Measure, Clean, Predict, Prevent: The Next Steps of The Plastic Tide.

What an amazing few months! On the 31st of March 2017 we set out to survey and clean 30 beaches around the UK to build an algorithm that will revolutionise our understanding of the growing plastic tide choking our coastlines!  

In this blog post, we'll outline our success so far, our challenges and next steps.

First and foremost, we to give a huge THANK YOU and recognition to those that have helped us complete our first phase and continue to support us.  These people include all those that donated on our Crowd Science page,  joined us on beaches cleans, the amazing efforts of the online taggers, our team of scientists and engineers - Stefan, Erik, Kayleigh, Dirk and Ben for all their efforts.  Even more thanks to friends and family and to the strangers that helped behind the scene and a massive thank you to Ellie for her drone piloting and mission director expertise!

Today, 5 months on, The Plastic Tide team have traveled over 4,200 miles, picked up 1 tonne of rubbish from 30 beaches with the help of over 80 volunteers around the UK while collecting over 12,000 drone images! Further still, since the launch of our Citizen Science project at the end of April, YOU, or over 4,500 of you, have made over 1 million classifications to teach our algorithm to find plastics!

Jumping for joy on our last beach clean of Phase I at Dovercourt Beach, Essex.

Jumping for joy on our last beach clean of Phase I at Dovercourt Beach, Essex.

Success! Version 1 of our Algorithm.

The team are super excited about the completion of the first version of our algorithm!  Ben Hahn, our engineer, has worked hard with your tagging data and has succeeded at getting the algorithm to detect a very impressive 25% of litter in images.  What is mind boggling is that these are images that the algorithm has not 'seen' before!  In other words, it is drawing on what it has been taught by you, our Citizen Science volunteers, to identify plastics in images.   Dr. Stefan Leutenegger, Ben Hahn's supervisor and Science Adviser to The Plastic Tide, sums up the results in his fantastic blog post here.

Tagging plastics has made it possible to teach our algorithm.

Tagging plastics has made it possible to teach our algorithm.

 

Behavioural Insights

Dr. Kayleigh Wyles, Environmental Psychologist also Science Adviser to The Plastic Tide, is studying the impact that tagging imagery with plastics and litter on beaches and online has on our well being.  Dr. Wyles conducted a scientific survey into these impacts online that we're excited to say is now finished.  We managed to achieve in excess of 700 responses, far bigger than the 200 we had anticipated originally!  Dr. Wyles is very excited about this, however, it will take some time to crunch the data.  Regardless, she believes the data set is an excellent and very robust dataset providing a great cross section of society.  Watch this space for her blog on this at the start of September! 

Drone in action over Gwithian Beach, Cornwall.

Drone in action over Gwithian Beach, Cornwall.

So, in short, a summary of our Objectives as on Our Mission page thus far:

  1. Algorithm - Version 1 completed and work started on Version 2!
  2. Public Engagement & Citizen Science - 1 million tags from 4,500+ volunteers online and 1 tonne of litter removed thanks to over 80 volunteers helping us on the beaches!  Our message reaching potentially hundreds of thousands thanks to the media interest in our work - see our In the Media page.
  3. Open Data & Collaboration - This will fall into Phase II, once we have got the algorithm detection rate up and can package it we release it, the images, and the framework behind the machine learning together under an Open Data license (license type TBC).
  4. Human Impacts - Dr. Wyles study has had over 700 responses, 500 or 250% increase, from a very representative slice of the UK population from many different backgrounds and geographic locations.

 

Next Steps:

We're all very excited about the performance of the first version of the algorithm and progress so far, but there is still much work to do!  With the results of Phase I we are fine tuning the detail of Phase II and now approach collaborators and potential funders.  A sneak peak at what is planned is provided below;

Pete with the branded items we found on South Landing Beach, East Riding.

Pete with the branded items we found on South Landing Beach, East Riding.

The Algorithm; Version 2.

