Hybrid Vineyard Digital Twin for Multi‑Objective Forecasting
Description
Position overview:
Develop a “digital twin” module that forecasts outcomes under climate and management scenarios and supports multi-objective optimisation. The goal is to quantify trade-offs (yield stability, water/input reduction proxies, soil carbon proxies) with uncertainty, enabling evidence-based planning in the Regenerative AI Engine.
Problem statement (why this exists):
Farm decisions are multi-objective and constraint-bound. Predictive models that optimise only accuracy often fail in stress years (heat/drought) and cannot communicate trade-offs. This track builds a scenario-capable model stack that generalises across seasons and produces decision-ready outputs with uncertainty.
Data you will work with:
- Open datasets relevant to vineyards/regenerative systems + Track A features (EO/climate/soil).
- Optional internal data: Symbioverse field datasets (Spain) for sanity checks and domain realism (not required).
Key responsibilities (linked to the innovation project):
- Build a predictive model stack using open datasets + Track A features.
- Compare pure ML vs hybrid approaches (mechanistic + ML), focusing on stress-year generalisation.
- Implement uncertainty-aware evaluation and robustness tests (missing-data, domain shift).
- Prototype a simple multi-objective optimiser that outputs ranked intervention packages (no proprietary product prescriptions).
Why your expertise is crucial:
We need someone who can design experiments that test generalisation and failure modes—not just headline accuracy.
Key outcomes for the Visiting Talent:
- End-to-end modelling pipeline with strong evaluation hygiene.
- A model card–style transparency report.
- Experience translating predictions into decision-support outputs under constraints.
Stream:
- Deep-Tech Talents (research-focused).
Skills
Sign in to read and join the discussion.