Scholarlink

<p>Geographical indications (GIs) operate as coupled socio-ecological systems in which product typicality derives from dynamic interactions among soil, climate, biota, management, and cultural practice. Their credibility increasingly depends on auditable mechanisms capable of linking territorial quality claims to verifiable environmental evidence. This study assesses the technical maturity of current machine learning (ML) approaches for supporting the operationalization of Digital Terroir, defined here as an Inferential Digital-twin layer for ecosystem -service auditing. Following PRISMA-ScR guidelines, an initial pool of 272 records was screened with an automated weighted-score system (94.2% thematic accuracy), producing a thematic corpus of 148 studies published between 2010 and 2025 for descriptive, multivariate, network, and meta-analytic analyses. Within this corpus, 25 studies met the full methodological-quality threshold (MMAT-adapted score ≥ 20; ICC = 0.87) and supported the in-depth qualitative synthesis. The analysis combined random-effects meta-analysis, funnel-plot inspection, leave-one-out sensitivity testing, multivariate statistics, and assessment against FAIR principles. Results indicate a generalization gap: classifiers frequently report high internal-validation accuracy (80–100%), whereas external robustness tests show greater performance degradation in models without spatially independent validation (11.8% drop versus 5.6% in spatially validated models; <i>d</i> = 0.95). Methodological fragmentation (modularity <i>Q</i> = 0.62; heterogeneity <i>I²</i> = 58%) and limited FAIR compliance (mean 34.2/100) are consistent with verification asymmetries that may currently restrict regulatory use. These findings suggest that Digital Terroir would benefit from a shift from static classifiers toward adaptive, spatially validated, and explainable models supported by proposed integrity benchmarks, including external degradation ≤ 8%, explainability (XAI) for territorial markers, and FAIR compliance ≥ 60/100. Such requirements may help convert GI sustainability claims into traceable evidence for third-party verification.</p> Graphical abstract <p></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

Digital terroir and ecosystem service auditability via machine learning in geographical indications

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月23日

查看全文
  • Catuxe Varjão de Santana Oliveira,
  • Luiz Diego Vidal Santos,
  • Gustavo da Silva Quirino,
  • Ana Karla de Souza Abud,
  • Francisco Sandro Rodrigues Holanda,
  • Renisson Neponuceno de Araújo Filho,
  • Tânia Cristina Azevedo

摘要

Geographical indications (GIs) operate as coupled socio-ecological systems in which product typicality derives from dynamic interactions among soil, climate, biota, management, and cultural practice. Their credibility increasingly depends on auditable mechanisms capable of linking territorial quality claims to verifiable environmental evidence. This study assesses the technical maturity of current machine learning (ML) approaches for supporting the operationalization of Digital Terroir, defined here as an Inferential Digital-twin layer for ecosystem -service auditing. Following PRISMA-ScR guidelines, an initial pool of 272 records was screened with an automated weighted-score system (94.2% thematic accuracy), producing a thematic corpus of 148 studies published between 2010 and 2025 for descriptive, multivariate, network, and meta-analytic analyses. Within this corpus, 25 studies met the full methodological-quality threshold (MMAT-adapted score ≥ 20; ICC = 0.87) and supported the in-depth qualitative synthesis. The analysis combined random-effects meta-analysis, funnel-plot inspection, leave-one-out sensitivity testing, multivariate statistics, and assessment against FAIR principles. Results indicate a generalization gap: classifiers frequently report high internal-validation accuracy (80–100%), whereas external robustness tests show greater performance degradation in models without spatially independent validation (11.8% drop versus 5.6% in spatially validated models; d = 0.95). Methodological fragmentation (modularity Q = 0.62; heterogeneity I² = 58%) and limited FAIR compliance (mean 34.2/100) are consistent with verification asymmetries that may currently restrict regulatory use. These findings suggest that Digital Terroir would benefit from a shift from static classifiers toward adaptive, spatially validated, and explainable models supported by proposed integrity benchmarks, including external degradation ≤ 8%, explainability (XAI) for territorial markers, and FAIR compliance ≥ 60/100. Such requirements may help convert GI sustainability claims into traceable evidence for third-party verification.

Graphical abstract