discussion by trying to make it shorter and clearer. We have also mentioned the results by discussing them. 309 the mention is Kriging is nice, but we have to read it in the methods subchapter. Maybe some of these lines would be better to put in the introduction/methodology chapters. Thank you for pointing that out. We agree. We have moved the reference to the kriging method to the introduction. The performance metrics are mentioned in several lines (302, 329, etc.). We apologise for the unnecessary repetition in the text. We have unified the mention of these metrics by avoiding repetition. Conclusion I suggest highlighting the evidence-based results, instead of general sentences. Thank you for your valuable suggestion. We have changed the text to refer more to the results obtained. I suggest rethinking the abbreviation of SS. We thank you again and confirm that we have accepted the suggestion to replace SS with SWS.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors have done a great job developing the manuscript and addressing all the issues and questions I raised previously. The manuscript has improved greatly.