Chasing deforestation through the clouds

Chasing deforestation through the clouds
.. or how we use Earth Engine and ESA imagery to easily see deforestation through clouds and (try to) stop deforestation during the Amazonian rainy season

By Juan Doblas Prieto, GIS specialist at Instituto Socioambiental

During the rainy season in the Brazilian Amazon the forests are blanketed with clouds. During these cloudy periods, it's difficult to detect deforestation using traditional alert systems, which rely on detecting the removal of forest canopy in satellite images. It can mean months before federal forces know that environmental crimes have occurred. Take the BR-163, for example, a very long road that connects the soy-producing Brazilian savanna with the heart of the Amazon basin, where large new illegal mining areas in national parks open during the rainy season, under the protection of the thick cloud cover.
Many efforts to monitor deforestation in the Brazilian Amazon rely on space-borne instruments, but the recent availability of SAR data on Google Earth Engine opens up new opportunities.
Space-borne optical instruments, like those from the Landsat program, provide a wealth of information about land surface changes to the ground-as long as the satellite captures an image of the ground and not an entirely cloud-filled image. This, unfortunately, is frequently the case during Brazil's cloudy seasons. Luckily, combining radar data with optical data allows us at Instituto Socioambiental to fill in the gaps and see through the clouds.
How all it began: with a big data indigestion
Starting in 2014, the European Space Agency (ESA) made available through their data portal all the data produced by the Sentinel satellites. As an excellent article by Sara Aparício pointed out in this blog, Sentinel is allowing us to observe how the earth changes, in different ways, very often. The Sentinel satellites already in orbit give us the power of combining optical and Radar instruments, producing very large amounts of data every single day, open for anyone to see and download.
But, as they say, fear the day your wishes come true. The much awaited Sentinel constellation flooded us with data. Now we had Terabytes of information to be downloaded, processed and integrated on our GIS platforms. This was literally melting our computers and storage facilities. After many sleepless nights fighting against downlinks, full HDs, CPU overheats and silly mistakes on the parametrization of the algorithms to process the Radar data, we started to fear that no one will be able to follow such data flow once the entire constellation was operational.
Earth Engine magic to avoid radar data indigestion
Enter Google Earth Engine. One fine day in 2015, Google Earth Engine engineers announced that GEE would digest all the data coming from ESA, and that Google datacenters will make most of the data preprocessing. And that changed everything.
We ran to our browsers to see how researchers from all over the world started building scripts that chewed and instantly processed Sentinel information. It took us some time, but we learned how to do it as well, and we were finally able to create a script that could be used by everyone with basic knowledge of remote sensing to compute large mosaics of Sentinel-1 data, and even to map recent deforestation or canopy damage. Now we can even map deforestation without leaving the browser, using then a second script to export the mapped polygons. All you need is an average-quality internet connection, and an Earth Engine account. We called that script SIRAD (Sistema de Indicador Radar de Desmatamento, or Radar Deforestation Indicator System).

SIRAD in action: that's a multitemporal Sentinel-1 mosaic over the entire western Amazon Basin. It took 10 seconds to obtain it on Google Earth Engine, instead of weeks or months doing it the 'old way'.
The most exciting of it all was that we were seeing deforestation going on under the clouds. That is the magic of radar. Unlike many satellites out there (notably Landsat), Sentinel-1 satellites emit cloud-piercing microwaves (yes, the same you use in your microwave oven at home!) to draw a picture of the planet's surface. This means every 11 days (6 days in the very near future) we can have a new image of the indigenous lands and natural parks we are watching, all year round!
Radar magic: comparision between a Landsat-8 image and a multitemporal Sentinel-1 mosaic. SIRAD script automatically selects the best (least cloudy) Landsat image of the period represented by the mosaic, but even the better image is heavily polluted with clouds.
At some point in February, while developing the SIRAD script, we saw a lineal deforestation feature growing deep into a National Forest. Some of our forest partners had told us of a remote, huge mining site in a protected forest, near their territory. That was it. We were seeing it growing in front of our eyes.
Proving a method never was so muddy
We kept an eye on that mining site, but at some point we decided that it was time to get out of the air conditioning and try to stop the illegal mining, and, at the same time, demonstrate that the script was working. So we did it. First, we took a pickup truck and drove until we were submerged in mud (see photo).
Then we tried to hire a boat in the nearest village, but no one would take us there, as it was too risky for them. So we hired a plane and flew there.
Although it wasn't easy, we reached the spot and saw it: an enormous, bleeding cut through the forest. More than 6 kilometers of trenches opened in a few weeks by hundreds of men working in appalling conditions to extract a few tons of ore. We were speechless. I was struck by mixed feelings: technically satisfied by the demonstration of our method and, at the same time, very worried, seeing illegal deforestation approaching the territories of our partners and friends.
Check out this in-flight camera video to see some detail of the illegal mining site. Beware of airsickness!

Looking to the future
We are now focused on two tasks:
Spreading the good news: We are beginning to teach the technical staff of different government agencies to use SIRAD to detect early signs of deforestation. This will allow enforcement bodies to catch those engaged in deforestation while it is taking place, instead of reaching an area only to verify it has been destroyed, as is often sadly the case.
Trying to automate deforestation detection: This is a tricky one. Unlike Landsat or Sentinel-2 optical images, Sentinel-1 images are quite noisy, and detecting changes is a challenge. Researchers are working hard, and we hope to do some progress based on that work before we go blind, that is, before the cloudy season arrives in September. Automation can be a huge step towards developing a regional, radar-based deforestation system such as those built by IMAZON or WRI, but operational year-round.
ESA Sentinel's Radar imaging is a powerful way to learn about changes on the land cover, but only trained specialists would dare to use such data directly. Google Earth Engine scripting is allowing anyone to process and use those data to monitor the forests and to watch Earth change almost in real time.
There is still a lot of work to be done, but the GEE community is up to the challenge!.

Thanks to Guido Lemoine, Rebecca Moore, Carlos Souza Jr., Gennadii Donchyts, Mort Canty, Simon Ilyushchenko, Noel Gorelick and other GEE 'heroes'.
Thanks to Chris Herwig and Juan Doblas Prieto.

https://medium.com/google-earth/chasing-deforestation-through-the-cloud…