Farmers' adaptive capacity towards soil salinity effects using hybrid machine learning in the Red River Delta

Nguyen, Huu Duy; Dang, Dinh Kha; Lai, Thi Anh Tam; Tran, Duc Dung; Shahabi, Himan; Bui, Quang-Thanh

doi:https://doi.org/10.5194/egusphere-2025-1051

Preprints

https://doi.org/10.5194/egusphere-2025-1051

Preprints

21 Mar 2025

| 21 Mar 2025

Farmers' adaptive capacity towards soil salinity effects using hybrid machine learning in the Red River Delta

Huu Duy Nguyen, Dinh Kha Dang, Thi Anh Tam Lai, Duc Dung Tran, Himan Shahabi, and Quang-Thanh Bui

Abstract. Soil salinity is a grave environmental threat to agricultural development and food security in large parts of the world, especially in the situation of global warming and sea level rise. Reliable information on the adaptive capacity of farms plays a key role in reducing the socioeconomic effects of soil salinization and helps policymakers and farmers propose more appropriate measures to combat the phenomenon. The aim of the research is to design a theoretical framework to assess soil salinity and farmers' adaptive capacity, based on machine learning, optimization algorithms (namely Xgboost (XGB), XGB- Pelican Optimization Algorithm (POA), XGB- Siberian Tiger Optimization (STO), XGB- Serval Optimization Algorithm (SOA), XGB- Particle Swarm Optimization (PSO), and XGB- Grasshopper Optimization Algorithm (GOA)), remote sensing, and interviews with local people. The geographical distribution of soil salinity was evaluated by applying machine learning Sentinel 1 and 2A. The adaptive capacity of farmers was evaluated through interviews with 87 households. The statistical indices, namely the mean absolute error (MAE), the root mean square error (RMSE), and the correlation coefficient (R²) were used to assess the machine learning models. The outcome of this study demonstrated that all optimization algorithms were successful in improving the accuracy of the XGB model. The XGB-POA was the most performance, with an R² value of 0.968, followed by XGB-STO (R² = 0.967), XGB-SOA (R² = 0.966), XGB-PSO (R² = 0.964), and XGB-GOA (R² = 0.964), respectively. The soil salinity map produced by the proposed models also indicated that the coastal and riverside regions were the most affected by soil salinity. The results also showed human and financial resources to be the two most important factors influencing the adaptive capacity of farmers. This study offers a key theoretical framework that supplements the previous studies and can support policy-markers and farmers in land resource management, for example accurately identifying areas affected by soil salinity for agricultural development in the context of climate change. In addition, this research highlights the importance of integrating machine learning, remote sensing, and socio-economic surveys in soil salinity management, which can support farmers for sustainable agricultural development.

Received: 05 Mar 2025 – Discussion started: 21 Mar 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 1763 KB)

Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Preprint (1763 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

22 Sep 2025

Predicting Soil Salinity in the Red River Delta (Vietnam) Using Machine Learning and Assessing Farmers' Adaptive Capacity

Huu Duy Nguyen, Dinh Kha Dang, Thi Anh Tam Lai, Duc Dung Tran, Himan Shahabi, and Quang-Thanh Bui

Nat. Hazards Earth Syst. Sci., 25, 3505–3524, https://doi.org/10.5194/nhess-25-3505-2025,https://doi.org/10.5194/nhess-25-3505-2025, 2025

Short summary

Huu Duy Nguyen, Dinh Kha Dang, Thi Anh Tam Lai, Duc Dung Tran, Himan Shahabi, and Quang-Thanh Bui

