Satellite Imagery Can Help Forecast Crop Yields
Researchers at St. Petersburg State Agrarian University have developed a neural network that will make it possible to assess crop yields across different sections of a field.
Space-based technologies are playing an increasingly important role in the development of Russian crop production. They provide real-time information about farmland, including the soil’s spectral characteristics. Analyzing these data with AI makes it possible to respond quickly to changes in environmental conditions. This marks a shift from assessing fields based on average values to detailed analysis and, eventually, the ability to manage crops individually, down to nearly every plant.
AI Analysis of Satellite Imagery
Researchers at St. Petersburg State Agrarian University (SPbSAU) have proposed an innovative approach to monitoring farmland using remote sensing and neural networks. The technology relies on AI analysis of satellite imagery.
Modern crop production still largely relies on practices that assess an entire field as a single unit. As a result, the same amount of fertilizer is often applied across the whole area, without accounting for variations in soil conditions. Yet even a relatively small field can contain distinct zones with different characteristics. Crop yields are also calculated for the field as a whole, which can produce inaccurate estimates.
“Average yield figures do not show how much individual sections differ from one another. If those differences are visible on a map, agronomists can account for them more easily when planning fieldwork and allocating resources,” said Omar Tsukaev, an SPbSAU student and one of the platform’s developers.
This approach has persisted because creating a detailed field map that records the characteristics of individual sections requires substantial resources and can therefore be expensive. The process involves agrochemical and soil surveys, as well as agroecological assessments. Such work is generally carried out no more than once every five years, even though conditions in the field can change much faster. These surveys are also virtually impossible to conduct across very large areas of farmland.
Neural Network to Assess Productivity Across Individual Field Sections
To evaluate field performance by analyzing its smaller sections, researchers have begun using AI to conduct retrospective analyses of satellite imagery. During training, the neural network uses data on vegetation conditions and moisture levels. Crop conditions are assessed using the Normalized Difference Vegetation Index (NDVI). Satellite data are then refined using yield maps generated by mapping systems installed on combine harvesters. In effect, this creates a historical record of the field, making it possible to assess the productivity of smaller areas and track yield trends over an extended period.
The analysis will use high-resolution multispectral and hyperspectral satellite imagery, including data from the Lobachevsky satellite, which was launched into orbit in late 2025. This small spacecraft was specifically designed for agroecological research across different regions of Russia.
“Satellite data help us monitor crop conditions efficiently and comprehensively, apply fertilizers and pesticides, and forecast agricultural crop yields,” said Anatoly Shevyrnogov, chief research scientist at the Institute of Biophysics of the Siberian Branch of the Russian Academy of Sciences.
The SPbSAU platform is therefore expected to improve the effectiveness of Russia’s precision agriculture systems by generating digital maps that account for the productivity of individual land sections and variations across fields. This makes it possible to apply fertilizer and water at variable rates according to local conditions. Integrating the platform with IT systems that manage agricultural machinery will also allow operators to upload fieldwork instructions tailored to the specific characteristics of each area.
Improving Land-Use Efficiency
Similar technologies are also being developed elsewhere in Russia. Researchers at the Krasnoyarsk Scientific Center of the Siberian Branch of the Russian Academy of Sciences have developed a method for creating soil fertility maps that divide fields into zones with different productivity levels. The analysis draws on data collected by satellites and drones. Using specialized software powered by AI, agronomists assess small areas of land and calculate how much fertilizer each one requires.
Russia’s IT sector has thus gained a new area for development: algorithms that turn Earth observation data into actionable information for crop production. This is particularly important for Russia, which has vast areas of agricultural land. Improving the efficiency of field operations could significantly increase food production while reducing costs by cutting the wasteful use of fertilizers and water.
Once the SPbSAU technology has been tested and validated in Russia, it could become an export product for regions that lack their own access to satellite data but face similar agricultural challenges.
