Advancing Agricultural Land Suitability in Urbanized Semi-Arid Environments: Insights from Geospatial and Machine Learning Approaches

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Please see attached 

Comments for author File: Comments.pdf

Comments on the Quality of English Language

The article is written well with minimal errors.

Author Response

Response to Reviewer 1

General Comments : This paper assesses the ML based standard algorithms used in agricultural land suitability studies. It discusses five different algorithms commonly used and employs them to find suitable agricultural land in Dharmapuri district , India. The article provides a discussion of the shortcomings or limitations of using a combination of different sets of data. The results and  discussion section are substantiated adequately. The paper has merit for land use managers to help identify sustainable land management practices by optimizing resources and ensuring food security in the regions studied. However, such models need to be further validated and need to be cautiously used in other geographic areas. The authors need to discuss this in the limitation or conclusion section(s). The following are the specific comments for further improvement of the paper.:

Comments 1: Lines 19-20 and 80-81 need additional references to justify the AGLS model's use. What were the representative locations where this model was used? Are the results consistent in irrigated and rainfed ag systems?

Response 1: Thank you for your constructive feedback. Agricultural land suitability (AgLS) involves the identification and recommendation of areas most appropriate for agricultural production. This can be assessed using various approaches, including statistical methods, multivariate analysis, and machine learning techniques. In the study area, the suitability of land for irrigated and rainfed agriculture is strongly influenced by seasonal rainfall patterns. Favorable rainfall can convert rainfed land into irrigated land and vice versa. The primary aim of this study is to evaluate agricultural land suitability by integrating historical precipitation data with other influencing factors in urbanized riverine regions. Additionally, we have highlighted the novelty of applying the Extra Trees Classifier (ETC) model for AgLS mapping for the first time. Please refer to lines 91–116 for further clarification.

Comments 2:  Lines 68-71.The authors need enough justification as to why AgLS is a reliable method. Has this method been widely tested in different scenarios? Are results consistent? Discuss additional results in this section (literature review).

Response 2: Thank you for your constructive feedback. The use of machine learning (ML) algorithms for agricultural land suitability analysis represents a cutting-edge and innovative approach in this field. The literature on the application of fine-tuned ML algorithms, such as SVM, KNN, RF, and Naïve Bayes, in this context remains limited. Notably, this study is pioneering in its application of the ETC algorithm for agricultural land suitability assessment within urbanized riverine ecosystems. The results, along with a comparative analysis and discussion of related studies, have been integrated into the Results and Discussion section. For further details, please refer to lines 99–116.

Comments 3:  Lines133-134-talk about the objective of finding challenges of different land suitability models used and improving them for optimal prediction. This, however, has not been discussed well in the discussion section.

Response 3: Thank you for your valuable feedback. We have conducted a comprehensive discussion of the findings for each model, both prior to and following hyperparameter tuning. Furthermore, we integrated satellite-derived LULC data to exclude non-agricultural features, including forests, water bodies, and settlements, by applying a masking process within the AgLS model. Please refer to lines 462–471 and 507–528 for details.

Comments 4:  Page 4-Study area-Can you add information on elevation and or slope? Also, what are the common agricultural practices and crops adopted so far, and what are the current land cover types? Include them in the study area characterization.

Response 4: Thank you for your constructive comments. The above details were added into the study area section.

Comments 5:  Figure 3-The human component is not integrated, as you discussed in the literature review; this component is imperative-Farmer's decisions play a crucial role that would not be captured by the variables you have stated in Figure 3.

Response 5: Thank you for this insightful comment. Farmers' decisions are inherently tied to inventory data. For this study, 200 farmers were selected from an initial pool of 500, based on the criterion of owning at least 2 acres of land. The participants were progressive and well-informed farmers, capable of comprehending and applying recommendations provided by the Department of Agriculture, Tamil Nadu, India. The evaluation centered on soil fertility, which was classified into two categories: suitable (indicating high productivity) and unsuitable (comprising degraded land, saline soils, and waterlogged areas). These classifications formed the basis of the inventory data and were validated through analysis conducted using Google Earth Pro (refer to lines 254–260).

Comments 6:   Line 179-184-explain how you interpolate the rainfall and temperature data.

Response 6: Thank you for the constructive feedback. The annual average climate data for temperature and rainfall were obtained in CSV format from the Global Weather website (https://globalweather.tamu.edu/). Monthly averages and annual mean values for rainfall and temperature were calculated. These data were subsequently interpolated using the Inverse Distance Weighting (IDW) method, as outlined in lines 187–190.

Comments 7:  Figure 4 caption-4(a) does not represent slope map.

Response 7: Thank you for your valuable feedback. The caption has been revised according to your suggestions, and the figure number has been updated to (3).

Comments 8:  Line 198-199-please discuss in detail the classification method, training samples used, and results of the accuracy assessment.