Dr. Leutenegger sums up the next steps in his blog here, but in short, the main focus of the next phase is to substantially increase the detection accuracy ideally into the order of 70-80% correct detections and maximally missing 10-20% of pieces.  This is a complex task and will involve studying the labeled images, including more tags and tweaking the algorithm parameters. Even so, we may find we need to collect more data or make changes to how the data is collected and/or labeled.  

But the coolest and most exciting capability is developing the ability for the algorithm to calculate the volume and mass of litter on any beach the drone images.  This would lead to the revolutionary capability of the algorithm to detect, measure and monitor the volume and weight of litter as well as the number of litter pieces! For example, it could quantify the effectiveness of beach cleans and assist in planning more effective cleans by knowing the mass of litter washing up.  It could also help land owners, such as Local Government, to baseline litter accumulation to commission more cost effective beach cleaning.

Version 1 of the algorithm detecting plastics and plastic types!

Version 1 of the algorithm detecting plastics and plastic types!

Plastics Research

Our plastics Science Advisors Dr. Erik Van Sebille and Dr. Arturo Castillo are exploring exciting research in two key areas that our algorithm opens up. 

  1. The Missing 99%: This problem highlights our lack of knowledge of what happens to litter when it enters our ocean.  For Dr. Van Sebille, as our algorithm develops in Phase II, it will allow the first opportunity to measure the litter washing up on our coastlines efficiently, cost effectively and rapidly.  This is the first step to coordinating more efficient beach cleans and measuring their effectiveness.   It will also allow us to identify areas where the litter might impact on sensitive wildlife and even where it could be a hazard to us!  But the research doesn't stop there, in the long term we hope to predict where litter will build up and understand how it accumulates on the seabed too!
  2. Plastics: The work in distinguishing between types of plastic will allow Dr. Castillo to eventually identify both the type of litter and the type of plastic it is made of.  Armed with this knowledge, repeat offenders could be identified, for example, and preventative action formulated by focusing plastic design and better packaging, but also by evidence basing local anti-litter policies.   
The ubiqutious Pot Noodle, Ellie and Pete found one on nearly every beach!

The ubiqutious Pot Noodle, Ellie and Pete found one on nearly every beach!

Open Platform

By the end of Phase II, version 2 of the algorithm will be completed and packaged up for anyone to download and run on their own drone images of beaches under an Open Software license.  

The results can then be uploaded upload to our online map allowing anyone to see the quantity of litter washing up on our beaches and when.  An online database will also allow anyone to download this data to study and share.

We also aim to modify existing drone apps to be able to perform the type of survey required to get maximum results from our algorithm. 

Citizen Science & Public Engagement

We are looking forward to running further beaches with YOU to gather more data and test our algorithm.  In fact, we envisaged setting up a series of trials for a fixed period of time at selected sites to test the robustness of the algorithm as it develops.  

Pete with the Widemouth Task Force and their secret beach cleaning weapon - 2 x dogs at Widemouth Beach, Cornwall.

Pete with the Widemouth Task Force and their secret beach cleaning weapon - 2 x dogs at Widemouth Beach, Cornwall.

However, an idea we're particularly excited about is combining the beach cleaning and litter tagging elements by 'gamifying' it.  In otherwords creating a game app, our thoughts are something similar to the wildly popular 'Pokemon Go' but for litter picking!  Users would be rewarded for identifying and picking up litter with prizes and trophies.  All this data could build a robust algorithm for not only beaches but also waterways, lakes and even streets and roads!  It could also be used to help coordinate local beach or perhaps even road cleans.  Coordinators could post news about cleaning events that might attract extract points, with the algorithm being used to do a before/after and estimating the effectiveness!

Timelines

We're working out the details of this as it will depend on how quickly we can get funding, but we hope that Phase II maybe completed by the end of 2018.  In the longer term we are looking at this project to develop a series of 4 or more phases over 3 years.  But more to come on that later!

How can you help?

Keep checking back for more news!