A Plastic Story

Using Earth observation to stop plastics before they enter the ocean.

Measured Not Modeled.

Each year, it is estimated that over 20 million tons of plastics finds its way into our oceans, directly causing over 10% of marine animal deaths.

Mismanaged plastic waste stems
from sources throughout the globe.

Models suggest that 122 river sinks contribute to over 90% of our oceans’ waste.

Once these plastics enter the oceans via rivers, they ride marine currents, making it nearly impossible to clean-up, impacting the lives of every animal along the way.

The scientific models show that the scale of the problem is immense. But the tools we have to fight it are outdated and imperfect, based on assumptions.

We need better information.

We need a clearer picture of these plastics
before we lose track of them.

After all, we can’t manage what we can’t measure.

We’ve built a reliable, scalable solution.

With the power of global, near real-time satellite imagery,
paired with data science and machine learning, we built
a first of its kind technology to identify terrestrial plastics waste from space.

Using it, we can now stop plastics at the source.

And we are starting with Bali.

We know that there are 10 legal landfill (TPA) sites across Bali.

Using satellites, our technology is monitoring how these sites are changing day by day.

With known TPAs as a source of truth, we have trained our algorithms to scour satellite imagery across all of Bali to find dozens of illegal or informal plastics dump sites across the island.

Here's how it works...

This is Temesi, a known TPA site.

Here, we have trained machines to find needles in a global haystack.

Our technology assesses a growing library of multispectral satellite
imagery - analyzing thousands of combinations for each and every
pixel throughout a massive
geographic target area.

In doing so, we leverage a number of powerful views on the data, including...

PCA

Linear projection of all 12 bands to 3 dimensions preserving the most information. Principle component axes mapped to RGB color channels.

TSNE

Non-linear mapping of 12-band signal embedding to 3-dimensional color channels. Closest visual representation of how a neural network might project the data.

False Color

Highlights vegetated vs. urban regions. Combination of 2186nm, 1610nm, 665nm bands.

Land Water

Clearly delineates boundaries between land and water and captures soil moisture. Combination of 833nm, 1610nm, and 665nm bands.

Atmosphere

Longwave imaging minimizes the influence of atmospheric absorption, but at a lower resolution.

Healthy Vegetation

Emphasizes chlorophyll concentration for vegetation health monitoring. Combination of 833nm, 1610nm, and 492nm bands.

Geology

Identifies geological features like faults, lithology, and formations. Combination of 2186nm, 1610nm, and 492nm bands.

RGB

Natural color representation of the scene. Data is not atmospherically corrected. Combination of 665nm, 559nm, 492nm bands.

Color Infrared

Highlights vegetated vs. non-vegetated regions. Combination of 833nm, 665nm, and 559nm bands.

Vegetation

Identifies vegetation, forest, and bare soils. Combination of 1610nm, 833, and 665nm bands.

SWIR

Used for monitoring drainage and soil moisture, as well as stages of crop growth. Combination of 2186nm, 864nm, and 665nm bands.

Agriculture

Useful for distinguishing agricultural varieties and health as well as barren lands. Combination of 1610nm, 833nm, and 492nm bands.

Leveraging the data hidden in each of these bands, our computers can interpret information that the human eye cannot. 

We break down the satellite imagery into individual pixels of hyperspectral data. This creates a map of unique spectral signatures for each 10x10 meter pixel.

Next, we employ a neural network that is trained to pull out meaningful information from these spectral signatures.

Based on what the algorithm sees, it then assesses the likelihood that the pixel contains plastic.

Finally, we recompile all of the findings to assess the potential of plastics in a given area.

And the results are accurate.

Collectively, this space-borne data is more than just pixels on a screen - its actual plastic on the ground and it can be verified by researchers on motorbikes.

This is also Temesi, as recorded by Google Street View.

While finding plastics is critical, monitoring it over time is the only way in which we can effectively drive change.

Lidar technology and synthetic aperture radar will allow us to monitor additional aspects of these sites over time.

WHATS HAPPENING HERE

This TPA site has increased in surface area by 5x (0.5 ha to 2.5 ha) over the last 11 months.

Temesi is just one TPA and Bali is just a small part of the world.

We have already found dozens of previously unknown dump sites across Bali and now Java.

But this technology is extensible for the entirety of the globe.

It can be used to

make predictions. inform decisions. drive policy. hold accountability. increase transparency. power a movement. save sea life.
Jan2019
Mar2019
Apr2019
May2019
Jun2019
Jul2019
Aug2019
Sep2019
Oct2019
Nov2019
Dec2019
Jan2020
Feb2020
Mar2020
Apr2020
May2020
Jun2020
Jul2020
Aug2020
Sep2020
Oct2020
Nov2020
Dec2020
plastics
no plastics
threshold
Classified Points
Classification

PCA

TSNE

False Color

Land Water

Atmosphere

Healthy Vegetation

Geology

RGB

Color Infrared

Vegetation

Swir

Agriculture

HIGH PROBABILITY OF PLASTICS
LOW PROBABILITY OF PLASTICS
00.0%

A Plastic Story

Using Earth observation to stop plastics before they enter the ocean.