The true cost of pancreas disease (PD) to the Norwegian aquaculture industry is difficult to quantify. Farms that are forced to slaughter 1 million fish early face NOK 30 to 45 million (US$3.4 to $5.2 million) in lost revenue, according to a December 2019 report by Nofima and Kontali.
The overall industry loses an estimated NOK 1 to 6 billion (US$116 to $698 million) each year to PD-associated costs like decreasing fish welfare, prevention tactics, insurance payments, and treatments. It’s becoming increasingly expensive to farm fish, and disease plays no small role.
Talented scientists and researchers have been working on PD forecasting models to help farmers better prepare for more than 10 years. Their studies have produced important findings, but limited data has led to limited on-farm applications.
Today, Manolin released breakthrough progress in aquaculture disease detection. Our machine learning models now predict early onset of PD and ISA with greater than 93% accuracy, and we’re proud to offer the only commercially available disease forecasting tool for farmers in Norway.
The last few months have been a culmination of many years of work. We’ve integrated numerous disparate data sources, filled the gaps in industry data, and thoroughly studied academia’s available disease research.
To forecast fish disease, we used a deep learning method of artificial intelligence known as a neural network. Neural networks are named after the neurons in the human brain and essentially behave the same way: a computer learns to perform a task by analyzing a set of training data.
In the human brain, we know that the more times a neural pathway is used, the stronger the connection becomes. This is how habits are formed and skills are developed. Similarly, the more times a neural network sees certain connections between factors in data sets—or for us, variables on the farm—the more precisely it can draw and weigh correlations.
A computer is able to ingest a constant stream of data and continually learn from it. This leads to an infinite number of connections between data points—patterns and insights that would take a lifetime to identify with human analysis alone.
When new data is collected, the model responds immediately. On the farm, this means that the moment a mortality event, dip in feeding, lice treatment, or storm occurs, the platform (and farmer) benefits from it.
Our model has had a lot to learn from. Manolin’s years of mining, manipulating, and formatting data points across Norway and combining them with anonymized private farm data has given us a powerful data set, and it’s only growing as we continue to build and connect with more farmers.
We ingest millions of data points throughout each day:
- Live disease outbreak reports
- Treatment activity across all 600 active salmon farms in Norway
- Data from government institutions like Fiskeridirektoratet, Veterinærinstituttet, Mattilsynet, IMR, and Meterologisk Institutt
- Oceanographic forecasts, marine sensors, and boat traffic data
- More than 50 daily farm production and environmental factors
- More than 20 years of historical industry data
This is constantly running behind the scenes of our dashboard, filling the gaps in industry data and automatically alerting farmers when something isn’t right.
Computers predict disease, but farmers prevent it. To us, the tools to make complex technologies fit simply and easily within a hectic day on the farm are just as important as the technologies themselves.
Farmers are inundated with information, technology, and new solutions. At Manolin, our goal is to use industry-leading data and technology behind the scenes of a dashboard that answers the simplest, most important questions: what do my fish need right now?