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-1051', Anonymous Referee #1, 13 May 2025
The manuscript entitled "Farmers' adaptive capacity towards soil salinity effects using hybrid machine learning in the Red River Delta" presents a timely and relevant study that integrates machine learning techniques with socio-economic analysis to assess both the spatial distribution of soil salinity and the adaptive capacity of farmers in a climate-vulnerable region. The study employs a suite of hybrid models combining XGBoost with various optimization algorithms (POA, STO, SOA, PSO, GOA), along with remote sensing data and household surveys, to deliver a comprehensive framework for analyzing the dual dimensions of environmental stress and human response. The article conforms to the journal-specific instructions and the topic fits well with the scope of the journal and proposes an innovative approach. While the article is generally well-structured and balanced, some key methodological details are missing. In particular, key details are missing regarding land use categorization in the study area and the rationale for selecting specific remote sensing indices and optimization algorithms. Additionally, although the findings related to farmers' adaptive capacity are insightful, they are presented largely in isolation from the machine learning analysis, with minimal integration between the two parts of the study. Furthermore, the Discussion section would benefit from a more critical engagement with existing literature, particularly studies that have applied similar optimization algorithms in environmental or agricultural contexts.
Detailed comments on each section are provided below.
Title
The current title may not accurately reflect the study’s output. In the present status, the study does not use machine learning to assess farmer’s adaptive capacity, but rather to predict soil salinization. The title should be reconsidered and rephrased to avoid any misleading interpretations.
Astract
The abstract is complete and gives a clear idea of the content without reading the paper.
Minor comments.
Introduction
Overall, the introduction covers the state of the art and explains the objectives of the study in a complete way. However, several acronyms and abbreviations are introduced here without first presenting their full forms. I recommend carefully reviewing the Introduction, and the manuscript as a whole, for consistency in defining all acronyms upon first use. Minor comments:
L42: Use “posing” instead of “poses”.
L51: Please rephrase “represent an extremely key role”.
L126: This passage would be more suitable for the final remarks (Conclusion) that the Introduction.
Materials and methods
The section is clearly structured into different sub-sections and easy to follow. However, some key information is unclear or missing:
Model selection and integration: The rationale behind the selection of the specific optimization algorithms (POA, STO, SOA, PSO, GOA) and how they are integrated with the XGBoost model is not clearly explained. It is also not fully clear how these hybrid models contribute to the generation of soil salinity maps. Clarifying this connection would strengthen the methodological transparency.

Land use consideration: It seems that the modeling process and salinity mapping does not account for different land use types. Applying models across the entire region without filtering by land use could lead to inaccurate interpretations, especially in heterogeneous agricultural landscapes. This should be explicitly addressed.

Irrigation practices: The manuscript would benefit from including contextual information on irrigation practices in the study area. Specifically, details on the main sources of irrigation, and the distribution of land use types (e.g., paddy fields, rainfed, and irrigated areas) would provide valuable background for understanding the drivers of soil salinity and its spatial variability.

Farmer interviews: It is recommended that the authors include the full list or at least a representative sample of the questionnaire items used in the household interviews, either within the main text or as supplementary material. Moreover, the socio-economic component of the study is presented independently from the machine learning analysis, with little discussion of how the two are connected. Figure 2 suggests that the selection of interview locations may have been informed by the salinity maps generated through the machine learning models, but this relationship is not clearly explained. Clarifying this linkage would enhance the coherence of the study and highlight the value of integrating spatial and social data.

Minor comments:
L133: Please remove “with the”.
L144: Replace “obtained at” with “reach”.
L165.166: Where are the soil sapling points located exactly?
L185: Please translate “extractés à partir de l’image” into English.
L221: Please define what a Tan commune is.
L223: There is an extra comma “is, often”.
L306 and onwards: Proposed by proposed by Kennedy and Eberhart (1995). Please check the reference style of similar citations throughout the manuscript.
Results
The results are clear and concise. As stated above, there is poor integration between the machine learning analysis and the socio-economic analysis. Minor comments:
L395: What questions are asked in the interviews? (see comment above)
L401: The passage “changing the crop structure” is unclear. Please rephrase.
L472: There is a typo here “the 2soil salinity”.
Discussion
The Discussion section addresses the main findings of the study, particularly the performance of the hybrid XGBoost models and the socio-economic insights from the farmer interviews. However, it falls short in a few critical areas that limit the depth and broader relevance of the study's conclusions:
Lack of comparative analysis: The discussion would benefit from a more comprehensive comparison with similar studies that have applied enhanced or hybrid XGBoost algorithms (or other machine learning approaches) in soil salinity mapping or related environmental modeling tasks. Including such references would help position the study within the existing body of literature and strengthen its contribution.

Limited interpretation of spatial variability: While the results highlight the spatial distribution of soil salinity, the discussion does not fully explore the potential environmental, agronomic, or anthropogenic drivers behind the observed variability. Possible contributing factors other than proximity to the coast or rivers should be discussed in more detail to provide context for the spatial patterns.

Integration between technical and social findings: The Discussion currently treats the machine learning results and socio-economic findings as separate components. A more integrated discussion that connects spatial variability in salinity with local adaptive capacity (e.g., explaining how different levels of salinization impact farmers' strategies or vulnerability) would enhance the coherence and practical relevance of the study.