Response 8: The authors sincerely appreciate this constructive suggestion regarding the LULC map. The entire classification process, including preprocessing, classification, and accuracy assessment, has been clarified and is detailed in lines 203–215. A total of 55 ground truth data points were utilized, with a minimum of 10 samples collected for each class. The LULC map was generated using the maximum likelihood supervised classification method, achieving an overall accuracy of 95% and a kappa coefficient of 0.89.

Comments 9:  Table 2-provide a unit for the last five parameters in the first column.

Response 9: Thank you for this constructive comment regarding the unit in the table 2. This has been updated in the manuscript.

Comments 10: Line 318-did you run any analysis (correlation or other) to avoid the selection of related data points?

Response 10: Thank you for your insightful comment. We conducted a correlation analysis for all input factors, employing Variance Inflation Factor (VIF) analysis as part of the feature selection and validation process. Random Forest techniques were used to determine feature importance, quantified as the average impurity decrease across all decision trees within the forest. However, due to the manuscript's length constraints, we excluded the detailed results from the final version.

Comments 11: Line 375-376-provide citation(s) Page 13, section 3.1.

Response 11: Thank you for your comment. The relevant citations have been incorporated into this paragraph.

Comments 12: Page 13, section 3.1, you also want to discuss a spatial variation of temperature and precipitation/rainfall in the study area as you discussed soil properties.

Response 12: Thank you for your feedback. The spatial variations in temperature and precipitation/rainfall have been incorporated and discussed in the revised manuscript, as detailed in lines 434–441.

Comments 13: Line 444-448- Have you validated your results to find if the suitable zones found by five different ML methods significantly differ from each other to ensure no presence of under or over-estimation?

Response 13: Thank you for your comment. The over- and under-predictions were evaluated using AUC, RMSE, and Kappa values. If these parameters are lower, it indicates over- or under-estimation by the model. The best-performing model, ETC, yielded AUC, RMSE, and Kappa values of 0.89, 0.69, and 0.15, respectively..

Comments 14: Line 521-I am glad you have discussed the limitation of the research, highlighting some issues. Can you add a paragraph on how you addressed these issues?

Response 14: Thank you for your constructive comments. To address this challenge, we applied a hyperparameter tuning step to all models to optimize performance, with the ETC model yielding the best results. The challenges related to this are discussed between lines 515-529. We plan to include future research that incorporates crop suitability analysis using ground truth data for various crops (e.g., mango, rice, paddy, ragi, pulses, and coconut) to directly improve the accuracy and reliability of the model, as mentioned in the study limitations between lines 558-577.

Response to Comments on the Quality of English Language.

Point 1: The English could be improved to more clearly express the research.

Response 1:  Your constructive feedback during the minor revision process has been invaluable. We sincerely appreciate the time and effort you dedicated to helping us improve our work for submission to the ISPRS International Journal of Geo-Information.

 

Reviewer 2 Report

Comments and Suggestions for Authors

I think the paper "Advancing Agricultural Land Suitability in Urbanized Semiarid Environments: Insights from Geospatial and Machine Learning Approach" is very valuable because it demonstrates the advantages of the AgLS-ETC model for land use analysis, which can better help farmers choose suitable types of cultivated land and provide a feasible solution for achieving sustainable agricultural productivity in an increasingly severe ecological environment. Although this paper is valuable enough to be published, there are still some issues that need to be addressed: as follows：

(1) The beginning of the introduction needs to be improved and more relevant to the study of land quality.

(2) What is the source of the water network in Figure 1? Please use a more detailed water network;How does Figure 2 help the research topic? What is its role?

(3) Please improve the clarity of the text in all images and check whether the image title in line 186 of Figure 4 corresponds to the image content.

(4) What is the specific implementation method of “A random sample of 200 farmers was selected to assess soil fertility across two suitability categories (suitable and unsuitable) designated as inventory data.”? How is soil fertility determined?

(5) The Landsat-8 satellite data used and the related processing were not specified.

 

(6) The accuracy of the generated model directly depends on the quality of the input data. It is recommended that the research results be supported by crop yields.

Comments for author File: Comments.pdf

Author Response

Response to Reviewer 2

General Comments: I think the paper "Advancing Agricultural Land Suitability in Urbanized Semiarid Environments: Insights from Geospatial and Machine Learning Approach" is very valuable because it demonstrates the advantages of the AgLS-ETC model for land use analysis, which can better help farmers choose suitable types of cultivated land and provide a feasible solution for achieving sustainable agricultural productivity in an increasingly severe ecological environment. Although this paper is valuable enough to be published, there are still some issues that need to be addressed: as follows:

Comments 1: The beginning of the introduction needs to be improved and more relevant to the study of land quality.

Response 1: Thank you for your constructive feedback. The introduction has been revised to first address the challenges of agricultural development in urbanized riverine regions, focusing on the impacts of climate change and anthropogenic activities globally. It then discusses the role of Agricultural Land Suitability (AgLS) analysis, the relevant condition factors, and the literature on modeling techniques. The rationale for using machine learning (ML) is presented, followed by a discussion of the research gap. Additionally, the introduction highlights agricultural challenges in India, specifically in the Dharmapuri district, and concludes with the objectives.

