AgroScope – Crop Production Agrotechnical Operations Support System

The goal of the project is to develop, test, and implement the AgroScope hardware and software system — an intelligent decision support system for agronomic practices. The project is aimed at creating an integrated drone-and-satellite system based on artificial intelligence that provides monitoring of the state of agricultural land, analysis of crop conditions, and generation of recommendations for agronomists. The key objective is to increase crop yields and the efficiency of crop production through timely and accurate analytics, as well as to reduce fertilizer and herbicide costs by 10–15% and lessen the environmental impact.

2026

Achieved Results:

• Successfully integrated 219 field boundaries covering a total area of 21,324.2 hectares without errors.
• Introduced digital agricultural monitoring using unmanned aerial vehicles (UAVs) and artificial intelligence algorithms for the first time.
• Automated aerial image processing, detection of problematic areas, and generation of analytical maps.
• Established a foundation for targeted field inspections and variable-rate application of agrochemicals.

Estimated Economic Impact: Up to KZT 28.18 million in annual savings across 18,787 hectares. Expected reductions in herbicide costs range from 9% to 15%, while fertilizer costs are projected to decrease by 7% to 14%.

Hardware:

• Unmanned aerial vehicles (UAVs) with mounted and ground-based equipment.

• A server cluster consisting of three types of servers (CPU and GPU computing): blade servers (Proxmox VE), bare-metal servers (Fujitsu) with NVIDIA N40S and Tesla K80 GPUs.

• FAS (NetApp) data storage system.

Software and infrastructure:

• ML and data analysis: ClearML (ML pipeline platform), JupyterLab, CVAT (image annotation).

• Orchestration and monitoring: Docker Swarm, Portainer, Ansible, Zabbix Agent.

• Storage and GIS: MinIO (S3-compatible storage), GeoServer, WebODM.

• Access management: Keycloak (SSO, roles).

• Satellite services: Integration with satellite imagery and geodata.

• Web platform: For management, visualization, and decision support.

The project does not use solutions from the Global CIO catalog

The complexity of the project stems from the need to integrate heterogeneous data sources (satellite imagery, UAV data, weather data) into a single system, as well as from the development and tuning of ML models for image segmentation and classification. A fault-tolerant server infrastructure supporting GPU computing and large data volumes is required. Additional complexity lies in addressing supporting tasks: data and machine learning lifecycle management, and system administration, which are critically important for the system's development and the adaptation of models to the customer's local conditions. The project also includes training specialists at the intersection of agricultural technology, AI, and UAVs, which requires methodological support and user training.

AgroScope is a software system designed to support agronomic practices in crop production. The system addresses the tasks of monitoring the state of agricultural land using satellite imagery and UAV data, analyzing crop conditions with the use of artificial intelligence methods (machine learning, segmentation, classification), determining crop growth stages, and identifying risks (diseases, stress, weeds). Based on this data, recommendations are generated for agronomists on optimizing the timing of treatments and fertilizer application.

The structural diagram of the system includes subsystems for identifying areas of interest, flight route planning, image storage, an ML platform, and a BI system. The architecture is deployed on a server cluster and uses modern tools for container orchestration, monitoring, and big data storage. The automated processes cover the segmentation of zones requiring increased attention, the processing of large volumes of photographic material, and the display of monitoring results. Mapping systems have been implemented for visualizing NDVI maps and derived metrics, as well as a field management subsystem with field cards, a 3D panorama, displaying area, crop type, NDVI index, and status.

Kazakhstan