🍇Grape.ag · 2020 Summer Intern Team2020 – 2021

Real-Time Powdery Mildew Detection for Vineyard Microclimates

Helped ship a production pipeline that runs UC Davis's powdery mildew risk model per vine, every 15 minutes. My piece: the Python algorithm on AWS Lambda, feeding the React Native iOS app that vineyard owners use in the field.

PythonpandasAWS LambdaReact Native

GitHub →App Store →

Grape.ag IoT sensor mounted on a vine — An IoT sensor mounted per-vine — measuring the temperature and humidity that feed the Powdery Mildew risk model.

The Story

The Powdery Mildew risk model from UC Davis was published in the 1960s. It's elegant but historically expensive to compute at field scale — for decades it was calculated regionally using weather stations. Computing it per vine block became feasible only when sensor and cloud-compute costs collapsed 60 years later. Grape.ag was the first platform to do it.

UC IPM Grape Powdery Mildew Risk Assessment Index →

The original UC Davis research our algorithm implements at sensor scale.

UC Davis Powdery Mildew risk assessment index table — The UC Davis risk index — a lookup table pairing temperature and humidity patterns to mildew risk. The algorithm I built computes this per sensor, every 15 minutes.

Alexander at Grape.ag — In the field during my Grape.ag internship.

What I Built

During my internship at Grape.ag (August 2020 – April 2021) I implemented the UC Davis algorithm in Python. The script ran every 15 minutes on AWS Lambda, ingesting live sensor data and populating the React Native iOS app that vineyard owners carried into the field.

Powdery Mildew risk is elevated when daily mean temperature > 10°C AND mean relative humidity has been > 85% for 10 consecutive days.

The Algorithm

→Ingest raw sensor readings (timestamp, temperature, humidity, sensor_id) from CSV exports
→Bucket into daily windows per sensor using pandas groupby
→Compute daily mean temperature → flag if > 10°C
→Compute rolling 10-day mean humidity → flag if > 85%
→Join conditions into a per-sensor, per-day risk table

Grape.ag iOS app — field dashboard — The Grape.ag iOS app — where the algorithm's output surfaces for vineyard owners.

Impact

→~20% improvement in crop protection efficiency (per team measurement)
→Shipped to App Store, currently v2.1.0 — still live and actively maintained in 2026
→Also contributed 15% revenue growth through market analysis identifying opportunities in Southern Mexico