Between 10-11 million tons of plastic leak into the oceans every year, a figure that is predicted to triple by 2040. Much could be prevented if littering in rivers could be curbed, long before the plastic reaches the oceans.

Together with the Minderoo Foundation in Australia, researchers at Berkeley and the University of Georgia have developed a system of neural networks that identifies landfills by analyzing spectral, spatial and temporal components of satellite images. Thanks to the network’s capabilities, it is possible to identify potential garbage stations from fairly blurry, pixelated images – which can then be verified using sharper photos of the same areas.

Illegal dumps

The researchers’ study confirms that there are many illegal dumps that are still unknown. So far, the technology has managed to find 347 waste sites in Indonesia, which turned out to be more than double what the authorities knew.

When the technique was then applied to a larger geographical area, with twelve countries in Southeast Asia – where this type of illegal dumping is most common in the world – a total of 996 waste sites could be identified. With the help of the neural networks, geographical areas can be studied over time, which enables the detection of garbage mountains already at a very early stage.

The data generation also makes it possible to gain more knowledge about how and where landfills occur. Analyzes show that 19 percent of these are located within 200 meters of watercourses, and that a large percentage are located directly adjacent to rivers, which significantly increases the risk of plastic leakage leading into the oceans. The researchers estimate that there are approximately 1,000 rivers that allow the plastic to reach the oceans – with the ten worst waterways located precisely in the studied area of ​​Southeast Asia.

Helped by previous research

The new technology is possible thanks to previous research done in the field of Earth observations, where neural networks have produced data sets based on image analysis.

The researchers began training the neural networks to recognize waste sites and find factors that distinguish them from ordinary farmland and withered vegetation. Thanks to the analyzes being carried out over a longer period of time, it is possible to distinguish them, as the appearance of fields and woodland varies over time.

The hope is that this will give authorities and non-profit organizations new and more cost-effective tools to combat plastic pollution.