Scholarlink

Background <p>Food security remains a persistent challenge in the Asia-Pacific region, where climate shocks and economic volatility interact to create complex and escalating risks. Traditional vulnerability assessments often rely on static or retrospective methods, failing to capture dynamic feedbacks, predict tipping points, or evaluate the resilience of policy interventions. This study addresses this gap by developing a forward-looking analytical framework that integrates predictive analytics with policy-relevant inference to assess and anticipate food security vulnerability under interacting climate and economic shocks.</p> Methodology <p>We construct a dynamic Food Security Vulnerability Index (FSVI) for 23 Asia-Pacific countries (2000–2023) using principal component analysis with three indicators that are conceptually distinct from explanatory variables: prevalence of undernourishment, cereal yield variability, and prevalence of moderate or severe food insecurity. We employ a hybrid sequential approach. First, ensemble machine learning models (Random Forest, XGBoost) are used to predict vulnerability trajectories and identify dominant, non-linear risk drivers via SHAP analysis. Second, key drivers inform a dynamic panel econometric model (Dynamic Common Correlated Effects) to estimate policy-relevant associations and shock interactions. Finally, the model is used for counterfactual policy simulations projecting vulnerability pathways to 2030.</p> Results <p>Machine learning models achieved high predictive accuracy (R²=0.893) and identified drought intensity and food import dependency as the primary drivers of vulnerability, with their interaction amplifying risk by approximately 40% in model-based predictions. Econometric results confirm high vulnerability persistence and show that public agricultural expenditure is associated with reduced baseline vulnerability and buffers an estimated 28% of drought impacts, with associations strongest in lower-middle-income countries. Simulations indicate business-as-usual trends would increase regional vulnerability by 8.9% by 2030, whereas integrated strategies combining climate-resilient investment, trade diversification, and inclusive development could reduce vulnerability by up to 21.6%.</p> Conclusion <p>The hybrid framework demonstrates that food security vulnerability is a dynamic, path-dependent process shaped by compound climate–trade risks. By linking machine learning prediction with policy-relevant econometric analysis, this study provides a scalable analytical foundation for anticipatory governance, supporting a shift from crisis response to proactive resilience building in the Asia-Pacific and beyond.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Vulnerability of agricultural food systems to climate and economic shocks: a hybrid machine learning and dynamic panel analysis of Asia-Pacific

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月20日

查看全文
  • Yudi Ferrianta,
  • Nuri Dewi Yanti

摘要

Background

Food security remains a persistent challenge in the Asia-Pacific region, where climate shocks and economic volatility interact to create complex and escalating risks. Traditional vulnerability assessments often rely on static or retrospective methods, failing to capture dynamic feedbacks, predict tipping points, or evaluate the resilience of policy interventions. This study addresses this gap by developing a forward-looking analytical framework that integrates predictive analytics with policy-relevant inference to assess and anticipate food security vulnerability under interacting climate and economic shocks.

Methodology

We construct a dynamic Food Security Vulnerability Index (FSVI) for 23 Asia-Pacific countries (2000–2023) using principal component analysis with three indicators that are conceptually distinct from explanatory variables: prevalence of undernourishment, cereal yield variability, and prevalence of moderate or severe food insecurity. We employ a hybrid sequential approach. First, ensemble machine learning models (Random Forest, XGBoost) are used to predict vulnerability trajectories and identify dominant, non-linear risk drivers via SHAP analysis. Second, key drivers inform a dynamic panel econometric model (Dynamic Common Correlated Effects) to estimate policy-relevant associations and shock interactions. Finally, the model is used for counterfactual policy simulations projecting vulnerability pathways to 2030.

Results

Machine learning models achieved high predictive accuracy (R²=0.893) and identified drought intensity and food import dependency as the primary drivers of vulnerability, with their interaction amplifying risk by approximately 40% in model-based predictions. Econometric results confirm high vulnerability persistence and show that public agricultural expenditure is associated with reduced baseline vulnerability and buffers an estimated 28% of drought impacts, with associations strongest in lower-middle-income countries. Simulations indicate business-as-usual trends would increase regional vulnerability by 8.9% by 2030, whereas integrated strategies combining climate-resilient investment, trade diversification, and inclusive development could reduce vulnerability by up to 21.6%.

Conclusion

The hybrid framework demonstrates that food security vulnerability is a dynamic, path-dependent process shaped by compound climate–trade risks. By linking machine learning prediction with policy-relevant econometric analysis, this study provides a scalable analytical foundation for anticipatory governance, supporting a shift from crisis response to proactive resilience building in the Asia-Pacific and beyond.