Technology

Detecting clouds, shadows, and snow

We analyze satellite images of the field to identify artifacts on it so we can use images without them in future work. This stage is necessary for our apps to work correctly because only clear pictures allow us to properly estimate field characteristics.

Dense clouds
Translucent clouds
Cloud shadows

We recognize thick and translucent clouds, shadows, and snow

Detecting field boundaries

We manually marked tens of thousands of fields and then trained an ML algorithm to define boundaries automatically. We show what happens with fields at any scale, from a whole region to a specific field plot. As a result, any farmer can get information about the state of their fields in our apps.



IoU 0.85
The accuracy of the automatic field delineation model
SVG Created with Sketch.
49 years is the time one person would need to manually mark these fields
2018
2019
2020
Our algorithms define field boundaries with a 5-meter accuracy
21,603,849
The number of fields marked in the United States
35,923,503
The number of fields marked in Europe
Our algorithms define field boundaries with a 5-meter accuracy
2018
2019
2020
21,603,849
The number of fields marked in the United States
Our algorithms define field boundaries with a 5-meter accuracy
35,923,503
The number of fields marked in Europe
49 years is the time one person would need to manually mark these fields

Detecting more than 50 crops mid-season

Satellite monitoring allows us to automatically identify a crop growing in a field. To clarify the information, we use data from the Sentinel-1 radar satellite.

We detect crops in almost all countries around the world: Australia and New Zealand, Russia, Turkey, China, Japan, Kazakhstan, Uzbekistan, South Africa, and the countries in Europe, North America, and South America.

376,835,301 ha

The area of the fields analyzed in Europe and the United States

Recognition accuracy F1score
0.92
0.96
Mid-season
End of season
Germany4,352,263 fields covering a total area of 18,200,467 ha
Wheat
24.3%714,729 fields4,415,611 ha
Corn
9.8%392,965 fields1,780,392 ha
Barley
2.5%80,840 fields451,446 ha
Beet
17.9%762,697 fields3,250,187 ha
Other
45.6%2,401,032 fields8,302,831 ha
Germany4,352,263 fields covering a total area of 18,200,467 ha
Wheat
24.3%714,729 fields4,415,611 ha
Corn
9.8%392,965 fields1,780,392 ha
Barley
2.5%80,840 fields451,446 ha
Beet
17.9%762,697 fields3,250,187 ha
Other
45.6%2,401,032 fields8,302,831 ha

Identifying the sowing date and plant growth stages

We identify the sowing date with Sentinel-1 satellite imagery. Then we determine plant growth stages using multispectral images. This helps select a time to apply fertilizers and pesticides. This feature is coming soon to our apps.

We’re grateful to the European Union for the free and open satellite data it makes available through the Copernicus Program and its fleet of Sentinel spacecraft.

7 days

The accuracy of identifying plant growth stages

Aiming to make agriculture more precise

OneSoil app users add information about their fields to the system, which helps us to constantly improve the machine learning algorithms we use. The more data there is, the more accurate our recommendations are. As a result, farming becomes faster and more efficient.

We believe that agriculture will be fully automated. Special programs and apps will collect information about the field, process it, and make data-driven decisions themselves. We’re already on the way to doing that.

We’ll soon be able to:

  • Predict
    crop yield
  • Plan and monitor fieldwork
  • Give recommendations to farmers at all stages of work
  • Predict plant diseases and the emergence of pests

We’ve harnessed these technologies to launch a service that provides statistics for businesses

OneSoil Business Insights