SEGES Innovation drives sustainable transformation in agriculture using AI and a new data science vision

The green, picturesque lands of Denmark are home to the highest-yielding cows in the world. 

A huge ecosystem equaled only by Israel and the US, the Danish cattle industry is a key revenue stream for the country’s economy – and a pillar of its flourishing agricultural sector. 

It is, however, an industry that’s also delicate and volatile.

Failure to spot and manage even the smallest issue – be it bovine metabolic disease, feed contamination and more – could prove catastrophic for both the animals’ welfare and the quality of the produce.

Yet, according to SEGES Innovation – Denmark’s leading Agriculture and Food R&D knowledge and innovation center - all of this can be managed and prevented using data and AI technologies. 

“Data is the single most powerful tool of modern agriculture,” says Ivar Ravn, Executive Director of Digital at SEGES Innovation. “From containing cattle disease to forecasting yield crops and testing new models, data is revolutionizing the way we work and conceive farming.” 

It’s a position that SEGES has fully embraced in recent years – combining data and machine learning to generate insights that help its customers make their operations more efficient and sustainable. All using Microsoft technology.

A data journey spanning decades 

SEGES Innovation is a non-profit R&D organization active in both the Danish and international agriculture and food chain industries. With more than 500 employees, the company specializes in advising farmers on running their businesses and being more environmentally conscious, while also prioritizing animal welfare. 

“Our mission is simple,” says Ravn. “To make Danish farmers and food producers the global leaders for quality and sustainable agriculture.”

That, he says, starts with data. Huge amounts of data. 

“Cows, dairy, beef, crops, farmers accounts, annual reports… we’ve been building our databases for decades,” he explains. “The basic structure in our cattle database alone dates back to 1996, while our field database is also more than 20 years old.” Years of data gathering have made SEGES unique within its sector – eventually leading them to apply intelligence and real-time analytics as a critical steppingstone to meet evolving business requirements.

Currently, the company’s source data is stored in different locations across legacy data, third-party applications, IoT devices, streaming data, and more. Raw data stored in relational databases such as Azure SQL DB is moved to an Azure Data Lake via a workflow manager. 

Within Azure Data Lake, data is cleaned and modeled, while Azure Databricks helps to transform it even further. Once this happens, the data is then orchestrated into Azure Synapse Analytics. Azure Landing Zone architecture was leveraged to ensure all of the data is secure, scalable, and governed. 

This entire orchestration of data is what SEGES refers to as SegesDataEstate.
“We’re the only ones in the world to have both the software and the R&D capabilities to handle datasets so deeply rooted in agriculture,” says Ravn.

“The challenge has been finding ways to truly make the most of them.”

Introducing machine learning  

In recent years, the volume of data SEGES has to manage has grown so much that the company has been looking for ways to support its data science team in their software and model development. 

That search led the company to machine learning.

“In a world dominated by IoT sensors, camera technology, spatial and radar data and more, we’ve put together an outstanding data science team capable of combining all our data and translating it into products and models we can sell to our farmers,” says Mogens Mikkelsen, Enterprise Architect at SEGES Innovation. 

“This has been game changing. It’s allowed us to look into predictive models, and opened our eyes to using advanced data management, data analytics, AI and machine learning on top of the heaps of data that we gather from Danish farmers.”

But leveraging these technologies has led to new challenges. Despite the success of the early machine learning models, SEGES was still being held back by its reliance on a legacy on-premises solution that was hard to maintain and work with. 

“What we needed was a solution that would allow us to train and test our models in the easiest, fastest, and most efficient way. So, we asked Microsoft and its partner twoday kapacity to help us build a new machine learning platform that would enable us to do just that.” 

Paving way for a new MLOPs Framework

Over three months, Microsoft and twoday kapacity built what Mikkelsen describes as “a clean and lean data science machine,” able to bring much-needed flexibility and speed to the SEGES data science team. 

Powered by Azure Machine Learning, the platform is used by the team to perform a wide range of tasks – from predicting crop yield to automating and optimizing account reporting and leading prevention and management of potato blight.  

“A key part of our offering is formulating hypotheses based on our data,” Mikkelsen explains. “Our teams usually test it, get results, and then use it to develop models and algorithms, which we then build into software products.

“This platform makes the entire process simpler, faster and more streamlined.”

The MLOps platform provides an intuitive and abstracted data science workflow for training, deployment, and governance. It is also connected to a data platform that allows it to incorporate Azure Synapse Analytics and Azure Databricks, as well as access data on Azure Data Lake Storage Gen 2. “We could see how painful it was for the SEGES to maintain all their on-prem clusters,” says Jes Ravnbøl, Senior Data Science Architect at twoday kapacity. “So, it was great to bring our technical knowledge to help them simplify their operations.” 

Alongside the platform, Microsoft and kapactiy also created a repeatable framework for MLOps – in alignment with the MLOps v2 Solution Accelerator – to tackle all the additional complexities the platform might have. 

“It's truly been a co-build effort between Microsoft, twoday kapacity and SEGES,” says Anne Kraght, Head of Strategic Alliances at twoday kapacity. 

A wide range of possibilities opening up

The new MLOps platform has already transformed SEGES’ operations and the way they provide their services. 

For example, the organization can now keep a closer look on its customers’ cattle, their health and yield – and is even able to predict potential issues. “We are experimenting with cameras monitoring cows in stables, providing us with data to train our models,” explains Ivar Ravn. 

“That means that we can now keep constant track of the cows, on behalf of the farmers, and spot early signs of disease or injuries in up to 90% of cases.”

On top of that, SEGES has developed a crop yield prediction model which uses multiple data sources - including satellite images, weather data, and terrain data - to accurately predict crop yield at harvest during growth season, both at field and at a 10x10m level. 

“This allows us to be more precise when forecasting expected crop yield,” he says. “In turn, this means that we can minimize the discharge of nitrogen pollutants to water bodies while still optimizing the return of the crop production.”

He stresses the significant cost reductions that the solution has helped to achieve. “Overall, across the ML model lifecycle, maintenance costs have been reduced by more than 95%,” adds Lasse Rose Malskær, Lead Data Scientist at SEGES. “Even the non-labor costs of deploying, running, and monitoring machine learning models in production have gone down by more than 80%.” 

But these savings are only modest compared to the reductions the company has seen in labor hours. “Before, all machine learning models were retrained and deployed manually,” he says. “With the MLOps platform, we have completely automated these processes – bringing the average time between retraining down from 6 months to a single day. This ensures that the best models are always in production for our users.” 

According to Ravn, interest and involvement in the platform now goes well beyond the data science team. 

“Even the business owners want to get more involved,” he says.  

“All of a sudden, we’re able to do more with our data than we ever thought possible. The more datasets we have, the more data we can put into the platform. It’s basically an entirely new way of doing business.”

Looking ahead

As labor time and costs continue to decline, SEGES can now invest more of its time in creating new and valuable data science products for its users. 

And with expansion plans already in the pipeline, the company is determined to continue its data journey even further – reaching more farmers and helping them to build a more sustainable agricultural sector. 

“We've come a long way, and these projects are a constant reminder of the difference we’re making in the world,” he concludes. 

“The farmers, the food producers, and the organizations trying every day to transform our sector and make it better: they are the ones we’re so proud to be adding value to.”