Agricultural Data Scientist
Predictive models for plant breeding and crop monitoring.
I build statistical and machine learning models that predict genotype performance from genomic, phenomic and field trial data, monitor crops from satellite imagery, and turn them into software that breeders and researchers use every day.
- 10+ years
- in applied statistics
- FielDHub
- on CRAN, peer-reviewed
- R · Python
- Shiny, APIs, Docker
How I work
From the field to a decision
01 · Field
Experimental design
Randomized, augmented and p-rep trials built so the data can answer the question.
02 · Data
Data pipelines
Phenotypic, genomic and environmental data cleaned and joined into one reproducible pipeline.
03 · Model
Predictive modeling
Linear mixed models with spatial adjustment, GAMs, machine learning and genomic prediction, validated honestly.
04 · Decision
Decision tools
Shiny apps, APIs and forecasting services that put the model in the hands of breeders and growers.
What I focus on
Digital agriculture
Field trials, plant breeding and agricultural statistics, turned into data products for research programs and the state's farm economy.
Predictive modeling
Spatial trends, genotype effects and forecasting: models that separate the genetic signal from the field noise.
Satellite crop monitoring
Satellite imagery and vegetation indices turned into field-level forecasts and crop-stress alerts.
Selected work
All projects →Production system · client work
Satellite crop monitoring
Daily satellite vegetation indices, forecasts and crop-stress alerts for every field.
Predictive modeling · write-up
Likelihood vs Accuracy
Why a higher REML likelihood doesn't guarantee better genotype predictions.
Latest writing
All posts →- Likelihood vs Prediction Accuracy in Mixed ModelsA higher REML log-likelihood does not guarantee more accurate predictions. We show why with simulated trials, where the true genetic values are known, and a small, transparent REML oracle written in base R.Predictive modeling · 53 min read
- Smoothers in the Field: GAMs for Capturing Treatment Effects and Spatial Trends in Plant Breeding TrialsHow Generalized Additive Models can model spatial trends and estimate genotype effects in plant breeding trials, compared side by side with a classical ASReml spatial model.Predictive modeling · 14 min read
- Digital agriculture
- Predictive modeling
- Genomic prediction
- Satellite crop monitoring
- Machine learning
- Experimental design
- R & Shiny
Work with me
Have field, genomic or satellite data and a decision to make?
I consult and collaborate on trial design and analysis, predictive modeling, ag data pipelines and apps, and training.