Machine Learning Scientist – Remote Sensing

GROW WITH TREEFERA We are a first-mile intelligence platform, delivering granular visibility into the point of origin in global ag & soft commodity supply chains - where risk, cost, performance and exposure are set. [https://app.ashbyhq.com/api/images/user-content/c732ad7a-0ae4-498f-b42a-bf284905cf4d/eaa3e7e4-29ff-4c00-86a8-f04fab8ec14a/Screenshot%202026-04-29%20at%2015.07.01.png] You’ll join a global, cross-functional team that values rigour, curiosity and working close to real-world challenges. Whether your focus is AI, climate, product or operations, you’ll have space to contribute meaningfully and make an impact from day one. If you’re excited by complex problems and want to help reshape how nature is valued in real-world decision-making, we’d love to hear from you. ROLE PURPOSE & RESPONSIBILITIES We are hiring a Machine Learning Scientist into the Science Team to contribute to the development of models — from classical statistics and gradient-boosted methods through to deep learning when warranted — that turn satellite, radar, and LiDAR observations into defensible, plot-level intelligence on the world's commodity supply chains. You will work across the full lifecycle — from research and prototyping through to validated, productionised models. A meaningful fraction of the team's work centres on building lightweight downstream models — classifiers, regressors, similarity-search workflows. Likely focus areas for this role include: - Commodity and plantation mapping by region - palm oil, cocoa, coffee, rubber, soy, timber and similar - in support of EUDR compliance and supply-chain due diligence. - Forest degradation and biomass / canopy-height estimation from multi-sensor fusion. - Develop ARR feasibility models that fuse climate, soil, and remote-sensing inputs to estimate site potential, forecast biomass and carbon trajectories, and quantify physical and permanence risk. RESPONSIBILITIES: - Design, train and evaluate ML models - from gradient-boosted methods to CNNs, U-Nets and vision transformers - for commodity and plantation mapping, land-cover classification, change and disturbance detection, and biomass / canopy-height estimation. - Build embedding-driven workflows on top of EO foundation models - few-shot classifiers, similarity search, downstream regressors. - Design validation strategies that benchmark outputs against plot inventories and third-party reference datasets, quantify uncertainty, and surface failure modes; produce QA artefacts (maps, plots, model cards, error analyses) that internal teams and clients can trust and defend. - Partner with Engineering to take models into scalable, reproducible inference pipelines across millions of plots, and contribute to a strong research culture across Science, AI and Engineering - reviewed code, shared tooling, and active engagement with EO/ML literature. WHO YOU ARE: MUST-HAVE REQUIREMENTS: - Degree in a quantitative field: environmental/earth science, computer science, physics, maths, engineering, or similar. - 3+ years of applied machine-learning experience, including time spent in an industry, product, or startup setting i.e. shipping models. - Expertise in geospatial Python tooling: rasterio, xarray, geopandas, GDAL, and the STAC ecosystem. - Hands-on experience training, evaluating, and debugging ML models across the modern Python stack - deep learning (CNNs, U-Nets, vision transformers) using PyTorch, as well as classical methods (gradient boosting, random forests) with scikit-learn. - Demonstrable experience with remote sensing data (optical, SAR) and an understanding of the sensor-specific quirks that matter for modelling. - Comfortable with Git, cloud compute (AWS or similar), and collaborative codebases. - Clear written and verbal communication: can explain modelling choices, uncertainties, and trade-offs to scientific and non-scientific stakeholders. - Domain exposure: deforestation, land-use change, biomass / canopy-height estimation, climate risk, or sup