04 — Publications / Quinoa

Quinoa Research

Quinoa is a grain from the Andes of South America. It is rich in nutrients and tolerates drought and cold, so interest in it is growing around the world.

We use drone images to estimate how quinoa grows and how much it yields. Below, we introduce three papers, each in a video of about two minutes.

This work is part of SATREPS Super Food Bolivia, an international research project with Bolivia.

2026

A simple way to estimate the yield of colorful quinoa in farmers' fields

On the high plateau of Bolivia, at 3,700 to 4,000 meters, we set out to predict quinoa yield in farmers' fields by analyzing drone images.

3D analysis is accurate, but it takes effort and a lot of computing. So we refined conventional 2D analysis to estimate yield instead. The method combines two vegetation indices that capture how far each plant spreads horizontally, and then adds a step that fills the holes inside each plant. With this, we reached accuracy comparable to the 3D method.

What does filling the holes in a plant mean?
Seen from above, each quinoa plant shows small gaps inside it. Filling those gaps before measuring how far the plant spreads let a simple 2D analysis match the accuracy of 3D.

UAV-based quinoa yield monitoring for smallholder fields via dual-vegetation-index hole-filling: A practical alternative to canopy surface modeling

Yamaguchi T.†, Yoshikawa Y.†, Kushino S., Villca M., Katsura K.

Remote Sensing Applications: Society and Environment 44, 102306 (2026)

† Co-first authors

2026

Predicting yield before harvest with machine learning

We grew eight quinoa varieties with diverse colors and plant forms for two seasons in a field at Tokyo University of Agriculture and Technology, in Fuchu, Tokyo. From drone images, we measured the shape of each plant, such as its height and volume, along with its color. Machine learning then predicted the yield of each plant.

Because quinoa varieties differ so much in color and form, shape information such as volume and surface area was more useful for prediction than color. Images taken at flowering gave the most accurate predictions. Even images from early growth predicted yield almost as well.

Can yield be predicted when the plants come in so many colors?
Shape, such as each plant's volume and surface area, helped more than color. Images taken at flowering were the most accurate, and even early images came close.

Plant-level yield prediction of quinoa in field plots across growth stages using UAV-derived 3D structural and spectral features with machine learning

Sesay S., Yamaguchi T., Kushino S., Yoshikawa Y., Adachi S., Katsura K.

Smart Agricultural Technology 13, 101940 (2026)

2025

Estimating growth by combining 3D data and color information

We grew eight quinoa varieties with diverse colors and plant forms in a field at Tokyo University of Agriculture and Technology, in Fuchu, Tokyo. We photographed each plant with a drone circling around it. From these photos we built a 3D model of the whole plant, then combined its volume with information from multispectral and ordinary color (RGB) images.

With this, we estimated the dry weight of the above-ground parts of the plant (biomass) and the leaf area per unit of ground area (leaf area index). Plant volume from the 3D model alone estimated biomass more accurately than plant height or surface area. Combining all three sources improved accuracy further, and the estimates held steady across varieties.

What do you learn by flying around a plant?
You get a 3D model of the whole plant. Its volume alone estimated the dry weight better than height or surface area did. Adding color information made it even more accurate.

Fusion of UAV-based 3D mesh and spectral features improves quinoa biomass and LAI estimation across genotypic and temporal variations

Sesay S., Yamaguchi T., Kushino S., Yoshikawa Y., Adachi S., Katsura K.

Smart Agricultural Technology 10, 100818 (2025)

← Back to Publications