Satellites reveal how Bolivian Lake Poopó is dying


I have always been fascinated by how we can observe the changes on the Earth through our eyes in the sky – the many hundreds of satellites which daily orbit our Earth. So, I recently undertook the NERC Undergraduate Research Experience Placement (virtually!) at the Natural History Museum with supervisors Peter Grindrod and Joel Davis observing & measuring temporal changes in shrinking lakes around the world. This was done by using satellite image data from Landsat satellites. This collection of satellites has been surveying the Earth and collecting data since the 1970s!

Landsat satellites have been surveying the Earth and collecting data since the 1970s. With a new launch date for Landsat 9 to be in April 2021. Image by USGS is in the public domain.

Many lakes have been decreasing in water level and size due to many factors from human removal to climate change. A lake that I was looking at is Lake Poopó, which is located in the south-west of Bolivia, next to Salar de Uyuni, the largest salt flat in the world.

Map of Lake Poopó located in Southwestern Bolivia. Lara Amusan, author provided.

Lake Poopó has been at the brink of extinction 3 times in the past 40 years (1980, 1994 and 2015). The government in Bolivia declared it a disaster zone in 2016. The lake has since returned, but is now shrinking back down into disappearing, and it may not recover this time. This has had dire consequences for human life as well as wildlife, as many people who live around the lake relied on it as a source of income through fishing and water for animals.

One of the main sources of the lake’s inflow is from Bolivia’s largest lake, Lake Titicaca. The lakes are connected by the Desaguadero River which has been affected by human activity. Diverting water for mining and agriculture has slowed down the inflow rate from the river. Factors like irregular rainfall has impacted the drastic rise & fall of water levels in the lake. El Niño helps to contribute to this. El Niño is a phenomenon that happens irregularly every 2-7 years causing heavy rainfall in some areas and warmer water/droughts in others. As climate change has increased the frequency of El Niño, and global warming has risen the temperatures around the area, the evaporation rate of Lake Poopó has tripled, causing the water to dry up faster than its replenished by rainfall or river inflow.

I decided to map this lake first. I used Landsat images from Earth Explorer, a site by the USGS that makes aerial and satellite image data free to access.

Next, I mapped the area of this lake. This is what first introduced me to QGIS – a free platform that supports the viewing, editing and analysis of geospatial data.

After a few tutorials I was ready to do this on my own, but I soon found a few cons of manual mapping – firstly, it takes a lot of time! Also, the colour RGB images made sometimes made it hard to determine boundaries for such a shallow lake as Poopó.

With the help of my supervisor I found out about index values, and a plugin called SCP.

These index values are calculated from remote sensing data, using Landsat bands to create an index image. They are usually based on surface reflectance and help to identify features in the image e.g. water has a lower reflectance in the near infrared compared to vegetation. I decided to use McFeeters’ Normalised Difference Water Index (NDWI) as I found it best in highlighting water content in the lake. This used images in the green and near infrared (NIR) wavelengths.


McFeeters’ Normalised Difference Water Index. Lara Amusan, author provided.

The Semi-Automatic Classification Plugin (SCP) within QGIS makes this relatively simple, with built in pre-processing and post-processing features through Python.

An NDWI image of Lake Poopó. Lara Amusan, author provided.

The NDWI method made things easier and more accurate, but not significantly quicker. After a bit more research on unsupervised analysis, I found something more semi-automated. What I found was an algorithm called K-Means Clustering, which is an algorithm used for the classification of data through vector quantisation. In regard to my work it orders satellite image pixels into a certain amount of k clusters, based on their spectral signature and groups using spectral angle mapping. With this unsupervised method a number of iterations need to be done to help get results. This uses results from the previous iteration to re-classify pixels. More iterations (classes) mean a longer process, but a good balance of both leads to a result that best matches the image.

Lake Poopó after using a K-Means Clustering algorithm, separating land from water for analysis. Lara Amusan, author provided.

Images of Lake Poopó comparing its surface area in 1986 to 2016. Lara Amusan, author provided.

The research and results from my project show how drastically features on Earth can change in such a short period of time as a result of climate change along with other human damages. It also shows how negatively it could impact wildlife, marine life and human life. Possibly with these methods, as a form of tracking land features, and patterns from previous research, we can help prevent the further destruction of these lakes, which in numerous cases are lifelines for many.

– A blog post by undergraduate REP Lara Amusan

Tagged with:

See more posts by

Stay informed

Click here to subscribe to our RSS newsletter by email.

Find Us

University College London is the administrative lead.

Pearson Building, UCL, Gower Street, London, WC1E 6BT

Follow us on Twitter