Minor comments:
L499-509: This paragraphs contains repetitions of already stated concepts. Perhaps it could be shortened.
L586: Please rephrase the sentence.
Conclusions
The conclusions are clear and well-balanced. However, I would recommend clearly stating the future steps to fill the existing gaps.
Citation: https://doi.org/10.5194/egusphere-2025-1051-RC1
- AC2: 'Reply on RC1', Nguyen Huu Duy, 12 Jun 2025
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-1051/egusphere-2025-1051-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-1051-AC2
RC2:
'Comment on egusphere-2025-1051', Anonymous Referee #2, 13 May 2025

Soil salinity, which significantly impacts agricultural activities worldwide, is considered one of the major environmental hazards caused by both natural and human-induced processes. This phenomenon has become increasingly severe due to the impacts of climate change, particularly rising sea levels. Therefore, evaluating soil salinity is regarded as a critical task for supporting sustainable agricultural planning. Assessing adaptive capacity is also regarded as a crucial instrument for reducing the impact of soil salinity on local livelihoods. One of the strengths of this article is the integration of physical data, machine learning models, and socio-economic data (through interviews with local populations). As such, this article is relevant and well-aligned with the journal's scope. I accept publishing this article with the condition of major revisions.
Abstract: Although the authors present the objectives, data, and results of the article, I would like to see the inclusion of quantitative results and the significance of the findings.
Introduction: It is necessary to point out the importance of this article. Additionally, it is important to emphasize the role of adaptive capacity in reducing the effects of soil salinity.
Study Area: The reasons for selecting this study area should be explained in more detail, especially the effects of soil salinity on agricultural activities.
Map 1: Please revise Map 1 for better clarity.
Map 2: Similarly, Map 2 should be revised for better quality.
Methodology: This study uses machine learning and optimization algorithms to construct the soil salinity map. However, I do not fully understand how the authors constructed these models. A more detailed explanation is needed.
Interviews with Local Populations: The inclusion of the interview methodology is necessary because adaptive capacity is a key outcome.
Discussion: Although this article clearly discusses the strengths and weaknesses of the machine learning models and also touches on the adaptive capacity of the populations, I believe it would be useful to add the methodology for addressing the effects of soil salinity at the community level.
Extrapolation Issues: In this section, the authors present issues of extrapolation. I would suggest expanding on this point, as it is a challenge not only in soil salinity but also in other types of natural hazards.

Citation: https://doi.org/10.5194/egusphere-2025-1051-RC2
- AC1: 'Reply on RC2', Nguyen Huu Duy, 12 Jun 2025
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-1051/egusphere-2025-1051-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-1051-AC1

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2025-1051', Anonymous Referee #1, 13 May 2025
The manuscript entitled "Farmers' adaptive capacity towards soil salinity effects using hybrid machine learning in the Red River Delta" presents a timely and relevant study that integrates machine learning techniques with socio-economic analysis to assess both the spatial distribution of soil salinity and the adaptive capacity of farmers in a climate-vulnerable region. The study employs a suite of hybrid models combining XGBoost with various optimization algorithms (POA, STO, SOA, PSO, GOA), along with remote sensing data and household surveys, to deliver a comprehensive framework for analyzing the dual dimensions of environmental stress and human response. The article conforms to the journal-specific instructions and the topic fits well with the scope of the journal and proposes an innovative approach. While the article is generally well-structured and balanced, some key methodological details are missing. In particular, key details are missing regarding land use categorization in the study area and the rationale for selecting specific remote sensing indices and optimization algorithms. Additionally, although the findings related to farmers' adaptive capacity are insightful, they are presented largely in isolation from the machine learning analysis, with minimal integration between the two parts of the study. Furthermore, the Discussion section would benefit from a more critical engagement with existing literature, particularly studies that have applied similar optimization algorithms in environmental or agricultural contexts.
Detailed comments on each section are provided below.
Title
The current title may not accurately reflect the study’s output. In the present status, the study does not use machine learning to assess farmer’s adaptive capacity, but rather to predict soil salinization. The title should be reconsidered and rephrased to avoid any misleading interpretations.
Astract
The abstract is complete and gives a clear idea of the content without reading the paper.
Minor comments.
Introduction
Overall, the introduction covers the state of the art and explains the objectives of the study in a complete way. However, several acronyms and abbreviations are introduced here without first presenting their full forms. I recommend carefully reviewing the Introduction, and the manuscript as a whole, for consistency in defining all acronyms upon first use. Minor comments:
L42: Use “posing” instead of “poses”.
L51: Please rephrase “represent an extremely key role”.
L126: This passage would be more suitable for the final remarks (Conclusion) that the Introduction.
Materials and methods
The section is clearly structured into different sub-sections and easy to follow. However, some key information is unclear or missing:
Model selection and integration: The rationale behind the selection of the specific optimization algorithms (POA, STO, SOA, PSO, GOA) and how they are integrated with the XGBoost model is not clearly explained. It is also not fully clear how these hybrid models contribute to the generation of soil salinity maps. Clarifying this connection would strengthen the methodological transparency.

