At 15 times the size of the entire country, Aotearoa’s Exclusive Economic Zone is our greatest natural resource.

But the vastness of our territorial waters poses a challenge for the commercial fishing fleets who are allocated quota each year to harvest certain species of fish within those waters.

Finding the fish is by no means an exact science. Until now, the greatest predictor is where they have been successfully fished in the past. But given the marine environment and its inhabitants are constantly changing, there is a drive to find smarter, more adaptive and efficient ways to fish.

And that’s exactly what Sealord — one of the largest seafood companies in the Southern Hemisphere — set out to achieve with Datacom as part of its ‘Advanced Fishing Analytics’ project.

By combining historical fishing data with satellite environmental data and real-time sensor readings from fishing vessels, the project is finding ways to more accurately predict where fish will be. Information that will be invaluable for Sealord’s own fishing operations and for the wider fishing industry.

Quantifying effort versus return

Armed with 30 years of trawl data provided by Sealord for two key species it harvests — hoki and jack mackerel — Datacom set about building a predictive model for the Catch Per Unit of Effort (CPUE). This is a key measure of the abundance of a target species and the accuracy of the fishing plan.

Datacom combined the trawling data with environmental, oceanographic data and satellite imagery from sources including NASA, the European Union’s Copernicus Earth Observation programme and Project Moana, a government-funded initiative that has created a 25-year hydrodynamic hindcast model of New Zealand waters.

These external sources of data helped fill gaps in the records and gave a more nuanced picture of the marine environment.

Included were data points such as surface temperature, ocean currents, salinity and levels of chlorophyll (the green pigment in the phytoplankton that forms the basis of the food chain in our oceans).

In preparation for the advanced analytics project, the Sealord team cleaned decades worth of data so it was in a usable format to feed into the predictive model.

Machine learning techniques were then applied to training the model. A preliminary analysis of the Sealord trawling data revealed a large disparity between effort and catch. More importantly however — populated with all of those data sets — the model proved to be an accurate predictor of past fishing success.

One of Sealord's fishing trawlers - the FV Rehua

The future of fishing

This modelling could be a game-changer for Sealord and for the economic and environmental sustainability of New Zealand’s fishing industry.

“When we are trying to catch fish, if we can avoid steaming off in an unproductive direction, we can reduce diesel consumption,” says Matthew Dodd, General Manager of Information Technology at Sealord.

“Another big part of the sustainability picture is minimising bycatch of non-quota fish species. We want to use the data to make better decisions about where we fish to reduce our chances of bycatch and focus our efforts on high quota species.”

Using these advanced analytics could inform more efficient harvesting strategies, help optimise route planning for trawlers to minimise damage to the seabed and reduce fuel consumption, and overall operating costs.

In the future the intention is to add more quality sets of data to the model, particularly for fish species where limited data is currently available.

Sealord’s Resources Manager - Fishing Operations, Charles Heaphy, says the data helps cut through the complexity.

“If we can spot trends early, we’ll be in a position to understand our fishery better. That’s why we need these big data sets and a way to look through the noise.”

Central to that is Sealord’s effort to share its analytics platform with the Deepwater Group, an industry partnership that represents quota owners of New Zealand deepwater fisheries and aims to optimise sustainable management of the fisheries.

Combining data sets from other industry players will improve the predictive power of the model and better account for bias in the data. Charles says a pilot project — part of the government-funded Project Moana scientific modelling effort — involved attaching sensors to fishing nets, which can supply real-time water temperature readings.

“We know surface temperature and historic temperature. But knowing the temperature of that column of water at a certain point in time is very useful,” says Charles.

Sealord is also adding sensors to its fishing vessels to more accurately monitor diesel usage and the performance of engines and boilers.

Matthew Polson, Datacom’s Head of Technology, Data & Analytics, says a critical next step in the project is refining the model and then finding ways to visualise the predictions that make it meaningful for people.

“It’s translating what this all means into a functional tool that can be used for decision making at Sealord,” Matthew says.

“The expertise, experience and commitment that the Datacom team brought to the project was pretty significant,” says Sealord's Matthew Dodd.

“At Sealord we are here for the long-term and fishing responsibly ensures we will have a business in 50 years’ time. Advanced analytics helps us do that by making valid predictions about the future.”

Technology-driven insights

Sealord’s Advanced Fishing Analytics project was based on the Azure Cloud and drew on several key Microsoft technologies including:

  • Azure Data Factory
  • Azure Data Bricks
  • Azure Data Lake
  • Power BI.

The project was delivered by a collaborative team of fishing experts from Sealord and machine learning specialists from Datacom.

"The model training was iterative, with a range of machine learning models and techniques tested and tuned to find which were the most effective for the dataset and problem at hand.”

Related industries
Agriculture
Related solutions
Data & analytics Cloud services