HORIZON-CL6-2027-03-GOVERNANCE-04 (IA) — AI supporting informed advice for farmers and foresters to improve competitiveness and sustainability

1) Excellence

Overall objective: Develop, test, validate, and pilot a multimodal AI advisory system (advisor cockpit + field app) that fuses trusted knowledge reservoirs and real-world observations to provide evidence-linked, impartial recommendations supporting competitiveness, sustainability, and resilience in agriculture and forestry.

We bring a coherent “trust + sensing + AI + decision” stack demonstrated in recent research outputs:

• Trust and accountability: Self-Sovereign Digital Identity (SSDI) for forestry stakeholders; blockchain smart contracts for timber traceability with performance/scalability analysis (throughput/latency/energy).

• Forest monitoring and sensing: adaptive sensor clustering for dynamic forest WSNs (energy efficiency, robustness, RL-based adaptation).

• AI for forest intelligence: forest sound classification (optimized MFCC + hybrid deep models); YOLO-based individual tree detection from satellite RGB imagery.

• Decision modelling under uncertainty: forest regeneration dynamics (probabilistic/temporal decomposition); stochastic MILP optimisation for wood supply chains.

• Knowledge-to-action: AI-driven knowledge management metamodel enabling the pipeline from Data → Information → Knowledge → Intelligence → Action, aligned with advisory work.

Scientific/technical concept
TRUST-AKIS.AI integrates four engines into one deployable advisory workflow:

1. Multimodal Field Interpreter (mobile)

   • Accepts text + images + short video + audio notes from end-users; checks data quality; extracts signals (symptoms, damage, operations context).

2. Context Fusion & Operational Digital Twin

   • Fuses EO, inventories, in-situ sensors, terrain/soil, management logs to build a lightweight farm/forest operational digital twin for scenario exploration.

3. RAG Knowledge Service for “current & local” guidance

   • Retrieval-augmented advice grounded in versioned, trusted sources (best practices, certification rules, relevant policies, market signals where permitted). Outputs include source links and timestamps.

4. Trust Layer (identity, consent, auditability)

   • SSDI + auditable event trails enabling traceable provenance: who provided which observation, under what consent, and how it informed the advice.

Impartiality-by-design (not a slogan)

• Evidence-linked advice (every recommendation shows the observations/sources used)

• Multi-option outputs (ranked alternatives with trade-offs, costs, risks)

• Uncertainty disclosure (confidence bands + “what additional data would change the decision”)

• Governance controls (conflict-of-interest rules; no hidden vendor prioritisation)

2) Impact

Expected outcomes addressed (all three):

1. Comprehensive, tailored, impartial advice improving competitiveness/sustainability/resilience.

2. Improved skills for advisors, farmers, foresters on effective and responsible AI use.

3. Increased use of reliable data through curated, interoperable datasets and connectors.

3) Implementation

Living labs across regions with advisory networks + farmer/forester organisations + SMEs: co-design sprints → prototype trials → validation → pilot deployment.

Work plan (WPs & deliverables)

• WP1 Co-design & adoption (SSH-led): user needs, behaviour, gender and inclusion, trust, training design; ethics-by-design.

• WP2 Data space connectors & curation: ingestion from public/private sources + in-situ; privacy-preserving processing; interoperable dataset release + metadata.

• WP3 Multimodal AI & fusion: perception models, context fusion, uncertainty estimation, explainability, robustness.

• WP4 RAG knowledge service: source governance, localisation, updates, citations, anti-hallucination controls.

• WP5 Trust layer & auditability: SSDI, consent, auditable logs, secure sharing; lightweight smart-contract patterns where appropriate.

• WP6 Pilots & validation (TRL advancement): farm + forest pilots; baseline comparisons; impact assessment; iteration to TRL target.

• WP7 Exploitation & dissemination: business models, standardisation alignment, training assets, scale-out plan.