Certifying the Sustainability of Herding Practices in Mongolia

Background Introduction

Globally, unsustainable agricultural practices have led to ecosystem degradation, particularly the destruction of grassland ecosystems due to overgrazing. Mongolia, as the world’s second-largest producer of cashmere, has significant economic, cultural, and ecological impacts from its nomadic herding practices. However, with the increase in goat numbers, overgrazing has become increasingly severe, leading to grassland vegetation degradation, soil erosion, and loss of wildlife habitats. To address this issue, the Mongolian government, in collaboration with the United Nations, declared 2026 as the “International Year of Rangelands and Pastoralists,” aiming to promote sustainable herding practices.

Certification schemes, as a means to incentivize sustainable land management, have been applied in various fields. However, existing certification schemes face numerous challenges in assessing sustainability, including the complexity of evaluation methods, disagreements among stakeholders, and issues such as “greenwashing.” Therefore, developing a transparent, repeatable, and representative assessment method that incorporates stakeholder opinions has become crucial for advancing sustainable herding practices.

Source of the Paper

This research was conducted by scholars from multiple institutions, including the Arthur Rylah Institute for Environmental Research (ARI), the Wildlife Conservation Society (WCS), and Agronomes et Vétérinaires Sans Frontières (AVSF), among others. The primary authors of the paper include Steve J. Sinclair, Khorloo Batpurev, Canran Liu, and others. The study was published in March 2025 in the journal Nature Sustainability, titled “Certifying the Sustainability of Herding Practices in Mongolia.”

Research Process

1. Research Design and Variable Definition

The study aimed to develop a sustainability assessment method based on stakeholder opinions and apply it to cashmere production in Mongolia. The research team first defined 19 variables covering the context (e.g., weather conditions), behavior (e.g., herding strategies), and outcomes (e.g., goat survival rate, vegetation cover) of herding practices. These variables were based on the existing Sustainable Cashmere Certification Committee (S3C) standards and were adjusted through preliminary interviews with 11 herders.

2. Scenario Design and Stakeholder Consultation

The research team created 245 hypothetical but plausible herding practice scenarios, each describing the behavior and outcomes of a herding family over one year. These scenarios were presented to stakeholders, including herders, scientists, and policymakers, on A5 paper cards. A total of 151 participants evaluated these scenarios, rating the sustainability of each scenario (0-100 points) and determining whether it warranted certification.

3. Data Modeling and Analysis

Based on the stakeholder evaluation data, the research team used two machine learning methods—Random Forest and Gradient Boosting—to build models predicting sustainability scores. The Random Forest model performed best and was used for subsequent analysis. By predicting median scores, the model was able to transform complex herding practices into a simple assessment tool.

4. Development of the Filtered Model

To reveal differences in opinion between producers and certifiers, the research team developed a “filtered model.” This model excluded scenarios that received high scores despite vegetation cover at the winter pasture falling below a certain threshold. The filtered model placed greater emphasis on environmental outcomes, particularly changes in vegetation cover, when assessing sustainability.

Key Findings

1. Diversity in Stakeholder Evaluations

Stakeholders’ evaluations of herding practice scenarios showed significant variation, with 13% of scenarios receiving scores spanning the full range of 0-100 points, indicating a lack of consensus on sustainable practices. However, when median scores were used as a representative measure, clear trends emerged in the data. For example, higher scores were positively correlated with goat survival rates and cashmere fiber acceptance rates.

2. Model Predictive Performance

The Random Forest model demonstrated good performance in predicting median scores (R²=0.46). The model predicted that scenarios with scores above 41.99 were more likely to be certified. The most influential variables in the model included cashmere fiber acceptance rate (fibre_accept) and the size of the goat flock at the end of the season (goat_end).

3. Results of the Filtered Model

The filtered model performed comparably to the full model in predicting scores (R²=0.47) but placed greater emphasis on environmental outcomes, particularly vegetation cover (veg_cov_dep). The filtered model no longer rewarded scenarios with increasing goat numbers unless other variables indicated sustainable practices.

Research Conclusions

The study developed a sustainability assessment method based on stakeholder opinions, transforming complex herding practices into a simple evaluation tool. This method introduces several innovations: (1) using stakeholder opinions as a source of information; (2) employing machine learning to navigate complex decision spaces; (3) integrating continuous scores and binary certification recommendations; and (4) revealing differences in opinion among different stakeholder groups through “filtering.”

The assessment tool can be applied to any production system, provided that the variables describing production practices can be clearly defined and appropriate stakeholders are available for evaluation. This method not only enhances the transparency and credibility of certification schemes but also provides a platform for negotiation between producers and certifiers.

Research Highlights

1. Stakeholder Involvement: The study ensured the representativeness and transparency of the assessment method by extensively consulting herders, scientists, and policymakers.
2. Application of Machine Learning: Through the Random Forest model, the study successfully transformed complex herding practices into an actionable assessment tool.
3. Filtered Model: The filtered model revealed differences in opinion between producers and certifiers, offering a new perspective on sustainability assessment.
4. Broad Applicability: This method is not only applicable to cashmere production in Mongolia but can also be extended to other production systems, demonstrating significant potential for widespread application.

Research Significance

This study provides a new assessment method for sustainable certification schemes, effectively addressing issues such as information asymmetry and “greenwashing.” By integrating stakeholder opinions and machine learning techniques, the research not only improves the accuracy and transparency of assessments but also offers an important tool for advancing global agricultural sustainability.