Machine learning models for onion harvest maturity and storage life prediction

Overview

This research initiative develops predictive models to optimize onion harvesting decisions and storage outcomes. By combining satellite-based weather and vegetation data with ground-level sensor measurements, the project aims to identify the optimal harvest window for onions and predict how long harvested bulbs will maintain marketable quality in storage. The work addresses a critical challenge in the onion supply chain: onions harvested at suboptimal maturity exhibit higher respiration rates, greater water loss, and earlier breakdown of dormancy, all of which shorten storage life and reduce commercial value.

The research builds on demonstrated expertise in Bayesian multilevel modeling for agricultural applications and leverages an existing commercial respiration monitoring platform already deployed across more than 200 apple storage facilities in the UK and US.

Technical specifications

Modeling approach:

• Bayesian multilevel models integrated with machine learning algorithms
• Satellite-derived weather data and vegetation indices (canopy greenness) combined with ground-based sensor inputs
• Farm-level measurements including temperature, light interception, and soil moisture collected via Terraprima sensors
• Phenology-based prediction of onion maturity stage and timing

Storage monitoring system:

• Real-time respiration monitoring (LabPods) adapted from a commercially proven platform for stored fresh produce
• Continuous tracking of bulb respiration rate, humidity, weight loss, and sprouting behaviour
• Adaptation of potato-oriented monitoring hardware with integrated scales for daily weight loss measurement during curing and storage
• Prior validation on calcium and potassium nutrition effects on internal sprout growth in stored yellow onions

Field validation design:

• Seven spatially distinct UK field sites in the first year
• 10 x 10 m demarcated plots per site with co-located Terraprima sensors
• Field assessment of growth stages synchronized with satellite imagery acquisition
• Harvest maturity assessment based on bulb size, time after transplanting, nutritional content, and dry matter content
• Post-harvest storage monitoring in LabPods to capture dormancy and respiration dynamics

Technology readiness level

The underlying respiration monitoring technology is commercially deployed at Technology Readiness Level 9 for apple storage, with over 200 operational units in UK and US facilities. The onion-adapted version, incorporating weight loss scales, is at an earlier stage of development. The Bayesian multilevel modeling framework has been validated in prior agricultural research applications. The integrated prediction system for onion harvest maturity and storage life is at an early-to-mid stage of development, with planned field validation across seven UK sites in year one followed by model refinement in subsequent phases. The project seeks research partners and collaborators to support field trials, data collection, model validation, and eventual commercialization of the predictive platform for the onion industry.