Agricultural Data Scientist — Crop Yield Modelling

Carl Bauer

Freelance · Mid-level · London

This role is no longer accepting applications

Similar open roles you can apply to now:

About the role

We make precision agriculture software for arable farms in the UK and Ireland. Our platform connects field sensors, satellite imagery, soil sampling data, and farm management records to give farmers and agronomists a single view of what is happening across their land. We are not a big company — 41 people, mostly engineers and agronomists, based in Lincolnshire and Cork. We have been building this for six years and we have 1,200 farm businesses on the platform. The modelling problem we have been unable to solve internally is crop yield prediction: given everything we know about a field at a given point in the growing season — soil composition, moisture levels, NDVI time-series from satellite, weather history and forecast, and the management interventions the farmer has applied — how well can we predict end-of-season yield, and how early can we make a prediction that is actually useful for decision-making? Our agronomists have built heuristic models that work reasonably well for winter wheat on well-characterised soils. They break down on less common crops and on fields with high spatial variability. We are commissioning a twelve-week project to build and evaluate a machine learning approach to yield prediction using three seasons of data we have across a subset of our field estate. If the approach is sound, we would like to continue with you on a retainer basis for ongoing model improvement.

Key Deliverables

List the expected deliverables for this project.

Audit and clean three seasons of crop performance, sensor, satellite, and weather data for the modelling dataset
Build and evaluate yield prediction models at field and sub-field resolution for the primary crop types in the dataset
Develop a framework for assessing how early in the season a reliable yield prediction can be made — and what the confidence interval looks like at each prediction window
Produce a written technical report with model evaluation, honest assessment of limitations, and recommendations for the data collection improvements that would most improve future model accuracy
Deliver a structured handover with the internal engineering team including documented code, training pipeline, and evaluation methodology

Requirements

Technical stack needed for this mission.

4+ years of data science with time-series modelling or geospatial analysis experience — agricultural domain knowledge is a genuine advantage but not required
Python for data processing, model development, and reproducible analysis workflows
Scikit-learn for regression, ensemble methods, and careful model evaluation across a heterogeneous field dataset
SQL for working with our farm management database and sensor data archive
Statistics: handling missing data in sensor time-series, spatial autocorrelation in field data, and communicating model uncertainty in terms agronomists can act on
NumPy for numerical computing and signal processing of satellite-derived vegetation indices
Familiarity with geospatial data formats — GeoTIFF, shapefiles, or equivalent — would accelerate the engagement significantly