Land use consideration: It seems that the modeling process and salinity mapping does not account for different land use types. Applying models across the entire region without filtering by land use could lead to inaccurate interpretations, especially in heterogeneous agricultural landscapes. This should be explicitly addressed.

Irrigation practices: The manuscript would benefit from including contextual information on irrigation practices in the study area. Specifically, details on the main sources of irrigation, and the distribution of land use types (e.g., paddy fields, rainfed, and irrigated areas) would provide valuable background for understanding the drivers of soil salinity and its spatial variability.

Farmer interviews: It is recommended that the authors include the full list or at least a representative sample of the questionnaire items used in the household interviews, either within the main text or as supplementary material. Moreover, the socio-economic component of the study is presented independently from the machine learning analysis, with little discussion of how the two are connected. Figure 2 suggests that the selection of interview locations may have been informed by the salinity maps generated through the machine learning models, but this relationship is not clearly explained. Clarifying this linkage would enhance the coherence of the study and highlight the value of integrating spatial and social data.

Minor comments:
L133: Please remove “with the”.
L144: Replace “obtained at” with “reach”.
L165.166: Where are the soil sapling points located exactly?
L185: Please translate “extractés à partir de l’image” into English.
L221: Please define what a Tan commune is.
L223: There is an extra comma “is, often”.
L306 and onwards: Proposed by proposed by Kennedy and Eberhart (1995). Please check the reference style of similar citations throughout the manuscript.
Results
The results are clear and concise. As stated above, there is poor integration between the machine learning analysis and the socio-economic analysis. Minor comments:
L395: What questions are asked in the interviews? (see comment above)
L401: The passage “changing the crop structure” is unclear. Please rephrase.
L472: There is a typo here “the 2soil salinity”.
Discussion
The Discussion section addresses the main findings of the study, particularly the performance of the hybrid XGBoost models and the socio-economic insights from the farmer interviews. However, it falls short in a few critical areas that limit the depth and broader relevance of the study's conclusions:
Lack of comparative analysis: The discussion would benefit from a more comprehensive comparison with similar studies that have applied enhanced or hybrid XGBoost algorithms (or other machine learning approaches) in soil salinity mapping or related environmental modeling tasks. Including such references would help position the study within the existing body of literature and strengthen its contribution.

Limited interpretation of spatial variability: While the results highlight the spatial distribution of soil salinity, the discussion does not fully explore the potential environmental, agronomic, or anthropogenic drivers behind the observed variability. Possible contributing factors other than proximity to the coast or rivers should be discussed in more detail to provide context for the spatial patterns.

Integration between technical and social findings: The Discussion currently treats the machine learning results and socio-economic findings as separate components. A more integrated discussion that connects spatial variability in salinity with local adaptive capacity (e.g., explaining how different levels of salinization impact farmers' strategies or vulnerability) would enhance the coherence and practical relevance of the study.

Minor comments:
L499-509: This paragraphs contains repetitions of already stated concepts. Perhaps it could be shortened.
L586: Please rephrase the sentence.
Conclusions
The conclusions are clear and well-balanced. However, I would recommend clearly stating the future steps to fill the existing gaps.
Citation: https://doi.org/10.5194/egusphere-2025-1051-RC1
- AC2: 'Reply on RC1', Nguyen Huu Duy, 12 Jun 2025
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-1051/egusphere-2025-1051-AC2-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-1051-AC2
RC2:
'Comment on egusphere-2025-1051', Anonymous Referee #2, 13 May 2025

