Making regenerative agriculture accessible worldwide through AI in agriculture 

Spatialise is taking a significant step toward more sustainable farming through the use of advanced AI technology. By developing a digital soil analysis tool that can accurately measure soil organic carbon (SOC) as well as essential nutrients like nitrogen, phosphorus, and potassium (NPK), the company provides a cost-saving and environmentally conscious solution for farmers. This collaboration with EuroCC Netherlands exemplifies how AI in agriculture can accelerate the transition to sustainable farming practices at scale.

Using digital soil mapping powered by satellite data and machine learning, Spatialise is enabling farmers to optimise fertiliser usage, reduce emissions, and achieve greater efficiency across agricultural operations.

High-performance AI for soil monitoring

At the core of Spatialise’s innovation is a soil nutrient monitoring tool that uses satellite imagery and machine learning to analyse agricultural land. Artificial neural networks are trained to predict soil nutrients on a global scale. These models are then fine-tuned using regional datasets to improve accuracy in specific areas.

However, training large neural networks with massive satellite datasets is computationally expensive. Fine-tuning for different regions requires repeated training cycles and extensive resources. Performing this type of advanced AI in agriculture at a global level posed a serious technical challenge in terms of scale and cost.

HPC Resources from EuroCC Netherlands

To meet these demands, Spatialise leveraged the expertise and infrastructure provided by EuroCC Netherlands. Using Snellius, the Dutch national supercomputer, Spatialise trains and refines its AI models efficiently and at scale.

The global model acts as a foundation and is enhanced through transfer learning with regional data. This process requires intensive hyperparameter tuning—a method of fine-tuning internal neural network settings for optimal accuracy. With Snellius’ high-performance computing capabilities and SLURM-based job scheduling, Spatialise can distribute workloads effectively, shorten training times, and scale its operations to accommodate complex models and large datasets.

The use of the super computing resources EuroCC Netherlands offers, showcases a successful example of AI in agriculture made possible by national HPC infrastructure.

Moving beyond costly cloud platforms

Before working with EuroCC Netherlands, Spatialise relied on Databricks on Azure—a comprehensive cloud-based platform that, while functionally robust, was prohibitively expensive for large-scale computational workloads.

By migrating to Snellius via EuroCC Netherlands, Spatialise significantly reduced computing costs while gaining access to state-of-the-art resources for neural network training and fine-tuning. The improved computational capacity led to more effective hyperparameter tuning, enhanced model performance, and a better understanding of key development areas. These included optimising satellite-based features, refining training methods, and enhancing transfer learning strategies.

This has directly contributed to faster development, greater research efficiency, and scalable AI deployment in the agricultural sector.

Benefits

• Cost savings: Reduced computational expenses compared to cloud-based solutions like Databricks on Azure.
• Improved efficiency: Accelerated training and fine-tuning of neural networks through hyperparameter optimisation using Snellius’ HPC resources.
• Optimised research: Identified key areas for enhancing training methods, satellite feature usage, and transfer learning techniques.
• Enhanced scalability: Enabled large-scale computations for accurate, region-specific soil nutrient predictions.

Technologies used

• High-Performance Computing (HPC)
• Artificial Intelligence (AI)

Sectors involved

• Agriculture
• Space
• Earth Observation Satellites