Comments 2:  What is the source of the water network in Figure 1? Please use a more detailed water network; How does Figure 2 help the research topic? What is its role?

Response 2: Thank you for your constructive comments. The river stream network was delineated using Google Earth Pro. Since the geological map (Figure 2) provides only general information for the study, it was not utilized in the analysis. As per your suggestion, Figure 2 was removed.

Comments 3:  Please improve the clarity of the text in all images and check whether the image title in line 186 of Figure 4 corresponds to the image content.

Response 3: Thank you for your valuable feedback. Revisions have been made to all figures in the manuscript as per your suggestions.

Comments 4:  What is the specific implementation method of “A random sample of 200 farmers was selected to assess soil fertility across two suitability categories (suitable and unsuitable) designated as inventory data.”? How is soil fertility determined?

Response 4: Thank you for your constructive comments. The study began with detailed field visits to a group of 200 farmers selected from an initial group of 500 participants, based on a minimum landholding requirement of 2 acres. The participants were experienced and progressive farmers, proficient in understanding and applying recommendations from the Department of Agriculture, Tamil Nadu, India. The evaluation focused on assessing soil fertility, categorizing the land into two suitability classes: suitable (high productivity) and unsuitable (degraded land, saline soils, and waterlogged areas). These categories served as inventory data, which were further validated through analysis using Google Earth Pro, as described between lines 254-261. 

Comments 5:  The Landsat-8 satellite data used and the related processing were not specified.

Response 5: The entire classification process, encompassing preprocessing, classification, and accuracy assessment, is detailed in lines 204–216. A total of 55 ground truth data points were used, with at least 10 samples per class. The LULC map was produced using the maximum likelihood supervised classification method, achieving an overall accuracy of 95% and a kappa coefficient of 0.89.

Comments 6:   The accuracy of the generated model directly depends on the quality of the input data. It is recommended that the research results be supported by crop yields.

Response 6: Thank you for your constructive comments. We agree with your insight regarding the importance of input quality. To address this challenge, we applied a hyperparameter tuning step to all models to optimize performance, with the ETC model yielding the best results. The challenges related to this are discussed between lines 515-529. Regarding crop yield, we plan to include future research that incorporates crop suitability analysis using ground truth data for various crops (e.g., mango, rice, paddy, ragi, pulses, and coconut) to directly improve the accuracy and reliability of the model, as mentioned in the study limitations between lines 558-577.

Response to Comments on the Quality of English Language.

Point 1: The quality of English does not limit my understanding of the research.

Response 1:  Your constructive feedback during the minor revision process has been invaluable. We sincerely appreciate the time and effort you dedicated to helping us improve our work for submission to the ISPRS International Journal of Geo-Information.

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript highlights that declining crop yields in the urbanizing riverine regions of Southern Asia pose a threat to food security. However, integrating geospatial data with machine learning models, such as the Extra Trees Classifier (ETC), provides potential solutions. In Dharmapuri, India, ETC demonstrated the best performance (RMSE = 0.15) in assessing agricultural land suitability, identifying 19.08% of the area as highly suitable after incorporating LULC data, paving the way for sustainable agricultural planning. The section is adequate. Minor corrections are needed in the figure captions, which should better explain elements like colors and presented points. The combined results and discussion section is relevant. I suggest the authors include study limitations and future perspectives.

Author Response

Response to Reviewer 3

General Comments: This manuscript highlights that declining crop yields in the urbanizing riverine regions of Southern Asia pose a threat to food security. However, integrating geospatial data with machine learning models, such as the Extra Trees Classifier (ETC), provides potential solutions. In Dharmapuri, India, ETC demonstrated the best performance (RMSE = 0.15) in assessing agricultural land suitability, identifying 19.08% of the area as highly suitable after incorporating LULC data, paving the way for sustainable agricultural planning. The section is adequate. Minor corrections are needed in the figure captions, which should better explain elements like colors and presented points. The combined results and discussion section is relevant. I suggest the authors include study limitations and future perspectives.

Comments 1: Minor corrections are needed in the figure captions, which should better explain elements like colors and presented points.

Response 1: Thank you for your valuable feedback. Revisions have been made to all figures in the manuscript as per your suggestions

Comments 2:  I suggest the authors include study limitations and future perspectives.

Response 2: Thank you for your constructive comments. To address this challenge, we applied a hyperparameter tuning step to all models to optimize performance, with the ETC model yielding the best results. The challenges related to this are discussed between lines 515-529. We plan to include future research that incorporates crop suitability analysis using ground truth data for various crops (e.g., mango, rice, paddy, ragi, pulses, and coconut) to directly improve the accuracy and reliability of the model, as mentioned in the study limitations between lines 557-577.

Response to Comments on the Quality of English Language.

Point 1: The quality of English does not limit my understanding of the research.

Response 1:  Your constructive feedback during the minor revision process has been invaluable. We sincerely appreciate the time and effort you dedicated to helping us improve our work for submission to the ISPRS International Journal of Geo-Information.