Optimised routing increased plastic collection by more than 60%

Summary

• Using one year of ocean and weather data, optimisation-based routing increased modelled plastic collection by more than 60% compared with the existing routing strategy, and by as much as 68% depending on the method.

  
  

• During some winter periods, the model roughly doubled collection by planning around rough seas and choosing better times to empty the collection system.

  
  

• The routing method was fast enough to be used in day-to-day planning, while still finding routes that performed close to the best routes identified by the full optimisation model.

Why is routing an ocean-cleanup system difficult?

Plastic in the Great Pacific Garbage Patch is not stationary. It moves with ocean currents, while waves and weather can limit when vessels can operate or empty the collection system.

The route also changes the next decision. Once plastic has been collected from an area, returning there soon may be less useful. A route that looks best for the next few days can therefore reduce the options available later.

Why is this a prescriptive analytics problem?

The operational decision is where the collection system should travel over time to remove as much plastic as possible within vessel, weather and extraction constraints.

This is different from ordinary route planning because the value of visiting a location changes after it has been visited. The researchers modelled path dependence directly rather than treating every location as having a fixed amount of plastic.

How did the team develop the routing method?

The project brought together Amsterdam Business School at the University of Amsterdam, London Business School and The Ocean Cleanup.

The team represented possible vessel locations, directions and times as a network of choices. It first solved a simplified version quickly using dynamic programming - a method that breaks a large decision problem into smaller linked decisions - and then tested the most promising routes against the full plastic-drift dynamics.

The faster method found routes that were, on average, within about 6% of the best possible result, while remaining fast enough for practical planning. For shorter planning periods, the researchers also developed a slower method that could confirm whether a route was the best possible one.

What did the findings show?

Across one year of real ocean and weather data, the optimisation-based approach increased modelled plastic collection by more than 60% compared with the existing routing strategy. Depending on the method, the increase reached 68%.

The gains were larger during some winter periods, when rough seas made the timing of extraction stops more important; in those periods, collection sometimes doubled in the model.

The direct operational result is more plastic collected with the same type of collection system, vessels and operating window. The study also showed that simply making the collection system larger can have diminishing returns if the rate at which it can be emptied does not improve at the same time.

How can the findings inform other applications?

The method is relevant to environmental operations where acting in one location changes the value of returning there later. Examples could include other cleanup, collection or monitoring systems in which material moves over time.

For The Ocean Cleanup, the framework can also be used to test future system designs before equipment changes are made, including the balance between collection capacity, extraction speed and weather constraints.