Soil salinity, which significantly impacts agricultural activities worldwide, is considered one of the major environmental hazards caused by both natural and human-induced processes. This phenomenon has become increasingly severe due to the impacts of climate change, particularly rising sea levels. Therefore, evaluating soil salinity is regarded as a critical task for supporting sustainable agricultural planning. Assessing adaptive capacity is also regarded as a crucial instrument for reducing the impact of soil salinity on local livelihoods. One of the strengths of this article is the integration of physical data, machine learning models, and socio-economic data (through interviews with local populations). As such, this article is relevant and well-aligned with the journal's scope. I accept publishing this article with the condition of major revisions.
Abstract: Although the authors present the objectives, data, and results of the article, I would like to see the inclusion of quantitative results and the significance of the findings.
Introduction: It is necessary to point out the importance of this article. Additionally, it is important to emphasize the role of adaptive capacity in reducing the effects of soil salinity.
Study Area: The reasons for selecting this study area should be explained in more detail, especially the effects of soil salinity on agricultural activities.
Map 1: Please revise Map 1 for better clarity.
Map 2: Similarly, Map 2 should be revised for better quality.
Methodology: This study uses machine learning and optimization algorithms to construct the soil salinity map. However, I do not fully understand how the authors constructed these models. A more detailed explanation is needed.
Interviews with Local Populations: The inclusion of the interview methodology is necessary because adaptive capacity is a key outcome.
Discussion: Although this article clearly discusses the strengths and weaknesses of the machine learning models and also touches on the adaptive capacity of the populations, I believe it would be useful to add the methodology for addressing the effects of soil salinity at the community level.
Extrapolation Issues: In this section, the authors present issues of extrapolation. I would suggest expanding on this point, as it is a challenge not only in soil salinity but also in other types of natural hazards.

Citation: https://doi.org/10.5194/egusphere-2025-1051-RC2
- AC1: 'Reply on RC2', Nguyen Huu Duy, 12 Jun 2025
  
  The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-1051/egusphere-2025-1051-AC1-supplement.pdf
  
  Citation: https://doi.org/10.5194/egusphere-2025-1051-AC1

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Reconsider after major revisions (further review by editor and referees) (13 Jun 2025) by Paolo Tarolli

AR by Nguyen Huu Duy on behalf of the Authors (13 Jun 2025) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (04 Jul 2025) by Paolo Tarolli

RR by Anonymous Referee #2 (06 Jul 2025)

RR by Aurora Ghirardelli (08 Aug 2025)

ED: Publish subject to technical corrections (14 Aug 2025) by Paolo Tarolli

AR by Nguyen Huu Duy on behalf of the Authors (18 Aug 2025) Author's response Manuscript

Journal article(s) based on this preprint

22 Sep 2025

Predicting Soil Salinity in the Red River Delta (Vietnam) Using Machine Learning and Assessing Farmers' Adaptive Capacity

Huu Duy Nguyen, Dinh Kha Dang, Thi Anh Tam Lai, Duc Dung Tran, Himan Shahabi, and Quang-Thanh Bui

Nat. Hazards Earth Syst. Sci., 25, 3505–3524, https://doi.org/10.5194/nhess-25-3505-2025,https://doi.org/10.5194/nhess-25-3505-2025, 2025

Short summary

Huu Duy Nguyen, Dinh Kha Dang, Thi Anh Tam Lai, Duc Dung Tran, Himan Shahabi, and Quang-Thanh Bui

Viewed

Total article views: 695 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
536	138	21	695	13	27

HTML: 536
PDF: 138
XML: 21
Total: 695
BibTeX: 13
EndNote: 27

Views and downloads (calculated since 21 Mar 2025)

Month	HTML	PDF	XML	Total
Mar 2025	50	12	1	63
Apr 2025	35	24	5	64
May 2025	56	20	4	80
Jun 2025	41	25	8	74
Jul 2025	31	12	0	43
Aug 2025	72	29	3	104
Sep 2025	251	16	0	267

Cumulative views and downloads (calculated since 21 Mar 2025)

Month	HTML	PDF	XML	Total
Mar 2025	50	12	1	63
Apr 2025	35	24	5	64
May 2025	56	20	4	80
Jun 2025	41	25	8	74
Jul 2025	31	12	0	43
Aug 2025	72	29	3	104
Sep 2025	251	16	0	267

Viewed (geographical distribution)

Total article views: 733 (including HTML, PDF, and XML) Thereof 733 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 22 Sep 2025

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (1763 KB)
Metadata XML

Short summary

Although several previous studies have been conducted to assess soil salinity in Vietnam, most have focused on assessing soil salinity and farmers' adaptive capacity in the Mekong Delta. Few studies have been conducted in the Red River Delta. The Red River Delta is one of the key agricultural regions in Southeast Asia. Therefore, assessing soil salinity and farms' adaptive capacity in this area is necessary.

Farmers' adaptive capacity towards soil salinity effects using hybrid machine learning in the Red River Delta

Journal article(s) based on this preprint

Interactive discussion

Interactive discussion

Peer review completion

Suggestions for revision or reasons for rejection

Suggestions for revision or reasons for rejection

Journal article(s) based on this preprint

Viewed

Viewed (geographical distribution)


Total:	0
HTML:	0
PDF:	0
XML:	0